shedding light on endocytosis with optimized super-resolution microscopy daniela monica leyton

Shedding light on endocytosis with optim

ized super-resolution microscopy

Daniela M

onica Leyton Puig

2017

Shedding light on endocytosis with optimized

super-resolution microscopy

Daniela Monica Leyton Puig

Uitnodiging

voor het bijwonen van de openbare verdediging van het

proefschrift van

Daniela Leyton Puig

op donderdag6 april 2017

om 12:00 uur

in de Agnietenkapelvan de

Universiteit van AmsterdamOudezijds Voorburgwal 229-231

Amsterdam

Receptie ter plaatse na afloop van de promotie

Paranimfen:Jonne Raaijmakers

Judith Haarhuis

Daniela Leyton [email protected]

ISBN: 978-94-6233-569-1

The studies described in this thesis were performed at the division of Cell Biology at the Netherlands Cancer Institute (NKI) in Amsterdam, and financially supported by the Dutch Technology Foundation STW.

Printed by: Gildeprint

Cover: single molecule localization super-resolution microscopy is a dotty techni-que. If we would make a super-resolution image of a corner of Amsterdam, this is what it would look like.

Copyright © 2017 D. M. Leyton Puig



ACADEMISCH PROEFSCHRIFT

ter verkrijging van de graad van doctoraan de Universiteit van Amsterdamop gezag van de Rector Magnificus

prof. dr. ir. K.I.J. Maexten overstaan van een door het College voor Promoties ingestelde commissie,

in het openbaar te verdedigen in de Agnietenkapelop donderdag 6 april 2017, te 12:00 uur

doorDaniela Monica Leyton Puig

geboren te Cochabamba, Bolivia

Promotiecommissie

Promotor: prof. dr. K. Jalink Universiteit van Amsterdam

Copromotor: dr. M. Innocenti NKI-AVL

Overige leden:

prof. dr. T.W.J. Gadella Universiteit van Amsterdam prof. dr. W.J. Wadman Universiteit van Amsterdam prof. dr. J. Borst Universiteit van Amsterdam dr. E.A.J. Reits Universiteit van Amsterdam prof. dr. W.H. Moolenaar NKI prof. dr. A. Sonnenberg RUL Leiden en NKI dr. M. Postma Universiteit van Amsterdam

Faculteit der Natuurwetenschappen, Wiskunde en Informatica

Introduction The fidelity of stochastic single-molecule super-reso-lution reconstructions critically depends upon robust background estimation PFA fixation enables artifact-free super-resolution imaging of the actin cytoskeleton and associated pro-teins Flat clathrin lattices are dynamic actin-controlled hubs for clathrin-mediated endocytosis and signalling of specific receptors. Perifosine inhibits EGFR signaling by inducing its in-ternalization

Summarizing discussion ReferencesEnglish SummaryNederlandse Samenvatting Curriculum Vitae List of Publications Dankwoord

Table of Contents

Chapter 1 Chapter 2

Chapter 3

Chapter 4

Chapter 5

Chapter 6 Addendum

7

23

53

71

93

109

119

7

Chapter 1

Introduction

8

1Cell signaling and endocytosis. In order to function, cells need to communicate with each other and get information about their environment. This information gets processed and induces a response, a process called cell signaling that is the central communication system in an organism. Cells can receive signals in the form of mechanical inputs or biochemical products, such as peptides, proteins, ions, lipids and gases.Signals from the extracellular space are received via a special type of proteins called receptors that are located inside or on the surface of the target cell. Upon binding of these signals, called ligands, receptors get activated and initiate signaling cascades inside the cell that can lead to short term alterations, e.g. changes in the cytoskeleton or adhesion to the substrate, or long term alterations, e.g. changed gene expression patterns. Therefore, signal transduction is a highly regulated process, and its deregulation can lead to diseases such as cancer and autoimmunity 1.

Receptor Tyrosine Kinases (RTKs).One type of cell surface receptors, RTKs, are enzymes that phosphorylate proteins by attaching a phosphate group to the amino acid tyrosine on the target protein. RTKs span the cell membrane a single time and have a catalytic kinase domain and regulatory C-terminal and juxtamembrane regions in the cytoplasm, and a ligand binding domain in the extracellular space 2,3. On the plasma membrane of cells inactive RTKs exist as monomers, dimers or oligomers 2. There, they bind polypeptide ligands which induce their dimerization and structural changes that lead to their activation 4. Ligand binding induces the dimerization of the extracellular regions, which in turn guide the dimerization of the intracellular domains. In some cases, a bivalent ligand interacts with two receptors, crosslinking their extracellular regions and thereby inducing dimerization. In other cases, ligand binding induces a conformational change that leads to dimerization of the extracellular regions. In this case, the ligand does not contribute to the dimerization interface. Finally, receptor dimerization can occur via the combination of both mechanisms 2. The dimerization of RTKs leads to the autophosphorylation in trans of tyrosine residues. The tyrosine residues in the tail of each receptor are phosphorylated by the tyrosine kinase domain of its counterpart in the dimer. Phosphorylation of tyrosine residues takes place in a precise order. First-phase autophosphorylation events enhance the catalytic kinase activity and maintain the active conformation of the dimer, second-phase events create the docking sites to recruit intracellular second messenger proteins containing Src homology-2 (SH2) and phosphotyrosine-binding (PTB) domains and, in some RTKs, third-phase autophosphorylation events further enhance the phosphorylation of downstream targets 2,4.There are 58 known types of RTKs in humans 2 and they activate important signaling pathways that lead to cell growth, survival, migration and differentiation; both during embryonic development and adult homeostasis 3,4.The Epidermal Growth Factor receptor (EGFR) belongs to the ErbB family of RTKs which includes Her1 (EGFR, ErbB1), Her2 (Neu, ErbB2), Her3 (ErbB3) and Her4 (ErbB4). The four members of the family share similar structure 4. Ligand binding to ErbB receptors induces their homo or heterodimerization and cross-phosphorylation. However, ErbB2 contains a unique extracellular domain and lacks a known ligand, and ErbB3 lacks intrinsic kinase activity making heterodimerization with other members of the family necessary in order to signal. EGFR was the first discovered member and is the most studied one. To date, there are 8 known EGFR ligands: epidermal growth factor (EGF), transforming growth factor-α (TGF-α), heparin-binding EGF-like growth factor (HB-EGF), betacellulin (BTC), amphiregulin (AR), epiregulin (EPI), epigen and neuregulin2-β. They differ in their affinity for the receptor

9

Introduction

15-7 and their capability to induce heterodimerization of EGFR with other members of the ErbB family 7,8. Moreover, different ligands produce different biological effects, even in the same cell line. For example, it has been shown that TGF-α stimulates cell proliferation more strongly than EGFR and HB-EGF, and that EGF stimulates cell migration and MAPK activation through different signaling proteins than TGF-α or BTC, probably by inducing different tyrosine phosphorylation patterns 7,9,10.Activation of EGFR starts signaling cascades that promote cell proliferation, survival and migration (Sorkin and von Zastrow, 2009). Therefore, it is not surprising that its signaling is commonly deregulated in cancers such as non-small cell lung, ovarian, breast, colorectal, head and neck and pancreatic cancer, where it is found overexpressed and/or mutated 5,6. Aberrant signaling of EGFR is currently clinically targeted with antibodies against the external ligand binding domain as well as with small molecules that inhibit the intracellular tyrosine kinase domain 5,6,11

G-protein coupled receptors (GPCR).GPCRs are another type of membrane receptors that sense biochemical (and other types of) cues. They comprise the largest group of membrane receptors with more than 900 members in humans 12 and they have diverse functions. GPCRs are divided in five families, based on sequence similarities: rhodopsin, adhesion, frizzled, glutamate and secretin families 13. They have seven transmembrane domains, linked by loops inside and outside of the cell –hence they are also called seven-transmembrane receptors. Their extracellular N-terminal portion and loops are ligand binding domains. However, some hydrophobic ligands bind to a pocket formed by the transmembrane domains 14. Their cytoplasmic C-terminal and loops are important for signal transduction and have high variation between GCPRs 14. Activation by their cognate ligand induces a conformational change that exposes cytoplasmic sequences that interact with G-proteins or increases their affinity for them. G-proteins are hetero-trimers present at the plasma membrane, composed of Gα, Gβ and Gγ subunits. Activated GPCRs act as guanine nucleotide exchange factors (GEFs) promoting the exchange of GDP for GTP in the Gα subunit of inactive G-proteins. The exchange for GTP leads to the dissociation of the GTP-bound Gα subunit from the βγ dimer. Both Gα and the βγ dimer stimulate downstream effectors that begin signaling cascades 12,15,16. Once the Gα subunit exchanges GTP for GDP it becomes inactive, and can reassociate with the βγ dimer 17.

P

Ras

ERK

JAK

STAT

PTENPI3K

AKT

P

EGF

Cytoplasm

EGFRFigure 1. Simplified scheme of EGF activated EGFR signaling pathways. a) Upon activation by its ligand EGF, EGFR dimerizes (if not already a dimer) and undergoes a conformational change that induces its auto-phosphorylation in trans. The phosphorylated residues (P) recruit proteins that trigger signaling cascades, leading to activation of the JAK-STAT, AKT and ERK pathways, which in turn activate transcription of genes encoding proteins involved in cell survival, proliferation and migration in a cell specific manner.

10

1

Some GPCRs, like the LysoPhosphatidic Acid (LPA) receptors, can couple to various types of G-proteins and trigger diverse signaling pathways. LPA is a bioactive phospholipid found in serum that plays roles in wound healing, cell proliferation and survival, neurite retraction, cytoskeletal changes, changes in cell adhesion, and cell migration 18,19. It binds and activates LPA receptors 1 (LPAR1) to 6. LPAR1 was the first identified LPA receptor and it is the most widely expressed GPCR 20,21. Activation of LPA receptors can stimulate tumorigenesis and metastasis in some types of cancers 22,23

Signaling and endocytosis. Activation of receptors and signaling cascades must be tightly regulated in order to avoid aberrant gene activation or metabolic alterations. Activation of signaling cascades can be controlled by regulating the composition and abundance of receptors at the membrane. Indeed, endocytosis of receptors from the plasma membrane is an important process by which cells regulate signal transduction. Activated receptors also recruit adaptor proteins of endocytic pathways that route them to special internalization domains. Internalized receptors are no longer available for activation by ligands 1,5. However, although the role of endocytosis of receptors has classically been considered the restriction of their signaling, it is now known that once inside the cell activated receptors sorted in internal endosomes can continue to signal to specific pathways 1,5,24. Endocytosis is therefore not only a mechanism by which certain signals are attenuated but also allows the spatial distribution of activation of some signaling pathways. For example, certain GPCRs remain associated to β-arrestins in endosomal compartments, where β-arrestins activate MAPK. In the same line, it has been shown that RTKs continue to signal to MAPK and Akt from early endosomes (Sorkin and von Zastrow, 2009. It has also been proposed that the shorter distance from endosomes to the nucleus would allow more efficient activation of transcriptional pathways 25.Receptors can be removed from the membrane through a number of different endocytosis mechanisms such as clathrin mediated endocytosis, caveolae dependent endocytosis, the CLIC/GEEC pathway, phagocytosis, micropinocytosis 26. The most common and best studied of these pathways is clathrin mediated endocytosis (CME) 26. There is evidence that all families of membrane receptors are internalized via CME 1. Some receptors, such as LPAR1, have only been shown to be internalized via CME 1,12,27, while others, such as EGFR, are internalized both via CME and clathrin-independent endocytosis mechanisms, depending on which ligand activates the receptor and its concentration 24,28-30.

Ras

ERK PI3KRho

Cytoplasm

LPAR1

PLC

DAG

PKCCa

IP3

++ROCK AKT

AC

cAMP

LPA

Gα12/13Gɣ

GβGαq/11

GɣGβ

Gαi/OGɣ

Gβ

Figure 2. Simplified scheme of LPAR1 signaling pathways. a) Upon activation by its ligand LPA, LPAR1 starts a signaling cascade by activating heterotrimeric G-proteins. LPAR1 can signal through various G-proteins: Gα12/13, Gαi/O and Gαq/11. Gα12/13 activates the Rho-associated kinase (ROCK) pathway, which is mainly involved in cell shape and migration. Gαq/11 activates the phospholipase C (PLC) – protein kinase C (PKC) pathway which leads to cell proliferation, and induces Ca++ release. Gαi/O also signals to the PLC-PKC and Ca++ pathways and in addition it activates the extracellular signal-regulated kinases (ERK, also known as mitogen-activated protein kinases, MAPK) and AKT pathways that promote cell proliferation and survival. Additionally, some crosstalk routes have been described which have been omitted in this scheme.

11

Introduction

1Clathrin mediated endocytosis. Forty years ago clathrin coated vesicles were purified for the first time from pig brain extracts 31. Since then, thanks to a combination of biochemical, structural and microscopy data, clathrin mediated endocytosis (CME) has become one of the best understood processes in cell biology. However, perhaps due to flexibility in the process, some uncertainty remains on how formation of endocytic sites is initiated (Godlee and Kaksonen, 2013). It is thought that the first step in the formation of a clathrin coated pit (CCP), called nucleation, is the interaction on the membrane of phosphatidylinositol-4,5-bisphosphate (PIP2), the clathrin adaptor AP2, and clathrin 32 Coccuci et al., 2012). However, some studies have shown that AP2 is not always necessary for the initiation of clathrin sites (Godlee and Kaksonen, 2013). Other clathrin adaptors, such as Eps15 and intersectins, and membrane curvature inducing proteins, such as the F-BAR domain proteins FCHo1 and 2, have also been proposed to initiate clathrin spots (Henne et al., 2010) and are thought to play a role in nucleation. There is a large list of clathrin adaptors identified to date and these may well be capable of initiating endocytic spots with different specificities and efficiency (Godlee and Kaksonen, 2013). On the other hand, the cargo, another important component of CCPs, appears to have no role in nucleation 33,34. Studies using viruses which bind to and cluster specific membrane receptors showed that CCP nucleation occurred at sites of binding, arguing in favor of a cargo dependent initiation process in this particular case. However, several more studies using fluorescent receptors showed that upon ligand binding receptors travel to pre-formed clathrin sites, acting merely as passive cargoes. Cargo might play a role at later stages in the maturation process of clathrin coated pits, defining the productivity of the endocytic event (Godlee and Kaksonen, 2013), as initiated endocytic sites that do not recruit cargo are generally aborted 35.

a) b)

c) d)

Chathrin heavy chainClathrin light chain

Adaptor protein

EGFEGFRLPAR1

Cytoplasm

Dynamin

Figure 3. Clathrin mediated endocytosis. a) Clathrin molecules assemble in trimers, called triskelia, that are composed of three clathrin heavy chains and three clathrin light chains. b) Activated receptors are targeted for endocytosis by the binding of adaptor proteins. b) Adaptor proteins in turn recruit clathrin triskelia that polymerize to form a cage that coats the region of the membrane that will be internalized. c) Once the coated vesicle is formed by invagination, the GTPase Dynamin is recruited to its neck where it induces the scission of the vesicle.

12

1After nucleation, the CCP grows by recruitment of more adaptor proteins and clathrin. The CCP recruits cargo and invaginates towards the cytoplasm of the cell until it forms a clathrin coated vesicle (CCV). CCVs pinch off the membrane with the help of the GTPase Dynamin, which forms a spiral ring that contracts around the neck of the vesicle 34. Once internalized, the clathrin coat dissociates from the vesicle and the internalized cargo is sorted in internal endosomes, from where it can be recycled back to the membrane or targeted for degradation 26.

Clathrin mediated endocytosis of EGFR. In general, phosphorylated RTKs are removed from the plasma membrane through different endocytic pathways. The most common RTK endocytosis mechanism is CME. Activation and auto cross-phosphorylation of RTKs creates interaction sites for ubiquitin ligase proteins that modify lysine residues. RTKs are ubiquitinated by ubiquitin ligase proteins, such as Cbl, which ubiquitinates EGFR, FGFR, PFGFR, VEGFR and others 36. Ubiquitination of RTKs serves as a sorting signal as well as a degradation targeting signal. Ubiquitinated RTKs are recruited to CCPs by adaptor proteins such as Epsin, Eps15 and Eps15R which bind to clathrin or AP2.Active EGFR recruits the adaptor protein growth factor receptor-bound protein (Grb2), which binds the phosphorylated residues Y1068 and Y1086 through its SH2 domain. In turn, Grb2 recruits Cbl 37 which leads to EGFR ubiquitination and recruitment to CCPs by Epsin and Eps15. Cbl is also recruited to EGFR independently of Grb2, via the phosphorylated tyrosine Y1045. Cbl recruitment to EGFR, and the level of its ubiquitination, depend directly on the dose of the ligand. Ultimately, the level of ubiquitination determines the fate of the receptor 29,38. Moreover, EGFR can also interact directly with the clathrin adaptor protein AP2 39, but AP2 is not required for CME of EGFR 40. Once internalized, EGFR can continue to signal to specific pathways from endosomes. Eventually, it is recycled to the plasma membrane or travels to late endosomes and lysosomes, where it gets degraded. The regulation of these pathways determines the continuation of signaling through endocytic signaling or receptor activation on the plasma membrane, or the attenuation or termination of signaling if the degradation route is followed 5.

Clathrin mediated endocytosis of LPAR1. Activated GPCRs are phosphorylated on multiple residues by GPCR kinases (GRKs) which induces their interaction with β-arrestins, due to a conformational change in both proteins 41. This interaction terminates the GEF activity of the receptors, and therefore further activation of G-proteins. Furthermore, β-arrestins bind to clathrin and AP2 41 recruiting the GPCR to CCPs. Other clathrin mediated endocytosis adaptors such as Epsins, CALM, HIP1 and HIP1R, Numb and Dab2 have been shown to play roles during the endocytic process of some GPCRs. After internal trafficking, GPCRs can continue to signal and they are eventually recycled to the membrane or targeted for degradation 12 .LPA bound LPAR1 is phosphorylated by GRK2, binds β-arrestin 20 and undergoes CME. Moreover, LPAR1, like some but not all GPCRs, can also be phosphorylated by PKC. This unconventional phosphorylation also induces endocytosis of LPAR1, although independently of β-arrestin binding. Both GRK and PKC induced LPAR1 endocytosis are AP2 dependent 21. Once in internal compartments, LPAR1 has been shown to signal to Akt from Adaptor protein, Phosphotyrosine interaction, PH domain and Leucine zipper-containing (APPL) endosomes 42, an early endosome subpopulation that contain Rab5 but lack EEA1 43.

Flat clathrin plaques (FCPs). It is a matter of debate if clathrin coated pits are always formed de novo or if they can also

13

Introduction

1pre-exist on the inner leaflet of the plasma membrane. In the early 80s, electron microcopy (EM) images revealed that clathrin triskelia not only coated CCPs and CCVs, but also large flat portions of the plasma membrane of cells 44,45. FCPs have been proposed to be endocytic hotspots 35,46, endocytic (non-curved) structures 47 or even inactive non-endocytic sites 48. Moreover, they have also been associated with substrate adhesion 33,49.

There are indications that FCPs could be sites of formation of clathrin vesicles. In electron microscopy images coated vesicles are often found juxtaposed to FCPs 44,45,50 and live cell studies using a system to detect bona-fide scission events have shown that discrete vesicle-like puncta invaginate from clathrin plaques 35,46,51, at least with the resolution of the conventional light microscope. Clathrin coats are lattices formed by combinations of hexagons and pentagons (much like the parts of a leather football) made by triskelia. Triskelia are native non assembled clathrin molecules that exist in the cytoplasm of cells as trimers. The relative combination of hexagons and pentagons on a clathrin lattice is related to the curvature of the coated membrane. On a flat surface, such as FCPs, the lattice is formed exclusively by hexagons. Some authors argue that the triskelia forming a FCP of hexagons would have to undergo too complex rearrangements to coat a curved vesicle, made of hexagons and pentagons, rendering this transition unlikely 32. However, computational modeling has shown that these rearrangements are indeed possible 52. Moreover, in the 80s Heuser 45,53 already proposed that the different forms of clathrin that can be observed in electron microscopy images, a few of them flat, corresponded to stages in the formation of CCVs. A recent study showed, using correlative electron microscopy and fluorescence microscopy, that clathrin endocytic sites begin with a small planar clathrin lattice that is continuously bent and remodeled into a curved vesicle as it matures 54, suggesting that flat clathrin plaques can also become curved and could therefore be sources of CCVs.

The role of actin in clathrin mediated endocytosis. Actin plays variable roles in CME 55. It is thought that polymerization of actin in the form of fibers (F-actin) against the membrane generates a force that drives the final CCV invagination towards the cytoplasm 56, counteracting membrane tension when it is high 57,58. Indeed, in yeast, where cell wall rigidity and internal pressure are high, it is well established that actin plays an essential role in CME 58,59. Moreover, actin might regulate CME indirectly by regulating local membrane tension, which determines the efficiency of the fission reaction 60.It this line, actin polymerization is not absolutely essential for CME in mammalian cells, but when needed it is thought to play a role in the late stages of vesicle maturation by helping vesicle budding and inward movement 34,58,59. Hindering the dynamic actin polymerization

a) b) c) Figure 4. Types of clathrin coated structures. a) Super-resolution images of clathrin coated pits and vesicles on the basal membrane of HeLa cells. Cells were fixed and stained for clathrin heavy chain b) Super-resolution images of clathrin coated plaques on the basal membrane of HeLa cells. Cells were treated as in (a). c) Super-resolution images of clathrin coated vesicles in proximity or juxtaposed to flat clathrin plaques on the basal membrane of HeLa cells. Cells were treated as in (a). Scale bar: 500 nm.

14

1process by genetic or biochemical means negatively affects CME 47,50,61,62, although with variable effects reported. Indeed, actin and some actin polymerization activators, including Neural Wiskot Aldrich Syndome Protein (N-WASP), cortactin and the Actin related protein (Arp) 2/3 complex, are recruited to clathrin coated structures (CCS) at different stages of the endocytic process 46,51,63. However, how actin and its regulatory proteins are recruited and regulated at clathrin endocytic sites is still under exploration. F-actin might be recruited to CCSs by the Huntingtin interacting protein 1 (HIP-1) and its relative HIP-1R. HIP-1 binds to clathrin heavy chain 55 and HIP-1R. In turn HIP-1R binds to F-actin. In fact, these proteins colocalize with actin in CCSs (Mooren et al., 2012). Dynamin II interacts with cortactin, an actin polymerization activator 64. Moreover, Dynamin binds N-WASP 58 and interacts with Intersectin, a RhoGEF for cdc42, which activates N-WASP 65. Finally, F-BAR proteins that are present in sites of clathrin endocytosis, FBP17, syndapin, FCHo and others have been linked to recruitment of actin regulatory proteins 58.

N-WASP and the Arp2/3 complex. Actin polymerization and de-polymerization into filaments is a tightly controlled process that involves many actin regulating proteins. The Arp2/3 complex is an important actin polymerization regulator that induces the formation of branched actin filaments on the sides of existing “mother” filaments 66,67. Arp2/3 complex induced actin filaments grow in a 70 degree angle on the sides of mother filaments. It has been shown that, at least in ruffles, the mother filaments are recently polymerized by mDia1 68. The Arp2/3 complex acts by mimicking the conditions that are necessary for the beginning of actin polymerization. Spontaneous polymerization of actin filaments is initiated by the formation of an actin nucleus consisting of three actin monomers, a process called nucleation. Two subunits of the Arp2/3 complex, the actin-related protein subunits ARP2 and ARP3, are structurally related to actin monomers. Binding of these subunits to an actin monomer mimics the actin nucleus facilitating the beginning of polymerization 69. For the polymerization to begin, the Arp2/3 complex needs to be activated by nucleation promoting factors (NPF). NPFs are divided in two subclasses, type I NPFs such as the Neural Wiskott–Aldrich syndrome protein (N-WASP) and WASP family verprolin-homologous protein (WAVE), and type II NPFs, such as cortactin 66,67,70. Type I NPFs are modular proteins that contain a C-terminal VCA (verprolin-homology, cofilin or central-homoly and acidic) region, where the Arp2/3 complex and the actin monomer bind to form the actin nucleus. Binding of Arp2/3 to the CA domain induces a conformational change that activates the complex. Furthermore, the V domain binds actin, thereby mimicking the actin nucleus 69. The Arp2/3 complex is also activated by the type II NPF cortactin. Type II NPFs bind Arp2/3 through acidic domains in their C-terminus and F-actin through tandem repeats (Rotty et al., 2013). Cortactin activates the Arp2/3 complex in synergy with N-WASP, by accelerating the release of the VCA from the junction of the growing branch which accelerates nucleation rates. Cortactin then remains in the junction of the growing branch 66 and blocks debranching (Rotty et al., 2013).The Arp2/3 complex plays important roles in the cell. In cell migration, Arp2/3 localizes in ruffles, lamellipodia and pseudopodia where it is activated by SCAR/WAVE. Moreover, in endocytosis, the Arp2/3 complex has a role in phagocytosis, activated by WASP which is the specific NPF expressed in haematopoietic cells; and a role in clathrin mediated endocytosis, where it is activated by N-WASP and cortactin. In invadosomes, the Arp2/3 complex is present in the core along with N-WASP and cortactin 69.

15

Introduction

1

Figure 5. Domains and binding partners of N-WASP. a) Depiction of domains of N-WASP, indicating their binding partners. N-WASP is a modular protein that contains four domains involved in its regulation. The N-terminal WASP homology-1 (WH1) domain is bound by WASP interacting protein (WIP) which stabilizes N-WASP in an auto-inhibited conformation. Next to WH1 is the basic region (B) that binds phosphoinositides (specifically, PtdIns(4,5)P2), followed by the G-protein binding domain (GBD) to which small GTPases bind. Binding of phosphoinositides and cdc42, or other small GTPases, induces the activation of N-WASP. Next to the GBD is the proline-rich region (PRD) that binds to Src-homology-3 (SH3) domain-containing proteins and profilin, further contributing to the activation of N-WASP. Finally, the C-terminal VCA (verprolin homology, central, acidic) region binds actin monomers and the Arp2/3 complex. b) N-WASP in the auto-inhibited state. In the closed conformation, the GBD and surrounding regions inhibit the binding of the Arp2/3 complex to the CA region hindering its activation.

In the resting state, N-WASP is in an auto-inhibited conformation, folded due to the interaction of the VCA region with the GBD and surrounding regions, such as the B region. Activation of N-WASP by binding of small GTPases, such as cdc42, to the GBD domain releases it from auto-inhibition. Other small GTPases bind the GBD domain, such as Tc10, RhoT and Chp (Rotty et al., 2013). The binding of phosphoinositides (such as PIP2) to the B region synergizes with cdc42 binding. Moreover, Src family proteins phosphorylate N-WASP close to the GBD domain and release the autoinhibition. This phosphorylation is enhanced by activation of cdc42 (Rotty et al., 2013). The Arp2/3 complex then becomes activated by binding simultaneously with actin to the VCA region of activated N-WASP. Moreover, activation of the Arp2/3 complex can be further enhanced by binding of SH3 domain-containing proteins such as Nck and Grb2 to the PRD, with different levels of N-WASP activity depending on the SH3 domain protein bound 70,71 (Rotty et al., 2013).WIP family proteins bind the WH1 domain in a stable manner, which stabilizes N-WASP in its inhibited conformation and suppresses its activity. However, WIP also links N-WASP to adaptor proteins such as CrkL and Nck. However, CrkL and Nck can also bind the PRD domain of N-WASP independently of WIP. Thus, auto-inhibited N-WASP is recruited to places of active actin polymerization 71 where it integrates signaling cascades that lead to actin polymerization 70. In summary, the release from auto-inhibition of N-WASP and the activation of the Arp2/3 complex is a consequence of the combined effects of various signaling molecules. Within the cell, activation of N-WASP is locally optimized for the many different processes in which it plays a role.

Super-resolution microscopy. Fluorescence microscopy is a type of optical microscopy that allows the study of the spatial distribution of molecules by labeling them with fluorescent dyes. Optical or light microscopes achieve magnification by using visible light and a system of lenses. Due to the wave nature of light and the lens system, two objects –e.g. organelles, vesicles or molecules- that are closer than ~250 nm cannot be seen as separate entities. This phenomenon, called the diffraction limit, is the reason why fluorescently labeled molecules inside a cell, which are discrete entities, merge together into a single blur 72. In the last few years, a number of approaches have been developed that circumvent this limit, and they are named together super-resolution microscopy. These techniques include stimulated emission depletion

a) b)

B GBD PRD V V C A

WIP

PIP2

Cdc42

SH3 domains

G-actin

Arp 2/3

WH1/EVH1 WH1/EVH1 B GBD

PRD

VVCA

16

1Microscopy (STED) 73, photo-activation localization microscopy (PALM) 74, stochastic optical reconstruction Microscopy (STORM) 75, direct STORM (dSTORM) 76 and ground state depletion followed by individual molecule return (GSDIM) 77, and others. Among the various super-resolution techniques, STORM, dSTORM, PALM and GSDIM can yield the best improvements in resolution. These techniques belong to the group of stochastic single molecule localization microscopy (SMLM) techniques and all of them use the same trick of separating molecules in time, as they cannot be separated in space. To achieve this separation, fluorescent molecules are stochastically switched on and off in time, like the lights of a Christmas tree. In this way, molecules that are “on” while the ones surrounding them are “off” can be seen as discrete entities. Due to diffraction of light, these discrete light spots still appear blurred, but their centers can be precisely determined by computer algorithms. The process of switching on molecules, localizing their centers and switching them off is repeated until many localizations have been collected. The resulting high resolution image is created by depicting all the localized centers.

GSDIM. In GSDIM and dSTORM, the switching of molecules is achieved by taking advantage of basic state transitions of conventional dyes 77. Excitation causes fluorescent molecules to jump from the resting state (S0) to the excited (S1) state, from where they return to S0 while emitting a photon: this is the principle of fluorescence. Besides to the S1 state, these molecules can also jump to additional states such as the triplet (T) and other metastable states. These states are longer-lived than S1 and more importantly, they are “dark”, i.e., no light is emitted when the molecules fall back to S0. In GSDIM molecules will switch on and off in time by stochastically jumping to and from S0, S1 or dark states, a process called blinking, that can be induced by exciting them with very high laser power.

Blinking and artifacts. The blinking characteristics of a fluorophore determine the super-resolution image quality, in particular the blink brightness and duty cycle. The brightness is the photon count of a fluorophore when it is on, and it determines the precision with which it can be localized and therefore the obtainable resolution 78. The on-off duty cycle is the fraction of time that a fluorophore spends in the on state 78. In a sample labeled with a fluorophore with a low duty cycle only a few molecules will be on per frame, enabling the proper localization of their centers 79-81. On the other hand, if the duty cycle of a fluorophore is high, two or more fluorophores that are close together might be on at the same time, causing overlap of their diffraction-limited images. Thus, these fluorophores will be seen as a single blink, leading to miss-localization at a point that is actually between them. This will lead to artefacts in the final reconstructed image. Moreover, high duty cycles also give rise to structured details in the background intensity. Structured background strongly affects the localization precision, reducing the resolution of the image and leading to artifacts in the final reconstructed image.Fortunately, the intrinsic molecular properties of fluorophores that allow us to switch them on and off can be manipulated. Fluorophore blinking is improved by the addition of a reducing agent, in general 2-Mercaptoethylamine (MEA), and for some types of fluorophores, removal of oxygen by addition of an oxygen scavenging system is beneficial 78,82. To date, only a limited number of fluorophores with good blinking properties is available. Moreover, different types of fluorophores benefit from different types of buffers, making the choice of combinations for two or more color images challenging. Large efforts are being made to develop buffer conditions that enhance blinking and are compatible with multiple

17

Introduction

1fluorophores 83,84. Moreover, new background subtraction and localization algorithms are developed to localize centers more accurately, even at suboptimal conditions.

Figure 6. Effect of suboptimal blinking on localization and image outcome. a) Two fluorophores that are closer together than the resolution limit must be sequentially switched on for their centers to be localized. If they blink simultaneously (for example due to high on-off duty cycles) they will be detected as one single event, and the calculated center will be localized in a position between the molecules. b) Depiction of the resulting calculated localizations of a circular object labeled with fluorophores with low (up) or high (down) on-off duty cycles. Fluorophores with high duty cycles will often be on simultaneously, causing multiple incorrect localizations. Super-resolution images (right) of a clathrin coated vesicle labeled with AlexaFluor647 (top, low duty cycle) and AlexaFluor532 (bottom, high duty cycle) show the different image outcomes of two fluorophores with different switching rates.

Sample preparation and fixation-induced artifacts. In most cases samples need to be fixed to be labeled for immunofluorescence. The purpose of fixation is to preserve a sample for observation without altering its morphological characteristics. Fixation artifacts, i.e. changes in the appearance of samples caused by imperfect fixation processes, are common. Therefore, optimization of the fixation process to preserve sample structure is of crucial importance. There are several ways to fixate cells that differ in how well they preserve different structures. Fixatives immobilize proteins by forming covalent chemical bonds or precipitating them. The most common type of fixatives are aldehydes, such as formaldehyde and glutaraldehyde, and they fixate cells by crosslinking proteins. Formaldehyde primarily links proteins via their lysine residues and glutaraldehyde crosslinks proteins in a stronger manner because it contains two aldehyde moieties that can link more distant pairs of proteins. The latter is predominantly used in electron microscopy (EM). However, although the fixating properties of glutaraldehyde are better than those of formaldehyde, its rate of diffusion across membranes is slower, making it necessary to pre-extract the sample before fixation. Pre-extraction can induce loss of soluble or weakly bound proteins. Moreover, glutaraldehyde often disrupts antibody binding sites (epitopes), making immunolabeling inefficient or unfeasible. Besides the type of fixatives, important factors to take into account for the fixation are the pH, temperature and osmolality of the fixation solution, which help maintain the characteristics of the sample during the process. The duration of the fixation process can also affect the outcome. Short fixation times can be insufficient, depending on the thickness of the sample, while long fixation times can affect epitope binding.In the electron microscopy community, fixation protocols have been optimized for many years, yielding excellent structure preservation in fixed samples. Protocols borrowed from EM for super-resolution microscopy therefore have the potential to provide excellent results, but they haven’t been optimized for immunofluorescence, and they commonly employ glutaraldehyde, which can damage antibody epitopes and causes sparse labeling.

b)a)

Object

High duty cycle

Low duty cycle

Simultaneus blinking

Sequentialblinking

Object

18

1In contrast, sample preparation guidelines from conventional fluorescence microscopy often lead to dense labeling but sample preservation is not sufficient for SR microscopy. While in the last few years much effort has been put into the optimization of blinking for SR microscopy, the effects of sample preparation on image quality have only recently begun to be explored 85,86.

a) b)

c) d)

e) f)

Figure 7. Super-resolution images showing the effect of different fixatives in the preservation of filamentous cell structures. a) Tubulin in HeLa cells fixed with Methanol. b) Tubulin in HeLa cells fixed with formaldehyde (in PBS, room temperature). c) Vimentin in HeLa cells fixed with glutaraldhyde. d) Vimentin in HeLa cells fixed with formaldehyde (in PBS, room temperature) e) F-actin in HeLa cells fixed with glutaraldehyde. d) F-actin in HeLa cells fixed with formaldehyde (in PBS, room temperature). In all panels: zoomed-in images on the right of the panel are depicted with dashed lines in original images on the left of the panel. Scale bar: 5 µm, scale bar zoomed-in images: 1 µm.

Quantitative super-resolution microscopy. Since the development of super-resolution microscopy a large number of studies have been published that provided beautiful high-resolution images as well as new scientific insights. To name a few, super-resolution microscopy has been applied to study the nuclear pore complex 87, mechanosensory podosomes 88, the ESCRT machinery at HIV assembly sites 89, the actin cytoskeleton in axons 90, among many others.Super-resolution microscopy can yield a large number of multicolor high-resolution images allowing quantitative analysis of protein colocalization. The several methods for quantification of color colocalization that already exist for conventional microscopy are not optimal for super-resolution microscopy since they are based per definition on quantification of 2-color signal overlap. In super-resolution, the signals of two molecules typically do not

19

Introduction

1overlap, but rather appear in close proximity. Therefore, new tools for the quantification of “colocalization” must be developed. Analysis of colocalization of proteins in super-resolution data has been replaced by modeling of relative distributions 87,89, normalized density or brightness profiles 48,88, the Hopkins index 48 and recently, a method developed to use the coordinates of the localizations to calculate their proximity 91,92. Another quantitative application of super-resolution microscopy is the study of protein clustering. In a cell, molecules often associate in groups to make processes more efficient 93. Signaling molecules assemble in small clusters to enhance their signals 94,95 or may become clustered by adaptors in order to be internalized 34, and so on. As with colocalization studies, many super-resolution publications focus on quantifying the arrangement of proteins in clusters by using different approaches, e.g.96-100. To quantify clustering, the analysis method is of crucial importance, since in stochastic SMLM, fluorophores may be localized more than once due to blinking and molecules are labeled with more than one fluorophore (6 to 7 fluorophores per secondary antibody, according to manufacturers). These multiple localizations per protein will form a small cluster, a phenomenon that we call intrinsic clustering. Intrinsic clustering will generate ‘false positives’ when conventional clustering analysis methods are used on super-resolution data, and it needs to be either controlled for, or factored out. Most published clustering analysis studies in super-resolution are based on Ripley’s K function (e.g., 100,101), which does not take into account the intrinsic clustering component and can thus produce erroneous results. Therefore, new methods that are better suited for data from stochastic SMLM and that take the intrinsic clustering component into account have been developed in the past few years 102,103. These methods are especially suited for the study of small clusters of proteins that could be masked by intrinsic clustering when using Ripley’s K function.

20

1Thesis outline

In the last years, after the commercialization of super-resolution microscopes, several studies have been published that apply super-resolution microscopy to answer biological questions. However, super-resolution microscopy is still in its infancy and optimization of labels, buffers, sample preparation methods, methods for visualization and correct data analysis are still being developed. In this thesis we present two studies that focus on the improvement of the quality of super-resolution images. We then apply improved super-resolution microscopy to study the processes of cell signaling and endocytosis.In Chapter 2 we focus on the correction of structured background that comes from cellular auto-fluorescence, out-of-focus fluorescence or fluorophores with high blinking duty cycles. We show that this type of background can be corrected for with a temporal median filter that preserves the short and bright blinking events. Subtracting the estimated background is a very simple procedure that can highly improve the quality of super-resolution data, especially when a fluorophore shows suboptimal blinking. This strategy therefore expands possibilities for multicolor super-resolution microscopy. When satisfactory blinking and data processing are achieved, the resolution obtained will depend ultimately on the level of conservation of the sample. In Chapter 3 we explore fixation methods that are compatible with immunolabeling and help achieving high resolution images of the actin cytoskeleton and its regulatory and binding proteins.The first biological application concerns endocytosis, which regulates the lipid and protein composition of the plasma membrane, and therefore the interaction of cells with their environment. Microscopy has played an important role in unraveling the mechanism of clathrin mediated endocytosis (CME), the most common and best studied of internalization routes. Live-cell microscopy and electron microscopy studies were instrumental for deciphering the different stages of clathrin vesicle formation and budding from the membrane. Electron microscopy also revealed the two existing types of clathrin coated structures: clathrin coated pits (CCPs) and flat clathrin lattices (FCLs). Moreover, the recruitment of actin and actin binding proteins to sites of endocytosis was shown using fluorescence microscopy. However, both the role of actin in CME and the difference between CCPs and FCLs are not yet completely understood. In Chapter 4 we use super-resolution microscopy to shed light into the function of FCLs. We find that they are hubs from which clathrin coated vesicles emerge, regulated by actin polymerization via N-WASP and the Arp2/3 complex. Moreover, we show that FCLs recruit receptors such as EGFR and LPAR1, and that they are crucial in the regulation of the signaling of LPAR1.Finally, in Chapter 5 we study the effect of Perifosine, a synthetic lipid, on EGFR. We show that Perifosine induces the internalization of EGFR via an unconventional mechanism, specifically in cells that highly express the receptor. To try to elucidate the rationale for this specificity of Perifosine, we examine the status of EGFR in the plasma membrane of these cells, and of cells with low EGFR expression, using an improved clustering analysis method suited for super-resolution data.

21

Introduction

1

23

Chapter 2

The fidelity of stochastic single-molecule super-resolution reconstructions critically depends upon robust background estimation

Eelco Hoogendoorn1*, Kevin C. Crosby1*, Daniela Leyton-Puig2*, Ronald M.P. Breedijk1, Kees Jalink2, Theodorus W.J. Gadella1 and Marten Postma1

1Section of Molecular Cytology and Van Leeuwenhoek Centre of Advanced Microscopy, Swammerdam Institute for Life Sciences, University of Amsterdam Science Park 904, NL-1098 XH Amsterdam The Netherlands

2Division of Cell Biology and Van Leeuwenhoek Centre of Advanced Microscopy, The Netherlands Cancer Institute, Amsterdam, The Netherlands.

* Authors contributed equally to this work.

Scientific Reports 4, 3854. 2014 Jan 24

24

2

INTRODUCTION

The diffraction limit, which has traditionally limited the ability of light microscopy to discern biological structures at nanometer resolution, has been circumvented by a number of different super-resolution techniques 104. In one class of widely used methods, including (F)PALM, STORM, dSTORM, and GSDIM 105, individual fluorescent molecules are stochastically switched to a temporary detectable state, during which the location of the individual molecules is determined at higher resolution using image analysis algorithms 106-108. Several different methodologies for performing stochastic single-molecule super-resolution reconstructions have been described and generally fall into two broad categories: localization based 106,107,109,110, and grid based reconstruction methods 108,111. Localization based methods typically utilize a Gaussian fit or a center of mass calculation, while grid based reconstruction methods rely on an inverse modeling approach by deconvolution or compressed sensing.However, a typical super-resolution -dataset may contain significant non-sparse, structured background components, complicating the analysis regardless of the method chosen for analysis. This background may accrue for a variety of reasons, such as weakly, continuously emitting fluorescent molecules attached to cellular structures or cellular auto-fluorescence 74,112. In order to accurately reconstruct a super-resolution image, all analysis algorithms require that the foreground signal from sparsely distributed emitters (containing the super-resolution information) is sufficiently separated from this background. For each data frame the observed fluorescence can be modeled as a sparse distribution of emitters that is convolved with a given or estimated point spread function and a spatio-temporal background:

( ),model convolve PSF distribution background= +

(Eq. 1)

The first term (foreground) contains the super resolution information and is fitted to the PSF model, given a certain estimated or fitted background. We find that the quality of this background estimate is critical to attaining reliable reconstructions; in many practical circumstances this can have a much greater impact on the fidelity of the final image than the specifics of the treatment of the foreground term. The vast majority of published super-resolution reconstruction algorithms utilize spatial filtering or local background fitting for background estimation. While foreground and background can be distinguished with some limited specificity on the basis of their spatial frequencies and intensity, there typically exists no clear band-gap between these spatial frequencies across the whole data set. This makes spatial filtering a limited tool for robustly separating foreground from complex, structured background. A key difference between non-specific (background) fluorescence and emitters of interest is that the latter appear and disappear over relatively rapid timescales. In the general signal processing literature, there are many different methods described for background estimation that exploit temporal information, which can greatly differ in computational complexity 113. Although a few previous super-resolution studies have mentioned, obliquely, some form of temporal filtering 114,115 for estimating the background component, the importance and effect of this type of background estimation has not been rigorously studied or reported on. For this reason, we have explored temporal background estimation methods in the context of super-resolution. We found that a running median filter applied to each pixel in the dataset along its temporal axis represents a straightforward and particular effective background estimator that greatly enhances the quality of the reconstruction. The logic behind the

25

The fidelity of stochastic single-molecule super-resolution reconstructions critically

depends upon robust background estimation

2

median filter as a background estimator is that super-resolution data is always somewhat sparse, and insofar as it is sparse, foreground contributions will tend be discarded by the median filter as outliers and therefore readily separated from background components. The running nature of the filter allows for gradual temporal changes in background and an arbitrary spatial shape of the background is permitted.We have applied temporal median filtering to data obtained from several different stochastic super resolution techniques, reconstruction methods, and probes. Furthermore, we have performed a number of simulations that mimic realistic conditions for stochastic super resolution data in order to validate and check the effect of the various techniques.

RESULTS

Estimation of background component using temporal median filter

The ability of a temporal median filter to separate background and foreground is illustrated in Figure 1. The panels in the left column (a,d,g) show raw data frames from LifeAct-mEos3.2, MyosinIIa-Alexa532 and MyosinIIa-Alexa647 data sets. The middle column (b,e,h) shows the background estimated for that frame using the temporal median filter (window size of 101 frames with 10 frame interpolation, see material and methods). The right column (c,f,i) shows the foreground obtained by directly subtracting the estimated background from the raw frames. Static or very slowly varying fluorescence in the image largely ends up in the background. Notable are the fiducial beads visible in panels a and b that are no longer apparent in the foreground image, as well as the ridge at the cell border (arrow). Panel

Figure 1. Background and foreground estimation by temporal median filtering. Panels in the left column (a,d,g) show raw data frames from LifeAct-mEos3.2, MyosinIIa-Alexa532 and MyosinIIa-Alexa647 data sets. Middle column (b,e,h) shows the background estimated for that frame using the temporal median filter (window size of 101 frames with 10 frame interpolation). Right column (c,f,i) shows the foreground calculated by subtracting the estimated background from the raw frames. For display purposes the values were clipped at zero in order to only show the fluorescence that is higher than the estimated background. Panel j shows the raw fluorescence trace (MyosinIIa Alexa 647 data set) at two adjacent pixels (arrow in panel h) and the corresponding background estimate.

26

2

j shows the raw fluorescence trace (MyosinIIa Alexa 647 data set) at two adjacent pixels (arrow in panel h) and the corresponding background estimate. The background level is different for these two adjacent pixels, but after correction for background the foreground traces, shown in panel k, are now strongly correlated. The two traces exhibit a number of switching events that are now more accurately separated from the background. The estimated background is relatively smooth compared to the noise in the raw data, which depends on the window size used for the running median filter (see material and methods). Note that the part of the signal that the median filter deems to be background, is not sparse, which would confound attempts at deducing its origin of emission with any super resolution algorithm. The implied foreground component by contrast, is very sparse. The dataset in the top row contains a very dominant and complex background. Even many of the strong foreground events are barely recognizable as such in a single raw data frame. The Myosin-Alexa dataset of the third row does not appear to contain much background upon inspection of the raw data, but the temporal median filter nonetheless reveals a substantial non-sparse signal component.

Dual-color GSDIM co-localization experiment

We performed several dual-color GSDIM co-localization experiments using the probes Alexa532 and Alexa647 attached to different secondary antibodies recognizing the same primary antibody that binds to myosinIIa in one experiment (Fig. 2) and vinculin in another experiment (Supplementary Fig. S1). Hence, a clear colocalization of the Alexa 532 and 647 color channels is expected in the super resolution reconstructions. However, for both experiments we observed large discrepancies between the reconstructed color channels (Fig. 2b,f, Supplementary Fig. S1a,c) using three representative (existing) super-resolution analysis algorithms, including Gaussian fitting 106, center of mass localization 107

Figure 2. Application of a temporal median filter prior to localization analysis improves fidelity of two-color GSDIM data. GSDIM imaging of myosinIIa independently labeled with Alexa532 (a-c) and Alexa647 (e-g). Without the utilization of the temporal median filter, the RapidSTORM reconstruction of the Alexa532 data set shows localizations that are skewed towards regions of high fluorescence (b) and exhibit poor co-localization with the Alexa647 (f) based on Pearson’s cross-correlation analysis (f, inset). Use of the temporal median filter prior to running the localization eliminates these artifacts in the Alexa532 reconstruction (c) and shows higher correlation with the Alexa647 reconstruction (g, inset). Intensity traces (d, h) are normalized to the area under the trace. Similar analysis for alternative reconstruction methods can be found in supplemental Figs S2,3. Scale bar: 3 µm.

27



2

(Supplementary Fig. S2a,d), and a (multi-fitting) grid based method 108 (Supplementary Fig. S3a,d). Specifically, in the Alexa532 data-sets, the presence of a moderately high background (Fig. 2a, supplementary video 1) results in a systematic bias such that the reconstruction is skewed towards regions with high fluorescence intensities. This gives the appearance of distinct foci along the fibers and subtle deformations, such as over-sharpening (compare Fig2. b and f, Fig2. b and c, Intensity profile, Fig. 2d). These artifacts were no longer observed when background was accounted for (see material and methods) using the temporal median filter prior to reconstruction (Fig. 2c, Supplementary Figs. S2, 3). Prior application of the temporal background-estimation filter results in a higher co-localization index as determined by cross-correlation analysis of the Alexa532 and Alexa647 data-sets (Fig. 2, insets).

Figure 3. Reconstructions for LifeAct-mEos3.2 HeLa cell using RapidSTORM, QuickPALM and deconvolution with and without the temporal median filter applied. An area that shows a high-degree of structured (heterogeneous) background fluorescence indicated with an arrow in Fig S1a-b leads to a spurious structure when using RapidSTORM (median smoothing 5 px setting) or deconvolution if the temporal median was not applied. When the temporal median filter prior to running the reconstruction analysis was applied (b, e, h) the effect of structured background is greatly reduced in the reconstruction such that the intensity profiles are now in much closer agreement for the different methods (c, f, i). This illustrates the relative importance of background estimation in the overall reconstruction process. Scale bar: 3 µm.

28

2

LifeAct-mEos3.2 PALM data with structured backgroundIn PALM data structured-background occurs quite frequently, as a slow buildup of cellular auto-fluorescence with complex spatial characteristics can develop during the acquisition 74,112. In a HeLa cell expressing LifeAct-mEos3.2 116,117 we observed this phenomenon (Fig. 1a,b, arrow, and Supplementary video S2), which gives rise to errant super resolution localizations yielding a spurious structure along the edge of the background (Fig 3a,c) and deformations. These types of artifacts are no longer present after improved background correction (Fig 3a-c), regardless of reconstruction method. Improper estimates of structured background of such large relative magnitude may easily lead to artifacts in the reconstruction on a micrometer scale, nullifying the intent of super resolution microscopy.

Intricate structures in LifeAct-Venus GSDIM data sets Figure 4 shows two additional GSDIM data sets obtained with LifeAct-Venus, a different fluorescent probe. Panel 4a shows the sum of all frames in the data sets; panel 4b, shows the sum of all background subtracted frames, both represent a diffraction limited image. Panels 4c and 4d show the RapidSTORM 106 reconstruction obtained without and with application of the temporal median filter. The analysis of these data sets revealed that temporal median filtering reduces the presence of strong foci at filament crossings, which appear to induce deformations that are not apparent in the diffraction limited images (arrow 4a,d). The fact that panel 4c has features that deviate from panels 4a, 4b and 4d can be attributed to the inappropriate application of a super resolution algorithm to a dataset that is not sufficiently sparse.

Figure 4. GSDIM data of two HeLA cells with LifeAct-Venus Panel a shows the sum of all frames in the data sets; panel b, shows the sum of all background subtracted frames, both represent a diffraction limited image. Panels c and d show the RapidSTORM reconstruction obtained without and with application of the temporal median filter. The analysis of these data sets revealed that temporal median filtering reduces the presence of strong foci at filament crossings, which appear to induce deformations that are not apparent in the diffraction limited images. Panels e and f show the RapidSTORM reconstruction obtained without and with application of the temporal median filter of another HeLa cell with an intricate F-actin network. The used threshold for all reconstructions was the same and the Gaussian smoothing filter was selected in this case (1 sigma). Color scale for images in panels c,d and e,f were chosen to be equal (scale bar 3 µm).

29



2

Panels 4e and 4f show the RapidSTORM reconstruction obtained without and with application of the temporal median filter of another HeLa cell with LifeAct-Venus. The background-corrected reconstruction reveals more intricate details in the F-actin structures, which are otherwise lost to background-related localization errors. Also the deformation induced by a hot-spot (arrow 4e,f) is greatly reduced in the background corrected reconstruction.

Analysis of synthetic datasetsA number of simulations that mimic realistic conditions for stochastic super resolution data were performed in order to validate and check the effect of the various techniques. Key parameters such as event amplitude, event density, and background conditions were varied. For the latter, uniform background conditions and structured background conditions were used. These parameter variations were applied to ring structures and filament structures, for which the ring size and the distance between filaments was varied respectively. Previous work in synthetic analysis has aimed to characterize the performance of a method in terms of accuracy, and a false positive and false negative rate. A different approach is taken here, as these measures are not necessarily adequate to characterize the effect of structured background on reconstruction quality. Specifically, the method described here is aimed to better quantify the systematic bias induced by correlated errors in the reconstruction, which is not conveyed by localization standard error of the mean alone. An image may appear very sharp, which in the absence of bias may be regarded as an indication of high accuracy, but in the presence of systematic biases, the result may in fact be substantially distorted.

Figure 5 shows the results of simulations of 10 nm wide filament pairs separated at decreasing distances (Fig. 5a) that were positioned at different distances (d) and placed on structured background (see material and methods). A simulated data set containing structured background (Fig. 5b) can result in reconstructions that show both artifacts and

Figure 5. Thin filaments at different inter-filament distances. (a) simulated with structured background (b). The structured background leads both artifacts and distorted structures (c), which are mitigated by the utilization of a temporal median filter prior to performing the localization analysis (d), resulting in a more accurate rendering of the structures. (e) Quantification of synthetic filaments shown in c-d. Pairs of filaments, each with thickness 10 nm (black bar). The profiles shown in red and green represent the quantified mean profile of the reconstructed filaments measured from the midline outward. Reconstructions were obtained from RapidSTORM, with settings median smoothing (5 px) and threshold 100 and rendered using the obtained amplitude blurred with a Gaussian with a SEM of 2 nm. Prior to reconstruction by RapidSTORM the background was accounted for using the temporal median filter, with a filter size of 101 frames and 10 frame interpolation.

30

2

distorted structures (Fig. 5c). These effects are mitigated by the utilization of a temporal median filter prior to performing the localization analysis (Fig. 5d), resulting in a more accurate rendering of the structures. Fig. 5e shows the quantification of synthetic filaments shown in Fig. 5c-d. The profiles shown in red and green represent the quantified mean profile of the reconstructed filaments measured from the midline outward. The red profiles were obtained using RapidSTORM without applying the temporal median filter prior to analysis and the green profiles were obtained using RapidSTORM after application of the temporal median filter. Application of the temporal median filter removes the artifactual structure in between the filaments (zero position), which was caused by the structured background. Furthermore, the application of the temporal median filter removed the inward bias visible in the top row panels (red), which is no longer present in the bottom row panels (green).A quantification of localization events in isolation makes it fundamentally difficult to consider correlations between localization errors. But by quantifying the appearance of structures as a whole, a measure of correlated errors can be derived. Line and ring patterns can be used to characterize the behavior of an algorithm. Line patterns (Fig. 5) test the ability to resolve nearby features, whereas ring patterns (Fig. 6) additionally test the influence of curved geometries as well. In order to quantitatively characterize the accuracy of a method, the symmetry axis of these 2D patterns is utilized, to obtain a 1D distribution of localizations (profiles). The repeatability of the localizations can be characterized in term of the width of these distributions and the systematic bias is given by the offset between the distribution of localizations and the underlying structure. It is in this measure of bias that the different methods investigated here show the biggest performance differences and where the necessity of accurate background estimation becomes most apparent.Figure 6 shows the effect of application of the temporal median filter prior to reconstruction with RapidSTORM on rings with a radius of 75 nm (see material and methods). The left column in both figures shows the reconstruction of the ring when there is no structured background. For all amplitudes, we find that without the presence of background the ring can be reconstructed reliably (left column), independent of the temporal median filter. However, if the structured background increases, localizations are increasingly biased by the structured background, not only skewing the localizations towards the center of the ring, yielding a smaller ring, but also generating false positives caused by the structured background. When the background, estimated using the temporal median filter, is accounted for prior to analysis with RapidSTORM, in all cases it was observed that the ring can be constructed without any bias towards the center of the ring giving a similar result independent of the background level. As expected from the Thompson equation 118, at higher background levels and lower amplitudes the accuracy is reduced, yielding a wider ring wall but the result remains unbiased and its position is conserved. We have also explored the effect of increasing event densities on the reconstruction of the rings with a radius of 150 nm using the same simulation parameters as used for the 75 nm ring. It was observed that an increasing overlap of events can introduce a reconstruction bias towards the center of the ring (Figs. S7). Application of the temporal median filter will compensate for this bias when using RapidSTORM (Fig. S8) because the temporal median filter removes part of the fluorescence that stems from overlapping events. However, this result is particular to RapidSTORM. The median filter in combination with a deconvolution reconstruction in fact shows a slight outward bias of the ring at extreme event density (Fig S9). This highlights the importance of choosing an appropriate combination of background estimator and localization algorithm (see Discussion).

31



2

Figure 6 RapidSTORM reconstructions of a ring with radius 75 nm with a threshold of 10 photons and using the standard spatial median filter without (a) and with (b) application of the temporal median filter prior to reconstruction. Background corrected reconstruction no longer show artifacts in the reconstructions and in all cases reliably reconstruct the ring. For these simulations the event cycle amplitude A = [200, 800, 3200] was varied and the structured background, where the peak of the structured background was varied using the values bs = [0, 10, 20, 30, 40, 50] photons using the same event list for each case with simulations settings as described in the supplementary information. The radial profiles of the rings were calculated for each panel and are shown together with the original ring in the right hand side column. The intensity scale in each panel was adjusted to show the full intensity range independent of the other panels. The localizations were obtained from RapidSTORM with settings median smoothing (5 px) and threshold 100 and rendered using the obtained amplitude blurred with a Gaussian with a SEM of 2 nm. Prior to reconstruction by RapidSTORM the background was accounted for using the temporal median filter, with a filter size of 101 frames and 10 frame interpolation.

32

2

DISCUSSIONWe have shown that appropriate background estimation is of pivotal importance for obtaining reliable super resolution reconstructions, which we illustrate with data obtained from (d)STORM, GSDIM, and PALM using several different probes representing a range of data types and qualities. Especially in case of spatially complex background the advantages of temporal median background filtering can be profound. Application of the proposed background estimator eliminates the large discrepancies that are otherwise observed when analyzing the same data with different super-resolution algorithms (Fig. 7), underlining the relative importance of background estimation in the overall reconstruction process. By using an appropriate background estimator, a wider range of imaging conditions, type of probes and samples are tolerated for stochastic super-resolution microscopy. Use of a temporal median filter enables background estimation for each pixel in a given data frame from the temporal distribution; an estimate of the background is obtained in this way without interfering with the spatial resolution of the original image. Ideally the use of temporal median filtering and estimation of the background is directly incorporated in the reconstruction software taking into account the appropriate statistical model. However, we observe that background subtraction with uniform offset can be applied as a preprocessing step and that reconstruction on the background corrected data set using existing software in general works robustly and that the added uncertainty generally is negligible compared to the detrimental effects structured background can have. In the case of localization based algorithms, special care could be taken when calculating the localization error as the background estimate will introduce some extra noise. The standard error of the median is by good approximation proportional to the standard error of the mean, which scales with α×σ/√N, where N denotes the median filter window size, σ² the variance, and α denotes a proportionality factor that depends on the distribution. For Poissonian data the variance scales with the mean intensity µ, σ²=k²×µ, where k denotes the detector gain. Then the added variance due to the background estimate becomes σb² = α²×σ²/N. The relative increase in the variance is approximately σb²/σ² = α²/N. The approximate proportionality factor for a normal like distribution is about 1.253, hence for a window size of ~100 frames the variance increases by 1.6%. We have compared a variety of existing widely used background estimation methods. The local fitting of background as employed in 106 generally turns out to be the least robust method (Fig 3). Understandably, this method cannot discern the peak of a background feature from a foreground event, nor can one effectively discriminate a peak on a slope, from a displaced peak. Therefore, using such a method leads to both false positives and biases, respectively.A spatial (Gaussian) filter as a background estimate 107 works somewhat better, but is sensitive to sharp background features, which get mistaken for foreground, and high density areas, which readily get mistaken for background in a distortive manner (Fig S2).The published temporal filters have been implemented and are found to be generally superior to the spatial filters. The method in 114 proposes a ten frame running mean filter. This works reasonably well for guarding against errors in localization accuracy, but we find that for high density high S/N data, the median filter captures up to four times as much of the foreground signal, compared to the short mean filter, which may result in substantial distortions of appearance (supplemental Fig. S10). The differences of frames method employed in 119 can be viewed as a 1-frame mean filter and consequently shares similar shortcomings, but to a stronger degree.The background estimation method employed in 112 may also be considered a temporal filter. In essence, the first principal component of the dataset is taken as a background

33



2

estimate. This may be regarded as a truncated eigenbackground 113. For datasets with simple background behavior, this method works well. But according to expectation, we find a single principal component as a background estimate breaks down for datasets where the background dynamics are too complex to be well represented by a single component (Supplemental Fig. S11).

We find that the temporal median filter generally works best over the range parameter conditions tested. That said, at extreme event densities background estimation becomes fundamentally more difficult and potentially significant errors may remain in the final reconstruction (Supplemental Fig. S9-11). Temporal median filtering can be easily integrated into existing workflows and consistently shows profound improvements in combination with several different classes of reconstruction algorithms and over a wide-range of data types and qualities. As our results demonstrate, without a conscious choice of background estimator appropriate for the given data quality, one is not guaranteed to achieve any super resolution at all. Because of the significant impact background can have on the fidelity of the final reconstructed image, it is highly recommended that an appropriate consideration of background estimation be a fundamental part of any stochastic super-resolution analysis workflow.

Figure 7. Line scan profiles for the LifeAct-mEos3.2, MyosinIIa-Alexa532 and MyosinIIa-Alexa647 data sets from (Fig. 2, 3,S2, S3) put side by side, using three reconstruction methods without and with application of the temporal median filter. The panels a, c and e reveal that the different methods give different results for the same data set when no temporal median filter is applied. The panels b, d and f reveals that application of the temporal median filter yields results that are in close agreement for the three reconstruction methods used.

34

2

METHODS

Estimation of the background component For most datasets with slow varying mean frame intensities the temporal median filter can be applied directly. However, because the average frame intensity can sometimes vary significantly due, for example, to an overall loss of fluorescence caused by depletion of fluorescent molecules or variations in the intensity of the excitation laser, a direct application of a temporal median filter on the raw data may be filtering other temporal signal components than event switching. To correct for entire frame intensity fluctuations, first the mean fluorescence intensity for each frame is determined and subsequently the data is scaled according to this mean fluorescence profile 112. A temporal moving median filter is then applied to the scaled data. The obtained median values are then rescaled by multiplying the median values with the mean frame intensity.

, , ,t x y t x yD D= (Eq. 2a)

, ,

, ,x y t

x y tt

DN

D=

(Eq. 2b)

{ }, , , , , , 1 , , , , 1 , ,median , ,..., ,..., ,x y t t x y t w x y t w x y t x y t w x y t wB D N N N N N− − + + − += × (Eq. 2c)

Where tD denotes the mean frame intensity for frame t, , ,x y tN denotes the normalized data frame and w denotes the window size for applying the temporal median filter. The size of the time window should be chosen such that it is significantly longer than the typical slow-switching events (>>10). For our datasets, we typically use a window size of about 100 frames to calculate the moving median. We have implemented two ways of computing the running temporal median. One way of efficiently computing the filter is by incrementally updating a list of sorted values for each pixel, and taking the center of the list. Another way of increasing efficiency is by means of computing the median only at certain keyframes, and linearly interpolating those. Such keyframe interpolation is well justified by the slowly varying nature of the background itself. Both methods produce virtual indistinguishable end-results. We find that the keyframe method is roughly equally efficient as the incremental method, given keyframes spaced 12 frames apart. With a 50 frame median filter radius (101 frames), of 512x512 pixels, this processing takes ~0.1 seconds per frame on an Intel i7-2700k at 3.8GHz. This makes the keyframe method somewhat preferable; it could be made faster still by increasing the distance between keyframes without significantly affecting quality, and it does not rely on low level language extensions to attain this speed, making it easier to integrate as a technique (a python script is provided). It is preferable that each stochastic super resolution software package integrates an optimized version of calculating the median filter, depending on the used programming language or hardware architecture (e.g. parallel computing or GPU) different approaches can be followed for calculating the running median 120 .

Constructs and sample preparationThe LifeAct-mEos3.2 was a kind gift from Tao Xu 116,117. HeLa cells were obtained from the American Tissue Culture Collection (ATCC). Cells were maintained in Dulbecco’s Modified Eagle Medium supplemented with GlutaMAX and 10% Fetal bovine serum (Invitrogen, Bleiswijk NL) and grown for at least 2 days in phenol-free media before imaging. Cells were plated on 24mm #1 round cover-glasses (Menzel-Gläser) in six well plates. Transfections

35



2

were done with Lipofectamine 2000 reagent according to the manufacturer’s protocol (Invitrogen, Bleiswijk NL). Twenty-four hours after transfection, cells were transferred to an Attofluor sample chamber (Invitrogen) and imaged live in microscopy medium (137mM NaCl, 5.4mM KCl, 1.8 mM CaCl2, 0.8 mM MgSO4, 20 mM D-glucose, 20mM HEPES).

Optical setup and imagingPALM imaging was performed on an inverted Nikon Eclipse Ti microscope equipped with a TIRF system using a 60x ApoTIRF 1.49 oil objective. A Coherent OBIS 488 50mW laser was used to locate cells producing pre-converted mEos3.2. Photo-conversion of mEos3.2 was elicited by continuous illumination with a Coherent 50mW 405 Cube laser with power settings of <1mW, while the converted FP was excited and bleached with a 1W 561 Coherent OPSL with power settings typically in the range of 200-300W/cm2. Excitation light was passed through a quad-band dichroic 405/488/561/640 (Chroma). Emission light was passed through a 561 nm RazorEdge ultrasteep long-pass edge filter (Semrock). Images were recorded with an Andor iXon 897 EMCCD with 50 ms exposure times at a frame rate at 12.2 Hz. Pixel size in the image was 67 nm.

Super-resolution imaging of Non-muscle MyosinIIA filaments and vinculinA7r5 cells were cultured on #1.5 coverslips. After 48 hours cells were washed briefly with PBS, fixed with 4% PFA for 10 min at room temperature and extracted with 0.1% Triton X-100. Samples were extensively washed with PBS and blocked with 5% BSA for 30 min at room temperature. MyosinIIA was labeled with a monoclonal primary antibody raised in rabbit (Sigma-Aldrich) diluted to a final concentration of 1mg/ml, for one hour at room temperature, washed, and incubated then with anti-rabbit IgG polyclonal antibody conjugated to Alexa Fluor 647 dye molecules (Invitrogen) and anti-rabbit IgG polyclonal antibody conjugated to Alexa Fluor 532 dye molecules (Invitrogen) both at a final concentration of 0.01 mg/ml, for 30 minutes at room temperature. Cells were imaged in the presence of an oxygen scavenging system (10% glucose, 0.5mg/ml glucose oxidase, 40 µg/ml catalase, 50mM MEA).HeLa cells were cultured on #1.5 coverslips. After 24 hours cells were washed briefly with PBS, fixed with 4% PFA for 10 min at room temperature, and extracted with 0.1% Triton X-100. Samples were extensively washed with PBS and blocked with 5% BSA for 30 min at room temperature. Vinculin was labeled with a monoclonal primary antibody raised in mouse (abcam) diluted 1:400, for one hour at room temperature, washed and incubated then with anti-mouse IgG polyclonal antibody conjugated to Alexa Fluor 647 dye molecules (Invitrogen) and-anti mouse IgG polyclonal antibody conjugated to Alexa Fluor 532 dye molecules (Invitrogen) both at a final concentration of 0.01 mg/ml, for 30 minutes at room temperature. Cells were imaged in the presence of an oxygen scavenging system (10% glucose, 0.5mg/ml glucose oxidase, 40 µg/ml catalase, 50mM MEA).Imaging of the samples was carried out on a Leica SR-GSD microscope. Images were taken in TIRF mode at 100 frames per second. Colors were sequentially imaged in order of decreasing wavelength. The setup consisted of the following components: an inverted microscope (DMI6000 B, Leica Microsystems GmbH), a 1.47-NA TIRF objective (HCX PL APO 100× NA 1.47), a tube lens providing an extra factor of 1.6× in magnification, a 488-nm fiber laser (2RU-VFL-P-300-488), a 532-nm fiber laser (2RU-VFL-P-1000-532-B1R, MPB Communications), a 642-nm fiber laser (2RU-VFL-P-1000-642-B1R, MPB Communications) and an EMCCD camera (iXon DU-897, Andor) with an effective EM gain of 148. Images were taken in TIRF mode at 100 frames per second for ~5100 time frames, giving a total measurement time of about 1 min for each color. Colors were imaged in order of decreasing wavelength. The filter cube (642HP-T) for imaging with the 642-nm laser consisted of an

36

2

excitation filter (zet405/642x), a dichroic mirror (t405/642rpc) and emission filters (et710 100lp and ET650LP). The epifluorescence filter cube (532HP-T) for imaging with the 532-nm laser consisted of an excitation filter (zet405/532x), a dichroic mirror (t405/532rpc) and emission filters (et600/100m and ET550LP). Pixel size in the image was 93.11 nm.

Super-resolution imaging of ActinHeLa cells were cultured on #1.5 coverslips. After 24 hours cells were transiently transfected using PolyEthylene Imine (PEI) using 1 µg of DNA and 3µg of PEI per well on a 6-well plate, with a plasmid bearing LifeAct tagged with the yellow fluorescent protein variant Venus. After 24 hours cells were washed briefly with PBS and fixed with PFA for 10 min at room temperature. Samples were extensively washed with PBS and imaged in the presence of PBS.Imaging of the samples was carried out on a Leica SR-GSD microscope. Images were taken in TIRF mode at 100 frames per second. The setup consisted of the following components: an inverted microscope (DMI6000 B, Leica Microsystems GmbH), a 1.47-NA TIRF objective (HCX PL APO 100× NA 1.47), a tube lens providing an extra factor of 1.6× in magnification, a488-nm fiber laser (2RU-VFL-P-300-488), a 532-nm fiber laser (2RU-VFL-P-1000-532-B1R, MPB Communications), a 642-nm fiber laser (2RU-VFL-P-1000-642-B1R, MPB Communications) and an EMCCD camera (iXon DU-897, Andor) with an effective EM gain of 148. Images were taken in TIRF mode at 100 frames per second for ~5100 time frames, giving a total measurement time of about 1 min for each color. Colors were imaged in decreasing wavelength order. The filter cube (642HP-T) for imaging with the 642-nm laser consisted of an excitation filter (zet405/642x), a dichroic mirror (t405/642rpc) and emission filters (et710 100lp and ET650LP). The epifluorescence filter cube (532HP-T) for imaging with the 532-nm laser consisted of an excitation filter (zet405/532x), a dichroic mirror (t405/532rpc) and emission filters (et600/100m and ET550LP). Pixel size in the image was 93.11 nm.

Synthetic dataWe have performed a number of simulations that mimic realistic conditions for stochastic super resolution data in order to validate and check the effect of the various techniques. Key parameters, such as event amplitude, event density, and background (both uniform and structured) were varied. These parameter variations have been applied to ring structures and filament structures, with variable ring size and the distance between filaments respectively (see Supplementary Material for more details).

Reconstruction algorithms and renderingFor analysis of stochastic real and simulated data two localization based methods and a direct reconstruction, deconvolution based method were used. For the localization techniques, we utilized RapidSTORM 3 106,110, which performs a Gaussian based fit, and QuickPALM 1.1 107, which performs a center of mass based calculation. Background was corrected for as a preprocessing step for these two algorithms, the corrected data was offset with a constant value to prevent negative values. For the grid based method we implemented the deconvolution variant without regularization as described in 108,121. For this algorithm the background estimate was directly incorporated into statistical model. Prior to application of the reconstruction algorithms any motion drift was corrected if present. The user defined parameters for each of these methods, such as width of the PSF and intensity thresholds, were estimated from the experimental data or obtained directly from the synthetic input. The threshold was chosen in such a way that the number of detected events was minimal outside the cell or in regions without structures. Other user defined settings are mentioned

37



2

in the appropriate figure legends. In order to ensure consistency and to allow direct comparison between the final images, we utilized our own rendering program written in Matlab R2010a (code made available in supplementary material). From the localization lists obtained from RapidSTORM or QuickPALM the location and amplitude of each event was extracted and rendered; the zoom factor of the final image was set to 8. Each localization was rendered using a Gaussian (integral normalized to unity) with a FWHM of 23.55 nm (σ = 10 nm) and localizations from simulated data were rendered with a FWHM of 4.71 nm (σ = 2 nm).AcknowledgmentsWe thank M.A. Hink for useful discussions. This research was supported by the Netherlands Research Organization (NWO-ALW VIDI 864.09.015 to MP and EH), a Middelgroot investment grant (834.07.003), the European Science Foundation (EuroMembrane, TraPPs to KCC and TWJG) and the Dutch Technology Foundation STW (to TWJG, DLP and KJ)

Author contributionsEH and MP conceived and implemented the proposed method in close collaboration with KCC. EH, MP and KCC participated in the data analysis. KCC, RMPB, TWJG, DLP and KJ designed and executed the experiments. All authors contributed to the preparation of the manuscript.

Competing financial interestsThe authors declare no competing financial interests.

38

2

SUPPLEMENTARY MATERIAL

Figure S1 RapidSTORM reconstruction of a two-color GSDIM data set without and with application of the temporal median filter. Panels a and b show a cell labeled with a secondary antibody attached to the probe Alexa 532, which binds a primary antibody for vinculin, a component of focal adhesion complexes. Panels c and d show the same cell but now with a secondary antibody attached to the probe Alexa 647, which binds the same primary antibody for vinculin. Because the primary antibody binds the same focal adhesion structures, both secondary anti-bodies are expected to yield the same distribution of molecules. Application of the temporal median filter yields Alexa532 and Alexa647 reconstructions (panels b,d) that are in closer agreement than the reconstructions without using the temporal median filter (panels a,c). In RapidSTORM 5px median smoothing setting was used and the color scale for images in the same row were chosen equal (scale bar 3 µm).

39



2

Figure S2 QuickPALM reconstruction (0 SNR setting) MyosinIIa-Alexa532 (a-c) and MyosinIIa-Alexa647 (d-f) without (a,d) and with (b,e) application of the temporal median filter. Panels c and f show the profiles obtained from the line scans performed on images reconstructed without (yellow) and with (blue) the temporal median filter. Despite the fact that QuickPALM includes a background estimate based on spatial filtering, the two-color myosin sets are in closer agreement when the temporal median filter is applied. This can be seen from the line scan profiles as well as the cross-correlation plots shown in the insets.

40

2

Figure S3 Deconvolution reconstructions for MyosinIIa-Alexa532 (a-c) and MyosinIIa-Alexa647 (d-f) without (a,d) and with (b,e) application of the temporal median filter. Panels c and f show the profiles obtained from the line scans performed on images reconstructed without (yellow) and with (blue) the temporal median filter. The two-color myosin sets are in closer agreement when the temporal median filter is applied, as can be seen from the line scan profiles as well as the cross-correlation plots shown in the insets.

41



2

Synthetic dataEach simulation starts with a density map at high spatial resolution that represents the object of interest. Given the event rate, which is represented by a probability that a molecule will be switched on during a frame, a random number of positions are generated from the density map. Given the mean event cycle duration, a random event cycle duration for each position drawn from an exponential probability distribution was subsequently generated. The event can turn on and off in the middle of a frame and, based on these on and off times, the fraction that the event is on during the frames it covers in time was calculated. The total number of photons for each random event cycle is also randomly drawn from an exponential probability distribution based on a mean event cycle photon count. The event amplitude for a given molecule and fame is then given by the event cycle photon count multiplied by the fraction of time it is on during that frame. This effectively will yield an exponentially shaped frame event amplitude distribution, which is often observed in real data. This procedure is repeated for every frame in the synthetic stack and in this way an event list, consisting of positions, amplitudes and frame index numbers is produced. Given the event list, each frame in the stack was rendered by positioning a predefined PSF, with spatial integral unity, at each of the positions in the event list corresponding to that frame index and subsequent multiplication of it by the amplitude of the respective event. The image with the background model is then added to each frame and subsequently photon shot noise is generated for each pixel by drawing random numbers from a Poissonian distribution.

The point spread function that was used in the simulations is depicted in Figure S4A, and was directly estimated from a GSDIM dataset with the fluorescent dye Alexa647. A composite Gaussian was used to represent the PSF, such that also the side lobes are included in the PSF (Dempsey et al., 2011):

( )2 2

2 21 2

1 12 21 2

2 21 22 2

r ra aPSF r e eσ σ

πσ πσ

− −

= + (Eq. S1)

Where a1 and a2 denote the amplitudes and σ1 and σ2 represent the width of the two Gaussian components. The estimated amplitudes were 0.61 and 0.39 and the widths were 1.26 and 2.88 pixels, where the latter width represents the side lobe (gray line Fig. S5a). The FWHM of this composite PSF was 3.16 pixels. The GSDIM data set had a pixel size of 93.1 nm, hence the widths in nanometers where 117.2 and 268.1 nm, with a FWHM of 294 nm. In our simulations we used slightly smaller pixel sizes, 74 nm (rings) or 80 nm (filaments), which gives slightly lower values for the widths and FWHM on the nanometer scale. The PSF was calculated on a grid with the same size as the data frame (see inset Fig. S4a for detail), and was normalized such that the integral (sum) was unity. The normalization constant directly depends on the pixel size; smaller pixels will lead to lower peak amplitudes of the PSF.For each simulation an event list based on a density map was generated, which is a representation of the object at high resolution. Figure S4b shows an example of a ring (inset) with a radius of 75 nm and a ring wall thickness of 10 nm and the corresponding event distribution that was generated based on this ring. The original frame size was chosen to be 32×32 pixels with a pixel size of 74 nm, the density map was constructed at a zoom level of 32, resulting in an image of 1024×1024 pixels with a pixel size of 2.31 nm. The average number of molecules that are switched on during a frame, Nswitch, is given by the product of the total number of molecules in the ring, Nmol, and the probability, pswitch, that a molecule is switched on during that frame: Nswitch = Nmol × pswitch. For this example the average number of

42

2

molecules that switch on during each frame was Nswitch = 0.19. Each molecule that is switched on is assumed to have an average event cycle duration, τcycle, of 2 frames. The expected total number of events, Nevents, for Nframes = 10,000 frames is then Nswitch × Nframes × 3 = 5673, which gives an average of 0.57 events per frame. Each event cycle contains on average A photons, for this example we used A = 800 photons. However, the average number of photons per event per frame is lower because the photons contained in an event cycle are distributed over a number of frames. For this particular event cycle duration the expected number of photons per event per frame is <N> = 800/3 = 266.67 photons. Multiplying this number by the peak intensity of the PSF gives the average peak amplitude of events per frame <Npeak> = 266.67 × 0.07 = 18.78 photons. If the pixel size is smaller, this peak amplitude will also become smaller.Figure S4c shows an example of a ring that is projected on top of structured background. The structured background was obtained by using a Gaussian blur filter with a sigma of 3 diffraction limited pixels, hence 74 × 3 = 222 nm applied to the density map with the object. The events generated from the ring are directly projected on top of this background, after which Poisson noise was generated. The ring is significantly smaller than the FWHM of the PSF and in the diffraction-limited image, as shown in Figure S4d the ring cannot be resolved. The diffraction-limited image was obtained by calculating the average of the simulated stack.In order to study the effect of the temporal median filter, we have varied the number of photons per event cycle, switch probability, ring size and background conditions. If the ring size and switch probability was the same, the same event list for all simulations was used. For each ring with a specific combination of event cycle amplitude and background level, a separate stack was created. These stacks were subsequently stitched together in a composite stack that was used for analysis.

http://stack.in/

43



2

Figure S4 Simulation of rings (a) Point spread function with lobes used in the simulations. (b) Distribution of event sizes for a ring. (c) Example of a ring projected on a structured background. (d) The diffraction limited image of the ring on the structured background. See text for more details.

44

2

Figure S5 RapidSTORM reconstruction of a ring with radius 75 nm with a threshold of 10 photons and using the standard spatial median filter. For this simulation the event cycle amplitude A = [200, 400, 800, 1600, 3200, 6400] was varied and structured background, where the peak of the structured background was varied using the values bs = [0, 10, 20, 30, 40, 50] photons using the same event list for each case with simulations settings as described in the text. The color scale in each panel was adjusted to show the full intensity range independent of the other panels. The localizations were rendered using the amplitude obtained from RapidSTORM and blurred with a Gaussian with a SEM of 2 nm. The radial profiles of the rings were calculated for each panel and are shown together with the original ring in the right hand side column. RapidSTORM settings were median smoothing (5 px)

and threshold 100.

45



2

Figure S6 RapidSTORM reconstruction of a ring with radius 75 nm with a threshold of 10 photons and using the standard spatial median filter. Prior to reconstruction by RapidSTORM the background was subtracted that was estimated using the temporal median filter, with a filter size of 101 frames and 10 frame interpolation. This simulation and data analysis is the same as in Figure S9. RapidSTORM settings were median smoothing (5 px) and threshold 100.

46

2

Figure S7 The effect of increasing event density on the reconstruction of a background free ring with radius 150 nm with RapidSTORM using a threshold of 100 and spatial median smoothing (5 px). For this simulation the event cycle amplitude A = (Berberoglu, 2009 1600, 3200, 6400) and switch probability (rate) pswitch ×106 = (Ahmed, 2011 120, 150, 180) were varied, other simulation parameters were the same as for the 75 nm ring. The same event list was used for each ring, which had the same rate. The average number of events per frame for the different rates was, 0.85, 1.70, 2.55, 3.37, 4.21 and 5.08. For the lowest rate value events will hardly overlap within each frame, however for increasingly higher event densities, events will start to overlap considerably, which biases the localization towards the center of the ring, as can be seen in the profiles shown in the right hand side column.

47



2

Figure S8 RapidSTORM reconstruction of the same simulation as shown in Figure S7, but now performed after application of the temporal median filter prior to reconstruction with RapidSTORM. The temporal median filter has the effect of removing event overlap in the center of the ring, which removes the localization bias towards the center of the ring as can be seen in the profiles shown in the right hand side column.

48

2

Figure S9: Deconvolution reconstruction of the same simulation as shown in Figure S7, using the temporal median filter. The temporal median filter has the effect of removing event overlap in the center of the ring and for deconvolution and other multi-fit methods this will have a slight widening effect of the ring and for these methods a more sophisticated background estimation method may be required.

49



2

Reconstruction with 101 frame temporal median filter

Same reconstruction with background estimate as described in [Baddeley 2009]

Figure S10: This figure demonstrates the differences between the ten frame mean filter described in [Baddeley et al. 2009] and the 101 frame median filter, on a detail of the same dataset as in Fig. 1g, 2. The mean filtered result is slightly noisier, but extreme errors in localization accuracy are avoided by this temporal filter. However, the mean filter allows substantially more foreground signal to bleed into the background. Moreover, the rate of this bleeding is not uniformly scaled over the image, but depends on local factors such as event density and duration. This may lead to substantial distortions in appearance (the triplet of bumps has all but disappeared in the right frame), and complicates quantitative analysis of features based on their absolute intensity. Note that both a running mean or median, of a background signal plus a positive foreground signal, will provide an estimate of the background which is always equal to or greater than the true background, proving that the median filtered reconstruction is the more accurate of the two, by up to a factor 4. While the median filter provides a superior reconstruction, it will still underestimate the true foreground intensity. If one seeks to make quantitative comparisons based on intensity, it is critically important to make an informed choice of background estimator.

50

2

Figure S11: This figure illustrates the difference between the method described in [Hess, S. T. et al., 2007] and a 100 frame median filter, on a cutout of the dataset in Fig. 1a, 3. The additional signal registered by the method described in Hess et al. 2007 is spurious and nonspecific; it does not correspond to switching foreground events, and therefore cannot contain super resolution information. For datasets with a simple background behavior, the method in [Hess, S. T. et al., 2007] may be expected to be very accurate, but for datasets with many different physical background components such as this one (regions of accumulating background, regions of depleting background, the behavior of fiducial beads), the method will fail to deliver super-resolution results. Note that the ridge artifact has a width on a micrometer scale. Note that this method may be regarded as a truncated version of the eigenbackgrounds described in [Piccardi 2004]. Extending this method to include a sufficient number of eigencomponents depending on the dataset is likely to provide a robust background estimate, but this would be computationally challenging.

Reconstruction with 101 frame temporal median filter

Same reconstruction with background estimate as described in [Hess, S. T. et al. 2007]

51



2

53

* Division of Cell Biology I, The Netherlands Cancer Institute, Plesmanlaan 121, Amsterdam, 1066 CX, The Netherlands. ǂ Division of Molecular Genetics, The Netherlands Cancer Institute, Plesmanlaan 121,

Amsterdam, 1066 CX, The Netherlands.

† Authors contributed equally to this work

Biology open 5, 1001-1009, 2016 Jun 29.

Chapter 3

PFA fixation enables artifact-free super-resolution imaging of the actin cytoskeleton and associated proteins

Daniela Leyton-Puig *, Katarzyna M. Kedziora *†, Tadamoto Isogai ǂ†, Bram van den Broek *, Kees Jalink *, and Metello Innocenti ǂ

54

3

SUMMARY STATEMENT

We show that proper PFA fixation allows high-quality super-resolution imaging of the actin cytoskeleton and can outperform gold-standard glutaraldehyde fixation for imaging of actin-binding proteins.

ABSTRACT

Super-resolution microscopy (SRM) allows to precise localization of proteins in cellular organelles and structures, including the actin cytoskeleton. Yet, sample preparation protocols for SRM are rather anecdotal and still being optimized. Thus, SRM-based imaging of the actin cytoskeleton and associated proteins often remains challenging and poorly reproducible.Here, we show that proper paraformaldehyde (PFA)-based sample preparation preserves the architecture of the actin cytoskeleton almost as faithfully as gold-standard glutaraldehyde fixation.We show that this fixation is essential for proper immuno-based localization of actin-binding and actin-regulatory proteins involved in the formation of lamellipodia and ruffles, such as mDia1, WAVE2 and clathrin heavy chain, and provide detailed guidelines for the execution of our method.In summary, proper PFA-based sample preparation increases the multi-colour possibilities and the reproducibility of SRM of the actin cytoskeleton and its associated proteins.

http://fixation.we/

55

PFA fixation enables artifact-free super-resolution imaging of the actin

cytoskeleton and associated proteins

3

INTRODUCTION

Single-molecule-based localization microscopy (SMLM) has circumvented the resolution limit of light microscopy, which prevents resolving details smaller than about half the wavelength of light 72. Consequently, SMLM provides up to 10 times more resolution than classical optical microscopy for biological studies. Over the last ten years, several SMLM variants have been developed, the most prominent being photo-activated localization microscopy (PALM) 74,124, (direct) stochastic optical reconstruction microscopy (STORM and dSTORM) 75 76 and ground-state depletion and single-molecule return (GSDIM) 77. Most SMLM methods rely on fluorophores that can be photoactivated or photoswitched by means of light. Depending on the photoactivation or the photoswitching method, SMLM techniques can be divided in “targeted switching and readout” and “stochastic switching and readout” 72,125. dSTORM and GSDIM belong to the “stochastic switching and readout” type of SMLM techniques and provide the highest possible resolution among the existing variants. Moreover, these techniques can be enhanced to provide details in the axial direction by inclusion of optical elements that discriminate molecules based on their Z-position 126. As dSTORM and GSDIM exploit basic transitions of standard chemical dyes to induce stochastic switching 77, they are rarely used in live-cell imaging studies.SMLM is receiving ever-increasing attention from biologists interested in capturing the distribution of individual molecules within the cell 127,128. Not surprisingly, SMLM is employed to study cellular processes depending on the assembly of monomeric actin into approximately 6 nm-wide filaments, such as formation of membrane protrusions, cell migration, cytokinesis, endocytosis, vesicle trafficking and organelle homeostasis 68,88,90,129. Although total internal reflection fluorescence microscopy (TIRFM) allows imaging single actin filaments in vitro under conditions that prevent actin-filament bundling, it has insufficient resolution to zoom in to the finest details of F-actin and its interacting proteins. Within cells, diffraction-limited microscopes can neither distinguish single actin filaments from bundles nor precisely map the position of actin-binding proteins along the side and the ends of actin filaments. Electron microscopy (EM) and cryo-electron tomography can readily resolve individual actin filaments in cells, but both methods are very laborious and time-consuming. SMLM bridges the gap between high-throughput, low-resolution conventional microscopy and low-throughput, high-resolution (cryo)EM by visualizing individual actin filaments in cells with intermediate throughput. However, SMLM is still in its infancy and step-by-step guidelines are only sparsely available 128. In particular, appropriate fixation of the structure of interest remains often very challenging and, more than any other step, defines both the quality and the reliability of SMLM images. The ideal fixative for SMLM should not only preserve the cellular structures faithfully but also allow dense labelling with fast switching (in)organic fluorophores 81,127,130. For preservation of cellular structure, crosslinking fixatives are usually superior to agents that precipitate and coagulate proteins, such as methanol, ethanol and acids. Crosslinking agents also permit binding of mushroom (Amanita phalloides) toxin phalloidin to the actin cytoskeleton for very dense labelling of actin 131,132. Two crosslinking agents are commonly used: paraformaldehyde (PFA) and glutaraldehyde (GA). PFA crosslinks amino groups without changing the tertiary structure of proteins, so that most epitopes remain available for specific antibodies 133,134; GA, on the other hand, cross-links proteins more efficiently than PFA but it has also two main disadvantages: it often makes tertiary structures unrecognizable by antibodies and it penetrates into cells slowly. Thus, cell permeabilization is required either before or during GA fixation, which

56

3

frequently causes the loss of both cytosolic and cytoskeleton-associated proteins 134. Although both PFA and GA have been successfully employed to study the actin cytoskeleton through SMLM 88,90,129, two recent studies claimed that GA should be the fixative of choice for SMLM of the actin cytoskeleton because PFA did not allow the detection of thin actin bundles and structures 85,86. As the type and concentration of fixative as well as the incubation time and the permeabilization method considerably influence the final outcome of the SMLM images, these contrasting results probably reflect the poor standardization of the sample preparation procedures. More importantly, the effects of different fixative agents have not been explored in detail and protocols for the localization of actin-binding and actin-regulatory proteins in SMLM are not available. Thus, anecdotal sample preparation protocols and the lack of a systematic optimization of multi-colour SRM seriously limit the flexibility and the reproducibility of SRM. Here, we show that proper PFA-based sample preparation preserves the architecture of the actin cytoskeleton almost as faithfully as GA and facilitates the localization of various actin-binding and actin-regulatory proteins by SMLM.

57



3

RESULTS

Proper PFA fixation enables high-quality SMLM imaging of the actin cytoskeletonWe initially set out to improve a fixation protocol that employs PFA dissolved in PBS (PFA-PBS) and preserves densely packed F-actin bundles but not thin and short actin filaments 85,86. We systematically varied fixation time and temperature, and obtained the best images when the specimen was fixed for 10 minutes with all washing buffers and PFA kept at 37°C. This allowed visualization of thin actin fibers at SMLM resolution (Fig. 1A,D). Nevertheless, both thin and thick fibers composing the dense cortical actin cytoskeleton of HeLa cells appeared pointillist with this fixation method (Fig. 1A). A slightly less dramatic loss of actin fiber integrity was visible in COS-7 cells, especially in areas with a low-density cortical actin cytoskeleton (Fig. 1D). We next tested PFA dissolved in PEM buffer, a well-known cytoskeleton-protective buffer (PFA-PEM) 135. This resulted in a better and more faithful preservation of the actin cytoskeleton with more uniformly stained thin crossing fibers and bundles in both HeLa and COS-7 cells (Fig. 1B,E). Strikingly, these results were comparable to those obtained with GA dissolved in a cytoskeleton-protective buffer (Fig. 1C,F), the gold-standard fixative for high-resolution studies of the actin cytoskeleton 129,136.

Figure 1. Proper paraformaldehyde fixation preserves the architecture of the actin cytoskeleton and is compatible with high-quality SMLM. HeLa (A-C) and Cos-7 (D-F) cells fixed with paraformaldehyde (PFA) dissolved in PBS (A and D), PFA in PEM buffer (B and E) or glutaraldehyde (GA) in cytoskeleton buffer (C and F). All cells were stained with AlexaFluor-647-labelled Phalloidin and imaged in parallel as described in the Materials and Methods. Representative SMLM images (left) and close ups of the boxed regions (right) are shown. Scale bar A-F: 10 µm, scale bar close ups: 1 µm.

PFA

-PEM

fixa

tion

GA

fixat

ion

PFA

-PB

S fix

atio

n

F

E

D

C

B

A

58

3

To compare quantitatively the effects of these fixation protocols on the actin cytoskeleton, we measured small actin fibers as they are most easily lost during fixation. The thinnest detectable fibers were segmented and their brightness profiles fitted to a Gaussian distribution in which full width at half maximum (FWHM) represents the diameter of the fiber. It turned out that the smallest fibers in cells fixed with GA had a diameter of about 35 nm and were thinner than those measured in cells fixed with PFA-PBS (about 45 nm in diameter). Interestingly, the FWHM measured in cells fixed with PFA-PEM was in between the above two values, namely approximately 40 nm (Fig. 2A). Visual inspection revealed that very thin fibers were only preserved in cells fixed with GA or PFA-PEM (Fig. 2B). Notably, mean uncertainty of the localization of the individual “blinks” of each fluorophore was equivalent for all fixations, showing that imaging conditions (dye, buffer, etc.) were kept constant during all experiments (Fig. 2C). Our results thus show that PFA fixation can yield excellent results to image the actin cytoskeleton at high resolution with SMLM, provided that optimal fixation temperature, time and buffer are employed.

PFA fixation expands the possibilities of multi-colour SMLM of the actin cytoskeleton and actin-binding proteinsWe next compared PFA-PEM fixation with GA fixation for preservation of actin-associated proteins and focussed on lamellipodia and ruffles, thin veil-like and ephemeral actin-based protrusions that present challenges for SMLM. We analysed the localization of the actin-regulatory proteins mDia1 and WAVE2, which are involved in the initiation of lamellipodia and ruffles 68.

B

A FWHM 40 nm FWHM 37 nmFWHM 47 nm

C

FWHM 29 nm FWHM 28 nmFWHM 45 nm

Loca

lizar

ion

prec

isio

n (n

m)

PFA-PEM fixation GA fixationPFA-PBS fixation

PFA-PEM fixation GA fixationPFA-PBS fixation

0

5

1 0

1 5

PFA-PEM fixation

GA fixation

PFA-PBS fixation

1.00 -

● ●

●●

●●

●

●

●

●

●

●

●

●● ● ●

● ● ●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0.00

0.25

0.50

0.75

0.0 0.1 0.2 0.3distance

Inte

nsity

Inte

nsity

0.75 -

0.50 -

0.25 -

0.00 -

distance0.00 - 0.1

-

0.2 -

0.3 -

● ●

●●

● ●●

●

●

●

●

●●

●

● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0.00

0.25

0.50

0.75

1.00

0.0 0.1 0.2 0.3distance

Inte

nsity

Inte

nsity

0.75 -

0.50 -

0.25 -

0.00 -

distance0.00 - 0.1

-

0.2 -

0.3 -

● ● ●

●● ●

●●

●

●

●

●

●●

● ●●

●● ●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●

●

●

●

●●●●●●●●●

●●●●

●

●

●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0.00

0.25

0.50

0.75

0.0 0.1 0.2 0.3distance

Inte

nsity

Inte

nsity

0.75 -

0.50 -

0.25 -

0.00 -

distance0.00 - 0.1

-

0.2 -

0.3 -

●● ● ● ● ●

●●

●

●

●

●

●

●●

● ●

●

● ●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0.0

0.2

0.4

0.6

0.0 0.1 0.2 0.3distance

Inte

nsity

Inte

nsity

0.60 -

0.40 -

0.20 -

0.00 -

distance0.00 - 0.1

-

0.2 -

0.3 -

● ● ● ● ●● ●

●

●

●

●

● ●●

●

●● ●

●

●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●

●

●

●

●

●

●●●●●●●●

●

●

●

●

●

●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0.0

0.2

0.4

0.6

0.8

0.0 0.1 0.2 0.3distance

Inte

nsity

Inte

nsity

0.60-

0.40 -

0.20 -

0.00 -

0.80 -

distance0.00 - 0.1

-

0.2 -

0.3 -

●●

● ●

●

●

●

●

●

●

●

●

● ●●

●● ●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●

●

●

●

●

●

●●●●●●●●

●

●

●

●

●

●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0.00

0.25

0.50

0.75

1.00

0.0 0.1 0.2 0.3distance

Inte

nsity

Inte

nsity

0.75 -

0.50 -

0.25 -

0.00 -

1.00 -

distance0.00 - 0.1

-

0.2 -

0.3 -

Figure 2. Proper PFA fixation preserves small actin fibers as faithfully as GA fixation. (A) PFA-PEM fixation preserves small actin fibers as faithfully as GA fixation. Individual actin fibers (5 to 10 fibers per condition) were manually segmented and their brightness profiles were plotted and fitted to a Gaussian distribution as described in Materials and Methods. Representative fibers, distribution profiles with Gaussian fit and FWHM are shown. Scale bar: 300 nm. (B) Only PFA-PEM and GA fixation preserve very thin actin fibers. Distribution profiles with Gaussian fit and FWHM of the thinnest detected fibers are shown for each fixation method. Scale bar: 300 nm. (C) Localization precision is not affected by the fixation method. Localization precision of images of actin obtained after different fixations was calculated using the Thunderstorm plugin 137 of Fiji 138,139. Bar graph shows localization precision and intensity (mean ± S.D) as obtained from 5 independent images (PFA-PBS fixation), 7 independent images (GA fixation) and 9 independent images (PFA-PEM fixation) (one-way ANOVA was employed to compare the obtained results and no significant differences were found).

59



3

Dual-colour SMLM imaging revealed that both GA fixation and PFA-PEM fixation preserved the characteristic actin mesh of lamellipodia and ruffles (Fig. 3A,B). However, density of mDia1 and WAVE2 particles was much higher in the cells fixed with PFA-PEM than in those fixed with GA (Fig. 3A,B). Strikingly, mDia1 signal in GA-fixed cells appeared almost identical to that of cells stained with secondary antibody alone, whereas that of WAVE2 was slightly higher (Fig. 3A). Conversely, PFA-PEM fixation resulted in WAVE2 and mDia1 labelling being much denser than the secondary antibody (Fig. 3B). Importantly, SMLM images from mDia1 knockdown and Nap1 knockdown HeLa cells ruled out that these localizations resulted from the unspecific binding of the primary antibodies (Fig. S1).

anti mouse IgG

mDia1

WAVE2

anti mouse IgG

mDia1

WAVE2

F-actin

F-actin

F-actin

F-actin

F-actin

F-actin

A

B

merge

merge

merge

merge

merge

merge

PFA

-PEM

fixa

tion

GA

fixat

ion

Figure 3. Paraformaldehyde fixation, but not glutaraldehyde fixation, enables detection of actin-binding proteins within lamellipodia and ruffles. HeLa cells were stimulated with epidermal growth factor (EGF) (100 ng/ml) for 3 minutes and fixed with either glutaraldehyde (GA) in cytoskeleton buffer (A) or paraformaldehyde (PFA) in PEM buffer (B) and then stained with either anti-mDIa1 (mDia1) or anti-WAVE2 (WAVE2) antibodies or mock stained. All samples were incubated with AlexaFluor-647-labelled Phalloidin and secondary goat anti-mouse antibodies (anti mouse IgG) labelled with AlexaFluor-532. Membrane ruffles were imaged in EPI mode. Representative SMLM images show the actin cytoskeleton in red and the actin-binding proteins mDia1 and WAVE2 in green. Dashed white areas depict the region of interest used for the quantification of green particles. Scale bar: 1 µm.

60

3

These observations were corroborated by quantitative image analysis of the number of particles per area (Table 1). For mDia1, we found that most of the particles imaged after fixation with GA derived from unspecific binding of the secondary antibody, whereas PFA-PEM fixation resulted in a 8-fold increase over the control (Table 1). This and the fact that only few mDia1 particles could be detected in the mDia1 knockdown cells stained with anti-mDia1 antibodies (Fig. S1) jointly indicate that PFA-PEM, but not GA, fixation enables visualizing mDia1 in SMLM. We found no specific mDia1 particles using a second anti-mDia1 antibody that recognizes a different and distant epitope (Fig. S3), thus it appears that mDia1 is washed out on GA fixation. For WAVE2, the number of localized particles per area was higher in the cells fixed with PFA than those fixed with GA (Table 1). Thus, GA fixation did permit specific, yet suboptimal, detection of WAVE2.

Leyton-Puig et al., Table 1Table 1. Quantification of the number of mDia1, WAVE2 and secondary antibody particles demonstrates the adverse effect of GA fixation on these proteins.Ruffles demarked by white dash lines in the F-actin channel were segmented and the number of mDia1 (mDia1), WAVE2 (WAVE2) and anti mouse IgG (Negative) particles present in the green channel was quantified using the particle analyser routine of Fiji 139 as described in Materials and Methods. For GA fixation, we analysed 11, 6 and 12 independent images for mDia1, WAVE2 and anti mouse IgG, respectively. For PFA-PEM fixation, we analysed 10, 14 and 4 independent images for mDia1, WAVE2 and anti mouse IgG, respectively. The number of particles was normalized to the area covered by F-actin and expressed as mean ± S.D. t-test was calculated with GraphPad Prism (p-value area WAVE 0.0128, p-value mDia1 <0.0001, p-value negative control 0.6321)

Figure 4. Paraformaldehyde and glutaraldehyde fixation allow comparable detection of focal adhesion proteins. HeLa cells were fixed with either glutaraldehyde (GA) in cytoskeleton buffer (A) or paraformaldehyde (PFA) in PEM buffer (B). Cells were stained for Vinculin or Paxillin followed by AlexaFluor-532-labelled secondary antibodies along with AlexaFluor-647-labelled Phalloidin. Basal membranes were imaged in TIRF mode as described in the Materials and Methods. Representative SMLM images depict F-actin in red and focal adhesion proteins in green. Dashed white boxes in the Paxillin and the Vinculin images mark the position of the close ups shown on the right. Scale bar: 1 µm.

F-actin Paxillin

Paxillin

A

B

Vinculin

Vinculin

merge

merge

merge

merge

F-actin

F-actin

F-actin

PFA

-PEM

fixa

tion

GA

fixat

ion

61



3

The different behaviour of WAVE2 and mDia1 likely reflects different binding kinetics and/or affinities for F-actin and/or other interacting proteins localizing within lamellipodia and ruffles. Consistent with this notion, focal adhesion-associated proteins Paxillin and Vinculin, which robustly bind to focal adhesion complexes 140, showed a similar localization pattern in both GA-fixed and PFA-PEM-fixed cells (Fig. 4). These results stress the importance of secondary antibody control experiments to establish labelling specificity. Furthermore, they show that our PFA fixation protocol represents a safe, robust and ready-made alternative for immuno-based multi-colour SMLM.

PFA fixation increases epitope preservation for high-quality SMLMA

B

C GA fixation PFA-PEM fixation

Pits/Vesicles

Pearson -0.68R 0.47

Plaques Plaques

Circ

ular

ity

Circ

ular

ity

PFA

-PEM

fixa

tion

GA

fixat

ion

0

2

4

6

**

GA fixation

PFA-PEM fixation

Are

a co

vere

d by

pla

ques

(%)

0 .0

0 .2

0 .4

0 .6

0 .8

Num

ber p

its /

cell

GA fixation

PFA-PEM fixation

**

D E

Pits/Vesicles

Pearson -0.76R 0.572 2

6.00

4.00

2.00

0.00

0.8

0.6

0.2

0.0

0.4

1.00

0.75

0.50

0.25

1.00

0.75

0.50

0.25

30002000100003000200010000Area (μm )2 Area (μm )2

Figure 5. Paraformaldehyde fixation, but not glutaraldehyde fixation, enables faithful detection of clathrin-coated structures. HeLa cells were serum starved overnight and then stimulated with EGF (100 ng/ml) for 5 minutes and fixed with PFA in PEM buffer (A) or glutaraldehyde (GA) in cytoskeleton buffer (B). Cells were stained with anti-clathrin heavy chain (CHC) antibodies and goat AlexaFluor-647-labelled secondary antibodies. Clathrin-coated structures (CCSs) at the basal plasma membrane were imaged in TIRF mode as described in the Materials and Methods. Representative TIRF images (left), SMLM images (middle) and SMLM close ups (right) corresponding to the dashed white areas are shown. Scale bar: 1 µm. Representative TIRF images (left), SMLM images (middle) and SMLM close ups (right) corresponding to the dashed white areas are shown. Scale bar: 1 µm. (C) GA fixation perturbs both circular and elongated CCSs. Size of circular and elongated CCSs was obtained by manual segmentation as explained in the Materials and Methods. Scatter plots depict each CCS found in four independent images as a colour-coded circle. Note that GA causes a dramatic reduction in the number of CCSs and that the few remaining ones have a smaller size compared to that of the PFA-fixed samples. Pearson coefficients and R2 of the correlation between Area and Circularity were obtained using GraphPad Prism (P-value GA fixation: <0.0001; P-value PFA-PEM fixation: <0.0001). (D and E) Area covered by plaques (D) and number of pits (E) are highly reduced in cells fixed with GA. The area and number of particles are expressed as mean ± S.E.M. t-test was calculated with GraphPad Prism (D, P-value area covered plaques 0.0025, E, P-value number pits 0.0065, n=3)

62

3

Clathrin-mediated endocytosis (CME) is a form of actin-based micro-pinocytosis 59,61,70 in which clathrin heavy chain (CHC) assembles around invaginations of the plasma membrane and guides the formation of diffraction-limited vesicles of 90-150 nm 34. Moreover, CHC has recently been shown to promote the formation of lamellipodia and ruffles through membrane recruitment of WAVE regulatory complex, independently of its role in vesicle trafficking and clathrin light chain 141. Using PFA fixation under optimal conditions, we observed two types of CCSs in cells by SRM (Fig. 5A): small and round CCPs and CCVs of about 150 nm in size and bigger clathrin patches of more heterogeneous shapes and sizes that likely represent the flat clathrin plaques previously described in HeLa and many other cell lines 142. Conversely, CCPs and CCVs could be hardly located in cells fixed with GA and the few remaining ones appeared very small and non-circular (Fig. 5B). Moreover, the bigger clathrin patches were either absent or lacked clearly defined boundaries (Fig. 5B). Manual and automated morphometric particle analyses of these images showed that GA causes a complete disorganization of the clathrin-coated structures (Fig. 5C and Fig. S2 respectively), with loss of both clathrin coated plaques (Fig. 5D) and pits (Fig. 5E). Surprisingly, only at very high resolution of SMLM were the detrimental effects of GA on CCSs visible as the corresponding TIRFM images appeared very similar (Fig. 5A,B). As it has been reported that GA destroys the CHC epitope recognized by the anti-CHC X-22 antibody 143, the above results suggest that PFA fixation should be preferred to GA fixation for all SMLM applications sensitive to epitope preservation.

63



3

DISCUSSION

During the last few years, several research articles and reviews have stressed that fluorophore switching and localization algorithms are potential sources of SRM image artefacts 78,127,144. Even though sample fixation and preparation are as important as the above factors for image quality and resolution, GA-based EM sample preparation protocols are typically employed in SMLM 129. As sample preparation protocols for SRM remain anecdotal and have not been optimized, SRM is often challenging and suffers from reproducibility problems. Here, we show that GA-based sample preparation protocols can negatively impact on SMLM imaging and may be severely flawed for immuno-based detection of certain proteins (Fig. 3 and Fig. 5). Instead, we present an optimal PFA fixation protocol that provides a simple, flexible and robust way to obtain high-quality super-resolution images of the actin cytoskeleton and associated proteins. This protocol enables proper immuno-based protein localization in two- (or three)-colour high-resolution SMLM images and quantitative morphometric and localization analyses. From the biological point of view, optimal PFA fixation has been instrumental in detecting mDia1 and WAVE2 within lamellipodia and ruffles, thereby providing crucial insight into the mechanism regulating the initiation of these actin-based protrusions. We also anticipate that optimized PFA fixation and SRM will shed new light on the role of CHC in the formation of lamellipodia and ruffles.In summary, washout and epitope masking of actin-binding and actin-associated proteins during GA fixation are unrecognized fundamental technical pitfalls of SMLM. As optimized PFA-based sample preparation represents a substantial technical advance that increases the possibilities and the reproducibility of multi-colour SRM, our protocol and an accompanying troubleshooting table will be deposited in a freely accessible public repository (Protocol Exchange DOI: 10.1038/protex.2016.042).

64

3

MATERIALS AND METHODS

ReagentsAlbumin bovine Fraction V. pH 7.0 was from Bio-Connect (Huissen, NL), 2-Butanol, Glutardialdehyde solution 25%, Magnesium chloride hexahydrate, Paraformaldehyde (PFA), Sodium hydroxide pellets, Triton X-100 were from Merck (Amsterdam. NL), Calcium chloride dehydrate, Catalase from Aspergillus niger, Dulbecco’s modified Eagle’s medium, EGTA, Gelatin from porcine skin (300g Bloom), Glucose oxidase Ethanol, Hydrochloric acid, MES hydrate, Phosphate buffered saline (PBS) tablets, Petroleum ether, Sodium borohydride, Cysteamine hydrochloride – MEA, D-(+)-Glucose anhydrous were from Sigma-Aldrich (Zwijndrecht, NL). Murine Epidermal growth factor (EGF), PIPES was from Thermo-Fischer Scientific (Breda, NL). Fetal bovine serum (FBS) was from APS (Bedford, UK). Precision tissue wipes were from Kimberly-Clark BV (Ede, NL).

AntibodiesPhalloidin-AlexaFluor 647, mouse anti-Clathrin heavy chain antibody (X22) was from Thermo-Fischer Scientific (Breda, NL), mouse anti-mDia1 and mouse anti-paxillin were from BD Biosciences (Breda, NL), mouse anti-mDia1 (D3) was from Santa Cruz (Heidelbarg, Germany), mouse anti-tubulin was from Sigma Aldrich (Zwijndrecht, NL), rabbit anti-vimentin was from GeneTex (Irvine, CA, USA), mouse anti-WAVE2 antibody was generated in house 68.

Inmunofluorescence Previous to all steps reagents were filter sterilized. 24 mm #1.5 coverslips were coated with 0.5% gelatin solution in a humidified incubator at 37⁰C for 30 minutes. Cells were plated on gelatin-coated coverslips and grown in DMEM supplemented with 10% FCS. The next day, they were either serum starved overnight in DMEM supplemented with 0.1% FCS or kept left untreated. For Fig. 2, cells were stimulated for 3 minutes with 100 ng/ml EGF. For Fig. 3, cells were stimulated for 5 minutes with 100 ng/ml EGF.For GA fixation cells were briefly rinsed with pre-warmed PBS supplemented with Magnesium and Calcium and incubated for 2 minutes in 0.3% GA and 0.25% Triton X-100 in cytoskeleton-preserving buffer (10 mM MES pH 6.1, 150 mM NaCl, 5 mM EGTA, 5 mM Glucose, 5 mM MgCl2) followed by 8 minutes in 0.5% GA in cytoskeleton-preserving buffer. Cells were then rinsed with and incubated for 7 minutes in freshly prepared 0.1% NaBH4 in PBS 129. For PFA fixation, a 20% (w/v) stock solution was prepared from powder as follows: PFA was dissolved in warm ddH2O under constant stirring keeping the temperature below 55-57 ⁰C. NaOH (10N) drops were added to help PFA dissolve. pH was adjusted to 7.2 by adding HCl (37%). The stock solution was filter-sterilized and kept at -20⁰C for up to a year 145. Cells were rinsed briefly with pre-warmed PBS containing calcium and magnesium and incubated for 10 minutes in freshly prepared pre-warmed 4% solution of PFA dissolved in PBS or cytoskeleton-preserving buffer (PEM) (80 mM PIPES pH6.8, 5 mM EGTA, 2 mM MgCl2) (optimal PFA fixation). Cells were rinsed twice with PBS and incubated for 10 minutes in 0.5% Triton in PBS.In all cases, cells where incubated in 5% BSA for at least one hour at 37⁰C or overnight at 4⁰C. Primary antibodies were diluted in 5% BSA and samples were incubated for 45 minutes at room temperature. Secondary antibodies and Phalloidin were incubated for 30 minutes. Phalloidin was used coupled with AlexaFluor-647 at a final concentration of 0.6 U of Phalloidin (Molecular probes for Life Technologies, Sigma Aldrich, Zwijndrecht, NL). Goat anti-Mouse and goat anti-Rabbit secondary antibodies coupled AlexaFluor-532 (Molecular

65



3

Probes for Life Technologies) were used a final concentration of 10 µg/ml.

Super resolution imaging and localization analysisSpecimens were imaged with a Leica SR-GSD 3D microscope using an oxygen scavenging system (GLOX: 10% Glucose + 0.5 mg ml-1 Glucose Oxidase + 40 µg ml-1 Catalase) supplemented with a reducing agent (Cysteamine hydrochloride - MEA) at 100 mM. Images were taken in TIRF or EPI mode at 10 ms or 28 ms exposure time for 10000-15000 frames. Structured background subtraction with a temporal median filter 144 was performed on the blinking movies using home-built software and the resulting movies were analyzed with Thunderstorm 137, an open source plugin for imageJ 146 (ImageJ; http://imagej.nih.gov/ij/), to locate the centre of blinking molecules, correct the XY drift and construct the final image. Images were rendered with 20 nm pixel size, with the Normalized Gaussian visualization option. For figure visualization purposes, images were convolved with a mean filter of 3x3 pixels. Images of filaments used for brightness profile plotting (Fig. 2) were rendered with 20 nm pixel size with the Histogram visualization option. Chromatic aberration (CA) was corrected using images obtained from 0.1 μm diameter Tetraspec microspheres (Invitrogen) embedded in a matrix. An affine transformation matrix was constructed from those data and affine correction of all images was carried out with an ImageJ macro, using the plugin Image Stabilizer 147.

Quantification of the FWHM of small actin fibersImages were rendered with 20 nm pixel size, with the Histogram visualization option. Small actin fibers were selected manually. Brightness profiles were acquired with an imageJ macro as follows: lines with thickness 20 pixels (straight or curved) were drawn manually on the fibers and the underlying pixels were excised from the images. Obtained Images from curved lines were straightened. Profiles of these straight filaments were fit with a 1D Gaussian function with offset. The FWHM was defined as 2.35 times the sigma (width) of the Gaussian fit.

Quantification of the distribution of particles of WAVE2, mDia1 and anti-mouse IgGRuffles were segmented using the Phalloidin images. WAVE2-, mDia1- or secondary antibody-positive particles were quantified with ImageJ 146 (ImageJ; http://imagej.nih.gov/ij/) using the analyze particles routine and selecting the particles bigger than 2 pixels with at least 2 localizations per pixel. The relative abundance of these particles was calculated by dividing the obtained number by the total area of the ruffle (particles/μm2).

Quantification of the distribution of pits and plaquesMorphometric analysis of clathrin-coated structures was performed by manual image segmentation. Selected particles were characterized with ImageJ 146 (ImageJ; http://imagej.nih.gov/ij/) using the analyze particles routine. For the automated analysis of all particles, particles bigger than 4 pixels were characterized for area and circularity with ImageJ 146 (ImageJ; http://imagej.nih.gov/ij/) using the analyze particles routine.

Generation of stable knockdown cellsStable knockdown cells were described and published before 68,148.

Statistical analysesGraphPad Prism (version 6.oh) was used to carry out all statistical analyses.

http://imagej.nih.gov/ij/

http://imagej.nih.gov/

http://imagej.nih.gov/ij/

http://6.oh/

66

3

COMPETING FINANCIAL INTERESTS

The authors declare that they have no competing financial interests.

AUTHOR CONTRIBUTIONS

M.I. and D.L.P designed the experiments and wrote the manuscript. D.L.P. and T.I. performed the experiments. D.L.P. performed data analysis and quantification. K.M.K. and B.B. quantified results. K.J. supervised the data analysis and edited the manuscript. M.I. conceived and coordinated the study.

FUNDING

This work was supported by STW Technology Foundation, grant number 12150 for Kees Jalink.

DATA AVAILABILITY

Protocol exchange DOI: 10.1038/protex.2016.042

67



3


Supplementary Figure 1. mDia1 and WAVE-complex knockdown cells prove the specificity of the anti-mDia1 and the anti-WAVE2 primary antibodies. (A) Control knockdown (control KD) and Nap1 knockdown (Nap1 KD) HeLa cells were stimulated with EGF (100 ng/ml) for 7 minutes and fixed with paraformaldehyde in stabilization buffer and stained with AlexaFluor-647 labelled Phalloidin, anti-WAVE2 antibodies and secondary goat anti mouse antibodies labelled with AlexaFluor-532. Membrane ruffles were imaged in EPI mode. Representative SMLM images show the actin cytoskeleton in red and WAVE2 in green. (B) Control knockdown (control KD) and mDia1 knockdown (mDia1 KD) HeLa cells were with EGF (100 ng/ml) for 7 minutes and fixed with paraformaldehyde in stabilization buffer and stained with AlexaFluor-647 labelled Phalloidin, anti-mDIa1 (mDia1) antibodies and secondary goat anti mouse antibodies labelled with AlexaFluor-532. Membrane ruffles were imaged in EPI mode. Representative SMLM images show the actin cytoskeleton in red and mDia1 in green.

WAVE2

WAVE2

mDia1

F-actin

F-actin

mDia1

F-actin

F-actin

cont

rol K

DA

B

Nap

1 K

Dco

ntro

l KD

mD

ia1

KD

merge

merge

merge

merge

68

3

Supplementary Figure 2. Two different antibodies fail to detect mDia1 molecules within lamellipodia and ruffles after glutaraldehyde fixation. HeLa cells were stimulated with Epidermal Growth Factor (EGF) (100 ng/ml) for 3 minutes and fixed with glutaraldehyde (GA) in cytoskeleton buffer and then stained with either anti-mDIa1 D3 antibody from SC (mDia1 D3) or anti-mDIa1 from BD (mDia1) antibodies (see Materials and Methods) or mock stained. All samples were incubated with AlexaFluor-647-labelled Phalloidin and secondary goat anti-mouse antibodies (anti mouse IgG) labelled with AlexaFluor-532. Membrane ruffles were imaged in EPI mode. Representative SMLM images show the actin cytoskeleton in red and the actin-binding protein mDia1 in green. Scale bar: 1 µm.

anti mouse IgG

mDia1

WAVE2

mDia1 D3F-actinmerge

mDia1 D3F-actinmerge

mDia1F-actinmerge

mDia1F-actinmerge

anti mouse IgGF-actinmerge

F-actinmerge anti mouse IgG

69



3

Supplementary Figure 3. Automated morphometric analysis of CCSs shows that only optimized paraformaldehyde fixation enables faithful detection of clathrin-coated structures. (A) The images analysed in Figure 4C were subjected to automated morphometric analysis using the particle analyzer routine of Fiji 139 as described in the Materials and Methods. Note the enrichment of small particles having circular and non-circular shapes in the GA fixed samples and the more heterogeneous distribution corresponding to big non-circular CHC assemblies and small circular CCPs and CCVs in the PFA fixed samples. (B) Particle area (C) Number of particles per cell. Pearson coefficients were obtained using Graph Pad Prism. p-value GA fixation: <0.0001, p-value PFA-PEM fixation: <0.0001.

Circ

ular

ityA GA fixation PFA-PEM fixation

0

2 0

4 0

6 0

GAfixation

PFA-PEM fixation

****

Num

ber s

truct

ures

/ ce

ll

GAfixation

PFA-PEM fixation

*

0

5 00

1 00 0

1 50 0B C

Pearson -0.67R 0.45

Pearson -0.71R 0.50

71

Chapter 4

Flat clathrin lattices are dynamic actin-controlled hubs for clathrin-mediated endocytosis and signalling of specific receptors

Daniela Leyton-Puig *†, Tadamoto Isogai ǂ†, Bram van den Broek *, Jeffrey Klarenbeek *, Hans Janssen §, Kees Jalink ¶* and Metello Innocenti ¶ǂ

* Division of Cell Biology I, The Netherlands Cancer Institute, Plesmanlaan 121, Amsterdam, 1066 CX, The Netherlands. ǂ Division of Molecular Genetics, The Netherlands Cancer Institute, Plesmanlaan 121, Amsterdam, 1066 CX, The Netherlands. § Division of Cell Biology II, The Netherlands Cancer Institute, Plesmanlaan 121, Amsterdam, 1066 CX, The Netherlands.

† Authors contributed equally to this work

Manuscript submitted

72

4

ABSTRACT

Clathrin lattices at the plasma membrane coat both invaginated and flat regions forming clathrin-coated pits and clathrin plaques, respectively. The function and regulation of clathrin-coated pits in endocytosis are well understood but clathrin plaques remain enigmatic nanodomains. Here we use super-resolution microscopy to show that clathrin plaques contain the machinery for clathrin-mediated endocytosis and associate with both clathrin-coated pits and filamentous actin. We also find that actin polymerization promoted by N-WASP through the Arp2/3 complex is crucial for the regulation of plaques but not pits. Flat clathrin plaques appear dynamic structures that undergo actin- and N-WASP-dependent disassembly upon activation of LPA receptor 1, but not EGF receptor. Most importantly, plaque disassembly promotes the endocytosis of LPA receptor 1 and down-modulates AKT activity. Thus, clathrin plaques serve as dynamic actin-controlled hubs for clathrin-mediated endocytosis and signalling that exhibit receptor specificity.

73

Flat clathrin lattices are dynam

ic actin-controlled hubs for clathrin-m

ediated endocytosis and signalling of specific receptors

4

INTRODUCTIONCells internalize membrane proteins, solutes and lipids through the formation of clathrin-coated vesicles (CCVs), a process referred to as Clathrin-mediated endocytosis (CME). CME entails five stages: initiation, cargo recruitment, clathrin coat assembly, scission of a clathrin-coated vesicle and subsequent uncoating thereof 34. In a widely accepted model, binding of clathrin adaptor proteins to the plasma membrane recruits clathrin triskelia, thereby promoting the self-assembly of a clathrin coat that marks de novo endocytic sites. At these sites, clathrin-coated pits (CCPs) mature, recruit cargoes and ultimately pinch off with the help of the GTPase dynamin to form small and roughly spherical CCVs of up to 200 nm in diameter 34,149.Pharmacological studies have suggested that actin polymerization optimizes CME of EGFR and some G-protein coupled receptors (GPCR) 12,150. However, actin has a non-obligatory and cell-type specific roles in CME of Transferrin Receptor 50,62. Knockdown studies showed that N-WASP and the Arp2/3 complex mediate the assembly of F-actin on CCPs and CCVs 61,70. Consistently, live-cell experiments demonstrated that actin appears in CCPs just before scission and only after N-WASP and the Arp2/3 complex 46,51,62,151,152 Thus, actin polymerization likely provides mechanical force for CCP remodeling and scission 46,47,153. In addition, recent data indicate that membrane tension may determine whether or not CME depends on actin 57. As CME is the main route for membrane protein internalization 154, it is not surprising that it affects signaling of receptor tyrosine kinases (RTKs) and G-protein coupled receptors 34. By removing activated receptors from the cell surface, CME can either attenuate or elicit the activity of specific downstream signaling pathways 1. Electron microscopy (EM) and total internal reflection fluorescence (TIRF) microscopy showed that, in addition to the curved CCPs and CCV, a second type of clathrin structures exists on the membrane of cells, namely large clathrin structures that are often referred to as flat clathrin plaques (FCPs) 44,46,135. The characteristic geometry and curvature of CCPs and FCPs arises from a different assemblage of clathrin triskelia 155. In fact, a combination of pentagons and hexagons determines the basket-like shape and curvature of the coat surrounding CCVs, whereas hexagonal only honey-comb-like structures give rise to FCPs 155.The function of FCPs is much debated: some studies concluded that FCPs are endocytically inactive, long-lived structures 48,49, whereas other studies found that FCPs can actively mediate internalization 35,44-46 or serve as focal sites of CCV formation 47. At any rate, CCVs are often found to surround the borders of FCPs in EM images 44,45,50,53. Light microscopy of clathrin tagged with e.g. Green Fluorescent Protein (GFP) has been instrumental to illuminate the spatiotemporal mechanics of CME 63,156-158. This approach has shown that convex clathrin coated pits (CCPs) and flat clathrin plaques (FCPs) exhibit distinctive persistence and brightness on the plasma membrane 35,48,153. However, the diffraction-limited resolution of the light microscope has hampered more detailed morphometric analyses and makes the discrimination between CCPs and FCPs challenging because of their small size. We exploited super-resolution (SR) microscopy to study the function and the regulation of FCPs. Here, we report that FCPs are dynamic structures associating with both actin filaments and the endocytic machinery, and that CCVs form from FCPs. By depleting N-WASP and the Arp2/3 complex, and using dominant negative mutants of N-WASP, we show that actin polymerization controls FCP dynamics. Finally, we demonstrate that FCPs are hubs for clathrin-mediated endocytosis and signalling of the LPA1 receptor (LPAR1). In summary, these data shed light on the enigmatic function of FCPs and unveil an actin-based mechanism regulating the lifecycle of these clathrin-coated structures (CCSs).

74

4

i

ii

i

ii

b c

0.00

0.50

1.00

0 500 1000 1500 2000

Circ

ular

ity

Area (µm )2

PlaquesPits

a

d

e

MergeCLC F-actinTIRF Merge

F-actin MergeCHCTIRF Merge

i

ii

N-WASP-GFP MergeCHCTIRF Merge

SRTIRFTIRF / SR

Figure 1. SR microscopy enables in-depth characterization of clathrin-coated structures. (a) Representative correlative TIRF-SR image of the CCSs on the basal membrane of HeLa cells. HeLa cells were fixed and stained with anti-clathrin heavy chain (CHC) antibodies as indicated in the Methods. Overlay of the TIRF (grey) and the SR (orange) images and zoomed-in regions (i, ii) are depicted. Scale bar, 1 µm. (b) Gallery of SR images showing the diversity of the CCSs (pits, plaques and combination thereof) found on the basal membrane of HeLa cells. Scale bar, 1 µm. (c) Morphometric analysis of CCSs on the basal membrane of HeLa cells. Circularity (1 = high circularity, 0 = high asymmetry) and surface area (µm2) of individual CCS were obtained as described in the Methods. (d) Actin associates with CCSs. Representative two-color TIRF and SR images of F-actin (red in merge) with either CHC or CLC tagged with mTurquoise2 (CLC-mTQ2) (green in merge). HeLa cells were fixed and stained with anti-CHC or anti-GFP antibodies and phalloidin to detect CCSs and F-actin, respectively. CLC-mTQ2 was transfected as described in the Methods. Scale bar, 1 µm. (e) N-WASP associates with CCSs. Representative two-color TIRF and SR images of CHC and N-WASP tagged with GFP (N-WASP-GFP). HeLa cells were transfected, processed and imaged as illustrated above. Scale bar, 1 µm.

RESULTS

SR microscopy enables in-depth characterization of clathrin-coated structuresWe used correlative TIRF and GSDIM SR microscopy 77 to characterize CCSs on the plasma membrane of cells. Given that HeLa cells assemble both pits and plaques 47,48, we imaged endogenous clathrin heavy chain (CHC) at the basal membrane of these cells. Small round structures (~150 nm diameter, hereafter referred to as CCPs) in which the lumen could often distinctly be resolved, and larger heterogeneous structures (hereafter referred to as FCPs) were visible in the SR images (Fig. 1a). Notably, CCPs that are juxtaposed to or overlapping with FCPs would be easily misclassified as FCPs using conventional light microscopy (Fig. 1a, b). In contrast, the higher resolving power of SR allowed reliable segmentation and morphometric analyses of both categories. Small and circular CCPs measured 120 +/- 25 nm in diameter (0.05 +/- 0.01 µm2) whereas FCPs were substantially larger and highly heterogeneous in shape (Fig. 1c), in line with earlier reports 48,142,159. Furthermore, CHC and GFP-tagged clathrin light chain (CLC) colocalized in dual-color SR images (Supplementary Fig. 1b), and both CCPs and FCPs could be detected in our HeLa cells also by EM (not shown). Thus, this SR-based approach appears well suited to study CCSs.

75




4

To obtain a molecular characterization of CCPs and FCPs, we assessed the presence of key components of the machinery regulating CME. SR clearly showed that CLC, adaptor protein complex 2 (AP2) and Dynamin II localized both in CCPs and FCPs (Supplementary Fig 1a-c). Two-color SR images of HeLa cells showed a clear enrichment of F-actin around both CHC-positive and CLC-positive structures (Fig. 1d). This is consistent with EM studies in which actin was found to associate with both CCPs and FCPs 143. Interestingly, N-WASP, which activates the Arp2/3 complex and mediates actin-dependence of CME 61,70, was present in both CCPs and FCPs (Fig. 1e). In addition, also N-WASP-regulator and actin-binding protein Cortactin 71 was enriched in both CCPs and FCPs (Supplementary Fig. 1d).In summary, SR shows that clathrin plaques are often juxtaposed to CCVs and contain both F-actin and the machinery mediating CCV formation. Thus, FCPs may be sites of actin-dependent formation of CCVs.

N-WASP is a key regulator of FCPs.To explore this idea, we imaged the CCSs in control knockdown (KD) and N-WASP KD cells (Fig. 2a). Consistent with our previous studies 61, the silencing of N-WASP resulted not only in CCSs devoid of F-actin (Fig. 2b) but also in a significant broadening of the peripheral CCS-rich ring at the basal membrane in HeLa cells, as visualized by TIRF imaging (Fig. 2c). Most importantly, SR images of the CCS-rich outer ring of the N-WASP KD cells showed that it was composed predominantly of FCPs (Fig. 2c, Supplementary Fig. 1e-f). Morphometric analyses revealed that plaques covered a much larger area of the N-WASP KD cells as compared to the control cells (Fig. 2c). In contrast, the knockdown of N-WASP had no effect on the number of CCPs (Fig. 2c). Rescue of the N-WASP KD cells with shRNA-resistant N-WASP reduced the area covered by plaques without perturbing the CCPs (Fig. 2d). BSC-1 cells reportedly do not form FCP at their membrane and instead only contain CCPs 149,160,161. Strikingly, upon silencing of N-WASP, BSC-1 cells presented large FCPs (Supplementary Fig. 2), the size and shape of which were comparable with those found in HeLa cells. These results confirm that the involvement of N-WASP in FCP regulation is common to different cell types.N-WASP mutants that lack the VCA domain do not bind Arp2/3 complex and consequently fail to induce actin polymerization. Overexpressed GFP-tagged N-WASP ΔVCA mutant strongly localized to FCPs and caused FCPs to cover a much larger area of the membrane than in control GFP-transfected cells (Fig. 3b-d). This is in line with the phenotype of the N-WASP KD cells and suggests the involvement of actin polymerization and the Arp2/3 complex in FCP regulation. Expression of the ΔWH1 and ΔPRD mutants exerted similar effects, even though they did not localize to FCPs (Fig. 3d). Hence, it is likely that these mutants deplete factors that are required for endogenous N-WASP to function at FCPs. Interestingly, overexpression of full-length N-WASP caused a significant reduction of the area covered by FCPs, whereas a functionally dead point mutant (H208D) that cannot bind Cdc42 70 did not localize or perturb FCPs (Fig. 3b-d). The observation that neither N-WASP nor its mutants affected the number of CCPs (Fig. 3e) and N-WASP’s localization to CCSs (Fig. 3f) jointly unmask a seemingly more significant role for N-WASP in the regulation of FCPs than in that of CCPs. Anyway, our results suggest that N-WASP exploits Arp2/3-complex-dependent actin polymerization to control FCPs.To strengthen this notion, we knocked down the core subunit Arpc2 162 to cause downregulation of the entire Arp2/3 complex in HeLa cells (Fig. 3g). Arp2/3 complex knockdown phenocopied that of N-WASP in these cells: both knockdowns exhibited a bigger area covered by FCPs but no change in the CCP number (Fig. 3h, i), Furthermore, over-expression of full-length N-WASP in the Arpc2 KD cells did not decrease the area covered

76

4

by FCPs (Fig. 3j), thereby showing that N-WASP controls the FCPs by activating the Arp2/3 complex.Taken together, these results indicate that actin polymerization controlled by N-WASP and the Arp2/3 complex has a key role in FCP regulation.

Figure 2. N-WASP is a key FCP regulator. (a) Characterization of control (CTR) knockdown (KD) and N-WASP KD HeLa cells. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown. (b) N-WASP regulates the association between CCSs and F-actin. Representative two-color SR images of control (CTR) KD and N-WASP KD cells stained for CHC (green in merge) and F-actin (red in merge). Dashed white boxes mark the position of the zoom-in. Scale bar, 1 µm. (c) Knockdown of N-WASP increases FCP presence on the basal membrane. Representative TIRF and SR images of CHC on the basal membrane of control (CTR) KD and N-WASP KD cells. Scale bar TIRF images, 10 µm. Scale bar SR images, 1 µm. Bar graphs show the number of CCPs per µm2 and percentage of total area of the ROI covered by FCPs (FCP-covered area). ROIs were defined as three µm-wide regions on the periphery of the cells and segmented for quantification, mean +/- SEM, ** p < 0.01, t-test. (d) Representative TIRF and SR images of CHC in N-WASP KD cells transfected with full-length shRNA-resistant N-WASP tagged with GFP (N-WASP-GFP).. Number of CCPs and FCP-covered area were obtained as above, mean +/- SEM, **** p < 0.0001, t-test. Scale bar TIRF images, 10 µm. Scale bar SR images, 1 µm.

a

**

n.s.

b

0

5

1 0

1 5

CTR KDN-WASP KD

0.4

0.2

0.0

c

CTR

KDN

-WAS

P KD

CHCN-WASP

CLCActin

CTR

KDN-

WAS

P KD

CTR

KD

N-W

ASP

KD

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

CTR KDN-WASP KD

CHC

F-actin

Merge Merge Merge

CHC

F-actin

Merge

d

****

n.s.

0 .0

0 .4

0 .8

0

5

1 0

1 5

GFPN-WASP FL

N-W

ASP

KD

N-WASP

CHC

GFPN-WASP FL

CHC

CHC / F-actin CHC / F-actin

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

CHC

77




4

dc

ba

CBC

FC

Ps w

ith G

FP

***

****

n.s.

e

h

i

#1 #2 #1 #2

**

*

j

n.s.n.s.

0

5

1 0

1 5

2 0

0 .0

0 .5

1 .0

1 .5

ArpC2 KD + GFPArpC2 KD + N-WASP FL

#1 #2

n.s.

n.s.

p34-ARC (ArpC2)

CHC

p21-ARC (ArpC3)

Arp2

Arp3

N-WASP

Vinculin

Actin

CTR

KD

#1 #2

ArpC2KD

CTR

KDAr

pC2

KD #

1Ar

pC2

KD #

2

GFP

N-WASP ΔVCA

N-WASP ΔPRD

N-WASP ΔWH1

****

****

**

N-WASP

***

n.s.

N-WASP H208D *

WH1/EVH1 PRDB V V C AGBD

N-WASP ΔVCA

N-WASP ΔPRD

N-WASP ΔWH1

N-WASP

GFP

H20

8DΔV

CA

Δ P

RD

ΔWH

1G

FP

N-W

AS

P

N-WASP H

208D

0 .0

0 .5

1 .0

1 .5

2 .0

2 .5

n.s.

g

0 .0

0 .5

1 .0

1 .5 n.s.

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

GFP

N-WASP ΔVCA

N-WASP ΔPRD

N-WASP ΔWH1

N-WASP

N-WASP H

208D

GFP

N-WASP ΔVCA

N-WASP ΔPRD

N-WASP ΔWH1

N-WASP

N-WASP H

208D

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

FCP

cove

red

area

0

5

1 0

1 5

2 0

178 254 391195

CTR KDArpC2 KD

-0. 1

0 .0

0 .1

0 .2

0 .3

0 .4

GFP

N-WASP ΔVCA

N-WASP ΔPRD

N-WASP ΔWH1

N-WASP

N-WASP H

208D

CBC

C

CPs

with

GFP

- 0. 1

0 .0

0 .1

0 .2

0 .3

0 .4

n.s.

****

****

n.s. n.s.

n.s.

f

Figure 3. Actin polymerization controlled by N-WASP and the Arp2/3 complex has a key role in regulating FCPs but not CCPs. (a) Schematic representation of N-WASP highlighting the mutants used in this study. A red asterisk marks the position of the H208D mutation. (b) Representative TIRF and SR images of CHC on the basal membrane of cells expressing the GFP-tagged N-WASP mutants described in (a). Scale bar TIRF images, 10 µm; SR images, 1 µm. (c) Active N-WASP regulates FCPs through the VCA, the WH1 and the PRD domains. Bar graph shows the normalized FCP covered area obtained as in Figure 2c, mean +/- SEM, ** p < 0.01, *** p < 0.001, **** p <0.0001, one-way ANOVA. (d) Localization of N-WASP at FCPs requires active N-WASP and the WH1 and the PRD domains. Bar graph shows coordinate-based colocalization (CBC) coefficient for the association between FCPs and the N-WASP mutants or GFP, mean +/- SEM, *** p < 0.001, **** p < 0.0001, one-way ANOVA. (e) Neither N-WASP nor its mutants perturb CCPs. Bar graph shows the normalized number of CCPs per µm2. ROIs were defined as in Figure 2c, mean +/- SEM, one-way ANOVA. (f) Localization of N-WASP at CCPs requires active N-WASP and the WH1 and the PRD domains. Bar graph shows the CBC for the association between CCPs and the N-WASP mutants or GFP, mean +/- SEM, **** p <0.0001, one-way ANOVA. (g) Knockdown of the Arp2/3 complex affects CCS morphology. Representative TIRF and SR images of CHC on the basal membrane of stable control (CTR) KD and ArpC2 KD HeLa cells obtained using two different hairpins (#1 and #2). Scale bar TIRF and SR images as in (b). (h) Characterization of the cells shown in (g). One of two experiments that were performed with similar results is shown. (i) Knockdown of the Arp2/3 complex phenocopies that of N-WASP. Bar graphs show number of CCPs per µm2 and the FCP covered area (percentage) defined as in (c), mean +/- SEM, * p < 0.05, ** p <0.01, t-test. (j) N-WASP regulates plaque dynamics only through the Arp2/3 complex. ArpC2 KD #1 and #2 cells were transfected with N-WASP-GFP. Bar graph shows the FCP covered area (percentage) defined and plotted as above, mean +/- SEM, t-test non-significant differences.

78

4

Flat clathrin plaques remodel in response to external stimuli.We noticed that withdrawal of Fetal Calf Serum (FCS) caused a significant increase in FCP (Fig. 4a). To determine whether FCS acts through N-WASP to affect the abundance of FCPs, we measured the area covered by plaques in control and N-WASP KD cells that were kept in either high (10%) or low (0.1%, serum starvation) FCS (Fig. 4b). Control cells showed 70% less FCPs when grown in the former than in the latter condition. Conversely, FCS had a more moderate effect in N-WASP KD cells (only 30% reduction in plaque covered area), most probably due to the residual N-WASP expression. To characterize how cells remodel plasma-membrane CCSs in response to FCS stimulation, we took serum-starved control KD cells and imaged them prior and after acute stimulation with FCS for 3, 7, 15, 30 and 60 minutes. SR revealed that the area covered by FCPs rapidly decreased (Fig. 4c), whereas the CCP number did not vary (Fig. 4c). These data show that FCPs are highly dynamic structures under control of external stimuli.

Live-cell TIRF microscopy and tracking of discrete CCSs in cells expressing RFP-tagged CLC confirmed this result, as CCS persistence after FCS stimulation was ~4 fold longer in N-WASP KD cells than in control ones (Fig. 4d). The bioactive lipid lysophosphatidic acid (LPA), a major component in FCS that strongly affects various cellular functions 163 fully recapitulated the effects of FCS. In contrast, epidermal growth factor (EGF) had no effect on FCPs (Fig. 4e, f). Of note, neither LPA nor EGF affected the number of CCPs significantly. Taken together,

time (min)time (min)

c

0

5

1 0

1 5

0 2 0 4 0 6 0

0 .0

0 .5

1 .0

1 .5

a

0 2 0 4 0 6 0

b

n.s.

n.s.

****

**

0 .0

0 .5

1 .0

1 .5

Pers

isten

ce (s

ec.)

CTR

KDN-

WAS

P KD

0 sec. 450 sec. 900 sec.d

****500

400

300

200

100

0

CTR KDN-WASP KD

10 %

FCS

0.1

% F

CS

0

5

10

15

20

e

NS

EG

F

FCS

f

NSEGF

LPA

FCS

******

0 .0

0 .5

1 .0

1 .5

0

2

4

6

8

n.s.

n.s.n.s. n.s.

LPA

CTR KDN-WASP KD

NSEGF

LPA

FCS

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

Num

ber o

f CC

Ps /

µm 2

FCP

cove

red

area

FCS (%)

0.1 10 0.1 10

FCS (%)

0.1 10 0.1 10

CTR KD N-WASP KD

Figure 4. FCPs remodel in response to external stimuli. (a) Representative SR images of the basal membrane of control (CTR) KD and N-WASP KD HeLa cells grown in either 0.1% or 10% FCS and stained with anti-CHC antibodies to detect CHC. Scale bar, 1 µm. (b) FCS affects the presence of FCPs, but not that of CCPs. Bar graph shows number of CCPs per µm2 and the FCP covered area (percentage) of cells grown in either 0.1% or 10% FCS. ROIs were defined as in Figure 2c, mean +/- SEM, ** P<0.01, **** p < 0.0001, t-test. (c) FCPs rapidly respond to FCS stimulation. Bar graph shows number of CCPs per µm2 and FCP covered area of serum-starved cells stimulated with 10% FCS for 3, 7, 15, 30 and 60 minutes. ROIs were defined as above, mean +/- SEM. (d) The knockdown of N-WASP increases the persistence of CCSs in cells stimulated with FCS. Representative images of live-cell TIRF movies of control (CTR) KD and N-WASP KD cells expressing CLC-RFP, stimulated with 10% FCS at time 0. Bar graph shows CCS track duration (s = seconds) , mean +/- SEM, **** p < 0.0001, t-test. (e) LPA recapitulates the effects of FCS on FCPs. Representative SR images control (CTR) KD cells that were serum starved and stimulated with 100 ng/ml EGF, 5 µM LPA or 10% FCS for 30 minutes or left untreated (NS) and subsequently stained as in (a). Scale bar, 1 µm. (f) Bar graphs show number of CCPs per µm2 and FCP covered area depicted as in (c). ROIs were defined as in (b), mean +/- SEM, *** p < 0.001, one-way ANOVA.

79




4

these findings shows that N-WASP-controlled FCPs are dynamic structures that respond to specific extracellular stimuli.

Figure 5. FCPs are sites of CCP formation. (a) CCSs exhibit different dynamics and fate. Three representative kymographs of large CCSs from live-cell TIRF movies of HeLa cells expressing CLC-RFP that were serum starved overnight and then stimulated with 10% FCS. CCS can dissociate into smaller structures (i and ii) or fluctuate (iii). Time (t) is in seconds (sec). Representative stages are shown below each kymograph. (b) FCPs are sites of CCP formation. Two representative 3D SR image slices of HeLa cells stained for CHC showing CCVs (whose position is marked by arrowheads in the Z stacks) in areas covered by FCPs. Scale bar, 1 µm. (c) Inhibition of Dynamin perturbs the morphology of CCSs. Representative 3D SR image slices of HeLa cells treated with DMSO or the dynamin inhibitor Dynasore (80 µM, 30 min) stained for CHC. Arrowheads mark CCPs in the Z stacks. Scale bar, 1 µm. (d) Representative SR images of HeLa cells treated with DMSO or the dynamin inhibitor Dynasore (80 µM, 30 min) stained for CHC. Scale bar, 1 µm. (e) Inhibition of Dynamin reduces FCPs and increases CCP number. Bar graph shows number of CCPs per µm2 and FCP covered area (percentage). ROIs were defined as in Figure 2c, mean +/- SEM, *** p < 0.001, **** p < 0.0001, t-test.

c

DM

SO

z: 0 nm z: 100 nm z: 200 nm

basal membrane cytoplasm

Dyn

asor

e

d DMSO Dynasore

DMSO

Dynas

ore

****

e

0 .0

1 .0

2 .0

***

0

5

1 0

1 5

DMSO

Dynas

oreNum

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

a

b

basal membrane cytoplasm

z: 0 nm z: 100 nm z:200 nm z:300 nm

FCS

FCS

FCS

CLC

CLC

CLC

(i)

(ii)

(iii)

900 sec.0 t :

80

4

FCPs are sites of CCP formation. To study the dynamics of CCSs, we generated kymographs of large CLC-positive structures detected in live-cell TIRF movies. In these kymographs, FCPs appeared as highly dynamic structures that showed quick and large fluctuations in shape and density (Fig. 5a). In some particular cases, clear disassembly into smaller structures could also be observed (Fig. 5a). Strikingly, visual inspection of 3D SR images of control HeLa cells revealed CCPs appearing on top of FCPs (Fig. 1b, Fig. 5b, Supplementary Movie 1). These observations are consistent with EM studies 44,53,164 that have captured CCPs seemingly emerging at the periphery of FCPs. Thus, FCPs may be intermediates in the formation of vesicles, a hypothesis that is also supported by computer simulation models 52. We reasoned that if CCVs emerge from FCPs, inhibition of their scission would produce accumulation of CCPs on the membrane at the cost of FCPs. Thus, we treated control HeLa cells with Dynasore, a small molecule inhibitor of dynamin that prevents vesicle scission and causes the accumulation of elongated CCPs on the plasma membrane 165. Strikingly, 3D SR pictures of Dynasore-treated cells displayed a substantial reduction of FCPs at the basal membrane and abundant CCPs, which often formed clusters (Fig. 5c). Morphometric analysis of 2D SR images of cells treated with DMSO (mock) or Dynasore further confirmed the markedly reduced area covered by FCPs and the increase in CCP number (Fig. 5d, e). As acute inhibition of dynamin results in the buildup of clumps of pits at the expense of plaques, FCPs appear to be hubs where CCPs can form and pinch off. Furthermore, the accumulation of FCPs upon depletion of either N-WASP or the Arp2/3 complex suggests an FCP-specific role for actin polymerization in the formation of these CCPs.

CTR

KD

EGFNS

N-W

AS

P K

D

CTR KDN-WASP KD

0 .0

0 .2

0 .4

0 .6

a

pits plaques

b

F

HI

NWKD SS

LPA

R1-

GFP

- C

HC

colo

caliz

atio

n

pits plaques

0 .0

0 .2

0 .4

EG

FR -

CH

Cco

loca

lizat

ion

CTR KDN-WASP KD

LPA

CTR

KD

N-W

AS

P K

D

0

1

2

3

4

EG

FR (E

PI/T

IRF

ratio

)

CTR KDN-WASP KD

c

0 20 40 60 (min.)0.0

0.5

1.0

1.5

0 20 40 60 (min.)

CTR KDN-WASP KD

LPA

R1

(EP

I/TIR

F ra

tio)

NSEGF

****

*

****

**

***

*******

****

NS

NSEGF NS

EGF NSEGF

NSLP

A NSLP

A NSLP

A NSLP

A

EGFR CHC

LPAR1CHC

********

****

****

*

*************

**

Figure 6. LPAR1 and EGFR are recruited to FCPs. (a) Activated EGFR is recruited to CCSs. Representative two-color SR images of control (CTR) KD and N-WASP KD cells stained for CHC (red in merge) and EGFR (green in merge). Cells were stimulated with 100 ng/ml EGF for 5 minutes. Scale bar, 1 µm. Bar graph shows CBC for the association of EGFR with either CCPs or FCPs prior to and after EGF stimulation, mean +/- SEM, * p < 0.05, ** p < 0.01, **** p < 0.0001, t-test. (b) Activated LPAR1 is recruited to CCSs. Representative two-color SR images of control (CTR) KD and N-WASP KD cells stably expressing LPAR1-GFP stained for CHC (red in merge) and GFP (green in merge). Cells were stimulated with 5 µM LPA for 5 minutes. Scale bar, 1 µm. Bar graph shows CBC for the association of EGFR with either CCPs or FCPs prior to and after LPA stimulation, mean +/- SEM, *** p < 0.001, **** p < 0.0001, t-test. (c) Internalization of EGFR and LPAR1 is impaired in cells lacking N-WASP. Bar graphs show internalization level of EGFR and LPAR1 measured as the ratio of EPI images vs. TIRF images of control (CTR) KD and N-WASP KD cells after stimulation with EGF or LPA, respectively, for 3, 7, 15, 30 and 60 minutes, mean +/- SEM, * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, one-way ANOVA.

81




4

Flat clathrin plaques are hubs for clathrin-mediated endocytosis and signalling of LPAR1Given that FCPs are sites of CCP assembly and disappear upon LPA stimulation, we reasoned that the FCPs could participate in ligand-induced CME. We determined the localization of endogenous EGF receptor (EGFR) and GFP-tagged LPA receptor 1 (LPAR1-GFP) in control and N-WASP KD cells counterstained for CHC using two-color SR. LPAR1 and EGFR attained a diffuse and homogeneous distribution on the basal plasma membrane of both cell lines before activation by the cognate ligand (Fig. 6a, b). Ligand-induced activation triggered the recruitment of both LPAR1 and EGFR to plaques and pits, supporting the notion that internalization of these receptors occurs primarily via CME 12,154. Consistently, ligand-induced internalization of EGFR and LPAR1 were impaired by downregulating N-WASP (Fig. 6c). However, only LPA caused the disappearance of FCPs from the basal membrane of control cells (Fig. 4e), which suggests that the recruitment of the LPAR1 to FCPs is followed by internalization. If FCPs were sites of CME of LPAR1, then reduced LPAR1 internalization in N-WASP KD cells could perturb, at least partly, the activity of the signal transduction pathways downstream of LPAR1. Therefore, we compared the activity of ERK and AKT pathways, the two main signaling axes downstream of LPAR1 and EGFR, in control KD and N-WASP KD cells. The activity of ERK and AKT induced by EGF did not depend on N-WASP, the area covered by FCPs or the dynamics thereof (Fig 7a, b). In contrast, LPA signaling to AKT was dramatically increased upon knockdown of N-WASP (Fig 7a, b). This suggests that in the absence of proper internalization, LPA receptors transmit stronger and/or more prolonged signals. To study this at higher temporal resolution, we made use of a specific biosensor to monitor the production of phosphatidylinositol 3,4,5 trisphosphate (PIP3), which is an upstream activator of AKT. Membrane relocalization of the pleckstrin homology (PH) domain of GRP1 166 induced by LPA stimulation was significantly higher in N-WASP KD cells than in the control cells (Fig. 7c). Instead, no significant difference was detected when EGF was used (Fig. 7c). Hence, increased LPA-induced PIP3 formation in N-WASP KD cells translates into higher levels of activated AKT (Fig. 7b). The effects of LPA are mediated by LPAR1 because the small molecule inhibitor Ki6425, which blocks both LPAR1 and LPAR3 167, abrogated AKT activation and LPAR3 is not expressed in HeLa cells 168 (Supplementary Fig. 3). These results suggest that LPAR1, but not EGFR, is internalized from FCPs and they raise the intriguing possibility that FCPs may function as signaling nanodomains.An independent line of evidence supports this view: live-cell TIRF experiments showed that the LPAR1 was rapidly recruited to CCSs after LPA stimulation in both control KD and N-WASP KD cells (Fig. 6b). In the control KD cells, the marked colocalization of LPAR1 and CCSs decreased progressively and was accompanied by LPAR1 endocytosis, as indicated by the accumulation of LPAR1-GFP-positive vesicles inside the cells (Fig. 7d, e). In the N-WASP KD cells, LPAR1-GFP persisted for several minutes in CCSs and it was not prominently internalized after LPA stimulation (Fig. 7d, e). Moreover, two-color kymographs of large CCSs in the control KD cells revealed co-fluctuation of CLC-RFP and LPAR1-GFP (Fig. 7f). In contrast, in kymographs from N-WASP KD cells CLC and LPAR1 colocalized more persistently (Fig. 7f). Notably, no evident difference was observed in the behavior of CCPs in the two cell lines (not shown). Therefore, increased PIP3 formation and AKT activation after LPA stimulation in cells with impaired FCP dynamics result from the lingering of the receptor in these structures.

82

40 50 100 150 200 (sec.)

0 100 200 300 400 (sec.)

↓

↓

↓

↓

EGFLPA

EGFLPA

M /

C ra

tioM

/ C

ratio

N-W

ASP

KDCT

R KD

LPA EGFNS

a

0

50

100

150CTR KDN-WASP KD

LPA

EGFLP

AEGF

Resp

onsiv

e ce

lls (%

)

***

CTR

KD

LPA

30 3350

LPA

N-WASP KD

CTR KD

LPA

LPA

b

fCTR KDN-WASP KD

0 2 4 6 (min.)0.0

1.0

2.0

LPAR

1en

richm

ent i

n CC

Ss

↓

LPA EGFNS

e

c

0 5 15 30 60120 0 5 15 30 60120

CTR KD N-WASP KD

pAKT

AKT

ERK

pERK

N-WASP

EGF (min.)

0.0

1.0

2.0

0.0

1.0

2.0

d

pAKT

AKT

ERK

pERK

N-WASP

LPA (min.) 0 5 15 30 60120 0 5 15 30 60120

CTR KD N-WASP KD

LPA-

GFP

Mer

geCL

C-RF

P

N-W

ASP

KD

LPA-

GFP

Mer

geCL

C-RF

P

335 sec.0

CLCLPAR1

CLCLPAR1

LPA

30 3350(sec.) (sec.)

Figure 7. FCPs are hubs for clathrin-mediated endocytosis and signalling of LPAR1. (a) Knockdown of N-WASP does not perturb the activation of AKT or ERK induced by EGF. Western blot analysis of AKT and ERK activation in control (CTR) KD and N-WASP KD HeLa cells after stimulation with 100 ng/ml EGF for 5, 15, 30, 60 and 120 minutes. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown (b) Knockdown of N-WASP induces hyper-activation of AKT after LPA stimulation. Western blot analysis of AKT and ERK activation in control (CTR) KD and N-WASP KD cells after stimulation with 5 µM LPA for 5, 15, 30, 60 and 120 minutes. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown. (c) Knockdown of N-WASP increased PIP3 formation after LPA but not EGF stimulation. Representative PIP3 formation tracking images using a PIP3 sensor (GRP1) tagged with GFP in live cell confocal images of control (CTR) KD and N-WASP KD HeLa cells stimulated with 100 ng/ml EGF and 5 µM LPA. Representative traces (one cell per trace) of brightness ratio between membrane and cytoplasm (M / C ratio). Bar graph shows percentage of responsive cells, mean +/- SEM, *** p < 0.001, t-test. (d) Activated LPAR1 clusters and is recruited to CCSs, and its internalization is impaired in cells lacking N-WASP. Representative images of two-color live cell TIRF movies of control (CTR) KD and N-WASP KD cells expressing CLC-RFP (red in merge) and LPAR1-GFP (green in merge), before and after stimulation with 5 µM LPA. (e) Graph shows LPAR1-GFP enrichment in CLC-RFP positive structures in control (CTR) KD and N-WASP KD cells. (f) Representative kymographs of non-diffraction limited structures selected in the CLC-RFP channel of two-color live cell TIRF movies showing CLC-RFP (red) and LPAR1-GFP (green) of control (CTR) KD and N-WASP KD HeLa cells, before and after stimulation with 5 µM LPA.

83




4

DISCUSSION

Since their discovery 44,45, the role and physiological relevance of flat clathrin-coated plaques has remained enigmatic and highly controversial. Taking advantage of the increased resolution provided by SR and combining it with molecular genetics, we have been able to unravel fundamental functional properties of FCPs that distinguish these CCSs from CCPs. We found that FCPs function as dynamic hubs for actin-dependent assembly of CCPs that regulate signaling and endocytosis of LPAR1, but not EGFR. Given that FCPs contain the machinery regulating CME, it is not surprisingly that a common mechanism relying on the WH1 domain, the GBD and the central PRD, controls the recruitment of N-WASP to CCPs and FCPs. Interestingly, these domains have a well-established role in regulating N-WASP auto-inhibition. Thus, the enrichment of N-WASP ΔVCA in both CCS types, vis a vis the absence of the H208D mutant, jointly suggest that only in its open conformation does N-WASP stably associate with CCPs and FCPs. Yet, at variance with the pits that form de novo on the plasma membrane, the maturation of pits at FCPs requires N-WASP and its ability to promote Arp2/3-complex-dependent actin polymerization. Consistent with the essential vs. dispensable roles of N-WASP in CCV formation at FCPs and de novo sites, respectively, we identified unique requirements for the action of N-WASP at FCPs. In fact, the ΔPRD mutant markedly increased the FCP-covered area, whereas it did not affect the CCP number (Fig. 3) or cell-surface EGFR 70. Furthermore, we also noted that downregulation of the N-WASP activator Abi1 in Nap1 KD cells, which impairs CME of the EGFR 70, has no effect on FCPs (data not shown). This indicates the existence of an FCP-specific set of N-WASP regulators.High quality EM studies previously proposed that the FCPs are intermediates in the formation of CCVs 45,46,53. At the heart of this hypothesis are micrographs depicting CCPs surrounding or even emerging from FCPs 44,45,50. Yet, based on thermodynamic and molecular considerations considerable doubt was raised regarding this interpretation. Because FCPs are formed by hexagons, they should undergo complicated and energetically unfavorable rearrangements to give rise to the basket-like combination of hexagons and pentagons that compose the clathrin coat of CCPs and CCVs. These considerations have led to a different, widely accepted model for CCV formation, which holds that pits are initiated de novo at endocytic sites and maintain a constant curvature throughout their maturation 32. However, this view has recently been challenged by computer simulations suggesting that the conversion of FCPs into CCPs is possible 52. Moreover, correlative light and electron microscopy (CLEM) showed that the clathrin coat may undergo continuous bending and remodeling at endocytic sites, starting from a rather flat configuration 54. In this arena, our finding that the LPAR1 is internalized upon recruitment to FCPs favors the idea that CCVs can form through a process that involves remodeling of FCPs. Although actin has an accessory role in mammalian CME, we show that it is in fact essential for CCV formation at FCPs. In these nanodomains, it may be hypothesized that actin polymerization provides mechanical force for the rearrangement of the flat clathrin lattice into a curved lattice and/or to counteract the resulting increase in membrane tension. Consistent with this notion, drugs that arrest actin dynamics inhibit the formation of CCVs, both de novo and from preexisting CCSs, and also prolong the lifetime thereof 50,62. Given that CLEM studies have found that the actin fibers visible in EM images are not always detected in the corresponding fluorescence images 143, the presence of F-actin at FCPs might be a general feature of these CCSs that has gone unnoticed using low-resolution microscopy approaches. In light of the above considerations, it is not surprising that cell-substrate adhesion and

84

4

membrane tension are key factors regulating CCS formation and dynamics 49,169. As strong adhesion delays CCP dynamics and CCV formation 49,169 and results in an increased abundance of FCPs on the ventral, but not the dorsal side of cells 32,49,169, it has been suggested that plaques are tissue-culture artifacts. However, FCPs have been observed in physiological contexts greatly varying for adhesion and tension, such as osteoclasts and bone, myocytes, non-adhering adipocytes and even on the non-adherent surface of cultured cells 48,170-172. Given that depletion of N-WASP results in the abundant presence of FCPs located between F-actin fibers (Fig. 2b), it seems unlikely that actin-based adhesion strengthening is the mayor determinant of FCPs.Most importantly, we show that FCPs are dynamic actin-controlled hubs for clathrin-mediated endocytosis and signalling of the LPAR1. AKT activation depends on PIP3, which is generated by activated PI3K using PIP2 as a substrate. PIP2 is available at high concentration at sites of CCP formation 173, and thus the prolonged residence of LPAR1 at FCPs in the N-WASP KD cells likely results in a higher amplitude of AKT activity. Given the importance of LPA for cancer cell chemotaxis 174,175 and the roles of AKT in cell migration, FCP-dependent control of LPA-induced AKT activity may be an unforeseen mechanism whereby N-WASP exerts its pro-metastatic function.

85




4

METHODS

Chemical and reagents High-glucose DMEM supplemented with pyruvate and GlutaMax® was from Invitrogen. Protease inhibitor EDTA-free cocktail, X-tremeGene 9 were from Roche. Gelatin with ~300 bloom, Glucose Oxidase Type VII from Aspergillus niger, Catalase from Aspergillus niger, EGTA, MES, Dynasore, DMSO, Sodium borohydride and all other chemicals if not otherwise specified were from Sigma-Aldrich. Fetal bovine serum was from APS. LPA (1-oleyl) was from Avanti Polar Lipids, EGF was from Invitrogen, Ki16425 was from Santa Cruz biotechnology. Cysteamine hydrochloride – MEA was from Fluka. PIPES was from Fisher scientific. Paraformaldehyde, Glutaraldehyde solution 25%, Magnesium chloride hexahydrate, Triton X-100 were from Merck. Albumin bovine Fraction V, pH 7.0 was from Serva.

Antibodies Antibodies were as follows: mouse anti-β-actin (AC-15), mouse anti-Clathrin Light Chain (CON.1) (Sigma-Aldrich), mouse anti-Clathrin Heavy Chain (X-22; ABR), goat anti-ARP2 (N-15) and anti-ARP3 (G-15) (Santa Cruz Biotechnology), goat anti-p34-ARPC2 (Imgenex), mouse anti-p21-ARPC3 (Transduction Laboratories), rabbit anti-RFP polyclonal (AB3216 Chemicon), mouse anti-vinculin (ab18058; Abcam), rabbit anti-GFP polyclonal was a kind gift from G. Kops, mouse anti-EGFR (Ab-1 528 Calbiochem), rabbit Akt (C67E7), rabbit pAkt (Ser473), rabbit p44/42 pERK and ERK were from Cell Signaling. Alexa-fluor-532 and Alexa-fluor-647 conjugated goat anti-mouse and anti-rabbit IgG (H+L) antibodies were from Invitrogen. Alexa-fluor-647 Phalloidin was from Invitrogen.

Expression Vectors Rat N-WASP (1-501) constructs were either previously described 61,70 or generated by PCR amplification and sequence verified. mRFP1-Clathrin Light Chain was from L. Lagnado, mTurquoise2-Clathrin Light Chain was from D. Gadella. LPAR1-GFP was from W. Moolenaar. pGRP1(PH)-EGFP was from A. Gray 166.

Cell Culture, Transfections and KnockdownsHeLa cells were cultured in DMEM GlutaMax® supplemented with 10% FCS. BSC-1 cells were cultured in MEM (PAA) supplemented with 10% FCS and GlutaMax®. Cells were transfected with X-tremeGene9 according to the manufacturer’s instructions. Stable N-WASP knockdown HeLa cells were previously described 70. Stable N-WASP BSC-1 cells were obtained by lentiviral infection with MISSION® TRC shRNA TRCN0000123061 (#1) and TRCN0000123063 (#2). ARPC2 knockdown HeLa cells were obtained by lentiviral infection with MISSION® TRC shRNA TRCN0000036502 (#1) and TRCN0000036503 (#2) as previously described 68,176. Stable LPAR1-GFP HeLa cells were obtained by retroviral infection and selected based on GFP expression levels by FACS sorting.

Western BlottingCells were cultured in 6-well plates, serum starved overnight in medium containing 0.1% FCS, treated with inhibitors (Ki16425 1 μM for 10 min) and stimulated with agonists as indicated. Whole-cell lysates were prepared by scraping PBS-washed cells in denaturing conditions in JS lysis buffer 177 supplemented with Na3VO4 5 μM, NaF 1 μM and protease inhibitor cocktail (Roche). Membranes were blocked in BSA 5% in TBS supplemented with Tween-X20 0.2% and incubated with primary antibodies, followed by HRP conjugated secondary antibodies.

86

4

Live-cell confocal and TIRF imagingHeLa and BSC-1 cells were seeded on 24 mm coverslips (#1.5) coated with 0.5% gelatin/PBS unless otherwise specified. Cells were transfected for 2-3 hours and serum starved overnight in medium containing 0.1% FCS. Cells were imaged in DMEM-F12 medium at 37⁰C in a humidified chamber with 5% CO2 level using a TCS SP8 confocal or a AM TIRF MC from Leica Microsystems.

GSDIM Super-Resolution Imaging HeLa and BSC-1 cells were seeded on 24 mm coverslips (#1.5) coated with 0.5% gelatine/PBS unless otherwise specified. Cells were serum starved overnight in medium containing 0.1% FCS. Cells were washed briefly with DPBS and fixed as described before 159. Briefly, 4% PFA in PEM buffer was added for 10 minutes, cells were then extracted in 0.5% Triton/PBS for 10 minutes and blocked for 1 hour in 5% BSA/PBS. Inmunofluorescence was performed at room temperature with primary antibodies followed by secondary Alexa-fluor coupled antibodies. For actin colocalization studies Phalloidin was used. Imaging was carried out on a Leica SR-GSD 3D microscope. Images were taken in TIRF mode at 100 frames per second and 8,000-15,000 frames were collected. For 3D images, a cylinder lens 126 was used in conjunction with EPI mode imaging. Images were taken sequentially in decreasing excitation/emission wavelength order in the presence of an oxygen scavenging system (10% Glucose, 0.5 mg/ml Glucose Oxidase, 40 μg/ml Catalase) and 100 mM MEA. Data obtained this way was background subtracted 144 and localization of events, filtering and drift correction was made with the Thunderstorm 137 plugin for ImageJ. Images were rendered with 20 nm pixel size, with the Normalized Gaussian visualization option. For figure visualization, images were convolved with a mean filter of 3×3 pixels.

Image AnalysisSuper-resolution image quantification: plaques and pits were manually segmented and analyzed with the particle analyzer routine from Fiji (ImageJ) 139,146. A 3 µm-wide region of interest (roi) localized at the periphery of the cell was selected for analysis. Clathrin coated pits and plaques were manually segmented and their characteristics, such as area, circularity and shape, were obtained with the Particle Analysis routine of Fiji. The density of pits per area was calculated as the number of pits in the roi against the roi’s area. The area covered by plaques was calculated as the total area of all plaques in the roi against the total roi’s area. Chromatic aberration correction: multicolor tetraspeck beads (Invitrogen) were imaged in orange (532 nm) and far-red (642 nm) channels. The results were aligned using the Image Stabilizer imageJ plugging 147 or the vec2dtransf R package. The obtained transformation coefficients were used to correct images (Fiji) or x,y pairs of coordinates (R).Coordinate based colocalization: a previously published algorithm based on super-resolution coordinates 91 included in the ImageJ plugin Thunderstorm 137 was used. Tracking of CCSs from live-cell TIRF movies: time-lapse movies of internalizing Clathrin-positive structures were recorded with 1 frame per second on a Leica TIRF system, with 488 nm and 561 nm lasers, using a green/red filter cube and exposure time of 900 ms. Laser power was adjusted to maximize signal-to-noise ratio while limiting bleaching to < 50% over 5 minutes. In order to allow faithful tracking of CCSs the time lapses were preprocessed in ImageJ as described below. Movies were smoothed in time by a temporal mean filter of 3 seconds (3 frames) and subsequently corrected for bleaching using a custom algorithm. In Brief, average fluorescence signal and background were calculated for every frame

87




4

using automatic thresholding. Images were divided by the difference between signal and background and multiplied by the difference found in the first frame. Next, a temporal median filter with a 10 seconds window was employed to eliminate apparent rapid intensity changes due to noise in individual CCSs. CCSs were tracked in a region bordering the plasma membrane and extending 3 µm into the cytoplasm using the ImageJ plugin TrackMate (http://imagej.net/TrackMate). Tracking was done using the Simple LAP tracker algorithm, with a maximum travel distance of 300 nm per frame. Local enrichment of LPAR1 in CCSs after LPA stimulation: two-color TIRF movies of cells expressing LPAR1-GFP and CLC-RPF with time resolution 3 seconds and exposure time 900 ms were first slightly smoothed in time by a temporal mean filter with a window of 6 seconds (2 frames). The CLC-RFP channel was converted to a binary mask by first applying a bilateral filter (with a spatial radius of 1 pixel and intensity range 3 times the standard deviation of the background), and subsequent thresholding using a semi-automatic iterative auto local threshold method. For the LPAR1-GFP channel, the median background values for each frame was determined by automatic thresholding and was subsequently subtracted. Next, the LPAR1-GFP signal was computed only in CLC-RFP-positive pixels, that were determined by the binary mask, as explained above. These data were then normalized to the mean LPAR1-GFP in CLC-RPF signal in the frames before stimulation. Kymographs are a sequence of line profiles in time and were obtained by manually selecting a region of interest from two-color TIRF movies pre-treated as explained above and tracking (and plotting) the singal in this area in time, for both colors. PIP3 sensor translocation: single confocal sections of cells expressing low levels of pGRP1(PH)-EGFP were acquired every 10 seconds. The resulting movies were subjected to image smoothing and thresholding in Fiji. To obtain membrane vs. cytoplasm fluorescence tracks, a user-selected region was subtracted from the cell’s edge and referred to as the membrane. Brightness levels pf the membrane and the rest of the cell were tracked over time and the membrane/cytoplasm ratio calculated for each time point. The number of responsive cells was obtained by blind scoring of responsive cells.

Data representation and statistical analyses R studio and GraphPad Prism (version 6.oh) were used to carry out all statistical analyses. GraphPad Prism (version 6.oh) was used to plot results always as mean +/- SEM of the data indicated below.

Figure 1c: n = 15 cells pooled from three independent experiments.Figure 2c: n = 15 control KD cells, n = 15 N-WASP KD cells, pooled from three independent experiments. Figure 2d: n = 13 control KD cells, n = 13 N-WASP KD cells, pooled from three independent experiments. Figure 3c: n = 117 cells for GFP, n = 54 cells for ΔVCA, n = 20 cells for ΔWH1, n = 20 cells for ΔPRD, n = 21 cells for H208D, pooled from three independent experiments. Figure 3d: n = 12 cells for GFP, n = 10 cells for ΔVCA, n = 12 cells for ΔWH1, n = 10 cells for ΔPRD, n = 7 cells for H208D, pooled from two independent experiments. Figure 3e: n = 8 cells for GFP, n = 10 cells for ΔVCA, n = 5 cells for ΔWH1, n = 10 cells for ΔPRD, n = 7 cells for H208D, pooled from two independent experiments. Figure 3f: n = 12 cells for GFP, n = 10 cells for ΔVCA, n = 12 cells for ΔWH1, n = 10 cells for ΔPRD, n = 7 cells for H208D, pooled from two independent experiments. Figure 3g: n = 11 CTR KD cells, n = 11 ArpC2 KD #1 cells, n = 10 ArpC2 KD #2 cells, pooled

http://imagej.net/TrackMate

http://6.oh/

http://6.oh/

88

4

from three independent experiments. Figure 3j: n = 8 ArpC2 KD #1 + GFP cells, n = 11 ArpC2 KD #1 + N-WASP cells, n = 10 ArpC2 KD #2 + GFP cells, n = 10 ArpC2 KD #2 + N-WASP cells, pooled from two independent experiments. Figure 4b: n = 22 control KD + 0.1% FCS cells, n = 23 control KD + 10% FCS cells, n = 22 N-WASP KD + 0.1% FCS cells, n = 23 N-WASP KD + 10% FCS cells, pooled from three independent experiments. Figure 4c: n = 19 cells for 0, n = 20 cells for 3, 7, 15, 30 and 60 minutes, pooled from three independent experiments. Figure 4d: n = 5048 CTR KD tracks, n = 3619 N-WASP KD tracks; n = 7 CTR KD cells, n = 7 N-WASP KD cells, pooled from two independent experiments. Figure 4f: n = 36 cells for NS, n = 36 cells for EGF, n = 36 cells for LPA, n = 34 cells for FCS, pooled from three independent experiments.Figure 5e: n = 23 cells for DMSO, n = 25 cells for Dynasore, pooled from two independent experiments.Figure 6a: n = 21 control KD NS cells, n = 14 control KD + EGF cells, n = 21 N-WASP KD NS cells, n = 14 N-WASP KD + EGF cells, pooled from three independent experiments. Figure 6b: n = 20 control KD NS cells, n = 21 control KD + LPA cells, n = 18 N-WASP KD NS cells, n = 20 N-WASP KD + LPA cells, pooled from three independent experiments. Figure 6c: EGF: control n = 51 cells for SS, n = 51 cells for 3 min, n = 53 cells for 7 min, n = 52 cells for 15 min, n = 55 cells for 15 min, n = 54 cells for 60 min; N-WASP KD n = 51 cells for SS, n = 50 cells for 3 min, n = 50 cells for 7 min, n = 50 cells for 15 min, n = 53 cells for 15 min, n = 50 cells for 60 min; pooled from three independent experiments, LPA: control n = 83 cells for SS, n = 87 cells for 3 min, n = 81 cells for 7 min, n = 83 cells for 15 min, n = 87 cells for 15 min, n = 87 cells for 60 min; N-WASP KD n = 81 cells for SS, n = 77 cells for 3 min, n = 95 cells for 7 min, n = 70 cells for 15 min, n = 83 cells for 15 min, n = 93 cells for 60 min; pooled from three independent experiments. Figure 7c: n = 12 control KD + LPA cells, n = 12 control KD + EGF cells, n = 11 N-WASP KD + LPA cells, n = 11 N-WASP KD + EGF cells, pooled from three independent experiments. Figure 7e: n = 24 control KD cells, n = 15 N-WASP KD cells, pooled from three independent experiments.Supplementary Figure 2d: n = 21 control NS cells, n = 14 control growing cells, n = 20 N-WASP KD 835 SS cells, n = 21 N-WASP KD 835 growing cells, n = 12 N-WASP KD 836 NS cells, n = 11 N-WASP KD 836 growing cells, pooled from three independent experiments.Supplementary Figure 4b: n = 11 control KD cells, n = 11 Nap1 KD cells, pooled from three independent experiments.

89




4

Competing interests

The authors declare no competing or financial interests.

Author contributions

T.I., D.L.-P., K.J. and M.I. designed the experiments. D.L.-P., T.I., J.K. and M.I performed the experiments. D.L.-P., T.I., B.v.d.B. and M.I. performed data analysis and/or quantified results. M.I. conceived and coordinated the study and K.J. supervised imaging and data analysis. M.I., D.L.-P. and K.J. wrote the manuscript.

Funding

This work was supported by Stichting voor de Technische Wetenschappen Technology Foundation [grant number 12150 to K.J.].

90

4

SUPPLEMENTARY MATERIALC

TR K

DN

-WA

SP

KD

CLC

a b

c d

AP2

CTR

KD

N-W

AS

P K

D

CHC Merge

CTR

KD

N-W

AS

P K

D

CLCCHC Merge

e f

CTR

KD

N-W

AS

P K

D

CLCCortactin Merge

CTR

KD

N-W

AS

P K

D

DynaminCHC Merge

0 nm

100 nm

200 nm

300 nm

400 nm

500 nm

600 nm

CTR

KD

N-W

AS

P K

D

Supplementary Figure 1. Molecular characterization of CCPs and FCPs. (a) CLC and CHC colocalize in CCSs. Representative two-color SR images of control (CTR) KD and N-WASP KD HeLa cells expressing CLC-mTQ2 and stained for GFP (green in merge) and CHC (red in merge). Scale bar, 1 µm. (b) CHC and AP2 colocalize in CCSs. Representative two-color SR images of control (CTR) KD and N-WASP KD cells expressing AP50-GFP and stained for CHC (red in merge) and GFP (green in merge). Scale bar, 1 µm. (c) Dynamin localizes in CCPs and FCPs. Representative two-color SR images of control (CTR) KD and N-WASP KD cells expressing Dynamin2-GFP and stained for CHC (red in merge) and GFP (green in merge). Scale bar, 1 µm. (d) Cortactin localizes in CCSs. Representative two-color SR images of control (CTR) KD and N-WASP KD cells expressing CLC-mTQ2 and stained for Cortactin (green in merge) and GFP (red in merge). Scale bar, 1 µm. (e) Representative TIRF and SR images of control (CTR) KD and N-WASP KD HeLa cells expressing CLC-mTQ2 and stained for GFP. Scale bar TIRF, 10 µm. Scale bar SR images, 1 µm. (f) Knockdown of N-WASP increases the presence of basal membrane FCPs. Representative projections of 3D SR images, color coded for depth, of control (CTR) KD and N-WASP KD cells stained for CHC. Scale bar, 1 µm.

91




4

Supplementary Figure 3. AKT hyper-activation in N-WASP KD cells is mediated by LPAR1. Western blot analysis of AKT activation in control (CTR) KD and N-WASP KD HeLa cells after stimulation with 5 µM LPA for 5 minutes with and without pre incubation with the LPA receptor inhibitor Ki16425. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown.

B

a

b c

HeL

a

BSC-1

CTR

KD

#1 #2

N-WASPKD

CHC

N-WASP

CLC

Actin

*******

0

2

4

6

N-W

ASP

KD #

2

N-W

ASP

KD #

1C

TR K

D

0 .6

#1 #2

0 .0

0 .2

0 .4

d

N-W

ASP

KD #

2

N-W

ASP

KD #

1C

TR K

D

CTR KDN-WASP KD

Num

ber o

f CCP

s / µ

m 2

FCP

cove

red

area

NSFCS NS

FCS NSFCS

#1 #2NS

FCS NSFCS NS

FCS

CTR KDN-WASP KD

Supplementary Figure 2. N-WASP is involved in FCP-regulatory mechanism in BSC-1 cells. (a) Characterization of control (CTR) KD and N-WASP KD (hairpins #1 and #2) BSC-1 cells. Total cell lysates were compared using the indicated antibodies. One of two experiments that were performed with similar results is shown. (b) Knockdown of N-WASP induces FCP presence in the plasma membrane. Representative TIRF and SR images of control (CTR) KD and N-WASP KD (#1 and #2) BSC-1 cells serum deprived overnight stained for CHC. Scale bar TIRF images, 10 µm. Scale bar SR images, 1 µm. (c) Knockdown of N-WASP inhibits FCPs response to FCS stimulation. Representative TIRF and SR images of growing control (CTR) KD and N-WASP KD (#1 and #2) BSC-1 cells stained for CHC. Scale bar TIRF images, 10 µm. Scale bar SR images, 1 µm. (d) Bar graph shows number of CCPs per µm2 and percentage of area covered by FCPs of cells serum starved overnight or kept in growing conditions, mean +/- SEM, *** P<0.001, **** P<0.0001, one-way ANOVA.

CTR KD N-WASP KD

Ki16425 1 uMLPA 5 uM - + +

- - +- +- - +

+

Actin

pAKT

AKT

93

Chapter 5

Perifosine inhibits EGFR signaling by inducing its

internalization

Daniela Leyton-Puig†, Katarzina Kedziora†*, Wouter Moolenaar, Kees

Jalink

Division of Cell Biology I, The Netherlands Cancer Institute, Plesmanlaan 121, Amsterdam, 1066 CX, The Netherlands. + Current address: Department of Genetics, University of North Carolina at Chapel Hill,

School of Medicine.

† Authors contributed equally

94

5

ABSTRACT

The epidermal growth factor receptor (EGFR) is a growth and survival signalling receptor that is commonly overexpressed in cancer. Anti-cancer therapies that target EGFR are widely used in the clinic, and many are under development. We show here that ALPs, unconventional chemotherapeutic drugs, can reduce the presence of EGFR on the plasma membrane of cells that overexpress the receptor. Moreover, Perifosine can inhibit the activation of EGFR by its natural ligand EGF and enhance its degradation. Using super-resolution microscopy we show that EGFR is pre-clustered in the membrane on cells that overexpress it, but not in cells with low expression levels. We propose that the differential effect of Perifosine in cells with high and low expression of EGFR is due to the different state of the receptor in the plasma membrane.

95

Perifosine inhibits EGFR signaling by inducing its internalization

5

INTRODUCTION

The epidermal growth factor receptor (EGFR) is a tyrosine kinase receptor whose signalling modulates cell growth, differentiation, survival, adhesion and migration 178. The dysregulation of EGFR signalling leads to tumorigenesis 5,24. EGFR is found both overexpressed and/or mutated in a variety of human cancers 5,24,179. Regulation of EGFR signalling can be achieved via internalization of the receptor or inhibition of its tyrosine kinase activity 1,24,28. Several anti-EGFR therapies using monoclonal antibodies and tyrosine kinase inhibitors have been developed, and are either in use or in clinical trials 6,11. Ligand induced endocytosis of EGFR is a well understood process. Upon ligand binding EGFR undergoes a conformational change that induces its dimerization. The dimerized receptor gets auto-phosphorylated due to its intrinsic kinase activity, allowing the recruitment of signalling proteins and adaptor proteins involved in its internalization 24. Phosphorylated EGFR is also ubiquitinated by the E3 ubiquitin ligase Cbl, a step that is also necessary for its endocytosis 24 and that targets the receptor for degradation 30. Alkylphospholipids and alkyl-lysophospholipids (ALPs) are synthetic analogues of lysophosphatidylcholine (LysoPC) that were synthesized for the first time in the 1960s 180. ALPs are promising anti-cancer agents as they inhibit cell proliferation in a tumour specific manner 181, they are metabolically stable and they have shown clinical efficacy in a variety of tumours 181. ALPs exert their effect by inserting into cell membranes leading to disturbance of their biophysical and biochemical properties. This leads to aberrant phospholipid metabolism and cholesterol homeostasis 182, inhibition of survival pathways 183, activation of apoptosis 184-186 and membrane micro-domain disruption 187,188. ALPs are translocated from the outer to the inner leaflet of the plasma membrane via a transmembrane phospholipid flippase 189-191 or are internalized by dynamin-dependent endocytosis 186,192. As anti-cancer therapy, ALPs are frequently combined with radiotherapy, and other anti-cancer agents that inhibit signalling pathways.Perifosine (D-21266) is a third generation ALP 193 that lacks the choline moiety and contains instead a cyclic aliphatic piperidyl group. It is not metabolized and shows slow elimination and high concentration in tumour cells 181,183 where it targets signal transduction pathways at the plasma membrane 183,194,195. Perifosine blocks AKT signalling by binding to its PH domain, which impedes AKT recruitment to the plasma membrane and its subsequent phosphorylation. It has been tested in a large number of clinical studies and it is frequently used in combination with other chemotherapeutic agents 181,194,196,197. In vitro, Perifosine shows anti-tumour effects against melanoma, lung, prostate, colon and breast cancer models 194 and it is already in clinical Phase 1, 2 and 3 trials for several types of cancer.In previous studies, we observed that internalization of EGFR in A431 cells can be triggered by ether-linked alkyl-lysophospholipids (ALPs) via an unknown mechanism that is independent of receptor activation 198. We here focus on Perifosine and show that it induces internalization of EGFR particularly in cells that overexpress the receptor. Moreover, we show that Perifosine reduces EGFR activation by its ligand but not its degradation. Finally, we explore the effect of Perifosine on EGFR micro-clustering in the plasma membrane by using super-resolution microscopy.

96

5

RESULTS

Perifosine induces EGFR internalization. We have previously described that ALPs can induce the endocytosis of EGFR in A431 epidermoid carcinoma cells 198, which highly overexpress this receptor 199. We here focus on Perifosine, a modern ALP that is commonly used for its AKT inhibiting properties 194,195. Perifosine induced EGFR internalization in A431 epidermoid carcinoma cells already at low doses (Figure 1a, d). Interestingly, Perifosine also induced the internalization of EGFR in MDA MB 468 (Figure 1b, d), triple negative breast cancer cells which also overexpress EGFR 200, albeit not as highly as A431 cells. In contrast, we did not observe internalization of the receptor upon Perifosine stimulation in cells that do not overexpress the receptor, such as HeLa 201 (Figure 1c) and HEK cells (data not shown). These results show that Perifosine induces internalization of EGFR in cells that highly express the receptor.

MDA MB 468

NT Perifosine 0.5 μM

A431

EGF

HeLa

0 .0

0 .4

0 .8

1 .2

***

***

0 .0

0 .4

0 .8

1 .2

*****

* **

a.

b.

c.

Perifosine 5 μM

NT Perifosine 0.5 μMEGF Perifosine 5 μM

NT Perifosine 0.5 μMEGF Perifosine 5 μM

NT

P 0.5 μ

MEGF

P 5 μM NT

P 0.5 μ

MEGF

P 5 μM

d. MDA MB 468A431

Figure 1. Perifosine induces EGFR internalization in cells that overexpress the receptor (a) Perifosine induces EGFR internalization in A431 cells. Representative confocal images of EGFR in A431 cells. A431 cells were left untreated (NT), treated with EGF (100 ng/ml) or Perifosine (0,5 µM and 5 µM as indicated) for 5 minutes, then fixed and stained with anti-EGFR antibodies as indicated in the Methods. Scale bar, 10 µm. (b) Perifosine induces EGFR internalization in MDA MB 468 cells. MDA MB 468 cells were treated as in (a). Scale bar, 10 µm. (c) Perifosine does not induce EGFR internalization in HeLa cells. Representative confocal images of EGFR in HeLa cells. HeLa cells were treated as in (a). Scale bar, 10 µm, two independent experiments. (d) Bar graph shows the normalized membrane EGFR level in A431 and MDA MD 468 cells as measured by FACS analysis (see Materials and Methods). A431 cells: four independent experiments; MDA MB 468 cells: two independent experiments. Mean +/- SEM, * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, ANOVA.

97


5

Perifosine does not induce activation of EGFR. Following activation of EGFR by its ligand, signalling is rapidly attenuated by endocytosis 1,28. Internalized EGFR may then be targeted for degradation or recycled back to the plasma membrane 30. Thus, conventional endocytosis and degradation of EGFR are a direct consequence of its activation, and they depend directly on its kinase activity 8,202. Therefore, therapeutic use of antibodies against EGFR that induce its endocytosis also commonly induces its activation 203,204. Remarkably, Perifosine did not induce phosphorylation or ubiquitination of EGFR in A431 cells (Figure 2a, 2b) or MDA MB 468 cells (Supplementary Figure 1a, 1b). This result suggests that Perifosine-induced EGFR internalization occurs via an unconventional mechanism that is not dependent on activation of its kinase domain.

Figure 2. EGFR internalization induced by Perifosine does not depend on phosphorylation. (a) Perifosine does not induce phosphorylation of EGFR. Western blot analysis of phosphorylated EGFR (in Tyrosine 1086) in A431 cells after stimulation with 100 ng/ml EGF, 0.5 µM and 5 µM Perifosine for 5 and 15 minutes. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown (b) Perifosine does not induce ubiquitination of EGFR. Immunoprecipitation of EGFR and western blot analysis of Ubiquitin in A431 cells after stimulation with 100 ng/ml EGF, 0.5 µM and 5 µM Perifosine for 2 and 15 minutes. EGFR was immunoprecipitated from 200 µg of total cell lysates and the product was compared using the indicated antibodies. One of three experiments that were performed with similar results is shown.

Perifosine inhibits the activation of EGFR by EGF. We hypothesized that the reduced availability of EGFR on the plasma membrane induced by Perifosine would translate into a reduced phosphorylation of the receptor after activation by its ligand. Indeed, in A431 cells pre-incubated with Perifosine for 30 minutes, phosphorylation of EGFR was highly inhibited, especially at higher concentrations of the ALP. For these experiments, EGFR activation was assayed by determining total tyrosine phosphorylation of the receptor using a specific monoclonal antibody, PY20 (Figure 3a, b). These results show that Perifosine inhibits EGFR activation and are in agreement with the inhibition of the EGFR signalling pathway in Mesothelioma cells by Perifosine 205. After internalization, activated receptors can be recycled to the plasma membrane or targeted for degradation 1,30. We found that in A431 cells, roughly half of the receptors were degraded after 5 minutes of stimulation with EGF (Figure 3a, b). Remarkably, in cells pre-treated with Perifosine, where much less EGFR was activated, still half of the receptors were degraded after 5 minutes of stimulation with EGF (Figure 3a, b). This result indicates that after treatment with Perifosine, while less receptors are activated, the same amount gets routed for degradation.

a. b.

5 15min 0

pEGFR(Y1086)

EGFR

Actin

P 0.5 µ

M

P 5 µM

EGFNT

5 15 5 15

IP: EGFRWB: UB

IP: EGFRWB: EGFR

2min 0

P 0.5 µ

M

EGFNT

15

P 5 µM

P 0.5 µ

M

EGFP 5

µM

98

5 EGFR detection by super-resolution microscopy. We have shown that Perifosine induces EGFR internalization only in cells with high receptor expression (Figure 1). On the plasma membrane of cells, EGFR may be present in a monomeric or dimeric state 206. In cells that overexpress EGFR the equilibrium between these two forms is shifted towards the dimeric state 206. We hypothesized that oligomerization of EGFR at the plasma membrane might explain, at least in part, the selective effect of Perifosine on cells overexpressing the receptor.

Figure 4. EGFR exhibits different conformational states in the membrane of cells with low and high expression. (a) EGFR is randomly distributed in the plasma membrane of HeLa cells. Representative super-resolution images of EGFR in unstimulated HeLa cells. Cells were fixed and stained with Alexa Fluor 647 labelled monoclonal anti-EGFR antibodies as indicated in the Methods. Scale bar, 1 µm. Graph shows the pair autocorrelation function of EGFR distribution on the membrane. Experimental data can be fitted with a Cauchy distribution based random model of the distribution of proteins (cyan). Three independent experiments, mean +/- SEM, n = 30 analysed regions from 15 individual cells. (b) EGFR is pre-clustered in the plasma membrane of A431 cells. Representative super-resolution images of EGFR in unstimulated A431 cells. Cells were treated as in (a). Scale bar, 1 µm. Graph shows the pair autocorrelation function of EGFR distribution on the membrane of unstimulated A431 cells. Cauchy distribution based- random model (cyan) fails to represent the data while clustered model (pink) can be successfully fitted. Three independent experiments, mean +/- SEM, n = 40 analysed regions from 20 individual cells.

101

102

100

g(r)

r [nm]

101 102

random model

101

100

g(r)

r [nm]

101 102

randomclustered model

HeLaa.

A431b.

NTNT

Figure 3. Perifosine inhibits the activation of EGFR by EGF but not its degradation. (a) Perifosine inhibits EGFR activation by EGF and enhances degradation of activated receptor. A431 cells were serum starved overnight and stimulated with 100 ng/ml EGF for 5 minutes; with or without 30 minute pre-incubation with Perifosine (0.5 µM and 5 µM). Western blot analysis shows total phosphorylated tyrosine (PY20) at the level of EGFR. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown. (b) Bar graph shows relative amount of EGFR activation (blue, total phosphorylated tyrosine signal) and the relative amount of total EGFR (red). Shown are mean +/- SEM from two independent experiments; * p < 0.05, ANOVA.

EGF + + +- - -

PY20

Actin

EGFR

PY20

P 0.5 µ

M

P 5 µM

P 0.5 µ

M

P 5 µM

a. b.

0 .0

0 .4

0 .8

1 .2

EGF + + +- - -

P 0.5 µ

MP 5

µM

P 0.5 µ

MP 5

µMNT NT

NT NT

PY20 (phospho EGFR)Total EGFR

*

99


5

To explore this possibility we used the single molecule localization super-resolution microscopy (SMLM) technique GSDIM 77 to obtain images of EGFR on the membrane of HeLa and A431 cells. The high resolution that can be achieved with GSDIM allows studying the distribution of proteins on the plasma membrane in much more detail. In conventional fluorescence microscopy, images of individual fluorophores overlap due to the diffraction-limited resolution of the light microscope. To avoid this overlap, in GSDIM fluorescent labels are switched between bright “on” and dark “off” states in time. In each time frame, only a few labels will be on and their localization can be determined with very high precision by computer algorithms. In time, the bright labels will turn off and other labels will be turned on randomly . This process of fluorophore switching, image acquisition and precise localization is repeated until the super-resolution image is build up from the coordinates of all the localized fluorophores 72,77. However, this is a stochastic process and in the final image each fluorophore, and therefore each molecule on the membrane is detected several times and thus it will be represented by a small cluster of dots. We call this effect “intrinsic clustering”, and it challenges the interpretation of SMLM images when studying protein distribution. Indeed, visual inspection of EGFR images on A431 and HeLa cells revealed no clear distribution pattern of EGFR beyond a clear difference in receptor density at the plasma membrane (Figure 4a, b).

Cluster analysis reveals that EGFR is pre-clustered in the plasma membrane of A431. To be able to assess receptor clustering from super resolution data, several clustering analysis methods have been developed that aim to discriminate receptor clustering from intrinsic fluorophore clustering 96,98,103,207,208. To analyse EGFR distribution from images of A431 and HeLa cells obtained with GSDIM, we used a method published by Sengupta and colleagues 103 that uses pair correlation analysis to estimate clusters in spatial data, but takes into account the intrinsic clustering component of SMLM. We applied a slight modification to the method that more reliably describes the distribution of the intrinsic clustering component we found in our experimental data (Supplementary Figure 1). The distribution of distances obtained from experimental data can then be described with one of two models, namely, a random distribution model and a model in which clusters are (additionally) present. We found that in unstimulated HeLa cells the distribution of EGFR on the membrane was well-described by the random model (Figure 4a). In contrast, EGFR molecules appeared already pre-clustered in unstimulated A431 cells (Figure 4b). These results show that EGFR oligomerization differs between cells with high and low receptor expression level, as has been suggested before 206. We also explored the effect of Perifosine on receptor oligomerization, for which we analysed clusters in cells treated with EGF or Perifosine. In both cases, EGFR was clustered on the membrane. Although in both cases clusters of EGFR seemed to become smaller (data not shown), the variability encountered in our preliminary data made it challenging to arrive at a conclusion. More analysis is necessary to determine if Perifosine can affect cluster conformation of EGFR on the plasma membrane of cells.

100

5

DISCUSSION

We show here that internalization of EGFR can be induced by the synthetic lipid Perifosine independently of activation or ubiquitination of the receptor, and we present preliminary data that suggest that receptor pre-clustering in cells with overexpression of the receptor might play a role in this process. Two different hypothesis may explain the intracellular accumulation of EGFR induced by Perifosine. 1. Perifosine triggers internalization. Perifosine could be inducing the internalization of EGFR in an unconventional manner, disrupting the lipid environment in which EGFR is contained in the plasma membrane. Although the effect of the interaction of EGFR with membrane lipids is still largely unknown 209, it has been proposed that the lipid environment of the cellular membrane can affect EGFR’s oligomerization state. In fact, depletion of cholesterol (which disrupts putative rafts) increases the percentage of oligomers of EGFR 210 Furthermore, Phosphatidic acid (PA), a product of the hydrolysis of phosphatidylcholine by Phospholipase D2 in the membrane, can regulate EGFR clustering 211 and PIP2 depletion has been found to impair cluster formation 209. We have shown that EGFR is pre-clustered in A431 cells, in accordance with previous studies 206,212,213 but randomly distributed in HeLa cells. The effect on lipid composition in a EGFR pre-clustered environment in high expressing cells could be the trigger for an internalization of the receptor that occurs independently of the induction of a conformational change associated with auto-phosphorylation. 2. Perifosine interferes with recycling of EGFR to the plasma membrane. Alternatively, the activity-independent accumulation of EGFR inside the cells could result from impaired recycling. It is known that even EGFR that are not bound to EGF are continuously internalized and recycled, a process that occurs with most plasma membrane proteins called tonic or constitutive recycling. In steady state, unstimulated EGFR are internalized slowly via clathrin-coated pits and rapidly recycled to the plasma membrane 214. It has been estimated that 1-2% of the total EGFR is internalized per minute 215, although the rate depends on the level of EGFR expression and cell type 216. It is possible that Perifosine acts to inhibit the normal fusion of the recycling vesicles with the membrane, thereby inducing accumulation of inactive EGFR in the cell. This hypothesis could explain the enhancement of stimulated EGFR degradation after incubation with Perifosine, where with the recycling route blocked, activated EGFR would predominantly follow the degradation route. The accumulation of small EGFR-containing vesicles apposed to the plasma membrane might explain the appearance of small sized clusters after Perifosine treatment, although more study is needed to confirm this preliminary finding.. EGFR and other members of the tyrosine kinase receptor family play various roles in the progression of cancer and therefore they are important candidates as targets for therapy 5,6,24. Many strategies have been developed during the last years to inhibit EGFR signalling in cancer cells, however resistance to these therapies is not uncommon and their efficacy as mono therapy agents is low 5,6,11. Our results with APLs suggest an alternative approach for EGFR signalling inhibition, in line with previous studies showing the enhancement of Cetuximab effectiveness with Perifosine 197 and the inhibition of the EGFR pathway in malignant pleural mesothelioma 205. A molecular understanding of ALP action may lead to new therapeutic strategies, and help to identify patients who may benefit from ALP treatment. Finally, we have used an improved clustering analysis method for super-resolution data that can be applied to the study of clusters of any type of protein in the membrane of cells. In the last few years several studies applying different methods for such analysis have

101


5

been published 99,100,217-221. One commonly used method is Ripley’s K-function 222, but the intrinsic clustering of single molecule localization microscopy (SMLM) data, that is, multiple localizations per molecule, cannot be properly controlled for with this method, leading to e.g. masking of small clusters or propagation of clusters towards longer distances. Using pair correlation analysis the intrinsic clustering component of SMLM can be estimated and factored out, allowing the detection of small protein clusters without the masking effect of multiple localizations per protein 102,103,207. We show here that the multiple localizations that arise from the same molecule, the intrinsic clustering component, follow a Chauchy distribution, as it was reported by Chu and colleagues 97, rather than the Gauss distribution used by Sengupta and colleagues. Moreover, we show that use of the Gaussian distribution assumption in the pair correlation analysis model can give rise to erroneous results, over-estimating clustering (Supplementary Figure 1). With the corrected method we confirm previous literature that reported that EGFR can be found on the plasma membrane of cells both as monomers or dimers, and that when its expression is high, the clustered state is dominant. Based on our preliminary data, we hypothesize that Perifosine might only induce endocytosis on cells with pre-clustered EGFR.

102

5

MATERIALS AND METHODS

Chemical and reagents High-glucose DMEM supplemented with pyruvate and GlutaMax® was from Invitrogen. Protease inhibitor EDTA-free cocktail, X-tremeGene 9 were from Roche. Perifosine, Glucose Oxidase Type VII from Aspergillus niger, Catalase from Aspergillus niger, Tris, EGTA, EDTA, MES, DMSO, NaCl, KCl, Sodium deoxycholate (NaDoc), SDS, NP40, Sodium borohydride and all other chemicals if not otherwise specified were from Sigma-Aldrich. Fetal bovine serum was from APS. EGF was from Invitrogen. LPC was from Avanti. Cysteamine hydrochloride – MEA was from Fluka. PIPES was from Fisher scientific. Paraformaldehyde, Magnesium chloride hexahydrate, Triton X-100 were from Merck. Albumin bovine Fraction V, pH 7.0 was from Serva. Alexa Fluor 647 Monoclonal Antibody labelling kit (A-20186) was from Molecular Probes.

Cell culture and antibodies HeLa, A431, MDA MB 468 cells were cultured in DMEM GlutaMax® supplemented with 10% FCS. Antibodies were as follows: mouse anti-β-actin (AC-15), mouse anti-EGFR (Ab-1 528 Calbiochem), rabbit anti-EGFR (sc-03 Santa Cruz Biotechnology), mouse anti-phosphotyrosine (BD Transduction), mouse anti-ubiquitin (P4D1 Santa Cruz Biotechnology), rabbit anti-pEGFR 1086 (Cell Signaling), mouse anti-pEGFR 1068 (1H12 Cell Signaling) , rabbit anti-pEGFR 1045 (Cell Signaling), rabbit anti-pEGFR 845 (Cell Signaling). Alexa Fluor 488 and Alexa Fluor 647 conjugated goat anti-mouse and anti-rabbit IgG (H+L) antibodies were from Invitrogen.

Biochemical assaysFor western blot cells were cultured in 6-well plates in DMEM with 10% FBS, serum starved overnight and treated with Perifosine, EGF, LPC in concentration indicated in each figure. Whole-cell lysates were prepared by scraping PBS-washed cells in denaturing conditions in modified RIPA lysis buffer (50mM Tris–HCl, 150mM NaCl, 1mM EDTA, 1% Triton X-100, 1% NaDoc, 1% SDS, 1%NP40) supplemented with protease inhibitor cocktail (Roche). Membranes were blocked in BSA 5% in TBS supplemented with Tween-X20 0.2% and incubated with primary antibodies, followed by HRP conjugated secondary antibodies. For IP experiments cells were cultures in 10cm dishes in DMEM with 10% FBS, serum starved overnight and treated with Perifosine or EGF at concentrations indicated in each figure. Whole-cell lysates were prepared by scraping PBS-washed cells in denaturing conditions in modified RIPA lysis buffer (50mM Tris–HCl, 150mM NaCl, 1mM EDTA, 1% Triton X-100, 1% NaDoc, 1% SDS, 1%NP40) supplemented with protease inhibitor cocktail (Roche). IP assays were performed using 200 µg of lysate. Immunoblot was performed as described above.

Immunofluorescence analysis and confocal imagingCells were seeded on glass uncoated coverslips (24mm, #1.5) in DMEM with 10% FBS for 24h, serum starved overnight, stimulated with Perifosine or EGF at concentrations indicated in each figure, washed with PBS, fixed with 4% paraformaldehyde in PEM buffer and permeabilized and blocked as previously described 159 and stained for endogenous EGFR. Cells were imaged using a TCS SP8 confocal from Leica Microsystems.

Flow cytometryFor surface EGFR analysis, cells were seeded in 10cm dishes in DMEM with 10% FBS. At 24 h after plating, cells were trypsinized and washed three times with cold PBS and blocked

103


5

in PBS with 5% BSA. Cells were incubated with anti-EGFR (Ab-1) antibody in washing buffer for 30 min on ice, washed twice, incubated with secondary antibody (Alexa Fluor 488-conjugated anti-mouse-IgG antibody) for 30 min on ice and washed twice again. Fluorescence measurements and data analysis were performed using BD FACSCalibur and CD CellQuest Pro software, respectively.

GSDIM Super-Resolution Imaging HeLa and A431 cells were seeded on uncoated 24 mm coverslips (#1.5). Cells were serum starved overnight, stimulated with EGF or Perifosine at the concentrations indicated in the each figure for 2 minutes and washed briefly with DPBS and fixed as described before 159. Briefly, cells were fixed in 4% PFA in PEM buffer, extracted in 0.5% Triton/PBS and blocked in 5% BSA/PBS. Inmunofluorescence was performed at room temperature with anti-EGFR antibody coupled with Alexa Fluor 647, home labeled with a kit (Alexa Fluor 647 Monoclonal Antibody labelling kit, A-20186 from Invitrogen). Imaging was carried out on a Leica SR-GSD 3D microscope. Images were taken in TIRF mode at 100 frames per second and 10,000 frames were collected. Images were taken in a buffer containing an oxygen scavenging system (10% Glucose, 0.5 mg/ml Glucose Oxidase, 40 μg/ml Catalase) and 100 mM MEA. Data obtained this way was corrected for structured background 144 and events were localized, drift corrected, filtered for events with uncertainty between 2 to 12 nm, merged (events closer than 50nm and up to one frame of separation were merged) with the Thunderstorm 137 plugin for ImageJ. Images were rendered with 20 nm pixel size, using the Histogram visualization option.

Clustering AnalysisWe used Python (x,y) 2.7.6.0 was used with following libraries: matplotlib 1.3.1-4, numpy 1.9.0-5, scipy 0.13.3-6, scikit-learn 0.17.1 and spyder 2.2.5-10 development environment.Images of EGFR antibody labeled with Alexa Fluor 647 and sparsely immobilized on a surface of a coverslip were analyzed with Density-Based Spatial Clustering Applications with Noise (DBSCAN) clustering algorithm as implemented with scikit-learn 223 . Minimal number of event within a cluster was 3 and maximum distance to the core sample was 50 nm. Clusters were aligned by their respective centres of mass and Gauss or Cauchy distributions were fitted to x and y positions of all analyzed events. For a Gauss distribution:

220 2/)(

21)( σ

σπxxexf −−= (1)

where σ is standard deviation.For a Cauchy distribution:

(2)

where γ is the scale parameter which equals half-width at half maximum (HWHM).Calculation of autocorrelation functions and fitting of modelsThe pair autocorrelation functions (g(r)) were calculated using Fast Fourier Transform (FFT) as described before 102. For data of antibodies immobilized on the coverslip, 9 square regions (4 µm) were selected for analysis within each image. For HeLa and A431 data 2 regions (3-4

])[()( 22

0 γπγ

+−=

xxxf

104

5

µm wide) were manually defined at the edge of each imaged cell. Pair autocorrelation functions report the probability of finding another event at a distance r from a given even in the form of:

(3)

where: * denotes a two dimensional convolution, g(r)PSF is a superresolved point-spread function, g(r)peaks is the pair auto-correlation function of detected events, g(r>0) represents the contribution to the correlation function of the distribution of molecules of interest (EGFR) and g(r=0) is a delta function with an amplitude related to the density of molecules of interest (ρ) :

ρ/1)0( ==rg (4)

Therefore, by combining eq.3 with eq.4:

PSFPSFpeaks rgrgrgrg )(*)0(/)()( >+= ρ (5)

In case of random distribution of events, the autocorrelation function is:

1/)()( += ρPSFpeaks rgrg (6)

While in the presence of clustering, the pair autocorrelation function of the distribution of molecules of interest can be approximated by an exponent:

(7)

where A is proportional to the density of molecules of interest within clusters and ξ is proportional to the cluster size.Distribution of localization events of a single protein was described as a 2D Gaussian 103,224 or Cauchy 97,225 surface. Therefore two forms of g(r)PSF were compared. For the Gauss distribution: (8)

where σ is the standard deviation of the distribution. For the Cauchy distribution:

5.122 )4()(

γπγ+

=r

rg PSF

(9)

where γ is the scale parameter (HWHM).Pair autocorrelation functions were calculated for each analyzed region and averaged. Models were fitted using standard deviation of averaged points as weights of individual points.

22 4/24

1)( σ

πσrPSF erg −=

)1(/)()( / ξρ rPSFpeaks Aergrg −++=

PSFpeaks rgrgrgrg )(*)]0()0([)( >+==

105


5

Statistical analysisFor determination of statistical significance, unpaired Student’s t-tests were performed using GraphPad Prism6 software. The indicated significance values are compared to control conditions. Box plots show median and 25th and 75th percentiles. P-values are indicated as follows: *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

106

5


Supplementary Figure 1. EGFR internalization induced by Perifosine does not depend on phosphorylation in MDA MB 468 cells. (a) Perifosine does not induce phosphorylation of EGFR. Western blot analysis of phosphorylated EGFR in MDA MB 468 cells after stimulation with 100 ng/ml EGF, 0.5 µM and 5 µM Perifosine for 5 and 15 minutes. Total cell lysates were compared using the indicated antibodies. One of three experiments that were performed with similar results is shown (b) Perifosine does not induce ubiquitination of EGFR. Immunoprecipitation of EGFR and western blot analysis of Ubiquitin in MDA MB 468 cells after stimulation with 100 ng/ml EGF, 0.5 µM and 5 µM Perifosine for 2 minutes. EGFR was immunoprecipitated from 200 µg of total cell lysates and the product was compared using the indicated antibodies. One of three experiments that were performed with similar results is shown.

totalEGFR

pEGFRY1086

Perifosine 5 µM

egf - +

2 5min 0 10 15

+ + +

- - - - -

- - - - -

- + + + +

2 50 10 15

IP: EGFR WB: UB

Perifosine 0.5 µM

egf - +

2 2min 0

-

- - +

IP: EGFRWB: EGFR

2

-

-

Perifosine 5 µM - - - +

a. b.

107


5

Supplementary Figure 2. Pair correlation analysis of SR data to determine clustering in the plasma membrane of cells (a) Multiple localizations of each fluorophore and antibody can be grouped. Example output of DBSCAN clustering of sparse sample (primary EGFR antibody labelled with Alexa Fluor 647 was immobilized on a surface of a coverslip). Localization points belonging to a single cluster (as identified by DBSCAN clustering algorithm) were visualized with circles of the same colour. Points that do not belong to any cluster (noise) were visualized as small black dots. Scale bar, 200nm. (b) Blinking points within identified clusters of a single molecule exhibit Cauchy distribution. 4097 clusters containing at least 3 blinking events were aligned by their respective centre of mass and plotted together (38461 points in total). Gaussian (red) and Cauchy (green) distribution were fitted to the distribution of the position of individual blinks on the x and y axis. (c) Sparsely distributed EGFR antibody localizations form clusters that fit a Cauchy model better than a Gauss model. Averaged autocorrelation function (black +/- SEM, n=18 analysed regions from 2 independent experiments) of sparsely distributed EGFR antibody (as in a) with fits based on Gauss (red) or Cauchy (green) distribution of blinks and random distribution of antibodies on the surface. (d-e) Importance of the selection of the right model of the distribution of blinks. Autocorrelation functions of EGFR (primary labelled antibody) distribution on the membrane of HeLa cells with fitted models random (solid line) or clustered (dashed line) distribution of molecules and Gauss distribution of individual blinks. Both models fail to correctly represent the experimental data. (e) Cauchy distribution-based random model correctly represents distribution of EGFR on the membrane of unstimulated HeLa cells (as shown in Figure 4). (f) Detection of clustering of EGFR upon EGF stimulation (2 min) in HeLa cells. In stimulated cells random model (solid line) fails to represent the data correctly, while the clustered model (dashed line) can be successfully fitted.

r [nm]

101 102 103

101

102

100

γ = 8.1 nmσ = 13.6 nm

g(r)

sparse antibodyon a coverslip

x [nm] count

coun

t

- 100 - 50 0 50 100

- 100

- 50

0

50

100

y [n

m]

0.040.00

0.04

0.00

γx = 8.3 nmσx = 23.3 nm

γy = 8.8 nmσy= 25.0 nm

CauchyGauss

distribution

101

102

100

g(r)

r [nm]

101 102


HeLa unstimulatedCauchy �ts

101

102

100

g(r)

r [nm]

101 102


HeLa unstimulatedGauss �ts

101

102

100

g(r)

r [nm]

101 102


HeLa + EGFCauchy �ts

a. b.

c. d.

e. f.

109

Chapter 6

Summarizing discussion

110

6

111

Summ

arizing discussion

6

The advent of super-resolution microscopy made it possible to surpass the diffraction limit of conventional optical microscopy. The increase in resolution that super-resolution microscopy enabled, like many of the breakthroughs in the history of science, did not happen through small continuous improvements but in one big step. From the 50s to the 80s a few concepts to improve the resolution in far-field optical microscopy were proposed, but they were not applicable or represented only a slight improvement in resolution 72. The leap came with the proposal of the use of fluorophore transitions in the early 90s with the concepts of STED and GSD 226,227. About ten years later, methods that proposed related types of on-off transitions for switching of fluorophores, known as stochastic single molecule resolution microscopy (SMLM), were published 74-77. The application of super-resolution microscopy has yielded a remarkable amount of publications, some with major impact in the understanding of cellular structures. Yet, as a relatively new technique, super-resolution microscopy is still under development and many aspects of the process of obtaining a super-resolution image can be further improved. In this thesis, we focused on the further development of stochastic SMLM and its application.

Good riddance: structured background subtraction.

In this thesis we first focused on the correction of suboptimal dye blinking. Blinking of fluorescent molecules is the basis of stochastic SMLM. Blinking properties, in particular photon yield and on-off duty cycle, are crucial for the final image outcome and vary among dyes. The brightness of a blink will determine how accurate its center can be localized. The duty cycle will determine the fraction of time a fluorophore spends in the on state 78 and therefore how many fluorophores are on at the same time 78,81,82. In samples labeled using fluorophores with high duty cycles a relatively high structured background intensity is usually present, caused by neighboring fluorophores that are on for prolonged times. Structured background (also called sample background 80 substantially reduces the localization precision and leads to a commonly seen image artifact. In Chapter 2 we show that structured background can be corrected in the original blinking movie, simply by calculating the median brightness per pixel over a window in time and subtracting it. Because bright and short “on” events clearly stand out, they are spared. This easy trick leads to blinking movies that contain only bright and short events that can be localized with higher precision. As a proof of principle, we use samples simultaneously labeled with the best known blinking dye, AlexaFluor647, and a commonly used dye that has high duty cycle in the same imaging buffer, AlexaFluor532. We show that application of a temporal median filter to AlexaFluor532 blinking movies corrects the sample background yielding images with similar quality as the ones obtained with AlexaFluor647. Furthermore, we show that this trick can also be applied to blinking movies of fluorescent proteins, which commonly have high duty cycles and are not too bright, to obtain remarkable improvements in image quality. Importantly, application of a temporal median filter expands the list of fluorophores that can be used for stochastic SMLM, which increases the possibilities for multicolor imaging.

Preserving the sample. Fixation of actin and its binding proteins.

Because of the importance of blinking, several reviews and studies have focused on fluorophore switching, and many localization algorithms have been developed that aim to improve final image quality 78,127. But even with blinking and localization optimized, the final SR image can only be as good as the preparation: image quality ultimately depends on the fidelity with which the sample is preserved during the fixation process. This aspect, which

112

6

has been well-studied in the electron microscopy field, has only recently been addressed for super-resolution microscopy 85,86. In Chapter 3 we focus on the optimization of the preservation of samples of the actin cytoskeleton and its binding proteins for super-resolution microscopy studies. We show that glutaraldehyde fixation procedures, which necessitate pre-extracting of the samples due to its low permeability, can result in loss of weakly bound or soluble proteins, such as mDia1 and WAVE2. Moreover, it is well known that glutaraldehyde can induce (small) changes in the tertiary structure of proteins, hindering epitope sites for antibodies and making immunofluorescence labeling ineffective. We show, using a commonly used anti-clathrin heavy chain (CHC) antibody, that the epitope hindering effect of glutaraldehyde detected in super-resolution images is not always obvious at lower resolution, making confocal or widefield microscopy inspection useless.

Of course, some of the adverse effects of glutaraldehyde were known to microscopy users. However, glutaraldehyde fixation preserves actin fibers better than any other procedure. How to image actin and its partners then, using fluorescence microscopy at high resolution? We show that with careful consideration of the fixation parameters, e.g. temperature, time and pH, during the fixation procedure, paraformaldehyde fixation can yield high quality super-resolution images of actin, while preserving weak binding proteins. Moreover, besides maintaining the fine actin structure, our fixation protocol is readily compatible with immunolabeling, enabling artifact free multicolor super-resolution imaging of actin and accessory proteins. Because super-resolution microscopy is a recent development, the field lacks extensive controls and experience, and therefore careful optimization of the fixation procedure is of upmost importance when new proteins are to be studied. To make a first step in the direction of specific sample preparation guidelines for super-resolution, we deposited a detailed protocol and troubleshooting table in Protocol Exchange.

Where are we and where are we going.

In this thesis we worked on optimizing sample preparation and post imaging background correction for super-resolution microscopy. Of course the quality of super-resolution microscopy experiments is determined by the sum of many parameters, each of which is under constant development. During the preparation of this thesis, several new developments for the improvement of super-resolution microscopy were published, including the improvement of photo-switching buffers to enhance blinking 84,228, buffers that are compatible with one or more dyes 83, fast switching brighter fluorescent labels 229-231, faster and/or more sensitive cameras 232, the introduction of a lens that made 3D imaging possible and remarkably easy 126, and so on. Finally, probably the most prolific area of developments for super-resolution microscopy has been the localization methods field, with several different mathematical approaches to calculate the center of blinks and several pre and post analysis tools, published in the last years (to many a few 108,110,137,147,233).

An important aspect is the size of the used labels. In a final super-resolution image, localizations are not at the center of the labeled molecules, but at the centers of the fluorophores that label them. These centers may be displaced with respect to the real position of the molecules. The displacement will vary depending on the type of labeling used. Fluorescent proteins are small 4 nm long barrels that in general are directly attached to the molecule of interest and therefore do not induce a large displacement. Also, organic fluorophores with their small size of about I nm are not a big concern. However, the linker

113

Summ

arizing discussion

6

moiety that is used for attaching the fluorophore to the molecule can be large, and this generally causes large displacements. Antibodies, the most commonly used linkers, are about 8-10 nm long and when a combination of primary and secondary antibody to label a molecule is used, this may induce a displacement of up to 20 nm 80. Many efforts have been put into developing smaller linkers between fluorophore and molecule. For example, nanobodies 234 and aptamers 235 are smaller options that bind the targeted molecules with high specificity while already reducing the displacement significantly. Other strategies to label molecules with very small linkers are Snap-tag 236, HaloTag 237, and click-chemistry 238. In essence, these are protein domains used to attach chemical fluorophores with high affinity and specificity but these approaches can induce high background and they require genetical modification of the molecule of interest.

Besides smaller linkers is also needed to design fluorophores that blink optimally for SMLM. That is, molecules that can yield a large number of photons in short times, improving the possibilities for live cell super-resolution microscopy studies; molecules that are independent of the environment for their blinking, such as fluorescent proteins in which the fluorophore is protected by the barrel; or that are linked to a reducing agent, and so on 72,127.

“There are no right answers to wrong questions”. What to use super-resolution microscopy for.

To obtain relevant information with SR microscopy, it is essential to define which questions it can answer. Stochastic SMLM should not be the first choice to study protein dynamics. Although some studies have used this type of super-resolution in live cells, there is a drawback implicit in the concept of super-resolution microscopy that makes live-cell studies challenging. The trick of switching fluorophores on and off is essential for surpassing the resolution limit, but it makes the imaging procedure very slow. Localization of fluorophores is impaired by movement of proteins and structures in living samples, and creation of the final SR image takes time. New faster CMOS cameras make the process of imaging shorter but often at the cost of photon collection, and therefore resolution is sacrificed 232. Fluorescent proteins are not as bright as small organic dyes, but the latter may be toxic for cells and usually are not too specific, yielding high background. Importantly, inducing photoswitching requires the use of high laser powers that cause phototoxicity 239.To overcome the challenge, bright fast-switching fluorescent proteins are still in development, and in the future, in combination with fast cameras will possibly make live-cell super-resolution studies possible.

Super-resolution microscopy should not be the choice for the study of large cellular structures. On the other hand, for very small structures, the still superior resolution of electron microscopy is much more suited. However, at an intermediate scale, there are many advantages in using super-resolution microscopy instead of electron microscopy. Firstly, electron microscopy studies require tedious and long sample preparation and acquisition is slow, typically yielding only a limited set of images. Super-resolution microscopy sample preparation is more accessible, and can be carried out at high throughput. Super-resolution microscopy is based on fluorescence microscopy and therefore molecules or structures of interest can be screened for and selected “by eye” before collecting the high-resolution image. Moreover, it is important that several different molecules can be imaged in the same cell. In electron microscopy, with immune-gold labeling, only two “colors” (small and large gold particles) can be discriminated. In super-resolution microscopy at least one more color can be added and large efforts are being made to separate emission of fluorescent dyes

114

6

to allow more than three color imaging. Even so, already in two color studies, the labeling density that can be achieved with super-resolution is still much higher than the one that can be achieved in electron microscopy. A very promising approach is the combination of super-resolution microscopy and electron microscopy, which allows the fluorescent tagging of molecules imaged at high resolution, which can later be resolved at Angstrom resolution. A few recent studies have shown the advantages of such promising combination 240-242.

Characterizing the intrinsic clustering component.

One application of quantitative super-resolution microscopy is the study of protein clustering, for which several methods are currently applied. The organization of entities in clusters, like plants or trees on a field, has been the subject of study for many years, and several methods to describe and quantify these organizations have been developed. Because of the achievable resolution, super-resolution microscopy prompted a series of studies revisiting membrane protein clusters and membrane composition organization, a topic long discussed in cell biology. Initially, several previously reported methods for analysis of clusters were directly applied, such as Ripley’s K 222,243,244. However, as explained earlier on in this thesis, in stochastic SMLM every protein is localized multiple times due to, first, the presence of multiple labels per antibody and secondly, multiple blinking of each label, a phenomenon we call intrinsic clustering of localizations.

In Chapter 5 we use a method that has already been employed for the analysis of protein clusters on cell membranes and adapt it to better describe our data with intrinsic clustering. To better analyze clusters in super-resolution data, Sengupta and colleagues 103 developed a model that aims to factor out the intrinsic clustering component in pair correlation analysis of super-resolution data. However, in this model, Sengupta and colleagues assume that intrinsic clustering adds a Gaussian distribution term to the analysis. In our experimental data, we found that the cluster formed by localizations of a single antibody is Cauchy distributed. A similar observation was also reported in an unrelated study 245. Furthermore, we show that using the wrong assumption on the distribution may lead to overestimation of protein clustering. Unlike the clustering methods used in many published SR studies that ignore intrinsic clustering, our modified model is especially suited for cluster analysis of stochastic SMLM data.

Helping understand endocytosis with super-resolution microscopy.

In this thesis, in Chapters 4 and 5 we apply our improved super-resolution microscopy to study the role of actin in clathrin mediated endocytosis (CME) and the state in which the epidermal growth factor receptor (EGFR) exists on the plasma membrane of different cells.Endocytosis controls the composition of lipids and proteins of the plasma membrane, and thus the manner in which cells respond to their environment. Endocytosis is a well-orchestrated process, and many proteins with different functions contribute at different points in time. Given the dynamic nature and small size of endocytic structures, a combination of normal resolution live cell microscopy and super resolution fixed cell microscopy provides the best tools to understanding how they are organized in space and time.

In Chapter 4 we study clathrin mediated endocytosis, the endocytosis process that uses clathrin as a coat protein to cover endocytic vesicles. Using super-resolution microscopy we explore the difference between flat clathrin plaques (FCP) and clathrin coated pits (CCP)

115

Summ

arizing discussion

6

and show that FCPs function as hubs for the formation of dynamic endocytic clathrin coated vesicle. Moreover, we show that the formation of these vesicles is controlled by N-WASP and Arp2/3 complex mediated actin polymerization. In this study we also map, for the first time, the N-WASP domains involved in the differential recruitment of the protein to CCPs and FCPs.

We found that actin controlled FCPs recruit the membrane receptors EGFR and LPAR1, and that they can be crucial in the regulation of signaling through these receptors. In cells in which actin polymerization was disrupted by means of N-WASP knockdown, internalization of both LPAR1 and EGFR was impaired. However, we found that only upon addition of LPA, and not EGF, stimulation of Akt signaling was enhanced, in an LPAR1-dependent manner. Are flat clathrin plaques signaling platforms?

The proteins that constitute a signaling pathway need to come in close proximity in order to interact. Their localization and availability is crucial for the signal propagation and intensity 246. Activated receptors and their signaling molecules may be sequestered together in clathrin coated structures (CCS). Recent studies have proposed that CCS are not only endocytic domains but also signaling platforms that are specialized in different pathways 247,248. Garay and colleagues 248 proposed that clathrin scaffolds are plasma membrane signaling microdomains required for certain signaling receptors. They showed that upon genetic and biochemical ablation of clathrin, the signaling of Akt was reduced upon stimulation with EGF. Because Akt signal was not affected when endocytosis was blocked by other means, such as inhibition of Dynamin II , they concluded that clathrin itself has a direct role in regulating Akt activation via EGFR. Such regulation of Akt by clathrin was independent of its endoctytic role. Along the same lines, Eichel and colleagues 247 showed that upon triggering of the B1AR receptor, activated β-Arrestin dissociates from the receptor and travels to CCS where it signals so as to activate MAPK, until vesicle scission occurs. Although β-Arrestin is known to activate MAPK signaling from endosomes 1, this study shows for the first that it can also signal to activate MAPK from CCS.

Our results also indicate that in actin-controlled FCPs, specialized LPAR1 to Akt signal transduction takes place. It has been shown using PIP2 probes 173 that there is a high concentration of PIP2 in sites of CCV formation that diminishes after fission 246. For the Akt pathway to be activated, activated PI3K needs to come in proximity with its substrate and in the same way PDKI and Akt need to be recruited together to specific sites 246. In fact, CCSs can exhibit PI3K activity 249. When endocytosis is slowed down due to reduced actin polymerization, this can give these proteins more time to interact in the signaling platforms, thereby strengthening the signal.

The view that CCS may function as signaling platforms is very recent. Much work needs to be done to understand how the different signaling CCS are regulated and what drives specific localization of different signaling platforms to different CCS.

Inducing endocytosis of overexpressed EGFR with Perifosine.

Aberrant receptor signaling is common in many types of cancer. A common clinical approach for regulating aberrant receptor signaling is the induction of its endocytosis. In Chapter 5 we investigate how Perifosine, a synthetic lipid drug, regulates EGFR signaling.

116

6

Perifosine induces the endocytosis of EGFR in a non-conventional mechanism that doesn’t depend on its activation. The endocytosed EGFR is not phosphorylated and it also doesn’t get ubiquitinated. However, Perifosine only exerts this effect in cells that overexpress EGFR. To investigate possible differences between cells that do and do not overexpress EGFR, we used our improved clustering analysis method. We show that EGFR is pre-clustered in cells with high expression and not in cells with low expression, even in the resting state. While these results are preliminary, we hypothesize that the presence of pre-clustered EGFR is a pre-requisite for Perifosine to exert it’s effect. One possibility is that Perifosine, as a non-natural lipid, disrupts the physicochemical properties of the membrane thereby inducing the endocytosis of EGFR. Stimulation of endocytosis and degradation of ErbBs is an attractive idea to inhibit tumor growth, especially in the cases in which the receptors are overexpressed.

In summary, in this thesis we have contributed towards the further development of important aspects of super-resolution microscopy and we have applied improved super-resolution microscopy to study the regulation of signaling and endocytosis of important membrane receptors.

117

Summ

arizing discussion

6

119

Addendum

120

&

121

Addendum

&

References1 Sorkin, A. & von Zastrow, M. Endocytosis and signalling: intertwining molecular networks.

Nature reviews. Molecular cell biology 10, 609-622, doi:10.1038/nrm2748 (2009).2 Lemmon, M. A. & Schlessinger, J. Cell signaling by receptor tyrosine kinases. Cell 141, 1117-

1134, doi:10.1016/j.cell.2010.06.011 (2010).3 Hubbard, S. R. & Miller, W. T. Receptor tyrosine kinases: mechanisms of activation and

signaling. Current opinion in cell biology 19, 117-123, doi:10.1016/j.ceb.2007.02.010 (2007).

4 Maruyama, I. N. Mechanisms of activation of receptor tyrosine kinases: monomers or dimers. Cells 3, 304-330, doi:10.3390/cells3020304 (2014).

5 Tomas, A., Futter, C. E. & Eden, E. R. EGF receptor trafficking: consequences for signaling and cancer. Trends in cell biology 24, 26-34, doi:10.1016/j.tcb.2013.11.002 (2014).

6 Seshacharyulu, P. et al. Tagreting the EGFR signaling pathway in cancer therapy. (2012).7 Macdonald-Obermann, J. L. & Pike, L. J. Different epidermal growth factor (EGF)

receptor ligands show distinct kinetics and biased or partial agonism for homodimer and heterodimer formation. The Journal of biological chemistry 289, 26178-26188, doi:10.1074/jbc.M114.586826 (2014).

8 Henriksen, L., Grandal, M. V., Knudsen, S. L., van Deurs, B. & Grovdal, L. M. Internalization mechanisms of the epidermal growth factor receptor after activation with different ligands. PloS one 8, e58148, doi:10.1371/journal.pone.0058148 (2013).

9 Wilson, K. J. et al. EGFR ligands exhibit functional differences in models of paracrine and autocrine signaling. Growth factors 30, 107-116, doi:10.3109/08977194.2011.649918 (2012).

10 Saito, T. et al. Differential activation of epidermal growth factor (EGF) receptor downstream signaling pathways by betacellulin and EGF. Endocrinology 145, 4232-4243, doi:10.1210/en.2004-0401 (2004).

11 Wheeler, D. L., Dunn, E. F. & Harari, P. M. Understanding resistance to EGFR inhibitors-impact on future treatment strategies. Nature reviews. Clinical oncology 7, 493-507, doi:10.1038/nrclinonc.2010.97 (2010).

12 Wolfe, B. L. & Trejo, J. Clathrin-dependent mechanisms of G protein-coupled receptor endocytosis. Traffic 8, 462-470, doi:10.1111/j.1600-0854.2007.00551.x (2007).

13 Kobilka, B. K. G protein coupled receptor structure and activation. Biochimica et biophysica acta 1768, 794-807, doi:10.1016/j.bbamem.2006.10.021 (2007).

14 Venkatakrishnan, A. J. et al. Molecular signatures of G-protein-coupled receptors. Nature 494, 185-194, doi:10.1038/nature11896 (2013).

15 Jalink, K. & Moolenaar, W. H. G protein-coupled receptors: the inside story. BioEssays : news and reviews in molecular, cellular and developmental biology 32, 13-16, doi:10.1002/bies.200900153 (2010).

16 Bar-Shavit, R. et al. G Protein-Coupled Receptors in Cancer. International journal of molecular sciences 17, doi:10.3390/ijms17081320 (2016).

17 Hanlon, C. D. & Andrew, D. J. Outside-in signaling--a brief review of GPCR signaling with a focus on the Drosophila GPCR family. Journal of cell science 128, 3533-3542, doi:10.1242/jcs.175158 (2015).

18 Liu, S., Murph, M., Panupinthu, N. & Mills, G. B. ATX-LPA receptor axis in inflammation and cancer. Cell Cycle 8, 3695-3701, doi:10.4161/cc.8.22.9937 (2009).

19 Noguchi, K., Herr, D., Mutoh, T. & Chun, J. Lysophosphatidic acid (LPA) and its receptors. Current opinion in pharmacology 9, 15-23, doi:10.1016/j.coph.2008.11.010 (2009).

20 Urs, N. M. et al. A requirement for membrane cholesterol in the beta-arrestin- and clathrin-dependent endocytosis of LPA1 lysophosphatidic acid receptors. Journal of cell science 118, 5291-5304, doi:10.1242/jcs.02634 (2005).

21 Urs, N. M., Kowalczyk, A. P. & Radhakrishna, H. Different mechanisms regulate lysophosphatidic acid (LPA)-dependent versus phorbol ester-dependent internalization of the LPA1 receptor. The Journal of biological chemistry 283, 5249-5257, doi:10.1074/jbc.M710003200 (2008).

22 Lappano, R. & Maggiolini, M. GPCRs and cancer. Acta pharmacologica Sinica 33, 351-362,

122

&

doi:10.1038/aps.2011.183 (2012).23 Lappano, R. & Maggiolini, M. G protein-coupled receptors: novel targets for drug discovery

in cancer. Nature reviews. Drug discovery 10, 47-60, doi:10.1038/nrd3320 (2011).24 Roepstorff, K., Grovdal, L., Grandal, M., Lerdrup, M. & van Deurs, B. Endocytic

downregulation of ErbB receptors: mechanisms and relevance in cancer. Histochemistry and cell biology 129, 563-578, doi:10.1007/s00418-008-0401-3 (2008).

25 Tsvetanova, N. G. & von Zastrow, M. Spatial encoding of cyclic AMP signaling specificity by GPCR endocytosis. Nature chemical biology 10, 1061-1065, doi:10.1038/nchembio.1665 (2014).

26 Doherty, G. J. & McMahon, H. T. Mechanisms of endocytosis. Annual review of biochemistry 78, 857-902, doi:10.1146/annurev.biochem.78.081307.110540 (2009).

27 Irannejad, R. & von Zastrow, M. GPCR signaling along the endocytic pathway. Current opinion in cell biology 27, 109-116, doi:10.1016/j.ceb.2013.10.003 (2014).

28 Sorkin, A. & Goh, L. K. Endocytosis and intracellular trafficking of ErbBs. Experimental cell research 315, 683-696, doi:10.1016/j.yexcr.2008.07.029 (2009).

29 Sigismund, S. et al. Clathrin-independent endocytosis of ubiquitinated cargos. Proceedings of the National Academy of Sciences of the United States of America 102, 2760-2765, doi:10.1073/pnas.0409817102 (2005).

30 Sigismund, S. et al. Threshold-controlled ubiquitination of the EGFR directs receptor fate. The EMBO journal 32, 2140-2157, doi:10.1038/emboj.2013.149 (2013).

31 Pearse, B. M. Coated vesicles from pig brain: purification and biochemical characterization. Journal of molecular biology 97, 93-98 (1975).

32 Kirchhausen, T. Imaging endocytic clathrin structures in living cells. Trends in cell biology 19, 596-605, doi:10.1016/j.tcb.2009.09.002 (2009).

33 Lampe, M., Vassilopoulos, S. & Merrifield, C. Clathrin coated pits, plaques and adhesion. Journal of structural biology, doi:10.1016/j.jsb.2016.07.009 (2016).

34 McMahon, H. T. & Boucrot, E. Molecular mechanism and physiological functions of clathrin-mediated endocytosis. Nature reviews. Molecular cell biology 12, 517-533, doi:10.1038/nrm3151 (2011).

35 Lampe, M., Pierre, F., Al-Sabah, S., Krasel, C. & Merrifield, C. J. Dual single-scission event analysis of constitutive transferrin receptor (TfR) endocytosis and ligand-triggered beta2-adrenergic receptor (beta2AR) or Mu-opioid receptor (MOR) endocytosis. Molecular biology of the cell 25, 3070-3080, doi:10.1091/mbc.E14-06-1112 (2014).

36 Goh, L. K. & Sorkin, A. Endocytosis of receptor tyrosine kinases. Cold Spring Harbor perspectives in biology 5, a017459, doi:10.1101/cshperspect.a017459 (2013).

37 Huang, F. & Sorkin, A. Growth factor receptor binding protein 2-mediated recruitment of the RING domain of Cbl to the epidermal growth factor receptor is essential and sufficient to support receptor endocytosis. Molecular biology of the cell 16, 1268-1281, doi:10.1091/mbc.E04-09-0832 (2005).

38 Sigismund, S. et al. Clathrin-mediated internalization is essential for sustained EGFR signaling but dispensable for degradation. Developmental cell 15, 209-219, doi:10.1016/j.devcel.2008.06.012 (2008).

39 Sorkin, A., McKinsey, T., Shih, W., Kirchhausen, T. & Carpenter, G. Stoichiometric interaction of the epidermal growth factor receptor with the clathrin-associated protein complex AP-2. The Journal of biological chemistry 270, 619-625 (1995).

40 Nesterov, A., Kurten, R. C. & Gill, G. N. Association of epidermal growth factor receptors with coated pit adaptins via a tyrosine phosphorylation-regulated mechanism. The Journal of biological chemistry 270, 6320-6327 (1995).

41 Tsvetanova, N. G., Irannejad, R. & von Zastrow, M. G protein-coupled receptor (GPCR) signaling via heterotrimeric G proteins from endosomes. The Journal of biological chemistry 290, 6689-6696, doi:10.1074/jbc.R114.617951 (2015).

42 Varsano, T., Taupin, V., Guo, L., Baterina, O. Y., Jr. & Farquhar, M. G. The PDZ protein GIPC regulates trafficking of the LPA1 receptor from APPL signaling endosomes and attenuates the cell’s response to LPA. PloS one 7, e49227, doi:10.1371/journal.pone.0049227 (2012).

43 Urbanska, A., Sadowski, L., Kalaidzidis, Y. & Miaczynska, M. Biochemical characterization of APPL endosomes: the role of annexin A2 in APPL membrane recruitment. Traffic 12, 1227-1241, doi:10.1111/j.1600-0854.2011.01226.x (2011).

00418-008-0401-3

123

Addendum

&

44 Maupin, P. & Pollard, T. D. Improved preservation and staining of HeLa cell actin filaments, clathrin-coated membranes, and other cytoplasmic structures by tannic acid-glutaraldehyde-saponin fixation. The Journal of cell biology 96, 51-62 (1983).

45 Heuser, J. Three-dimensional visualization of coated vesicle formation in fibroblasts. The Journal of cell biology 84, 560-583 (1980).

46 Merrifield, C. J., Perrais, D. & Zenisek, D. Coupling between clathrin-coated-pit invagination, cortactin recruitment, and membrane scission observed in live cells. Cell 121, 593-606, doi:10.1016/j.cell.2005.03.015 (2005).

47 Saffarian, S., Cocucci, E. & Kirchhausen, T. Distinct dynamics of endocytic clathrin-coated pits and coated plaques. PLoS biology 7, e1000191, doi:10.1371/journal.pbio.1000191 (2009).

48 Grove, J. et al. Flat clathrin lattices: stable features of the plasma membrane. Molecular biology of the cell 25, 3581-3594, doi:10.1091/mbc.E14-06-1154 (2014).

49 Batchelder, E. M. & Yarar, D. Differential requirements for clathrin-dependent endocytosis at sites of cell-substrate adhesion. Molecular biology of the cell 21, 3070-3079, doi:10.1091/mbc.E09-12-1044 (2010).

50 Fujimoto, L. M., Roth, R., Heuser, J. E. & Schmid, S. L. Actin assembly plays a variable, but not obligatory role in receptor-mediated endocytosis in mammalian cells. Traffic 1, 161-171 (2000).

51 Taylor, M. J., Perrais, D. & Merrifield, C. J. A high precision survey of the molecular dynamics of mammalian clathrin-mediated endocytosis. PLoS biology 9, e1000604, doi:10.1371/journal.pbio.1000604 (2011).

52 den Otter, W. K. & Briels, W. J. The generation of curved clathrin coats from flat plaques. Traffic 12, 1407-1416, doi:10.1111/j.1600-0854.2011.01241.x (2011).

53 Heuser, J. E., Keen, J. H., Amende, L. M., Lippoldt, R. E. & Prasad, K. Deep-etch visualization of 27S clathrin: a tetrahedral tetramer. The Journal of cell biology 105, 1999-2009 (1987).

54 Avinoam, O., Schorb, M., Beese, C. J., Briggs, J. A. & Kaksonen, M. ENDOCYTOSIS. Endocytic sites mature by continuous bending and remodeling of the clathrin coat. Science 348, 1369-1372, doi:10.1126/science.aaa9555 (2015).

55 Loebrich, S. The role of F-actin in modulating Clathrin-mediated endocytosis: Lessons from neurons in health and neuropsychiatric disorder. Communicative & integrative biology 7, e28740, doi:10.4161/cib.28740 (2014).

56 Merrifield, C. J. Seeing is believing: imaging actin dynamics at single sites of endocytosis. Trends in cell biology 14, 352-358, doi:10.1016/j.tcb.2004.05.008 (2004).

57 Boulant, S., Kural, C., Zeeh, J. C., Ubelmann, F. & Kirchhausen, T. Actin dynamics counteract membrane tension during clathrin-mediated endocytosis. Nature cell biology 13, 1124-1131, doi:10.1038/ncb2307 (2011).

58 Mooren, O. L., Galletta, B. J. & Cooper, J. A. Roles for actin assembly in endocytosis. Annual review of biochemistry 81, 661-686, doi:10.1146/annurev-biochem-060910-094416 (2012).

59 Kaksonen, M., Toret, C. P. & Drubin, D. G. Harnessing actin dynamics for clathrin-mediated endocytosis. Nature reviews. Molecular cell biology 7, 404-414, doi:10.1038/nrm1940 (2006).

60 Morlot, S. et al. Membrane shape at the edge of the dynamin helix sets location and duration of the fission reaction. Cell 151, 619-629, doi:10.1016/j.cell.2012.09.017 (2012).

61 Galovic, M., Xu, D., Areces, L. B., van der Kammen, R. & Innocenti, M. Interplay between N-WASP and CK2 optimizes clathrin-mediated endocytosis of EGFR. Journal of cell science 124, 2001-2012, doi:10.1242/jcs.081182 (2011).

62 Yarar, D., Waterman-Storer, C. M. & Schmid, S. L. A dynamic actin cytoskeleton functions at multiple stages of clathrin-mediated endocytosis. Molecular biology of the cell 16, 964-975, doi:10.1091/mbc.E04-09-0774 (2005).

63 Merrifield, C. J., Qualmann, B., Kessels, M. M. & Almers, W. Neural Wiskott Aldrich Syndrome Protein (N-WASP) and the Arp2/3 complex are recruited to sites of clathrin-mediated endocytosis in cultured fibroblasts. European journal of cell biology 83, 13-18, doi:10.1078/0171-9335-00356 (2004).

64 McNiven, M. A. et al. Regulated interactions between dynamin and the actin-binding protein cortactin modulate cell shape. The Journal of cell biology 151, 187-198 (2000).

124

&

65 Humphries, A. C., Donnelly, S. K. & Way, M. Cdc42 and the Rho GEF intersectin-1 collaborate with Nck to promote N-WASP-dependent actin polymerisation. Journal of cell science 127, 673-685, doi:10.1242/jcs.141366 (2014).

66 Helgeson, L. A. & Nolen, B. J. Mechanism of synergistic activation of Arp2/3 complex by cortactin and N-WASP. eLife 2, e00884, doi:10.7554/eLife.00884 (2013).

67 Rotty, J. D., Wu, C. & Bear, J. E. New insights into the regulation and cellular functions of the ARP2/3 complex. Nature reviews. Molecular cell biology 14, 7-12, doi:10.1038/nrm3492 (2013).

68 Isogai, T. et al. Initiation of lamellipodia and ruffles involves cooperation between mDia1 and the Arp2/3 complex. Journal of cell science 128, 3796-3810, doi:10.1242/jcs.176768 (2015).

69 Goley, E. D. & Welch, M. D. The ARP2/3 complex: an actin nucleator comes of age. Nature reviews. Molecular cell biology 7, 713-726, doi:10.1038/nrm2026 (2006).

70 Innocenti, M. et al. Abi1 regulates the activity of N-WASP and WAVE in distinct actin-based processes. Nature cell biology 7, 969-976, doi:10.1038/ncb1304 (2005).

71 Takenawa, T. & Suetsugu, S. The WASP-WAVE protein network: connecting the membrane to the cytoskeleton. Nature reviews. Molecular cell biology 8, 37-48, doi:10.1038/nrm2069 (2007).

72 Hell, S. W. Microscopy and its focal switch. Nature methods 6, 24-32, doi:10.1038/nmeth.1291 (2009).

73 Klar, T. A., Jakobs, S., Dyba, M., Egner, A. & Hell, S. W. Fluorescence microscopy with diffraction resolution barrier broken by stimulated emission. Proceedings of the National Academy of Sciences of the United States of America 97, 8206-8210 (2000).

74 Betzig, E. et al. Imaging intracellular fluorescent proteins at nanometer resolution. Science 313, 1642-1645, doi:10.1126/science.1127344 (2006).

75 Rust, M. J., Bates, M. & Zhuang, X. Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM). Nature methods 3, 793-795, doi:10.1038/nmeth929 (2006).

76 Heilemann, M. et al. Subdiffraction-resolution fluorescence imaging with conventional fluorescent probes. Angewandte Chemie 47, 6172-6176, doi:10.1002/anie.200802376 (2008).

77 Folling, J. et al. Fluorescence nanoscopy by ground-state depletion and single-molecule return. Nature methods 5, 943-945, doi:10.1038/nmeth.1257 (2008).

78 Dempsey, G. T., Vaughan, J. C., Chen, K. H., Bates, M. & Zhuang, X. Evaluation of fluorophores for optimal performance in localization-based super-resolution imaging. Nature methods 8, 1027-1036, doi:10.1038/nmeth.1768 (2011).

79 Small, A. & Stahlheber, S. Fluorophore localization algorithms for super-resolution microscopy. Nature methods 11, 267-279, doi:10.1038/nmeth.2844 (2014).

80 Deschout, H. et al. Precisely and accurately localizing single emitters in fluorescence microscopy. Nature methods 11, 253-266, doi:10.1038/nmeth.2843 (2014).

81 van de Linde, S., Wolter, S., Heilemann, M. & Sauer, M. The effect of photoswitching kinetics and labeling densities on super-resolution fluorescence imaging. Journal of biotechnology 149, 260-266, doi:10.1016/j.jbiotec.2010.02.010 (2010).

82 Vogelsang, J. et al. Make them blink: probes for super-resolution microscopy. Chemphyschem : a European journal of chemical physics and physical chemistry 11, 2475-2490, doi:10.1002/cphc.201000189 (2010).

83 Nahidiazar, L., Agronskaia, A. V., Broertjes, J., van den Broek, B. & Jalink, K. Optimizing Imaging Conditions for Demanding Multi-Color Super Resolution Localization Microscopy. PloS one 11, e0158884, doi:10.1371/journal.pone.0158884 (2016).

84 Lee, S. F., Verolet, Q. & Furstenberg, A. Improved super-resolution microscopy with oxazine fluorophores in heavy water. Angewandte Chemie 52, 8948-8951, doi:10.1002/anie.201302341 (2013).

85 Whelan, D. R. & Bell, T. D. Image artifacts in single molecule localization microscopy: why optimization of sample preparation protocols matters. Scientific reports 5, 7924, doi:10.1038/srep07924 (2015).

86 Bachmann, M., Fiederling, F. & Bastmeyer, M. Practical limitations of superresolution imaging due to conventional sample preparation revealed by a direct comparison of CLSM,

125

Addendum

&

SIM and dSTORM. Journal of microscopy, doi:10.1111/jmi.12365 (2015).87 Szymborska, A. et al. Nuclear pore scaffold structure analyzed by super-resolution

microscopy and particle averaging. Science 341, 655-658, doi:10.1126/science.1240672 (2013).

88 van den Dries, K. et al. Dual-color superresolution microscopy reveals nanoscale organization of mechanosensory podosomes. Molecular biology of the cell 24, 2112-2123, doi:10.1091/mbc.E12-12-0856 (2013).

89 Van Engelenburg, S. B. et al. Distribution of ESCRT machinery at HIV assembly sites reveals virus scaffolding of ESCRT subunits. Science 343, 653-656, doi:10.1126/science.1247786 (2014).

90 Xu, K., Zhong, G. & Zhuang, X. Actin, spectrin, and associated proteins form a periodic cytoskeletal structure in axons. Science 339, 452-456, doi:10.1126/science.1232251 (2013).

91 Malkusch, S. et al. Coordinate-based colocalization analysis of single-molecule localization microscopy data. Histochemistry and cell biology 137, 1-10, doi:10.1007/s00418-011-0880-5 (2012).

92 Tarancon Diez, L. et al. Coordinate-based co-localization-mediated analysis of arrestin clustering upon stimulation of the C-C chemokine receptor 5 with RANTES/CCL5 analogues. Histochemistry and cell biology 142, 69-77, doi:10.1007/s00418-014-1206-1 (2014).

93 Sydor, A. M., Czymmek, K. J., Puchner, E. M. & Mennella, V. Super-Resolution Microscopy: From Single Molecules to Supramolecular Assemblies. Trends in cell biology 25, 730-748, doi:10.1016/j.tcb.2015.10.004 (2015).

94 Monks, C. R., Freiberg, B. A., Kupfer, H., Sciaky, N. & Kupfer, A. Three-dimensional segregation of supramolecular activation clusters in T cells. Nature 395, 82-86, doi:10.1038/25764 (1998).

95 Palczewski, K. Oligomeric forms of G protein-coupled receptors (GPCRs). Trends in biochemical sciences 35, 595-600, doi:10.1016/j.tibs.2010.05.002 (2010).

96 Nan, X. et al. Single-molecule superresolution imaging allows quantitative analysis of RAF multimer formation and signaling. Proceedings of the National Academy of Sciences of the United States of America 110, 18519-18524, doi:10.1073/pnas.1318188110 (2013).

97 Pertsinidis, A. et al. Ultrahigh-resolution imaging reveals formation of neuronal SNARE/Munc18 complexes in situ. Proceedings of the National Academy of Sciences of the United States of America 110, E2812-2820, doi:10.1073/pnas.1310654110 (2013).

98 Notelaers, K. et al. Membrane distribution of the glycine receptor alpha3 studied by optical super-resolution microscopy. Histochemistry and cell biology 142, 79-90, doi:10.1007/s00418-014-1197-y (2014).

99 Zhang, J., Carver, C. M., Choveau, F. S. & Shapiro, M. S. Clustering and Functional Coupling of Diverse Ion Channels and Signaling Proteins Revealed by Super-resolution STORM Microscopy in Neurons. Neuron 92, 461-478, doi:10.1016/j.neuron.2016.09.014 (2016).

100 Truan, Z. et al. Quantitative morphological analysis of arrestin2 clustering upon G protein-coupled receptor stimulation by super-resolution microscopy. Journal of structural biology 184, 329-334, doi:10.1016/j.jsb.2013.09.019 (2013).

101 Oszmiana, A. et al. The Size of Activating and Inhibitory Killer Ig-like Receptor Nanoclusters Is Controlled by the Transmembrane Sequence and Affects Signaling. Cell reports 15, 1957-1972, doi:10.1016/j.celrep.2016.04.075 (2016).

102 Veatch, S. L. et al. Correlation functions quantify super-resolution images and estimate apparent clustering due to over-counting. PloS one 7, e31457, doi:10.1371/journal.pone.0031457 (2012).

103 Sengupta, P., Jovanovic-Talisman, T. & Lippincott-Schwartz, J. Quantifying spatial organization in point-localization superresolution images using pair correlation analysis. Nature protocols 8, 345-354, doi:10.1038/nprot.2013.005 (2013).

104 Gould, T., Hess, S. & Bewersdorf, J. Optical nanoscopy: from acquisition to analysis. Annual Review of Biomedical Engineering 14, 231-254, doi:papers://E34C3E74-68C2-4C19-8BBC-5B4A4BEEBBFB/Paper/p3117 (2012).

105 Ewers, H. Nano Resolution Optical Imaging Through Localization Microscopy. Cellular Imaging Techniques for Neuroscience and Beyond, 1-20, doi:10.1016/b978-0-12-385872-6.00004-0 (2012).

106 Wolter, S. et al. rapidSTORM: accurate, fast open-source software for localization

00418-014-1206-1

126

&

microscopy. Nature methods 9, 1040-1041, doi:10.1038/nmeth.2224 (2012).107 Henriques, R. et al. QuickPALM: 3D real-time photoactivation nanoscopy image processing

in ImageJ. Nature methods 7, 339-340, doi:10.1038/nmeth0510-339 (2010).108 Mukamel, E. A., Babcock, H. & Zhuang, X. Statistical deconvolution for superresolution

fluorescence microscopy. Biophysical journal 102, 2391-2400, doi:10.1016/j.bpj.2012.03.070 (2012).

109 Cheezum, M. K., Walker, W. F. & Guilford, W. H. Quantitative comparison of algorithms for tracking single fluorescent particles. Biophysical journal 81, 2378-2388, doi:10.1016/S0006-3495(01)75884-5 (2001).

110 Wolter, S. et al. Real-time computation of subdiffraction-resolution fluorescence images. Journal of microscopy 237, 12-22, doi:10.1111/j.1365-2818.2009.03287.x (2010).

111 Zhu, L., Zhang, W., Elnatan, D. & Huang, B. Faster STORM using compressed sensing. Nature methods 9, 721-723, doi:10.1038/nmeth.1978 (2012).

112 Hess, S. T., Gould, T. J., Gunewardene, M., Bewersdorf, J. & Mason, M. D. Ultrahigh resolution imaging of biomolecules by fluorescence photoactivation localization microscopy. Methods in molecular biology 544, 483-522, doi:10.1007/978-1-59745-483-4_32 (2009).

113 Piccardi, M. Background subtraction techniques: a review. 2004 IEEE International Conference on Systems, Man and Cybernetics (2004).

114 Baddeley, D., Jayasinghe, I. D., Cremer, C., Cannell, M. B. & Soeller, C. Light-induced dark states of organic fluochromes enable 30 nm resolution imaging in standard media. Biophysical journal 96, L22-24, doi:10.1016/j.bpj.2008.11.002 (2009).

115 Hess, S. T. et al. Dynamic clustered distribution of hemagglutinin resolved at 40 nm in living cell membranes discriminates between raft theories. Proceedings of the National Academy of Sciences of the United States of America 104, 17370-17375, doi:10.1073/pnas.0708066104 (2007).

116 Riedl, J. et al. Lifeact: a versatile marker to visualize F-actin. Nature methods 5, 605-607, doi:10.1038/nmeth.1220 (2008).

117 Zhang, X. et al. Rational design of a tetrameric protein to enhance interactions between self-assembled fibers gives molecular hydrogels. Angewandte Chemie 51, 4388-4392, doi:10.1002/anie.201108612 (2012).

118 Thompson, R. E., Larson, D. R. & Webb, W. W. Precise nanometer localization analysis for individual fluorescent probes. Biophysical journal 82, 2775-2783, doi:10.1016/S0006-3495(02)75618-X (2002).

119 Cremer, C. et al. Superresolution imaging of biological nanostructures by spectral precision distance microscopy. Biotechnology journal 6, 1037-1051, doi:10.1002/biot.201100031 (2011).

120 Juhola, M., Katajainen, J. & Raita, T. Comparison of algorithms for standard median filtering. Signal Processing (1991).

121 Wang, G. & Smith, S. J. Sub-diffraction limit localization of proteins in volumetric space using Bayesian restoration of fluorescence images from ultrathin specimens. PLoS computational biology 8, e1002671, doi:10.1371/journal.pcbi.1002671 (2012).

122 Berberoglu, M. Precocious puberty and normal variant puberty: definition, etiology, diagnosis and current management. Journal of clinical research in pediatric endocrinology 1, 164-174, doi:10.4008/jcrpe.v1i4.3 (2009).

123 Ahmed, S. Nanoscopy of cell architecture: The actin-membrane interface. Bioarchitecture 1, 32-38, doi:10.4161/bioa.1.1.14799 (2011).

124 Gould, T. J., Verkhusha, V. V. & Hess, S. T. Imaging biological structures with fluorescence photoactivation localization microscopy. Nature protocols 4, 291-308, doi:10.1038/nprot.2008.246 (2009).

125 van de Linde, S. et al. Direct stochastic optical reconstruction microscopy with standard fluorescent probes. Nature protocols 6, 991-1009, doi:10.1038/nprot.2011.336 (2011).

126 Huang, B., Wang, W., Bates, M. & Zhuang, X. Three-dimensional super-resolution imaging by stochastic optical reconstruction microscopy. Science 319, 810-813, doi:10.1126/science.1153529 (2008).

127 Endesfelder, U. & Heilemann, M. Art and artifacts in single-molecule localization microscopy: beyond attractive images. Nature methods 11, 235-238, doi:10.1038/

007/978-1-59745

127

Addendum

&

nmeth.2852 (2014).128 Patterson, G., Davidson, M., Manley, S. & Lippincott-Schwartz, J. Superresolution imaging

using single-molecule localization. Annual review of physical chemistry 61, 345-367, doi:10.1146/annurev.physchem.012809.103444 (2010).

129 Xu, K., Babcock, H. P. & Zhuang, X. Dual-objective STORM reveals three-dimensional filament organization in the actin cytoskeleton. Nature methods 9, 185-188, doi:10.1038/nmeth.1841 (2012).

130 Ha, T. & Tinnefeld, P. Photophysics of fluorescent probes for single-molecule biophysics and super-resolution imaging. Annual review of physical chemistry 63, 595-617, doi:10.1146/annurev-physchem-032210-103340 (2012).

131 Wulf, E., Deboben, A., Bautz, F. A., Faulstich, H. & Wieland, T. Fluorescent phallotoxin, a tool for the visualization of cellular actin. Proceedings of the National Academy of Sciences of the United States of America 76, 4498-4502 (1979).

132 Capani, F. et al. Phalloidin-eosin followed by photo-oxidation: a novel method for localizing F-actin at the light and electron microscopic levels. The journal of histochemistry and cytochemistry : official journal of the Histochemistry Society 49, 1351-1361 (2001).

133 Fujiwara, K. Techniques for localizing contractile proteins with fluorescent antibodies. Current topics in developmental biology 14, 271-296 (1980).

134 Robinson, R. W. & Snyder, J. A. An innovative fixative for cytoskeletal components allows high resolution in colocalization studies using immunofluorescence techniques. Histochemistry and cell biology 122, 1-5, doi:10.1007/s00418-004-0656-2 (2004).

135 Heuser, J. E. & Kirschner, M. W. Filament organization revealed in platinum replicas of freeze-dried cytoskeletons. The Journal of cell biology 86, 212-234 (1980).

136 Urban, E., Jacob, S., Nemethova, M., Resch, G. P. & Small, J. V. Electron tomography reveals unbranched networks of actin filaments in lamellipodia. Nature cell biology 12, 429-435, doi:10.1038/ncb2044 (2010).

137 Ovesny, M., Krizek, P., Borkovec, J., Svindrych, Z. & Hagen, G. M. ThunderSTORM: a comprehensive ImageJ plug-in for PALM and STORM data analysis and super-resolution imaging. Bioinformatics 30, 2389-2390, doi:10.1093/bioinformatics/btu202 (2014).

138 Abreu Ribeiroa, S. et al. A super-resolution map of the vertebrate kinetochore. Proceedings of the National Academy of Sciences of the United States of America (2010).

139 Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nature methods 9, 676-682, doi:10.1038/nmeth.2019 (2012).

140 Geiger, B., Spatz, J. P. & Bershadsky, A. D. Environmental sensing through focal adhesions. Nature reviews. Molecular cell biology 10, 21-33, doi:10.1038/nrm2593 (2009).

141 Gautier, J. J. et al. Clathrin is required for Scar/Wave-mediated lamellipodium formation. Journal of cell science 124, 3414-3427, doi:10.1242/jcs.081083 (2011).

142 Traub, L. M. Clathrin couture: fashioning distinctive membrane coats at the cell surface. PLoS biology 7, e1000192, doi:10.1371/journal.pbio.1000192 (2009).

143 Collins, A., Warrington, A., Taylor, K. A. & Svitkina, T. Structural organization of the actin cytoskeleton at sites of clathrin-mediated endocytosis. Current biology : CB 21, 1167-1175, doi:10.1016/j.cub.2011.05.048 (2011).

144 Hoogendoorn, E. et al. The fidelity of stochastic single-molecule super-resolution reconstructions critically depends upon robust background estimation. Scientific reports 4, 3854, doi:10.1038/srep03854 (2014).

145 Hrckova, G. & Olson, P. OLSON LAB PROTOCOL: PFA Preparation from Powdered PFA <http://olsonlab.com/resources/Data-Files/OlsonLabProtocol.PFA_Preparation.pdf> (2013).

146 Schneider, C. A., Rasband, W. S. & Eliceiri, K. W. NIH Image to ImageJ: 25 years of image analysis. Nature methods 9, 671-675 (2012).

147 Li, K. The image stabilizer plugin for ImageJ, <http://www.cs.cmu.edu/~kangli/code/Image_Stabilizer.html> (2008).

148 Beli, P., Mascheroni, D., Xu, D. & Innocenti, M. WAVE and Arp2/3 jointly inhibit filopodium formation by entering into a complex with mDia2. Nature cell biology 10, 849-857, doi:10.1038/ncb1745 (2008).

149 Ehrlich, M. et al. Endocytosis by random initiation and stabilization of clathrin-coated pits. Cell 118, 591-605, doi:10.1016/j.cell.2004.08.017 (2004).

00418-004-0656-2

http://olsonlab.com/resources/Data-Files/OlsonLabProtocol.PFA_Preparation.pdf

http://www.cs.cmu.edu/~kangli/code/Image_

128

&

150 Laroche, G., Rochdi, M. D., Laporte, S. A. & Parent, J. L. Involvement of actin in agonist-induced endocytosis of the G protein-coupled receptor for thromboxane A2: overcoming of actin disruption by arrestin-3 but not arrestin-2. The Journal of biological chemistry 280, 23215-23224, doi:10.1074/jbc.M414071200 (2005).

151 Kaksonen, M., Sun, Y. & Drubin, D. G. A pathway for association of receptors, adaptors, and actin during endocytic internalization. Cell 115, 475-487 (2003).

152 Kaksonen, M., Toret, C. P. & Drubin, D. G. A modular design for the clathrin- and actin-mediated endocytosis machinery. Cell 123, 305-320, doi:10.1016/j.cell.2005.09.024 (2005).

153 Taylor, M. J., Lampe, M. & Merrifield, C. J. A feedback loop between dynamin and actin recruitment during clathrin-mediated endocytosis. PLoS biology 10, e1001302, doi:10.1371/journal.pbio.1001302 (2012).

154 Bitsikas, V., Correa, I. R., Jr. & Nichols, B. J. Clathrin-independent pathways do not contribute significantly to endocytic flux. eLife 3, e03970, doi:10.7554/eLife.03970 (2014).

155 Ungewickell, E. & Branton, D. Assembly units of clathrin coats. Nature 289, 420-422 (1981).156 Cocucci, E., Aguet, F., Boulant, S. & Kirchhausen, T. The first five seconds in the life of a

clathrin-coated pit. Cell 150, 495-507, doi:10.1016/j.cell.2012.05.047 (2012).157 Gaidarov, I., Santini, F., Warren, R. A. & Keen, J. H. Spatial control of coated-pit dynamics in

living cells. Nature cell biology 1, 1-7, doi:10.1038/8971 (1999).158 Rappoport, J. Z., Simon, S. M. & Benmerah, A. Understanding Living Clathrin-Coated Pits.

Traffic 5, 327-337, doi:10.1111/j.1398-9219.2004.00187.x (2004).159 Leyton-Puig, D. et al. PFA fixation enables artifact-free super-resolution imaging of the actin

cytoskeleton and associated proteins. Biology open 5, 1001-1009, doi:10.1242/bio.019570 (2016).

160 Traub, L. M. & Bonifacino, J. S. Cargo recognition in clathrin-mediated endocytosis. Cold Spring Harbor perspectives in biology 5, a016790, doi:10.1101/cshperspect.a016790 (2013).

161 Loerke, D. et al. Cargo and dynamin regulate clathrin-coated pit maturation. PLoS biology 7, e57, doi:10.1371/journal.pbio.1000057 (2009).

162 Gournier, H., Goley, E. D., Niederstrasser, H., Trinh, T. & Welch, M. D. Reconstitution of human Arp2/3 complex reveals critical roles of individual subunits in complex structure and activity. Molecular cell 8, 1041-1052 (2001).

163 Eichholtz, T., Jalink, K., Fahrenfort, I. & Moolenaar, W. H. The bioactive phospholipid lysophosphatidic acid is released from activated platelets. The Biochemical journal 291 ( Pt 3), 677-680 (1993).

164 Hoffmann, A. et al. A comparison of GFP-tagged clathrin light chains with fluorochromated light chains in vivo and in vitro. Traffic 11, 1129-1140, doi:10.1111/j.1600-0854.2010.01084.x (2010).

165 Macia, E. et al. Dynasore, a cell-permeable inhibitor of dynamin. Developmental cell 10, 839-850, doi:10.1016/j.devcel.2006.04.002 (2006).

166 Gray, A., Van Der Kaay, J. & Downes, C. P. The pleckstrin homology domains of protein kinase B and GRP1 (general receptor for phosphoinositides-1) are sensitive and selective probes for the cellular detection of phosphatidylinositol 3,4-bisphosphate and/or phosphatidylinositol 3,4,5-trisphosphate in vivo. The Biochemical journal 344 Pt 3, 929-936 (1999).

167 Ohta, H. et al. Ki16425, a subtype-selective antagonist for EDG-family lysophosphatidic acid receptors. Molecular pharmacology 64, 994-1005, doi:10.1124/mol.64.4.994 (2003).

168 Jongsma, M., Matas-Rico, E., Rzadkowski, A., Jalink, K. & Moolenaar, W. H. LPA is a chemorepellent for B16 melanoma cells: action through the cAMP-elevating LPA5 receptor. PloS one 6, e29260, doi:10.1371/journal.pone.0029260 (2011).

169 Ferguson, J. P. et al. Deciphering dynamics of clathrin-mediated endocytosis in a living organism. The Journal of cell biology (2016).

170 Vassilopoulos, S. et al. Actin scaffolding by clathrin heavy chain is required for skeletal muscle sarcomere organization. The Journal of cell biology 205, 377-393, doi:10.1083/jcb.201309096 (2014).

171 Bellve, K. D. et al. Plasma membrane domains specialized for clathrin-mediated endocytosis in primary cells. The Journal of biological chemistry 281, 16139-16146, doi:10.1074/jbc.M511370200 (2006).

129

Addendum

&

172 Akisaka, T., Yoshida, H., Suzuki, R., Shimizu, K. & Takama, K. Clathrin sheets on the protoplasmic surface of ventral membranes of osteoclasts in culture. Journal of electron microscopy 52, 535-543 (2003).

173 Zoncu, R. et al. A phosphoinositide switch controls the maturation and signaling properties of APPL endosomes. Cell 136, 1110-1121, doi:10.1016/j.cell.2009.01.032 (2009).

174 Muinonen-Martin, A. J. et al. Melanoma cells break down LPA to establish local gradients that drive chemotactic dispersal. PLoS biology 12, e1001966, doi:10.1371/journal.pbio.1001966 (2014).

175 Kedziora, K. M. et al. Rapid Remodeling of Invadosomes by Gi-coupled Receptors: DISSECTING THE ROLE OF Rho GTPases. The Journal of biological chemistry 291, 4323-4333, doi:10.1074/jbc.M115.695940 (2016).

176 Isogai, T. et al. Proteomic analyses uncover a new function and mode of action for mouse homolog of Diaphanous 2 (mDia2). Molecular & cellular proteomics : MCP 14, 1064-1078, doi:10.1074/mcp.M114.043885 (2015).

177 Isogai, T., van der Kammen, R. & Innocenti, M. SMIFH2 has effects on Formins and p53 that perturb the cell cytoskeleton. Scientific reports 5, 9802, doi:10.1038/srep09802 (2015).

178 Yarden, Y. & Sliwkowski, M. Untangling the ErbB signaling network. Nature reviews. Molecular cell biology (2001).

179 Harris, T. J. & McCormick, F. The molecular pathology of cancer. Nature reviews. Clinical oncology 7, 251-265, doi:10.1038/nrclinonc.2010.41 (2010).

180 Arnold, D., Weltzien, H. U. & Westphal, O. [Concerning the synthesis of lysolecithin and its ether analogs]. Justus Liebigs Annalen der Chemie 709, 234-239 (1967).

181 van Blitterswijk, W. J. & Verheij, M. Anticancer mechanisms and clinical application of alkylphospholipids. Biochimica et biophysica acta 1831, 663-674, doi:10.1016/j.bbalip.2012.10.008 (2013).

182 Jimenez-Lopez, J. M., Rios-Marco, P., Marco, C., Segovia, J. L. & Carrasco, M. P. Alterations in the homeostasis of phospholipids and cholesterol by antitumor alkylphospholipids. Lipids in health and disease 9, 33, doi:10.1186/1476-511X-9-33 (2010).

183 Hideshima, T. et al. Perifosine, an oral bioactive novel alkylphospholipid, inhibits Akt and induces in vitro and in vivo cytotoxicity in human multiple myeloma cells. Blood 107, 4053-4062, doi:10.1182/blood-2005-08-3434 (2006).

184 Gajate, C. et al. Intracelullar triggering of FAS, independently of FASL, as a new mechanism of antitumor ether lipid-induced apoptosis. Int. J. Cancer (2000).

185 Gajate, C. et al. Intracellular triggering of Fas aggregation and recruitment of apoptotic molecules into Fas-enriched rafts in selective tumor cell apoptosis. The Journal of experimental medicine 200, 353-365, doi:10.1084/jem.20040213 (2004).

186 van der Luit, A. H. et al. A new class of anticancer alkylphospholipids uses lipid rafts as membrane gateways to induce apoptosis in lymphoma cells. Molecular cancer therapeutics 6, 2337-2345, doi:10.1158/1535-7163.MCT-07-0202 (2007).

187 Gomide, A. B. et al. Disrupting membrane raft domains by alkylphospholipids. Biochimica et biophysica acta 1828, 1384-1389, doi:10.1016/j.bbamem.2013.01.017 (2013).

188 Zaremberg, V., Gajate, C., Cacharro, L. M., Mollinedo, F. & McMaster, C. R. Cytotoxicity of an anti-cancer lysophospholipid through selective modification of lipid raft composition. The Journal of biological chemistry 280, 38047-38058, doi:10.1074/jbc.M502849200 (2005).

189 Munoz-Martinez, F., Torres, C., Castanys, S. & Gamarro, F. CDC50A plays a key role in the uptake of the anticancer drug perifosine in human carcinoma cells. Biochemical pharmacology 80, 793-800, doi:10.1016/j.bcp.2010.05.017 (2010).

190 Bryde, S. et al. CDC50 proteins are critical components of the human class-1 P4-ATPase transport machinery. The Journal of biological chemistry 285, 40562-40572, doi:10.1074/jbc.M110.139543 (2010).

191 Chen, R., Brady, E. & McIntyre, T. M. Human TMEM30a promotes uptake of antitumor and bioactive choline phospholipids into mammalian cells. Journal of immunology 186, 3215-3225, doi:10.4049/jimmunol.1002710 (2011).

192 Van der Luit, A. H. et al. Resistance to alkyl-lysophospholipid-induced apoptosis due to downregulated sphingomyelin synthase 1 expression with consequent sphingomyelin- and cholesterol-deficiency in lipid rafts. The Biochemical journal 401, 541-549, doi:10.1042/

130

&

BJ20061178 (2007).193 Fensterle, J., Aicher, B., Seipelt, I., Teifel, M. & Engel, J. Current View on the Mechanism of

Action of Perifosine in Cancer. Anti-Cancer agents in Medicinal Chemestry (2014).194 Richardson, P. G., Eng, C., Kolesar, J., T, H. & Anderson, K. C. Perifosine, an oral, anti-

cancer agent and inhibitor of the Akt pathway: mechanistic actions, pharmacodynamics, pharmacokinetics, and clinical activity. (2012).

195 Gradziel, C., Wang, Y., Stec, B., Redfield, A. G. & Roberts, M. F. cytotoxic Amphiphiles and Phosphoinositides Bind to Two Discrete Sites on the Akt1 PH Domain. Biochemestry (2014).

196 Verheij, M., Moolenaar, W. H. & van Blitterswijk, W. J. Combining Anti-tumor Alkyl-Phospholipid Analogs and Radiotherapy Rationale and Clinical Outlook. Anti-Cancer agents in Medicinal Chemestry (2014).

197 Li, X., Luwor, R., Lu, Y., Liang, K. & Fan, Z. Enhancement of antitumor activity of the anti-EGF receptor monoclonal antibody cetuximab/C225 by perifosine in PTEN-deficient cancer cells. Oncogene 25, 525-535, doi:10.1038/sj.onc.1209075 (2006).

198 Ruiter, G. A., Verheij, M., Zerp, S. F., Moolenaar, W. H. & Van Blitterswijk, W. J. Submicromolar doses of alkyl-lysophospholipids induce rapid internalization, but not activation, of epidermal growth factor receptor and concomitant MAPK/ERK activation in A431 cells. International journal of cancer 102, 343-350, doi:10.1002/ijc.10741 (2002).

199 Haigler, H., Ash, J. F., Singer, S. J. & Cohen, S. Visualization by fluorescence of the binding and internalization of epidermal growth factor in human carcinoma cells A-431. Proceedings of the National Academy of Sciences of the United States of America 75, 3317-3321 (1978).

200 Cailleau, R., Olive, M. & Cruciger, Q. V. Long-term human breast carcinoma cell lines of metastatic origin: preliminary characterization. In vitro 14, 911-915 (1978).

201 Berkers, J. A., van Bergen en Henegouwen, P. M. & Boonstra, J. Three classes of epidermal growth factor receptors on HeLa cells. The Journal of biological chemistry 266, 922-927 (1991).

202 Heukers, R. et al. Endocytosis of EGFR requires its kinase activity and N-terminal transmembrane dimerization motif. Journal of cell science 126, 4900-4912, doi:10.1242/jcs.128611 (2013).

203 Ferraro, D. A. et al. Inhibition of triple-negative breast cancer models by combinations of antibodies to EGFR. Proceedings of the National Academy of Sciences of the United States of America 110, 1815-1820, doi:10.1073/pnas.1220763110 (2013).

204 Spangler, J. et al. Combination antibody treatment down-regulates epidermal growth factor receptor by inhibiting endosomal recycling. Proceedings of the National Academy of Sciences of the United States of America, doi:10.1073/pnas.0913476107/-/DCSupplemental (2010).

205 Pinton, G., Manente, A. G., Angeli, G., Mutti, L. & Moro, L. Perifosine as a potential novel anti-cancer agent inhibits EGFR/MET-AKT axis in malignant pleural mesothelioma. PloS one, doi:10.1371/journal.pone.0036856.g001 (2012).

206 Chung, I. et al. Spatial control of EGF receptor activation by reversible dimerization on living cells. Nature 464, 783-787, doi:10.1038/nature08827 (2010).

207 Sengupta, P. & Lippincott-Schwartz, J. Quantitative analysis of photoactivated localization microscopy (PALM) datasets using pair-correlation analysis. BioEssays : news and reviews in molecular, cellular and developmental biology 34, 396-405, doi:10.1002/bies.201200022 (2012).

208 Mattila, P. K. et al. The actin and tetraspanin networks organize receptor nanoclusters to regulate B cell receptor-mediated signaling. Immunity 38, 461-474, doi:10.1016/j.immuni.2012.11.019 (2013).

209 Wang, Y. et al. Regulation of EGFR nanocluster formation by ionic protein-lipid interaction. Cell research 24, 959-976, doi:10.1038/cr.2014.89 (2014).

210 Saffarian, S., Li, Y., Elson, E. L. & Pike, L. J. Oligomerization of the EGF receptor investigated by live cell fluorescence intensity distribution analysis. Biophysical journal 93, 1021-1031, doi:10.1529/biophysj.107.105494 (2007).

211 Ariotti, N. et al. Epidermal growth factor receptor activation remodels the plasma membrane lipid environment to induce nanocluster formation. Molecular and cellular

131

Addendum

&

biology 30, 3795-3804, doi:10.1128/MCB.01615-09 (2010).212 Clayton, A. H., Tavarnesi, M. L. & Johns, T. G. Unligated epidermal growth factor receptor

forms higher order oligomers within microclusters on A431 cells that are sensitive to tyrosine kinase inhibitor binding. Biochemistry 46, 4589-4597, doi:10.1021/bi700002b (2007).

213 Gadella, T. W., Jr. & Jovin, T. M. Oligomerization of epidermal growth factor receptors on A431 cells studied by time-resolved fluorescence imaging microscopy. A stereochemical model for tyrosine kinase receptor activation. The Journal of cell biology 129, 1543-1558 (1995).

214 Rush, J. S. & Ceresa, B. P. RAB7 and TSG101 are required for the constitutive recycling of unliganded EGFRs via distinct mechanisms. Molecular and cellular endocrinology 381, 188-197, doi:10.1016/j.mce.2013.07.029 (2013).

215 Herbst, J. J., Opresko, L. K., Walsh, B. J., Lauffenburger, D. A. & Wiley, H. S. Regulation of postendocytic trafficking of the epidermal growth factor receptor through endosomal retention. The Journal of biological chemistry 269, 12865-12873 (1994).

216 Wiley, H. S. Trafficking of the ErbB receptors and its influence on signaling. Experimental cell research 284, 78-88 (2003).

217 Kaufmann, R., Muller, P., Hildenbrand, G., Hausmann, M. & Cremer, C. Analysis of Her2/neu membrane protein clusters in different types of breast cancer cells using localization microscopy. Journal of microscopy 242, 46-54, doi:10.1111/j.1365-2818.2010.03436.x (2011).

218 Scarselli, M., Annibale, P. & Radenovic, A. Cell type-specific beta2-adrenergic receptor clusters identified using photoactivated localization microscopy are not lipid raft related, but depend on actin cytoskeleton integrity. The Journal of biological chemistry 287, 16768-16780, doi:10.1074/jbc.M111.329912 (2012).

219 Magenau, A. et al. Discreet and distinct clustering of five model membrane proteins revealed by single molecule localization microscopy. Molecular membrane biology 32, 11-18, doi:10.3109/09687688.2014.990997 (2015).

220 Shelby, S. A., Holowka, D., Baird, B. & Veatch, S. L. Distinct stages of stimulated FcepsilonRI receptor clustering and immobilization are identified through superresolution imaging. Biophysical journal 105, 2343-2354, doi:10.1016/j.bpj.2013.09.049 (2013).

221 Bar-On, D. et al. Super-resolution imaging reveals the internal architecture of nano-sized syntaxin clusters. The Journal of biological chemistry 287, 27158-27167, doi:10.1074/jbc.M112.353250 (2012).

222 Kiskowski, M. A., Hancock, J. F. & Kenworthy, A. K. On the use of Ripley’s K-function and its derivatives to analyze domain size. Biophysical journal 97, 1095-1103, doi:10.1016/j.bpj.2009.05.039 (2009).

223 Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research (2011).

224 Sengupta, P. et al. Probing protein heterogeneity in the plasma membrane using PALM and pair correlation analysis. Nature methods 8, 969-975, doi:10.1038/nmeth.1704 (2011).

225 Pertsinidis, A. et al. Ultrahigh-resolution imaging reveals formation of neuronal SNARE/Munc18 complexes in situ. Proceedings of the National Academy of Sciences of the United States of America (2013).

226 Hell, S. W. & Kroug, M. Ground-state-depletion fluorscence microscopy: a concept for breaking the diffraction resolution limit. Applied Physics B (1995).

227 Hell, S. W. & Wichman, J. Breaking the diffraction resolution limit by stimulated emission: stimulated-emission-depletion-microscopy. . Opt Lett. (1994).

228 Olivier, N., Keller, D., Rajan, V. S., Gonczy, P. & Manley, S. Simple buffers for 3D STORM microscopy. Biomedical optics express 4, 885-899, doi:10.1364/BOE.4.000885 (2013).

229 Andresen, M. et al. Photoswitchable fluorescent proteins enable monochromatic multilabel imaging and dual color fluorescence nanoscopy. Nature biotechnology 26, 1035-1040, doi:10.1038/nbt.1493 (2008).

230 Tiwari, D. K. & Nagai, T. Smart fluorescent proteins: innovation for barrier-free superresolution imaging in living cells. Development, growth & differentiation 55, 491-507, doi:10.1111/dgd.12064 (2013).

231 Wombacher, R. et al. Live-cell super-resolution imaging with trimethoprim conjugates.

132

&

Nature methods 7, 717-719, doi:10.1038/nmeth.1489 (2010).232 Huang, F. et al. Video-rate nanoscopy using sCMOS camera-specific single-molecule

localization algorithms. Nature methods 10, 653-658, doi:10.1038/nmeth.2488 (2013).233 Holden, S. J., Uphoff, S. & Kapanidis, A. N. DAOSTORM: an algorithm for high- density

super-resolution microscopy. Nature methods 8, 279-280, doi:10.1038/nmeth0411-279 (2011).

234 Ries, J., Kaplan, C., Platonova, E., Eghlidi, H. & Ewers, H. A simple, versatile method for GFP-based super-resolution microscopy via nanobodies. Nature methods 9, 582-584, doi:10.1038/nmeth.1991 (2012).

235 Opazo, F. et al. Aptamers as potential tools for super-resolution microscopy. Nature methods 9, 938-939, doi:10.1038/nmeth.2179 (2012).

236 Jones, S. A., Shim, S. H., He, J. & Zhuang, X. Fast, three-dimensional super-resolution imaging of live cells. Nature methods 8, 499-508, doi:10.1038/nmeth.1605 (2011).

237 Los, G. V. et al. HaloTag: a novel protein labeling technology for cell imaging and protein analysis. ACS chemical biology 3, 373-382, doi:10.1021/cb800025k (2008).

238 Raulf, A. et al. Click chemistry facilitates direct labelling and super-resolution imaging of nucleic acids and proteinsdaggerElectronic supplementary information (ESI) available. See DOI: 10.1039/c4ra01027bClick here for additional data file. RSC advances 4, 30462-30466, doi:10.1039/c4ra01027b (2014).

239 Waldchen, S., Lehmann, J., Klein, T., van de Linde, S. & Sauer, M. Light-induced cell damage in live-cell super-resolution microscopy. Scientific reports 5, 15348, doi:10.1038/srep15348 (2015).

240 Johnson, E. et al. Correlative in-resin super-resolution and electron microscopy using standard fluorescent proteins. Scientific reports 5, 9583, doi:10.1038/srep09583 (2015).

241 Sochacki, K. A., Shtengel, G., van Engelenburg, S. B., Hess, H. F. & Taraska, J. W. Correlative super-resolution fluorescence and metal-replica transmission electron microscopy. Nature methods 11, 305-308, doi:10.1038/nmeth.2816 (2014).

242 Loschberger, A., Franke, C., Krohne, G., van de Linde, S. & Sauer, M. Correlative super-resolution fluorescence and electron microscopy of the nuclear pore complex with molecular resolution. Journal of cell science 127, 4351-4355, doi:10.1242/jcs.156620 (2014).

243 Hsu, C. J. & Baumgart, T. Spatial association of signaling proteins and F-actin effects on cluster assembly analyzed via photoactivation localization microscopy in T cells. PloS one 6, e23586, doi:10.1371/journal.pone.0023586 (2011).

244 Itano, M. S. et al. Super-resolution imaging of C-type lectin and influenza hemagglutinin nanodomains on plasma membranes using blink microscopy. Biophysical journal 102, 1534-1542, doi:10.1016/j.bpj.2012.02.022 (2012).

245 Pertsinidis, A., Zhang, Y. & Chu, S. Subnanometre single-molecule localization, registration and distance measurements. Nature 466, 647-651, doi:10.1038/nature09163 (2010).

246 Naguib, A. Following the trail of lipids: Signals initiated by PI3K function at multiple cellular membranes. Science signaling 9, re4, doi:10.1126/scisignal.aad7885 (2016).

247 Eichel, K., Jullie, D. & von Zastrow, M. beta-Arrestin drives MAP kinase signalling from clathrin-coated structures after GPCR dissociation. Nature cell biology 18, 303-310, doi:10.1038/ncb3307 (2016).

248 Garay, C. et al. Epidermal growth factor-stimulated Akt phosphorylation requires clathrin or ErbB2 but not receptor endocytosis. Molecular biology of the cell 26, 3504-3519, doi:10.1091/mbc.E14-09-1412 (2015).

249 Jiang, X., Huang, F., Marusyk, A. & Sorkin, A. Grb2 regulates internalization of EGF receptors through clathrin-coated pits. Molecular biology of the cell 14, 858-870, doi:10.1091/mbc.E02-08-0532 (2003).

133

Addendum

&

Shedding light on endocytosis with optimized super-resolution microscopy

English summaryIn this thesis we present the optimization of two aspects that affect the outcome of super-resolution microscopy images and the application of this optimized technique on the study of the fundamental cell process of endocytosis.

Super-resolution microscopy is one of the latest additions to a large list of microscopy techniques developed through the years, since the Dutch draper and scientist Antoni van Leeuwenhoek, and the English scientist Robert Hooke, built and used microscopes for the first time in the 17th century. Since then, and until the concepts of super-resolution were published and applied, the resolution limit of optical microscopy was ~200 nm, due to the diffraction barrier. In an optical microscope, a point source, such as a small emitting fluorophore in fluorescence based microscopy, appears as a disc of at least 200 nm of diameter, called Airy disk. When two point sources are closer together than this distance, their Airy disks merge and they cannot be seen as separate entities. In a biological sample, the many copies of the same protein labeled with a fluorescent dye also look like an Airy disk, making the distinction between single molecules impossible.

In the last few years, a number of approaches were developed to overcome this challenge, and they are grouped together as super-resolution microscopy. In this thesis, we focus on the further development and use of one of these approaches, called stochastic single molecule localization microscopy (SMLM). The base of stochastic SMLM is separating fluorescing molecules in time since they cannot be separated in space. For this, stochastic SMLM techniques exploit a chemical property of dyes that causes them to turn on and off under certain conditions. The on and off turning of molecules in time is colloquially called blinking, and it occurs stochastically. When only one fluorophore is on, the center of the Airy disk represents the original position of the molecule. Therefore, every time a fluorophore is on, its center is calculated to determine this original position. This process is repeated until all the centers of all the blinks/fluorophores have been calculated. In the end, an image is created by depicting all the calculated centers in a plane.

Here, we present the optimization of two aspects of stochastic SMLM in Chapters 2 and 3. First, we focused on improving the outcome of images obtained with fluorophores with suboptimal blinking. Blinking of fluorophores is the basis of stochastic SMLM and it determines how accurate they can be localized. However, every fluorophore blinks with different characteristics. In some cases, suboptimal blinking of fluorophores that are on during long times gives rise to a type of background that is structured, because it comes from a structure in the sample that is labeled. Background of any kind can hamper the precision with which the centers of blinks are localized and can lead to images with artifacts. We show that structured background can be estimated and corrected with the application of a temporal median filter. This simple trick removes structured background, allowing the precise localization of molecules and generation of high-resolution artifact-free images.

We then focused on sample preparation for stochastic SMLM. The stochastic SMLM technique

134

&

used in this thesis, GSDIM or dSTORM, is typically used on fixed samples, due to the long image acquisition times and the type of fluorophores used. Sample fixation procedures have been optimized and are well known to users of other microscopy techniques, but are only now beginning to be particularly evaluated for SMLM. Taking as a model actin and its binding proteins, we show the importance of specific fixation optimization for super-resolution microscopy. From electron microscopy experts, we know that the best fixative for ultra structural actin studies is glutaraldehyde. However, glutaraldehyde can induce loss and/or conformational changes in proteins, hampering the binding of antibodies used to label them for multicolor studies. We show that using paraformaldehyde, another type of aldehyde fixative, under proper conditions of temperature, pH and fixation times, the fine structure of actin is still maintained. Moreover, we found no detrimental effect of paraformaldehyde on any of the actin binding proteins tested. Because of the infancy of super-resolution microscopy, the preparation of every new sample should be carefully optimized. For SMLM studies of actin and its binding partners, we show that paraformaldehyde fixation, and not glutaraldehyde fixation yields high quality artifact-free images.

The final goal of optimizing super-resolution microscopy is its application for the study of fundamental processes in cell biology. In Chapters 4 and 5, we use our optimized super-resolution microscopy to study endocytosis.

Endocytosis is a process through which cells internalize parts of their membrane, including proteins, many times by forming vesicles. There are many types of endocytosis. In clathrin mediated endocytosis, the internalized vesicles are coated by a protein named clathrin. In chapter 4, we present the results of the study of different clathrin coated structures, clathrin coated pits and flat clathrin plaques, that co-exist in cells. Using optimized super-resolution microscopy, we show that flat clathrin plaques are controlled by N-WASP and the Arp2/3 complex, actin polymerization activators. Moreover, we found that flat clathrin plaques are platforms from where clathrin vesicles can be formed. Interfering with the dynamic actin polymerization process via knock down of N-WASP or the Arp 2/3 complex, interfered with the dynamics of these vesicle platforms, and obstructed the internalization of two important membrane receptors: the Lysophosphatidic acid receptor 1 (LPAR1) and the epidermal growth factor receptor (EGFR). Interestingly, although endocytosis of both receptors was delayed, only the signaling pattern of LPAR1, and not that of EGFR, was affected, indicating that there is some level of specialization on plaque mediated endocytosis and signaling of particular receptors. More needs to be done to fully understand the role of actin controlled flat clathrin plaques in these important processes.

Internalization of receptors occurs after they are activated by their cognate ligands. Internalized receptors can no longer be activated on the membrane, which contributes in part to the cell´s signaling control. Therefore, inducing the internalization of overexpressed or mutated membrane receptors is an interesting therapeutic approach for cancer. A receptor that is commonly overexpressed or mutated in cancer is EGFR.

In Chapter 5, we show that endocytosis of EGFR in cells that overexpress it can be induced by Perifosine, a synthetic lipid. This internalization occurs in an unconventional manner,

135

Addendum

&

since EGFR does not get activated or ubiquitinated, steps necessary for its endocytosis. Moreover, Perifosine´s induced EGFR endocytosis prevents its activation by its cognate ligand, EGF. Interestingly, Perifosine does not induce EGFR endocytosis when the receptor is not overexpressed. In order to elucidate the mechanism of action of Perifosine and its specificity, we studied the organization of EGFR in the plasma membrane of these cells with super-resolution microscopy. We found that in cells that overexpress it, EGFR is oligomerized, while in cells without overexpression, EGFR is randomly distributed on the membrane. Although more studies need to be done, our preliminary results indicate that the specificity of Perifosine might be due to the contrasting distribution of EGFR on cells that express it at different levels.

136

&

Een nieuw licht op endocytose door middel van geoptimaliseerde super-resolutie microscopie.

Nederlandse samenvattingIn dit proefschrift beschrijven we de optimalisatie van twee aspecten van super-resolutie microscopie en het gebruik van deze geoptimaliseerde techniek in het bestuderen van het fundamentele cellulaire proces endocytose.

Super-resolutie microscopie is één van de jongste toevoegingen aan een lange lijst door de jaren heen ontwikkelde microscopie-technieken. Deze lijst begint in de 17e eeuw, toen de Nederlandse lakenhandelaar en wetenschapper Antoni van Leeuwenhoek en de Engelse wetenschapper Robert Hooke microscopen begonnen te bouwen en te gebruiken. Vanaf die begintijd, totdat de concepten van super-resolutie voor het eerst werden gepubliceerd, was de limiet van de haalbare resolutie van optische microscopie ~200 nm. Deze limiet komt door de diffractie barrière. In een optische microscoop wordt een puntbron, zoals bijvoorbeeld een enkel oplichtend fluorofoor in fluorescentiemicroscopie, afgebeeld als een schijf met een diameter van tenminste 200 nm, de zogeheten Airy schijf. Het centrum hiervan benadert de positie van de puntbron. Als twee puntbronnen echter dichter bij elkaar staan dan 200nm dan versmelten hun Airy schijven, met als gevolg dat ze niet afzonderlijk kunnen worden gedetecteerd. In een biologisch preparaat zijn de fluorescent gelabelde exemplaren van eenzelfde eiwit zo dicht bij elkaar dat ze overlappende Airy schijven vormen. Dit heeft als gevolg dat afzonderlijke moleculen niet kunnen worden onderscheiden.

In de laatste paar jaar zijn verschillende manieren ontwikkeld om deze uitdaging te overwinnen, allen onder dezelfde noemer: super-resolutie microscopie. In dit proefschrift focussen we op het verder ontwikkelen en gebruiken van één van deze benaderingen, namelijk stochastische single molecule localization microscopy (SMLM). Stochastische SMLM is gebaseerd op het scheiden van de fluorescente signalen in tijd, omdat dit op plaats niet mogelijk is. Om dit te bereiken benut SMLM een chemische eigenschap van specifieke kleurstoffen (dyes) die ervoor zorgt dat ze in bepaalde condities op een stochastische manier aan- en uitschakelen, in de volksmond ook wel blinking genoemd. Van elke fluorofoor die ‘aan’ is wordt het centrum van de Airy schijf berekend. Dit proces wordt herhaald totdat alle centra van alle blinks/fluoroforen bepaald zijn. Uiteindelijk wordt met al deze posities een super-resolutie beeld geconstrueerd.

In de hoofdstukken 2 en 3 presenteren we de optimalisatie van twee aspecten van stochastische SMLM. In de eerste plaats hebben we gefocust op het verbeteren van super-resolutie beelden die gemaakt zijn met fluoroforen die suboptimaal blinken. Het blinken van fluoroforen staat aan de basis van de stochastische SMLM techniek en bepaalt hoe nauwkeurig ze gelokaliseerd kunnen worden. Elke andere fluorofoor blinkt weer met verschillende karakteristiek. In sommige gevallen resulteert het blinken van fluoroforen die lang ‘aan’ staan tot een achtergrondsignaal. Omdat dit signaal niet uit het preparaat komt maar door het fluorofoor geproduceerd wordt,maakt dat niet willekeurig is, maar gestructureerd. Elk achtergrondsignaal vermindert de nauwkeurigheid van de localisatie van het echte signaal en kan leiden tot super-resolutiebeelden met artifacten. We laten zien dat gestructureerde achtergrond kan worden beoordeeld en gecorrigeerd door middel van

137

Addendum

&

een temporal median filter. Deze eenvoudige truc verwijdert gestructureerde achtergrond, waardoor precieze lokalisatie van moleculen mogelijk is en hiermee dus het genereren van een hoge-resolutie artefact-vrije beelden.

Daarna hebben we gefocust op sample preparatie voor stochastische SMLM. De stochastische SMLM techniek die gebruikt is in dit proefschrift, GSDIM of dSTORM, wordt vanwege de lange acquisitietijden en het type gebruikte fluoroforen meestal toegepast op gefixeerde preparaten. Algemeen bekende fixatieprocedures zijn geoptimaliseerd voor andere microscopietechnieken, maar worden pas recentelijk geëvalueerd voor SMLM. Door actine en actine-bindende eiwitten als model te nemen tonen we het belang aan van optimalisatie van de fixatie voor super-resolutie microscopie. Van elektronenmicroscopie-experts weten we dat het beste fixeermiddel voor ultrastructuurstudies glutaaraldehyde is. Echter, glutaaraldehyde kan conformationele veranderingen in eiwitten veroorzaken die de binding van antilichamen belemmeren. Terwijl deze antilichamen zijn nodig om de eiwitten te labelen met eerder genoemende dyes. We demonstreren dat bij fixatie met paraformaldehyde, een ander fixatief, onder de juiste condities (temperatuur, pH en fixatietijd) de fijnstructuur van actine behouden blijft. Bovendien vonden we geen nadelig effect van paraformaldehyde op elk van de geteste actine-bindende eiwitten. Super-resolutietechnieken staan nog in de kinderschoenen, en daarom dient de preparatie van elk nieuw sample zorgvuldig te worden geoptimaliseerd. Voor SMLM studies aan actine en bindingspartners laten we zien dat paraformaldehyde-, en niet glutaaraldehydefixatie, resulteert in hoge kwaliteit artefact-vrije afbeeldingen.

Het ultieme doel van het optimaliseren van super-resolutie microscopie is het toepassen ervan bij het bestuderen van fundamentele processen in de celbiologie. In Hoofdstukken 4 en 5 gebruiken we onze verbeterde super-resolutietechnieken voor onderzoek naar endocytose.

Endocytose is een proces waarbij cellen een gedeelte van hun membraan samen met eiwitten internaliseren, veelal door het vormen van vesicles. Er bestaan vele soorten endocytose. Bij één daarvan is het eiwit clathrin een belangrijke schakel, de geinternaliseerde vesicles zijn dan bekleed met clathrin, ook clathrin-coated genoemd. In Hoofdstuk 4 presenteren we de resultaten van een studie naar verschillende clathrin-coated structuren, clatrin-coated pits en platte clatrin plagues, die naast elkaar bestaan in de cel. Met behulp van geoptimaliseerde super-resolutie microscopie laten we zien dat platte clathrin plagues gereguleerd worden door N-WASP en het Arp2/3 complex, beide activators van actin polymerisatie. Daarnaast hebben we gevonden dat clatrin plagues platformen zijn van waaruit clathrin vesicles kunnen worden gevormd. Het beïnvloeden van de dynamische actine polymerisatie via het uitschakelen van N-WASP of het Arp2/3 complex leidde tot veranderingen in de dynamica van deze platformen. Daarnaast blokkeerde het de internalisatie van twee belangrijke membraanreceptoren: de lysofosfatidinezuur-receptor 1 (LPAR1) en de epidermale groeifactor-receptor (EGFR). Hoewel endocytose van beide receptoren werd vertraagd, veranderde opmerkelijk genoeg alleen het signaleringspatroon van LPAR1, en niet dat van EGFR. Dit resultaat impliceert een zekere specificiteit in plague-gemedieerde endocytose en signaaltransductie van bepaalde receptoren. Verder onderzoek zal moeten uitwijzen wat precies de rol is van de door actine gereguleerde platte clathrin plagues bij deze belangrijke processen.

Internalisatie van receptoren vindt plaats nadat ze geactiveerd zijn door hun verwante

138

&

liganten. Reeds geïnternaliseerde receptoren kunnen niet meer op het membraan worden geactiveerd, wat deels bijdraagt aan de regulatie van celsignalering. Het induceren van internalisatie door middel van overexpressie van mutante membraanreceptoren is daarom een interessante therapeutische benadering voor kanker. Een receptor die vaak tot overexpressie komt of gemuteerd is bij kanker is EGFR.

In Hoofdstuk 5 laten we zien dat endocytose van EGFR, in cellen waarin deze tot overexpressie komt, veroorzaakt kan worden door perifosine, een synthetisch lipide. Deze internalisatie geschiedt op een onconventionele manier, omdat EGFR niet wordt geactiveerd of geubiquitinileerd; iets wat normaliter voor endocytose benodige stappen zijn. Bovendien verhindert perifosine-geinduceerde endocytose activering door zijn verwante ligant, EGF. Wanneer de receptor niet tot overexpressie wordt gebracht leidt dit interessant genoeg niet tot internalisatie van EGFR. Om het werkingsmechanisme en de specificiteit van perifosine op te helderen hebben we de organisatie van EGFR in het plasmamembraan van deze cellen onderzocht met super-resolutie microscopie. We hebben gevonden dat in cellen met overexpressie EGFR oligomeren vormt, terwijl in cellen zonder overexpressie EGFR willekeurig is verdeeld over het membraan. Hoewel er meer onderzoek zal moeten worden gedaan, impliceren onze eerste resultaten dat de specificiteit van perifosine wellicht wordt veroorzaakt door de uiteenlopende verdeling van EGFR op het membraan van cellen met verschillende expressieniveaus.

139

Addendum

&

Curriculum VitaeDaniela Monica Leyton Puig was born on September 22, 1979 in Cochabamba, Bolivia. She obtained her high school diploma from Saint Andrew´s School in La Paz, Bolivia, after which she enrolled in the Computer Science Bachelor program at the “San Pablo” Catholic Univer-sity in La Paz, Bolivia. In 2002 she moved to Madrid, Spain where she obtained a Bachelor in Biology and Masters´ Degree in Biomedicine at the Complutense University. She continued her education and obtained a second Masters’ degree in Molecular Biotechnology at the University of Barcelona in Barcelona, Spain. In Barcelona, she worked as a researcher for the Spanish Council for Scientific Research (CSIC) in the Institute of Molecular Biology of Barce-lona (IBMB) under the supervision of dr. Isabel Uson, where she worked in the crystallization and functional characterization of AtzR, a bacterial DNA binding protein. During this period she collaborated with the group of dr. Fernando Govantes in the Department of Molecular Biology and Biochemical Engineering at the University Pablo de Olavide, in Sevilla, Spain, where she worked for a short period. In December 2011, she began her PhD training in the Biophysics group of prof. dr. Kees Jalink at the Netherlands Cancer Institute (NKI) in Amster-dam. Her project was part the STW nanoscopy consortium funded by the STW technology foundation, which included several universities and companies committed to further devel-op super-resolution microscopy. The results of her contribution to the development of the technique and its application in cell biology research are described in this thesis.

140

&

Publication listLeyton-Puig D*, Isogai T*, van den Broek B, Klarenbeek J, Janssen H, Jalink K, Innocenti M.* These authors contributed equallyFlat clathrin lattices are dynamic actin-controlled hubs for clathrin-mediated endocytosis and signalling of specific receptorsManuscript submitted

van Veen M, Matas-Rico E, van de Wetering K, Leyton-Puig D, Kedziora KM, Sidenius N, Ja-link K, Perrakis A, Moolenaar WH.GDE3 suppreses urokinase receptor activity through GPI-anchor cleavageManuscript submitted

Matas-Rico E, van Veen M, Leyton-Puig D, van den Berg J, Koster J, Kedziora KM, Molenaar B, Weerts MJ, de Rink I, Medema RH, Giepmans BN, Perrakis A, Jalink K, Versteeg R, Moo-lenaar WH.Glycerophosphodiesterase GDE2 Promotes Neuroblastoma Differentiation through Glypi-can Release and Is a Marker of Clinical Outcome.(2016) Cancer Cell. 0;30(4):548-562. doi: 10.1016/j.ccell.2016.08.016.

Leyton-Puig D, Kedziora KM, Isogai T, van den Broek B, Jalink K, Innocenti M.PFA fixation enables artifact-free super-resolution imaging of the actin cytoskeleton and as-sociated proteins.(2016) Biol Open. 5(7):1001-9. doi: 10.1242/bio.019570.

Kedziora KM, Leyton-Puig D, Argenzio E, Boumeester AJ, van Butselaar B, Yin T, Wu YI, van Leeuwen FN, Innocenti M, Jalink K, Moolenaar WH.Rapid Remodeling of Invadosomes by Gi-coupled Receptors: Dissecting the role of Rho GT-Pases.(2016) J Biol Chem. 291(9):4323-33. doi: 10.1074/jbc.M115.695940.

Isogai T, van der Kammen R, Leyton-Puig D, Kedziora KM, Jalink K, Innocenti M.Initiation of lamellipodia and ruffles involves cooperation between mDia1 and the Arp2/3 complex.(2015) J Cell Sci. 128(20):3796-810. doi: 10.1242/jcs.176768.

Argenzio E, Margadant C, Leyton-Puig D, Janssen H, Jalink K, Sonnenberg A, Moolenaar WH.CLIC4 regulates cell adhesion and β1 integrin trafficking.(2014) J Cell Sci. 127(Pt 24):5189-203. doi: 10.1242/jcs.150623.

Hoogendoorn E*, Crosby KC*, Leyton-Puig D*, Breedijk RM, Jalink K, Gadella TW, Postma M.* These authors contributed equallyThe fidelity of stochastic single-molecule super-resolution reconstructions critically depends upon robust background estimation. (2014) Sci Rep. 4:3854. doi: 10.1038/srep03854.

Nieuwenhuizen RP, Lidke KA, Bates M, Leyton-Puig D, Grünwald D, Stallinga S, Rieger B. Measuring image resolution in optical nanoscopy.(2013) Nat Methods. 557-62. doi: 10.1038/nmeth.2448.

141

Addendum

&

DankwoordThere are many people that helped me during these past five years, science and non-science wise, and without whom it would have been a lot more difficult and much less fun! First of all, my promotor, Kees. Thank you for giving me the opportunity to do this and to learn so much. I take with me many good memories of my time in the lab, as much of the stressy as of the fun times. I probably won´t miss the jokes though ;). I hope we keep in touch. My co-promotor, Metello. Thank you for investing so much time and patience in teaching me. I really admire your determination and persistency. Wouter, I learned a lot from our Monday meetings with you :). Thank you for helping me and trusting me with the Perifosine project.To the best paranimfs ever, Jonne and Judith. Jonne, its great that you are my paranimf, neighbor, friend and mother of my foster cat. Thank you for being such a good friend, for always being genuine, for all the fun times and especially for the best crashed PhD party ever! Judith, I love your enthusiasm! No matter what you always have a smile and find a positive side to everything. Thanks for making my days in the lab happier :). You irradiate joy and I hope to have that always close by! Kasia, without you I wouldn’t have made it, literally. We went through all our PhDs together and I miss you, our philosophical conversations and (small) discussions. It’s amazing how two completely opposite characters can manage to grow such a great friendship. Michiel, Veeni! Jojo, I really enjoyed the many good times we had, our conversations –also when we were just complaining-, and even how much you mocked me and complained about my latina-ness. Thank you for always listening, also to the eventual latina drama ;). Rita, my riri, I was so lucky to find this sweet Portuguese girl when I was new in the lab, who became such a good friend. I miss party Rita though ;). Thank you for always being there for me! Bram, brami, the bramster… I learned sooo much from you! Thank you for being so incredibly patient (and knowledgeable), and always willing to help me! Elisa, desde ya echo de menos tus dichos malagueños y cómo no paras de hablar! Nos hemos reído mucho! Gracias por enseñarme tanto en el lab! Boeki, little one, it was always so much fun with you. Even at dutch lunch when you told me all those gossips and I couldn’t understand… hope to have you back in NL soon! Mar, pequeña, llegaste más tarde pero llenaste huecos! Qué suerte que todavía nos tocó algo de tiempo juntas para conocernos! A por más cafes y parties! Jerry, you are such a great presence in the lab, from the moment you arrived you made all of us laugh a lot more. Me for sure! Pablo, me hizo falta tu tontería diaria en el lab desde el día en que te fuiste! Femke, all the parties we had together were super fun! Too bad we never finished our awesome foci project. Maaike, I always enjoyed our daily office talks and gossip :). Jeffrey, thanks for all the help in the lab! Mariet, thanks for all the help, and the fun parties. Marcel, when I arrived you definitely showed me the scary dutch character, but in the end you were actually a softy ;). Francisca, thank you for your help with the clathrin project! To the rest of B5, old and new: Dalila, Elisabetta, Rene, Roy, Lenno, Melinda, Mihoko, Marion, Marianne, Anja, Rob, Iraia, Anna, Louise, Alessandra, Babet, Leila, Daan, Fransisca, Anja B, Benjamin, Ahmed, Marjon, Arnoud, Alba, Lisa, Veronika, Wei, we had great times, beers, conversations. Thank you all for making everyday a bit more special.And to all the rest NKI people who made these years a lot more fun: Jacqueline, after many c-breaks, beers, “nice” photos and PhD retreats… the fun has just started ;). Fabricio, primero que nada gracias por llevar el nombre de Bolivia tan alto jajaja, al menos uno que nos hace quedar bien! Aunque me hagas quedar mal a mi ;). Ahora que no voy al NKI tenemos que

142

&

quedar a comer fuera! Tada, your incredibly peaceful and patient attitude taught me a lot. Thank you for all the help with the clathrin project. Tessy and Chiara, to many more fun dinners and parties! Tao, I laughed a lot with you, I am sure you will become a great scientist, Rui, Philip, Joppe, Vincent, hola como estas?, Lukas, Markus, Lorenzo, Gozde, Alejandro, thanks! Lauran, Lenny, Marjolijn, Martijn, Frank and Anita, thanks for all the help! To all the people from the STW nanoscopy consortium, thank you for a very nice collaboration. Robert, it was super nice to work with you. Erik, thanks for all the input and your constant enthusiasm with our projects and collaborations. Marten, thank you for having me in the background project :). Daphne and Linsday, it was great to work with you in your FP project! A mis amigas hermosas que siempre están ahí para oírme quejar, llorar o reir: Mariana, mi gordi, siempre ahí conmigo, eres lo máximo! Gina, soy feliz de tenerte medio cerca!, Yvonne, Noelia, Sofia, Carolina, Andrea, Ana Carola, las extraño mucho! Anna, me encanta lo amigas que hemos terminado siendo desde aquel día que te dije que tomáramos un café ;). Gracias por tu apoyo y tus ánimos constantes en todo! Eres super importante para mi! Niene, thank you for deciding to be my friend as soon as we met! And for all the good times… for sure many more to come! Lucy and Serena, something good had to come from actually surviving that dutch survival course. I am sure we have many more years of laughing about it together.Frans and Cathy, thank you for all your support and care over these years, and for taking me in as family since day one. Bram, Geertje, Myrthe, Martijn, thank you for all your support and interest all these years!Elisa y Javier, mis padres putativos. Os quiero y echo de menos siempre!! Aunque Javier sea tan... exigente ;). Gigi, Lucia, Esa e Paolino, grazie per il sostegno che mi avete dato fin dall´inizio. E per considerarmi parte della vostra familia. Papi, gracias por estar ahí y confiar ciegamente en mi capacidad. Hermanas, como hubiese hecho yo sin ustedes, en la vida entera? Ustedes son mi pilar y mi soporte de cada día. Además de mi risa y alegría. Las extraño cada minuto de mi vida!! Pronto, las cuatro juntas en algún rincón del mundo! Momi, gracias por repetirme cada día que yo lo podía todo, y por enseñarme a vivir la vida a tope. Sé que esto te hubiese hecho muy feliz, y por eso es para tí. Me haces falta.Marijn, during all these years you have always been my number one supporter. In my good, my bad, and my worse days. Thank you for still choosing me on all of those. And always managing to make me laugh anyway :). You are the best partner in life and I am looking forward to lots more laughter, and our next adventure together!

143

Addendum

&

144

&

Shedding light on endocytosis with optim

ized super-resolution microscopy

Daniela M

onica Leyton Puig

2017




Uitnodiging

voor het bijwonen van de openbare verdediging van het

proefschrift van

Daniela Leyton Puig

op donderdag6 april 2017

om 12:00 uur

in de Agnietenkapelvan de

Universiteit van AmsterdamOudezijds Voorburgwal 229-231

Amsterdam

Receptie ter plaatse na afloop van de promotie

Paranimfen:Jonne Raaijmakers

Judith Haarhuis

Daniela Leyton [email protected]

shedding light on endocytosis with optimized super-resolution microscopy daniela monica leyton

Documents