serotonergic regulation of melanocyte conversion: a ...containing the pineal gland, the pituitary,...

R E S E A R C H A R T I C L E

D E V E L O P M E N T A L B I O L O G Y

Serotonergic regulation of melanocyte conversion:A bioelectrically regulated network for stochasticall-or-none hyperpigmentationMaria Lobikin,1 Daniel Lobo,2 Douglas J. Blackiston,1 Christopher J. Martyniuk,3

Elizabeth Tkachenko,1 Michael Levin1*

http://stke.scienD

ownloaded from

Experimentally induced depolarization of resting membrane potential in “instructor cells” in Xenopuslaevis embryos causes hyperpigmentation in an all-or-none fashion in some tadpoles due to excessproliferation and migration of melanocytes. We showed that this stochastic process involved sero-tonin signaling, adenosine 3′,5′-monophosphate (cAMP), and the transcription factors cAMP responseelement–binding protein (CREB), Sox10, and Slug. Transcriptional microarray analysis of embryos takenat stage 15 (early neurula) and stage 45 (free-swimming tadpole) revealed changes in the abundance of 45and 517 transcripts, respectively, between control embryos and embryos exposed to the instructor cell–depolarizing agent ivermectin. Bioinformatic analysis revealed that the human homologs of some of thedifferentially regulated genes were associated with cancer, consistent with the induced arborization andinvasive behavior of converted melanocytes. We identified a physiological circuit that uses serotonergicsignaling between instructor cells, melanotrope cells of the pituitary, and melanocytes to control the pro-liferation, cell shape, and migration properties of the pigment cell pool. To understand the stochasticityand properties of this multiscale signaling system, we applied a computational machine-learning methodthat iteratively explored network models to reverse-engineer a stochastic dynamic model that recapitu-lated the frequency of the all-or-none hyperpigmentation phenotype produced in response to variouspharmacological and molecular genetic manipulations. This computational approach may provide insightinto stochastic cellular decision-making that occurs during normal development and pathologicalconditions, such as cancer.

ce
on F
ebruary 14, 2020m

ag.org/

INTRODUCTION

Bioelectrical mechanisms contribute cell-to-cell communication and large-scale pattern formation processes. Slowly changing ion currents and theresulting spatial gradients of plasma membrane resting potential across cellfields affect the signaling pathways that control cell proliferation (1, 2), dif-ferentiation (3–5), and migration (6, 7). Spatiotemporal gradients of trans-membrane potential (Vmem) contribute to the regulation of many patterningprocesses during embryogenesis and regeneration (8–12). Bioelectricsignaling is also implicated in cancer biology (13–16), with the recognitionof ion channels as oncogenes (17–19) and as drug targets (20–22), aswell aswith the identification of bioelectrical states that are associated with neo-plastic transformation and tumor prevention (23–26).

One aspect of embryogenesis is the coordination of migration, prolif-eration, and morphogenesis of specific subpopulations of cells. To studythe role of bioelectric cues in coordinating these processes, we use themodel organism Xenopus laevis, which is a system in which biophysicalcues, molecular genetic signaling (signaling by gene products), and mor-phogenetic outcomes can be readily characterized. A proportion of tad-poles in which instructor cells are experimentally depolarized acquire ahyperpigmented phenotype as a result of altered melanocyte morphologyand behavior (6, 27): The melanocytes overproliferate, acquire a highlyabnormal arborized cell shape, and invade the neural tube, brain blood

1Biology Department and Center for Regenerative and Developmental Biolo-gy, Tufts University, Medford, MA 02155, USA. 2Department of BiologicalSciences, University of Maryland, Baltimore County, Baltimore, MD 21250,USA. 3Center for Environmental and Human Toxicology and Department ofPhysiological Sciences, UF Genetics Institute, University of Florida, Gaines-ville, FL 32611, USA.*Corresponding author. E-mail: [email protected]

ww

vessels, and gut, properties that are characteristic of dysregulation of thenormal epithelial-to-mesenchymal transition that occurs during embryo-genesis (28, 29). Instructor cells are characterized by the presence of theglycine-gated chloride channels GlyR (also known as GlyCl), are presentthroughout the somatic tissues of the body, and control melanocyte pro-liferation, shape, andmigration. The hyperpigmentation phenotype resultsfrom increased numbers, altered cell shape, and ectopic localization ofmelanocytes, not changes in melanin production or pigment granule be-havior. This conversion process involves thewhole body, not only the skin(6, 27), and is induced by a bioelectrical signal, not requiring DNA mu-tation or chromatin-modifying drug exposure.

Depolarization of the instructor cells stimulates a serotonin-dependentsignal, involving several serotonin receptors, the serotonin transporter SERT,and the vesicular monoamine transporter VMAT (6, 30). Induction of hyper-pigmentation requires serotonin receptors 5HT-R1, 5HT-R2, and 5HT-R5(6, 30). These serotonin receptors are heterotrimeric guanine nucleotide–binding protein (G protein)–coupled receptors (GPCRs) that signalthrough either the second messenger systems of adenylyl cyclase (AC)and adenosine 3′,5′-monophosphate (cAMP) or phospholipase C (PLC)and inositol trisphosphate (31). Because both 5HT-R1 and 5HT-R5 areGPCRs coupled to Gi/o, which inhibits cAMP signaling, and cAMPsignaling plays a role in melanoma (32), we used pharmacological toolsto explore the involvement of cAMP signaling in control of melanocyte be-havior and found that depolarization of instructor cells signaled to melano-cytes through cascades involving cAMP production and the transcriptionalactivity of the cAMP response element–binding protein (CREB), as well asthe transcription factors Sox10 and Slug, all of which are implicated in de-velopmental regulation of melanocytes (33) and pathogenesis of melanoma(32, 34–38).

w.SCIENCESIGNALING.org 6 October 2015 Vol 8 Issue 397 ra99 1

http://stke.sciencemag.org/


Dow

n

We also found that conversion of melanocytes by instructor cellsinvolved melanocyte-stimulating hormone (MSH)–secreting melanotropecells of the pituitary. To uncover previously unknown targets of long-rangebioelectrical signaling, we performed a genome-wide analysis of transcrip-tional changes in embryos in which the instructor cells were depolarized.

One puzzling aspect of this phenotype is that hyperpigmentation isstochastic, yet binary, at the level of the organism: A given treatment affectsonly somepercentage of themanipulated embryos, and every animal is eitherhyperpigmented or not. Because currently available reverse-engineeringmethods for steady-state or time-series concentration data cannot be usedto infer stochastic models directly from phenotypes (39–44), we developedan automated method based on evolutionary computation to derive astochastic network model that predicted how divergent body-wide pheno-types can stochastically arise. Our method reverse-engineered the necessarycomponents, regulatory interactions, and parameters of a stochastic ordinarydifferential equation model directly from training data sets and accuratelypredicted the stochastic phenotypes resulting from pharmacological manip-ulation of the embryos. Our approach revealed the molecular steps andcircuits by which change in the membrane voltage of dispersed, specific cellpopulations can affect the behavior of widely distributed cells of other types.

on February 14, 2020


loaded from

RESULTS

Ivermectin-induced instructor cell signaling is mediatedby cAMP, PKA, and CREBTo depolarize instructor cells of 1-day [stage 10; note that all stages aredefined as Nieuwkoop and Faber (NF) (45)] Xenopus embryos, we usedthe GlyR-opening drug ivermectin, which causes the efflux of chloride ionsfrom the GlyR-positive instructor cells (6, 46). Because melanocytes do notexpress the gene encoding GlyR (6), ivermectin does not directly affect themelanocytes. We measured the percent of the population that becamehyperpigmented at stage 45, 7 days of ivermectin exposure (Fig. 1, A toC). Inspection of tadpole morphology, behavior, and survival indicated noovert toxicity or teratogenesis due to hyperpigmentation, consistent withprevious studies (6).

We used a suppression screen strategy (6, 30, 47) to identify signalsinvolved in mediating the hyperpigmentation phenotype.We applied inhib-itors of various aspects of cAMP signaling together with ivermectin and as-sessed the effect on the frequency of hyperpigmentation. Treating stage10embryoswith the cell-permeable cAMPantagonist 2′5′-dideoxyadenosineand ivermectin significantly reduced the incidence of hyperpigmentation thatoccurred with ivermectin treatment alone (87% hyperpigmented with 1 mMivermectin alone and 67%hyperpigmentedwith 500mM2′5′-dideoxyadenosineand ivermectin) (Fig. 1C).We also exposed stage 10 embryos to ivermectinand forskolin, which directly stimulates AC to increase cAMP, or to for-skolin alone and found that forskolin alone led to a significant incidence ofhyperpigmentation compared with the occurrence in control animals (Fig.1C). These results suggested the involvement of cAMP in Vmem-inducedchanges in melanocyte behavior.

Changes in intracellular cAMP abundance regulate the activity of thecAMP-dependent PKA. To determine whether PKA was involved in thehyperpigmentation phenotype, we exposed stage 10 Xenopus embryos toivermectin and the PKA antagonist H89 dihydrochloride. Coapplicationof H89 resulted in a significant reduction in the incidence of hyperpigmen-tation compared with embryos exposed to ivermectin alone (Fig. 1D), sug-gesting that PKA is involved in this signaling process.

cAMPandPKAsignaling affect gene expression throughCREB.Trans-location of activated PKA to the nucleus phosphorylates Ser133 in CREB,thereby activating it (48). To address whether CREB is involved in

ww

Wild

typ

e

Hyp

erp

igm

ente

d

D

C

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Control IVM 2'5'-DDA 2'5'-DDA + IVM Forskolin

Perc

ent

hyp

erp

igm

enta

tio

n

Dep

ola

rize

d

Dep

ola

rize

d +

cAM

P an

tag

on

ist

cAM

P ag

on

ist

323 255

177 178 254

**

cAM

P an

tag

on

ist

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Control IVM H89 H89 + A-CREB A-CREB +IVM

XlCreb1 VP16-XlCreb1

Perc

ent

hyp

erp

igm

enta

tio

n

**

258 313 169 163

131 110

Dep

ola

rize

d

CA

CRE

Bin

ject

ed

DN

CRE

B i

nje

cted

WT

CRE

B i

nje

cted

*PK

A a

nta

go

nis

t

Dep

ola

rize

d +

PK

A a

nta

go

nis

t

93

*

166

IVM

A B

DN

CRE

B i

nje

cted

+D

epo

lari

zed

Fig. 1. cAMP and CREB are involved in mediating instructor cell signaling.(A and B) Frogs treated with pharmacological agents at stage 10 are
scored for pigmentation phenotypes at stage 45. (C) Percent of the popu-lation exhibiting hyperpigmentation in embryos at stage 45 after treatmentat stage 10 with the cAMP antagonist 2′5′-dideoxyadenosine (2′5′-DDA)with or without ivermectin (IVM) or the AC activator forskolin. (D) Percentof population exhibiting hyperpigmentation in embryos at stage 45 aftertreatment at stage 10 with the protein kinase A (PKA) antagonist H89 withor without ivermectin or injected into one cell at the four-cell stage with themRNA for the indicated CREBwith or without ivermectin.A-CREB encodesdominant-negative (DN) CREB; XlCreb1 encodes wild-type (WT) CREB;and VP16-XlCreb1 encodes constitutively active (CA) CREB. In (C) and(D), sample sizes (number of embryos) are noted for each condition. Errorbars represent 1 SEM. *P < 0.0001 based on Pearson’s c2 test.



mediating the hyperpigmentation phenotype in Xenopus, we injected onecell of a stage 5 (32-cell) embryo with synthetic mRNAs encoding eitherwild-type CREB (XlCreb1), dominant-negative CREB (A-CREB) (49), or aconstitutively active version of CREB (VP16-XlCreb1) (50). Injection of theRNA for wild-type CREB had no effect on the percent of hyperpigmentedanimals, whereas injection of the RNA for constitutively active CREBresulted in hyperpigmentation of ~42% of tadpoles (Fig. 1D). Injection ofthe RNA for the dominant-negative CREB into one cell of a stage 5 embryohad no significant effect on the percent of hyperpigmented animals wheninjected alone and compared to control uninjected embryos or whencombinedwith ivermectin and comparedwith embryos exposed to ivermec-tin alone (Fig. 1D). Together, these results implicate the involvement ofcAMP-PKA-CREB pathway in transducing signals from the depolarizedinstructor cells to melanocytes.

Hyperpigmentation phenotypeinvolves the pituitary gland



Dow

nloaded from

Instructor cells are sparse yet widely distrib-uted throughout the entire embryo and theresponding neural crest–derived melano-cytes, which can be many cell lengths away,suggesting that the signal may not be directfrom the instructor cells to the melanocytes.Noticing that rare embryos with spontane-ously missing midline structures neverunderwentmelanocyte conversion after iver-mectin expression, we suspected involve-ment of one of the unpaired (centrallylocated) endocrine organs. Because mela-notrope cells of the pituitary gland secreteaMSH, which promotes the dispersion ofmelanin granules in the Xenopus melano-cytes (51), we hypothesized that the pituitarymay also be involved in controlling other as-pects of melanocyte behavior.

We individually removed regionscontaining the pineal gland, the pituitary,or, as a control, a region below the cementgland (Fig. 2, A and B) from tail bud–stage(stage 32) embryos that had been exposedto ivermectin continuously starting at theneurula stage (stage 10). The first melano-cytes develop at stage 33 or 34; therefore,we removed the pineal or pituitary glandsbefore the appearance of the first melano-cyte. The embryos were reared to free-swimming tadpole stage and scored forhyperpigmentation. Removing the pinealgland had no significant effect on ivermectin-inducedhyperpigmentation,whereas removingthe pituitary gland decreased hyperpig-mentation incidence by ~50% comparedwith ivermectin-exposed controls (Fig. 2,C and D). These results indicated that hy-perpigmentation involved the pituitary andrevealed that depolarization-induced hyper-pigmentation can be interrupted by remov-ing the pituitary until at least stage 32.

Because the pituitary releases aMSH, aknown regulator of melanocyte function,

ww

we exposed Xenopus embryos to aMSH agonists and antagonists. Nearlyall of the embryos exposed to the aMSH agonist SHU 9119 from the neu-rula to tadpole stages developed a hyperpigmented phenotype (Fig. 2D).Embryos exposed to MSH-RIF and depolarizing ivermectin displayed32% less hyperpigmentation incidence than did those exposed to ivermectinalone (Fig. 2D).

To determine whether serotonin signaling and aMSH were part ofthe same pathway, we exposed the embryos to combinations of inhibitorsof aMSH and 5HT signaling. Either MSH-RIF or altanserin, a 5HT-R2–specific antagonist, significantly decreased ivermectin-induced hyper-pigmentation when applied to embryos individually (Fig. 2D). Embryosexposed to all three pharmacological agents (ivermectin, MSH-RIF, andaltanserin) showed a similar proportion of hyperpigmented tadpoles asthose exposed to either ivermectin and MSH-RIF or ivermectin and altan-serin, indicating that both serotonin and aMSH signaling are in the samepathway.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Control cut + IVM

Pituitary cut + IVM

Pineal cut + IVM

Control IVM SHU 9119 IVM + MSH-RIF IVM + altanserin

IVM + MSH-RIF + altanserin

Perc

ent

hyp

erp

igm

enta

tio

n

1 mm

Cement gland

Forebrain

Pineal

Pituitary

Spinal cordHindbrainMidbrain

A B

Pineal cut

Pituitary cut

Control cut

C

Control cut +

Pituitary cut +

Pineal cut +

HP

HP

WT

** * *

NS NS

NS

NS

*D

Dep

ola

rize

d

Dep

ola

rize

d +

MSH

an

tag

on

ist

Dep

ola

rize

d +

Dep

ola

rize

d +

MSH

+

MSH

ag

on

ist

NS

IVM treated

IVM treated

IVM treated

5H

T-R

anta

go

nis

t

5H

T- R

an

tag

on

ist

Fig. 2. The pituitary gland is necessary for ivermectin-mediated hyperpigmentation. (A) Side view showingthe main regions of Xenopus tadpole brain. (B and C) Xenopus embryos were treated with ivermectin from
neurula stage (stage 10), cut at tail bud stage (stage 32), raised to tadpole stage (stage 45), and scored forhyperpigmentation (HP). Cuts were performed on tail bud stage (stage 32), removing the pineal gland,pituitary gland, or a control region below the cement gland and away from both the pituitary and pineal.(D) Effect of control cuts (n = 54 embryos), pineal cuts (n = 46 embryos), or pituitary cuts (n = 59 embryos)on the percent of ivermectin-induced hyperpigmented tadpoles. Effect of inhibition of MSH with aMSHrelease–inhibiting factor (MSH-RIF), an MSH agonist SHU 9119, or a 5HT receptor antagonist altanserinon the percent of hyperpigmented tadpoles. Error bars represent 1 SEM. *P < 0.0001, Pearson’s c2 test;NS, not significant.



Instructor cell depolarization results in transient increasein Sox10 expressionSox10 is a transcription factor that promotes the specification of neural crestprogenitors to themelanocyte lineage (33). Slug is amember of theSnail family



Dow

nloaded from

of transcription factors that are important inthe epithelial-to-mesenchymal transition inneural crest cells (37, 52). Depolarizationof specific regions of developing Xenopusembryos by forced expression of genes en-coding depolarizing ion channels increasesthe expression of genes encoding severaltranscription factors, including Sox10 (27).Whereas injection of mRNA encoding de-polarizing ion channels into blastomeres re-sults in depolarization of all of the injectedcells and daughters thereof, ivermectin de-polarizes only the instructor cells.Therefore,we compared the effects of ivermectin andinjectionofXminK,which encodes a subunitof a depolarizing potassium channel sub-unit, onSox10 andSlug expressionbyquanti-tative real-time polymerase chain reaction(RT-qPCR) of tail bud–stage embryos (Fig. 3,A and B). We used RT-qPCR of whole em-bryos to obtain quantitative results that couldnot be achieved by in situ hybridization inthe ivermectin-treated embryos. Either injec-tion ofXminK at the one- or two-cell stages orexposure to ivermectin at stage 12 resulted ina similar induction of Slug in the entire tailbud–stage embryo. Both also induced Sox10,but injection of XminK produced a largerincrease in Sox10 abundance (Fig. 3B).

Because ivermectin resulted in a lowerthan expected increase in Sox10 expressionin the tail bud–stage embryos, we analyzedSox10 by RT-qPCR from embryos exposedto ivermectin at stage 10 and then fixed atdifferent stages. This revealed that Sox10mRNA increased in abundance by stage15, which occurs ~12 hours after the addi-tion of ivermectin under our experimentalconditions, and then decreased at laterstages (Fig. 3C). To determine the spatialprofile of Sox10 expression, we analyzedstage 15 embryos that had been treatedwithivermectin by in situ hybridization. In con-trol embryos, Sox10 expressionwas limitedto symmetrical regions along the neuralfold (Fig. 3D). In contrast, ivermectin-treatedembryos displayed ectopic Sox10 expres-sion in a punctate pattern throughout theembryo (Fig. 3, E and F), in a pattern simi-lar to that of the GlyR-expressing instruc-tor cells (6). Moreover, in contrast to theconsistent expression pattern of Sox10 incontrols, the ivermectin-treated embryosexhibited a patchier signal, suggesting a re-duction in expression in the regionswhereSox10 is normally expressed.

ww

These data indicated that similar to regional depolarization throughmRNA injection, ivermectin, which produces abnormally depolarizedVmemonly in the instructor cells, can alter gene expression, in particular,producing foci of ectopic Sox10 expression throughout the embryo in

Control 0

2

4

6

8

10

12

14

16

Ivermectin minK

Fo

ld c

ha

ng

e

Sox10 Slug

B

CTRL ST15IVM ST15

D E F

XminK

minK injection

Sox10A

0

5

10

15

20

25

30

35

40

ST20 ST25 ST30 ST35 ST40 ST15

Fo

ld c

ha

ng

e

NF stage

C

0.1 mm 0.1 mm0.1 mm

Sox10

**

** *

NS

Fig. 3. Both localized depolarization and sparse, widely distributed depolarization increase Sox10 tran-scripts. (A) Expression of Sox10 in embryos injected with a mixture of XminK and b-gal mRNAs at the
four-cell stage that were fixed at tail bud stages. Far left shows injection, and middle shows the sectionplane for the data shown at the right. (B) Effect of ivermectin treatment or XminK injection on the expressionof Sox10 and Slug in tail bud stage embryos (stage 25) as assessed by RT-qPCR. XminK-injected animalswere injected into one cell at the four-cell stage. Ivermectin-treated animals were exposed to the drug fromneurula stage (stage 10) onward until processing for RT-qPCR at stage 25. Control animals were uninjectedand untreated. Red dashed line denotes no fold change compared to control. All experimental treatmentsresulted in significant increase in expression of both Sox10 and Slug compared to control (P < 0.05, Stu-dent’s t test; n = 10 embryos per sample, samples run in triplicate, three biological replicates). (C) Effect ofivermectin exposure started at neurula stage on Sox10 expression as assessed by RT-qPCR in embryoscollected at the indicated stages. ST, stage. NF stages 15 to 35, n = 10 embryos per sample, samples runin triplicate, three biological replicates; NF stage 45, n = 5 embryos per sample, samples run in triplicate,three biological replicates. P < 0.05, Student’s t test. (D to F) In situ hybridization for Sox10 performed onstage 15 control (CTRL) embryos (D) or embryos that had been exposed to ivermectin treatment starting atstage 10 (E and F). (F) is an enlargement of the area boxed in (E), and arrowheads indicate positive stainingoutside the main Sox10-positive area.





Dow

nloaded from

a pattern similar to that of instructor cells, butSox10 induction is transient compared withthat achieved by regional depolarization bymRNA injection.

Instructor cell depolarizationproduces a distinct pattern ofgene expressionGiven that instructor cell depolarizationincreased the expression of genes encodingSox10 and Slug, two neural crest–specifictranscription factors, we performed tran-scriptional microarray analysis of wholeembryos at early neurula (stage 15) and tad-pole (stage 45) stages, comparing stage-matched controls and embryos treated withivermectin from stage 10.

We found that ivermectin exposure re-sulted in 45 transcripts (Fig. 4A and tableS1) that were differentially expressed bystage 15 embryos and 517 transcripts thatwere differentially expressed in stage 45embryos (Fig. 4A and table S2).Hierarchicalclustering of the data revealed that develop-mental stage, rather than ivermectin, was themain driver of the differences in gene expres-sion (fig. S1). Of the five mRNAs that weredifferentially expressed in both stage 15 andstage 45 embryos exposed to ivermectin,three are unknown and represent expressedsequence tags (one for UniProt ID Xl.52612and two for Xl.52879). The two genes thatwere differentially expressed in embryosof both stages in response to ivermectin ex-posure were HIG1 (encoding hypoxia-inducible domain family, member 1A) witha 2.4-fold increase in the early embryos rela-tive to the controls and a 4.5-fold increase inlate embryos relative to the controls, andANKRD37 (encoding ankyrin repeat do-main 37) with a 2.9-fold increase in theearly embryos relative to the controls and9-fold increase in the late embryos relative tothe controls. Too few differentially ex-pressed genes were identified in the earlyembryos for functional enrichment analysis;nevertheless, our data showed that fivemRNAs were altered in abundance withina few hours of depolarization of instructorcells (in the stage 15 embryos) and remaineddifferentially expressed from controls at thesecond time point (in the stage 45 embryos).

We identified sufficient differentiallyexpressed transcripts in the later stage em-bryos (the same stage at which the hyper-pigmentation phenotype is strongly evident)to perform enrichment analysis using theGOrilla database (53) with Homo sapiensas the reference species. We analyzed thetranscripts for enrichment of Gene Ontology

C

40 5 512

Early embryo Late embryo

A

3 sequence tags2 genes

BEarly embryos

NF ST15

Late embryosNF ST45

Fig. 4. Human homologs of the genes with increased expression resulting from ivermectin-induced de-polarization are associated with neoplasms and cancer. (A) Venn diagram of the transcripts that were differen-
tially regulated (increasedordecreased) in transcriptionalmicroarraysofXenopus tadpoles treatedwith ivermectinstarting at stage 10 and collected at early (stage 15) and late stage (stage 45). See fig. S1 for the differentiallyexpressed transcriptsclusteredbystageandcondition.See tablesS1andS2 fora list of the transcriptsandgenes.(B and C) Pathway analysis of the differentially expressed transcripts in stage 15 embryos (B) and stage 45embryos (C). Proteins are red shapes, diseases are purple boxes, stimulatory regulatory events are indicatedby an arrow and a plus sign on the relationship line, inhibitory regulatory events are indicated by a blunt line,and arrows without any sign indicate direct binding of proteins.
www.SCIENCESIGNALING.org 6 October 2015 Vol 8 Issue 397 ra99 5





Dow

nloaded from

(GO) for “biological process” and “molecular function,”which identified theimmune response, metabolism, and ion regulation (table S3) as the majorenriched processes.

Pathway analysis (54) using human homologs of the frog genes revealedthat 45 mRNAs that were differentially expressed in response to ivermectinin the early stage 15 embryos were involved in the cell processes related tochromatin remodeling, apoptosis, cell differentiation, and mitosis (data S1).We also used the program Pathway Studio to determinewhether the humanhomologs of the differentially expressed genes were associated with dis-eases, which revealed that 6 of the 45 differentially expressed genes wereassociated with neoplasms (Fig. 4B).

Pathway analysis by subnetwork enrichment analysis (54) indicatedthat 9% of the ivermectin-affected genes (45 of 517) in stage 45 embryoswere associated with cancer (Fig. 4C and table S4). The results of thepathway and disease analyses implied that the transcriptional alterationsthat occurred in response to depolarization of the instructor cells maybe similar to those associated with disease states and developmental dys-regulation in humans.

Modeling predicts both the large-scale dynamics and thequantitative stochastic response to instructorcell depolarizationThe hyperpigmentation phenotype is both bistable at the level of the indivi-dual (no partially hyperpigmented “salt and pepper” tadpoles occur) andstochastic at the level of the population (different conditions result in differentpercentages of hyperpigmented and normal tadpoleswithin a treated cohort).To better understand this stochastic process, we undertook an automatedcomputational modeling approach to identify amodel of a signaling networkthat could characterize all of the known necessary signaling pathwayscontrolling this behavior as a result of the embryo-wide depolarization thatproduces embryo-wide hyperpigmentation.To identify amodel that accurate-ly predicted hyperpigmentation frequency in the population, (i) we generateda simple arrowdiagramnetworkwith aminimal set of regulatory connectionsand (ii) we automated the remaining steps in the process—the simulation of agiven mathematical model with a specific set of parameters, the comparisonof the results with data obtained experimentally, and the assignment ofspecific link parameters, new connections, and yet-to-be-discovered nodeswithin the network to discover new mathematical models (text S1). Themodel included both intracellular and intercellular signaling events butdid not include parameters representing the spatial location of each of theregulatory events and elements; thus, it is a nonspatial, dynamic model.

Using this process, we discovered amathematicalmodel of ordinary dif-ferential equations representing every known and three unknown elementsin the network with a defined starting concentration value that was eitherspecified (for the drugs) or inferred by the automated method. Elementsin the network could be a signaling molecule (such as cAMP or serotonin),a protein (such as SERT), or a pharmacological compound (such as iver-mectin). The changes in their concentrations over time (dynamics) caninvolve regulatory interactionswith other elements in the network, exponen-tial decay, and fluctuations due to random noise (text S1). Once we haddefined the elements in the starting network, we defined a set of requiredregulatory links connecting the elements in the network controlling de-polarization through melanocyte activation, as suggested by the functionalexperiments in this and previous literature (6, 30, 38, 55–58). Note that thestarting network did not include GlyR, the target of ivermectin, because theautomated method could identify and add this element reliably if necessary,and we preferred to start with the minimal number of humanly definedelements and regulatory links in the model. These required links couldnot be removed during exploration of the network model space, yet theirparameters needed to be defined by the automated method. Using an auto-

ww

mated iterative simulation and modeling process (described in text S2), wereverse-engineered a network that exhibited bistable and stochastic behavior,resulting in wild-type or hyperpigmented outputs with frequencies matchingexperimental data (Fig. 5A). To reverse-engineer diverse regulatory phe-nomena, we used Hill kinetics (59) as a generalization of Michaelis-Mentenkinetics and an algorithm based on evolutionary computation (44, 60).

The evolutionary algorithm maintains a population of signaling net-works, including the initial set of fixed regulatory links in addition tonew randomly generated links and random parameters based on bio-chemically plausible rules. The algorithm periodically refines new networkmodels by crossing and mutating existing network models and by usingrandomoperations to add and delete nodes and links and to alter parameters,which enables the exploration of a large set of possiblemodels. By applyingerror analysis, the algorithm identified the best models in each evolution,discarding the poor scoring network models and proceeding iterativelyfor a defined number of generations to identify the network model withthe best fit to the experimental data.

We generated a set of data using various pharmacological manipulationswith or without ivermectin, similar to those described in Fig. 1, and scoredthe frequency of hyperpigmentation in each condition (data S2). We used asubset of these pharmacological experiments as a training data set using theevolutionary algorithm and error analysis to identify a network model (Fig.5A and text S3 for the ordinary differential equations) and a second set as atest data set to explore the predictive power of the model.

The model identified using the training data set contained threeyet-to-be-characterized elements (labeled a, b, and c in Fig. 5A). Simula-tions using the model network parameters and various test conditions con-verged to a steady state in which the hyperpigmentation was complete(values close to 1) or nonexistent (values close to 0); no simulation producedan intermediate pigmentation value, thus matching the all-or-none pheno-type. Simulations of the no-treatment condition resulted in a hyperpigmen-ted phenotype in 3% of the cases (Fig. 5B), which is consistent with thefrequency of hyperpigmentation that occurred in the control embryos (dataS2). Simulations of the effect of cyanopindolol (an inhibitor of 5HT-R1)resulted in the hyperpigmentation phenotype in 32% of cases (Fig. 5B),whichmatcheswithin 10% error the frequencywe observed (Fig. 6A). Sim-ulations of the effect of combined ivermectin and cyanopindolol treatmentproduced hyperpigmented phenotypes 89% of the cases (Fig. 5B), whichmatches within 5% error the frequency that we observed (Fig. 6A).

The performance (as the percentage of correct outcomes) of the automat-ically reverse-engineered network compared to the experimental results invivo was greater than 85% for all the 20 experiments that we used in thetraining data set during the network search by the evolutionary algorithm(Fig. 6A). Thus, the inferred model uncovered by this process correctlypredicted both the large-scale dynamics and the quantitative stochasticoutcomes of the data set.We also assessed the performance of the model bycomparing the outcome to data from experiments that were not in thetraining set. All the outcomes predicted by the model for this test data setcorrectly predicted the outcome, with the model exhibiting a performancegreater than 75% (Fig. 6B).

We examined the dynamics of the model with simulations of differenttreatments to gain an understanding of the all-or-none behavior. Projectingthe trajectories of several experiments in the phase space (the state of differ-ent components in the network through time) revealed different dynamicattractors (stable outcomes, in terms of product concentrations and melano-cyte properties, to which the system can converge) for the hyperpigmentedand nonhyperpigmented phenotypes, along with phase bifurcations (aglobal change in the system behavior) after applying different treatments.We assessed the effect of no treatment (Fig. 7A), ivermectin (Fig. 7B), orforskolin (Fig. 7C) on the trajectories of two serotonin receptors (5HT-R1






Dow

nloaded from

and 5HT-R2) and the resulting degree of hyperpigmentation through time in100 simulations of the inferred model, starting from the initial state (earlyembryo, yellow dot) and ending in the final stable state (hyperpigmentedand nonhyperpigmented tadpole, red and blue dots, respectively). Theattractors defined by the dynamical system, representing the final stablehyperpigmented and nonhyperpigmented phenotypes (red and blue dots,respectively, in Fig. 7,A toC),were always located in anyof the two extremepigmentation states (1, completely hyperpigmented; 0, completely nonhy-perpigmented), explaining the bistability observed for all treatments.

The phase-space trajectory diagrams revealed two additional propertiesof the networkmodel: the presence of a separatrix (a boundary between two

ww

major modes of behavior in the system) and bifurcations in the state space.In each of the examples shown, the trajectories passed near a separatrix. Forexample, Fig. 7A shows a single trajectory resulting in the hyperpigmentedstate (red dot), and the rest of trajectories end in the nonhyperpigmentedstate; a separatrix must lie between these two types of trajectories, definingthe border between the basins (regions of influence) of the two attractors.The presence of a separatrix derives from the dynamics of the networkmodel and can explain how trajectories that start at the same initialconditions can converge to different attractors and, hence, produce the ob-served stochastic penetrance of the hyperpigmented phenotype. The trajec-tory diagrams also revealed the creation of bifurcations in the state space due

A

Time (arb. unit)Time (arb. unit)



Co

nce

ntr

atio

n (

arb

. un

it)

Co

nce

ntr

atio

n (a

rb. u

nit

)C

on

cen

trat

ion

(arb

. un

it)

No treatment97% 3%

Cyanopindolol68% 32%

IVM + cyanopindolol

11% 89%

B

0.5

1

1.5

2

0.5

1

1.5

2

0.5

1

1.5

2

0.5

1

1.5

2

0.5

1

1.5

2

0.5

1

1.5

2

HP

5 H-T

R1

R2 R5

SERT

MSH

PKA

a

IVM

CREB

b

cAMP

Cyanopindolol

Altanserin SB 699551

Med 5HT

FluoxetineReserpine

VMATAA

SHU 9119MSH-RIF

MTP

Forskolin2'5'-DDA

VP16-XlCreb1

H89

A-CREB

c

Pigmentation level

5HT receptor 5

5HT receptor 1

5HT receptor 2

SERT

VMAT

MSH

cAMP

PKA

CREB

5HT

Unknown a

Unknown b

Unknown c

Normal pigmentation Hyperpigmentation

Fig. 5. Inferred dynamic stochasticmodel recapitulating the observed pheno-type frequencies. (A) Reverse-engineered network that produces hyperpig-mentation phenotype at frequencies matching those observedexperimentally. Starting network elements were the drugs (text without ashape) and the named nodes andblue interaction lines, representing knownelements and relationships in the hyperpigmentation (HP node) pathway.Elements added algorithmically included three unknown required compo-nents (nodes a, b, and c) and green interaction lines. The model also de-scribed the kinetic parameters (see text S3 for details). Multiple signalinginteractions are combined together in a necessary (dashed lines) or suffi-cient (solid lines) fashion (see Materials andMethods for details). Activatingregulatory interactions are indicated with arrows, and inhibiting interactionsare indicated with blunt lines. (B) Simulations of the model showing the dy-namic changes in concentration of each colored node after running themodel to steady state and the proportion of normally pigmented (pigmen-tation value of 0) and hyperpigmented (pigmentation value of 1) tadpoles atthe end of the simulation. Simulations of control (no treatment), the pres-ence of cyanopindolol, an inhibitor of 5HT-R1, and ivermectin and cyano-pindolol are shown.




to the pharmacological treatments, which produced a variation in the loca-tion and basin of the attractors (note the different positions of the red andblue dots, which are especially evident when comparing Fig. 7, B and C).Thus, by altering the dynamics of the network, pharmacological treatmentsproduced different trajectories to the attractors, which can explain thechange in frequencies of the resultant stochastic hyperpigmented and non-hyperpigmented phenotypes.

ww

A qualitative summary of the trajectories and attractors under differ-ent treatments highlights the bifurcations produced in the state space, thephenotypes associated with the different attractors, and their relative fre-quency for each treatment (Fig. 7D). Starting with an identical embryo(identical initial conditions), the system defined by the inferred modelcan stochastically converge to either a hyperpigmented or a nonhyperpig-mented phenotype within a specific frequency, and the frequency can be

http://stke.sciencemag.

Dow

nloaded from

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Mo

del

per

form

ance

A

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Mo

del

per

form

ance

B

Control

IVM

Cyanopindolo

l

IVM

+ cy

anopindolo

l

Altanse

rin

IVM

+ alta

nserin

SB 699551

IVM

+ SB 699551

Med 5HT

MTP

IVM

+ M

TP

IVM

+ flu

oxetine

IVM

+ re

serp

ine

MSH-R

IF

SHU 9119

2'5'-DDA

Forskolin

VP16-XlC

reb1

A-CREB

H89

IVM

+ M

SH-RIF

SHU 9119 + Flu

oxetine

SHU 9119 + 2'5'-D

DA

IVM

+ 2'5'-D

DA

IVM

+ alta

nserin

+ M

RIF

IVM

+ A

-CREB

IVM

+ H

89

Fig. 6. Inferred model performance with the experiments in the trainingdata set and the experiments in the validation data set. (A) Percent-

indicates 85% performance accuracy. (B) Percentage of correctpredictions of the model for a set of experiments not used in model

on org/
age of correct outcomes of the model for each of the experiments
used to reverse-engineer the model (training data set). Dashed line
generation (validation data set). See data S2 for details. MTP,methiothepin.
February 14, 2020

5HT-R1 5HT-R2

Hyp

erp

igm

enta

tio

n

IVM

B

5HT-R1 5HT-R2

Hyp

erp

igm

enta

tio

n

Forskolin

CNo treatmentIVMForskolin

5HT-R1

SERT

Low Medium High

Low

Med

ium

Hig

hD

Hyp

erp

igm

enta

tio

n

5HT-R1 5HT-R2

Initial stateFinal hyperpigmented stateFinal nonhyperpigmented state

No treatment

A

Fig. 7. Phase space of the inferredmodel showing stochastic developmentaltrajectories and pharmacological treatment bifurcations. (A to C) Trajectories

proportional to the number of trajectories that converge to that phenotype.Note that in the ivermectin condition, there are two attractors for the hyper-

of the state of two serotonin receptors (5HT-R1 and 5HT-R2) and degree ofhyperpigmentation in the inferred model during 100 simulations for each ofthree representative treatments. “0” represents inactive on the 5HT-R axes;“2” or “2.5” represents full activity. “0” on the hyperpigmentation axis repre-sents normally pigmented outcome; “1” represents a fully hyperpigmentedoutcome. The initial state of the embryo (yellow dot) is the same for all thesimulations for a given treatment. The attractors represent hyperpigmentedphenotypes (red dots) and nonhyperpigmented phenotypes (blue dots) andare the stable states to which the system converges. The attractor dot size is

pigmented state, a representation of how different trajectories can producethe same phenotypic outcome. (D) Qualitative phase-space trajectoriessummarizing the dynamics of the inferred model during three different treat-ments. In the model, each treatment represents a change to the initial con-centration of the indicated pharmacological drug, resulting in a bifurcation (aglobal change in the dynamics of the system) in the signaling network, a shiftin the attractors, and different frequencies for the resultant phenotypes. Thetrajectory linewidths are proportional to the frequency of simulationsproducing the trajectory.






Dow

nloaded from

precisely altered with the application of certain pharmacological treatments(which change the initial conditions of the corresponding drugs in thenetworkmodel). This analysis revealed howan apparently similar hyperpig-mented phenotype can result from two different molecular states: either ahigh level of 5HT-R1 activity combined with a low level of SERT activity(bottom right in Fig. 7D) or a low level of 5HT-R1 combined with a highlevel of SERT (top left in Fig. 7D).

DISCUSSION

Previous studies have examined the role of Vmem in regulating the pigmen-tation of X. laevis embryos and have suggested serotonin signaling and theinvolvement of multiple serotonin receptors in translating the bioelectricalsignal to a change in melanocyte behavior (6, 30). Here, we present aspectsof the molecular mechanism of the pathway and evidence for a physiolog-ical circuit that controls melanocyte behavior and ultimately tadpole pig-mentation (Fig. 8). Furthermore, we identified a dynamic model thatquantitatively predicted the stochastic effects of pharmacologically alteringvarious control points of this process. The reverse-engineered modelaccounted for the bistable all-or-none hyperpigmentation phenotype andfor the stochastic penetrance observed under multiple conditions, correctlyreproducing the results from experiments used to identify the model andpredicting results from independent experimental conditions

At the level of the molecular pathway, our pharmacological and geneticmanipulations indicated a cAMP-dependent transcriptional pathway asinvolved in the depolarization-induced hyperpigmentation phenotype.Our results also showed that the depolarization of a sparse cell population(instructor cells) stimulated a similar increase in transcription factor geneexpression as that associated with hyperpigmentation mediated by theregional injection ofmRNA encoding the depolarizingKCNE1 ion channel(27), whichwould be expected to depolarize only the injected cells and theirprogeny, not necessarily instructor cells.We found that the expression of thegenes encoding the transcription factors Sox10 and Slug was increased inembryos after ivermectin-induced depolarization. Sox10 is expressed in pre-migratory neural crest cells, and its expression becomes gradually restrictedto cells in the glia or melanocyte lineages (61, 62). In themelanocyte, Sox10plays an indispensable role inmelanocyte survival, proliferation, andmigration(63) and is also expressed in primary andmetastatic melanoma (34–36, 64).We detected increased transcripts for Slug at the tail bud stage after exposureto ivermectin at stage 15, suggesting a persistent increase in expression. Incontrast, we found that transcripts for Sox10 only transiently increased afterdepolarization, gradually returning to baseline by stage 35. These resultsindicate that these two transcriptional regulators may serve differentfunctions and are subject to different regulatory inputs.

Microarray analysis of stage 15 and stage 45 embryos that had been treatedwith ivermectin compared to control embryos revealed differential expres-sion of a relatively small set of genes (44) at stage 15 and a larger set (517) atstage 45. Pathway analysis of homologs of the 517 genes in humans indi-cated that several of them are implicated in different disease conditions, in-cluding melanoma (table S4). For example, ivermectin exposure resulted inan increase in the transcripts for POMC. POMC, a precursor to aMSH andmelanocortin, has been linked to a number of metabolic disorders (65) andseveral cancers, including lung cancer (66) and melanoma (67–70). The re-ceptor for melanocortin, a peptide derived from POMC, is associated withdepressive disorder and is implicated in serotonergic signaling in humandisease (71), consistent with our model’s unification of serotonergic,melanocyte regulatory, and cancer-related pathways. Microarray analysisalso suggested that transcripts encoding transmembrane serine 2 protease,which is encoded by a gene commonly expressed in prostate cancer (72),were increased by ivermectin exposure and that transcripts for albumin, a

ww

soluble, monomeric protein that comprises about one-half of the blood ser-um protein and that has been implicated in breast cancer (73–75), weredecreased. Other responsive genes remain to be identified, because we ana-lyzed pooled embryonic cells for RT-qPCR, which likely obscured changesin transcripts that were limited to small populations of cells or that increasedin some cells and decreased in others. Application of region- or cell-specificnext-generation sequencing approaches should reveal such differentiallyregulated transcripts.

One curious aspect of this phenotype is its all-or-none character: Tad-poleswere never seen to be partially hyperpigmented: Each tadpole is eithernormal or is covered with abnormal melanocytes. This type of bistablephenomena has been reported in bacteria (76), in synaptic plasticity (77),in developmental processes, including the epithelial to mesenchymaltransition (78), the specification of neuron cell types (79), and the matura-tion of Xenopus oocytes (80), and in pathological situations, such as cancer(81). Signaling networks with feedback loops can produce irreversibleswitch-like responses (82),which could explain these all-or-nonephenomenaat the level of individual cells. Creating switch-like responses is a goal ofsynthetic biology, and the approach that we took may be useful for devel-opingmodels that can be used by synthetic biologists to create systemswiththis all-or-none property.

However, at a population level, the hyperpigmented phenotype isstochastic in that many of the conditions produced less than 100% pene-trance of the phenotype. The decision of hyperpigmentation occurred notat the level of individual cells but at the level of the individual. Similartissue-level decision-making has been reported in development duringleft-right patterning with an entire region becoming left or right (83–87),but the mechanisms that achieve this coordinated response are unknown.

Here,we presented and provided a proof of principle for a computationalmethod for reverse-engineering dynamic signaling networks that can ex-plain both bistable and stochastic resultant phenotypes. Starting with a dataset containing the incidence data on hyperpigmentation under 20 differentpharmacological perturbations, the automatedmethod discovered a dynam-ic network that recapitulated the bistability and the stochasticity shown inthe set of experiments performed in vivo. The automated method evaluatedmore than 20 million different networks, comprising 40 billion virtualexperiments, to reverse-engineer the dynamic network that could reca-pitulate the stochastic phenotypes in 20 pharmacological assays. Thenetwork model is multiscale, including candidates (the diffusible factorsMSH, 5HT, and cAMP) for the global coordination of cell statewithin eachembryo and intracellular signaling relationships, and has only three un-known nodes. Because the network model connects these unknown nodesto other nodes into the model, candidate approaches, for example, bysearching protein-protein interaction databases, or untargeted screeningmethods could be used to identify these unknown nodes.

The model correctly predicted the outcomes of new experiments thatwere not used in generating the model, validating the automated computa-tional method and the reverse-engineered network. The model also pre-dicted that cAMP signaling did not occur directly through the serotoninreceptors (5HT-R1, 5HT-R2, and 5HT-R5) involved in the hyperpigmenta-tion response (the discovered network does not contain any direct link fromthe serotonin receptors toward cAMP; Fig. 5A), which signal through Gproteins that either inhibit cAMP production or are coupled to inositol tri-sphosphate and diacylglycerol signaling or protein kinase C. In summary,the method that we presented here can lead to the discovery of dynamicsignaling networks that provide testable predictions and mechanisms ofcomplex stochastic resultant phenotypes.

The discovered network is a model of signaling cascades mediating thecontrol of cell behaviors by an initial bioelectric signaling event; it isconsistent with proposals that some disease-promoting stimuli, such as






Dow

nloaded from

ultraviolet exposure, mediate their effects through changes of resting mem-brane potential (88, 89).Moreover, themodel suggests a number of tractablecontrol points for regulating aberrant cell behaviors, for example, the use ofserotonin reuptake inhibitors, which are used in the treatment of depression,a disease believed to involve aberrant serotonergic signaling (6,90). If thediscovered model reflects a network that is similar to that controlling suchcell behaviors inmammals, then themodelmay represent a starting point forin silico testing of proposed biomedical interventions.

MATERIALS AND METHODS

Animal husbandryXenopus embryos were maintained according to standard protocols (91) in0.1× Marc’s Modified Ringers (MMR; pH 7.8). Xenopus embryos werestaged according to Nieuwkoop and Faber (92). All experimental proce-dures involving Xenopus embryos were approved by the institutionalanimal care and use committees and Tufts UniversityDepartment of Lab-oratory Animal Medicine under protocol M2014-79.

www

MicroinjectionCapped, synthetic mRNAswere dissolved in nuclease-free water (Ambion)and injected into embryos at cleavage stages in 3% Ficoll using standardmethods (91). The mRNA injections were made using borosilicate glassneedles calibrated to bubble pressures of 50 to 70 kPa in water, delivering50- to 70-ms pulses. After 30 min, embryos were washed and cultured in0.1× MMR until desired stages. Constructs used included XlCreb1 (geneID, BC041206), VP16-XlCreb1 (50, 93), injected into one cell of NF stage5 (32-cell) embryos, and XminK [also known as KCNE1 and isK (27, 94)],injected into one cell of NF stage 3 (4-cell) embryos.

Drug exposureStocks of ivermectin (Sigma) were stored at 10 mM concentration in di-methyl sulfoxide. Embryos were exposed in 0.1× MMR (91) from stage10 to stage 45 in the presence of 1 mM ivermectin (Sigma), 500 mM 2′5′-dideoxyadenosine (Sigma), 5 mMforskolin, 25 mMMSH-RIF, 500 nMSHU9119, 50 mMcyanopindolol, 10 mMaltanserin, 2.5 mMSB 699551, 5 mM5HT added to medium, 10 nM methiothepin, 15 mM fluoxetine, 100 mMreserpine, and 20 mM H89 (sources are listed in data S2).

MelanocyteGlyR-expressingcell

Pituitary

MSHCl−

SERT

5-H

T

GlyR

5-H

T

5-H

T

5-H

T

VMAT

MSH

POM

C

No

rmal

Pola

rize

d in

stru

cto

r cel

lIv

erm

ectin

Dep

olar

ized

inst

ruct

or c

ell

Melanocyte

MC

1R

MSHCl−

Cl−

SERT

GlyR

5-H

T

Instructorcell

depolarized

Instructorcell

polarized5-

HT

5-H

T

5-H

T

5-H

T

Iver

mec

tin

VMAT

Pituitarymelanotrope

cell

Pituitarymelanotrope

cell

MSH

5HT-

R5

5HT-

R15H

T-R1

POM

C

MC

1R

Melanocyte

MSH

MSHMSH

MSH

MSH

MSH

MSHMSH

Cl−

5HT-

R25H

T-R2

5HT-

R5

Fig. 8. Schematic pathways for melanocyte control downstream of instructor ceptors stimulates AC activity (not shown), thereby enhancing MSH release.
cell signaling.Under normal conditions thepolarized instructor cells producearelatively small serotonergic signal (stars) due to reuptakeand retentionof 5HTby the SERT into the instructor cells. At this relatively low concentration ofserotonin, only a high-affinity serotonin receptor 5HT-R5 on the pituitary mela-notrope cells may become activated, thereby producing normally pigmentedanimals. Right: When the instructor cell population is depolarized, SERT ex-ports serotonin, resulting in increased serotonin concentrations in the micro-environment of themelanocytes and the pituitary. At this higher concentration,serotonin may activate 5HT-R5 and 5HT-R2 on both the pituitary melanotropecells and on melanocyte cells. In the melanotrope cells, activation of 5HT re-
Serotonin signaling increases the abundance of pro-opiomelanocortin(POMC), a precursor of MSH. MSH binding to melanocortin 1 receptors(MC1R) on melanocytes stimulates AC activity (not shown), increasing cAMPproduction, PKA activity (not shown), and CREB phosphorylation and activity(not shown), thereby increasing the expression of Sox10 and Slug. Byincreasing cAMP in melanocytes, serotonin may also contribute proliferation,invasive migration, and the altered morphology of the melanocytes. Althoughmelanocytes can also respond to serotonin directly, this pathway is not shownhere because the most parsimonious network that explains all of the results,including the long-rangesignalingby instructor cells, doesnot require this link.

.SCIENCESIGNALING.org 6 October 2015 Vol 8 Issue 397 ra99 10





Dow

nloaded from

All compounds were obtained from Tocris, unless otherwise noted, andprepared in Millipore (18 megohms) water, unless indicated. All drug treat-ments were performed using embryos from mixed batches of fertilizations,using at least three biological replicates. Control experiments were per-formed using embryos in normal media (0.1× MMR).

MicrosurgeryBefore microsurgery, stage 32 embryos were anesthetized in a 0.02% tri-caine solution (pH 7.5) in 0.1× MMR. Using a surgical blade (Feather#11), tissue was excised from one of the three regions depending on thetreatment. In the first experimental group, a wedge-shaped cut was madebetween the eye and the cement gland, corresponding to the region of thedeveloping pituitary. In the second experimental group, a single cut removedthe entire anterodorsal region of the head (above the eye), corresponding tothe region of the developing pineal gland. In the third experimental group,awedge-shaped cut removed the tissue between the cement gland and stom-ach, which served as a control for wounding. After tissue removal, animalswere kept anesthetized for 30min, after which they werewashed twicewith0.1× MMR. After washing, operated animals were raised at 16°C under a12-hour light/12-hour dark cycle and were scored for the presence or ab-sence of hyperpigmentation at stage 46.

Expression analysisIn situ hybridization was performed as previously described (95). Xenopusembryoswere collected and fixed inMEMFA for 30min (91). Before in situhybridization, embryos were washed in phosphate-buffered saline (pH 7.4)plus 0.1% Tween 20 and then transferred to methanol through a 25%/50%/75% series. Probes for in situ hybridization were generated in vitro fromlinearized templates using a digoxigenin-labeling mix (Roche). Chromo-genic reaction times optimized signal/background ratio. Analyses representconsistent patterns from 50 to 60 embryos for each marker. Probes used forin situ hybridization include Sox10 (GenBank accession no. AY149116)and Slug (GenBank accession no. AF368040).

RNA extraction and cDNA synthesisEmbryos (collected n = 10 per Eppendorf tube, five biological replicates)were washed in ribonuclease (RNase)– and deoxyribonuclease (DNase)–free water and homogenized in TRIzol reagent (Life Technologies).Homogenized samples were stored at −80°C for up to 1 month. TotalRNAwas extracted using TRIzol according to the manufacturer’s instruc-tions (Life Technologies). RNAyield and quality were assessed by spectro-photometry (ND-1000, NanoDrop) and gel electrophoresis, respectively, toassess integrity of 28S and 18S RNA.

Reverse transcription was performed using ThermoScript RT-PCRSystem (Life Technologies). Each in vitro reverse transcription reactionwas performed using 1 mg of total RNA and 50 mg of oligo(dT)20 primers(Life Technologies). RNA and primers were mixed, denatured for 5 min at65°C, and placed on ice before adding the reaction mix according to themanufacturer’s instructions. Reverse transcription reaction was carried outat 50°C for 45min. The reaction was terminated by incubating at 85°C for5 min, followed by RNA degradation using 1 mg of RNase H for 20min at37°C. The complementary DNA (cDNA) was stored at −20°C until use.The quality and quantity of cDNAwere validated using Advantage 2 PCRkit (Clontech) on cDNA samples using a-tubulin primers (96).

Quantitative real-time PCRPrimers (table S5)were designed usingPrimer3Plus enhancedweb interface(97) for Sox10 (AY149116) and Slug (AF368040).Ornithine decarboxylase(ODC), awidely used endogenous control forXenopus, was used to normalizetarget gene expression, and primer sequences have been previously published

www

(98). The PCR specificity was verified by BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi) using the National Center for Biotechnology InformationX. laevis reference sequence.Desalted primerswere obtained from Invitrogenby Life Technologies.

For each primer pair, standard curve primer analysis was performedusing serial dilutions of cDNA from stage 25 control embryos [1 (un-diluted), 10−1, 10−2, 10−3]. Formation of primer-dimer and amplificationspecificity were assessed by efficiency and melt curve analysis.

The cDNA from validated RNAwas used to perform RT-qPCR assays.For each biological sample, three technical replicates were run in eachRT-qPCR experiment. Each treatment contained five biological replicates.Triplicate negative controls lacking template were also run for each cDNAsample for each reaction.

PCRs were assembled manually. Samples were prepared by adding 1 mlof cDNA (diluted 1:5 in ddH2O), 10 ml of 2× Power SYBR Green PCRMaster Mix (Applied Biosystems), and 0.5 ml of each primer (diluted to10 mM) in a final volume of 20 ml. Reactions were incubated in 96-wellMicroAmp Optical Reaction plates at 95°C for 10 min followed by 40cycles at 95°C for 15 s and at 60°C for 1 min in a StepOnePlus qPCRinstrument (Applied Biosystems).

The RT-qPCR data were analyzed using the StepOne software v.2.3, andDDCT values were calculated (Applied Biosystems). Fold change of targetgenes relative to the amount of the control geneODCwas calculated as−2DDCT.

Microarray analysisGene expression analysis was performed using samples treated with iver-mectin fromNF stage 10 onward and collected at two developmental stages:stage 15 (early neurula) and stage 45 (free-swimming tadpole). Embryoswere collected in Eppendorf tubes (n = 50 for stage 15 and n = 5 for stage45) and frozen at −80°C. RNA extraction andmicroarray analysis were per-formed by the Beth Israel DeaconessMedical Center Genomics and Proteo-mics Center (Harvard) according to the standardAffymetrix protocol, usinga high-throughput hybridization and scanning system. Microarray hybrid-ization was performed using the Affy 3′ IVT Express Kit. Fragmented andbiotin-labeled and amplifiedRNAwas hybridized to theGeneChipXenopuslaevis Genome 2.0 Array as per the manufacturer’s protocol. The Affyme-trix GeneChip Xenopus laevis Genome 2.0 Array has 32,400 probe setsrepresenting more than 29,900 X. laevis transcripts.

All microarray data were analyzed using Bioconductor packages in R.The quality of hybridized arrays was assessed using Affymetrix guidelineson the basis of scaling factor, background value, mean intensity of chip, and3′ to 5′ ratios for spike-in control transcripts. The outlier analysis wasperformed using unsupervised clustering and principal components analy-sis. All high-quality arrays were normalized using the MAS5 algorithm de-veloped by Affymetrix. The absent or present calls for the transcripts werecalculated using the MAS5 algorithm. The differentially expressed tran-scriptswere identified on the basis of fold change andAffymetrix transcriptscalls. Transcripts with increased abundancewere considered to be those thatchanged more than twofold in the experimental group compared to thecontrol group, with present calls in the experimental group. Transcriptswithdecreased abundance were considered to be those that changed more thantwofold in the experimental group compared to the control group, withpresent call in the control group. Differentially expressed transcripts wereidentified using customized script in R. The list of differentially expressedgenes was annotated using the Affymetrix Xenopus platform.

BioinformaticsFunctional enrichment for differentially expressed genes was conductedusing the GOrilla database (53) (updated on 2 August 2014). X. laevisgene names were uploaded and mapped using the official gene symbol.






Dow

nloaded from

H. sapiens was used as the reference species for gene ontology assign-ments. Pathway Studio 9.0 (Ariadne) was used for pathway analysis toconstruct gene interaction networks. For early embryos, 21 genes weremapped to an annotated pathway in the database (data S1), whereas inthe late embryo experiments, 210 genes were successfully mapped to anannotated pathway (data S1) [see (99) for additional details on performingpathway analysis].

Model definitionWe modeled dynamical signaling networks as systems of stochastic ordi-nary differential equations. A signaling network was composed of a set ofinterconnected elements, including the products perturbed during theexperiments, the pharmacological drugs used, an element representing thedegree of pigmentation in the tadpole, and any additional unknownproductsnecessary to achieve the correct stochastic behavior of the system and theresultant all-or-none states. Each element in a network is representedwith anequation, modeling its production rate as the linear relation between a pro-duction term, a decay term, and a noise term. The production term ismodeled with a combination of Hill functions, each of them representingthe regulatory interaction between two products.Whenmultiple interactionsregulate a product, they can be grouped as sufficient, necessary, or a com-bination thereof. Sufficient interactions can regulate the production of thetarget product independently of other sufficient interactions (modeledwith amaximum operator, similar to a logic OR interaction; hence, activation haspreference over inhibition). In contrast, all the necessary interactions need tobe coordinated to regulate the production of the target product (modeledwith a minimum operator, similar to a logic AND interaction; hence, inhi-bition has preference over activation). This scheme allows the modeling ofmany different regulatory interaction combinations.

The following equation illustrates amodel of the production of product aas regulated by two necessary products (activator b and inhibitor c) and twosufficient products (activator d and inhibitor e):

da

dt¼ ra min

bh1

ah11 þ bh1;

ah22ah22 þ ch2

;

�

maxdh3

ah33 þ dh3ah44

ah44 þ eh4

� ��− laaþ x tð Þ

where ra is the production constant, hi are the Hill coefficients, ai arethe dissociation constants in the Hill functions, la is the decay constant,and x(t) is a Gaussian random noise of zero mean (see text S1 for furtherdetails).

Model simulationGiven a signaling model network described as a system of ordinary differen-tial equations, an experiment can be simulated in silico by numerically inte-grating the system with a set of initial conditions. The initial conditionsinclude any pharmacological element used in the experiment as a constantequal to 1.0 in the case of its presence in the experiment or equal to 0.0 inthe case of its absence. Because of the stochastic nature of the system of equa-tions, the pigmentation level resultant from a simulation for a given set of ini-tial conditions can vary among different runs (see text S1 for further details).

Model evaluationTo calculate the predictive power of a given signaling network model, eachexperiment (defining a set of pharmacological treatments) was simulated100 times. The resultant phenotype of a simulation was considered hyper-pigmented if the level of the corresponding product was above 0.9 and non-hyperpigmented if it was below 0.1. Then, we intuitively define the error of

www

a signaling networkmodelM for a setE of n experiments as themean squareerror between the in silico and invivo outcomes for each of the experiments:

Error M ;Eð Þ ¼ 1

n∑n

i¼1

�Hi − H ′

i

�2 þ �Ni − N ′

i

�2� �

whereHi andNi are the resultant frequency of hyperpigmented and non-hyperpigmented phenotypes in the in silico simulation of experiment i, andH′i andN′i are the resultant frequency of hyperpigmented and nonhyperpig-mented phenotypes in the in vivo assay of experiment i. In addition, themodel error used during the search penalizes phenotypes that are not similarto any of the observed resultant phenotypes in the in vivo experiment, aswell as models that do not converge to a steady state. See text S1 for adetailed description of the error calculation.

Automated reverse-engineering methodWe implemented a computational method based on (44) to automate thediscovery of the topology (regulatory links) and parameters of a signalingnetwork that minimize the error for a given set of training experiments. Inaddition to the training data set of experiments, the algorithm takes as inputa set of regulatory links from known interaction pathways that will con-strain the generation of candidatemodels—these sets of linkswill be includedin any model considered during the search. The method uses an evolution-ary algorithm approach, in which a set of signaling networks evolves insilico by crossing over and mutating them and replacing those withworse error (100, 101). After a fixed number of generations, in our case10,000, the best network is returned by the algorithm. See text S2 for fur-ther details.

Inferred model of bistable andstochastic hyperpigmentationWeapplied our automatedmethod to a training data set of 20 experiments toreverse-engineer a model represented with a system of ordinary differentialequations that could explain the bistability and exact stochasticity reportedin the experiments with the least error. The experiments included theapplication of 15 different pharmacological drugs affecting 10 knownproducts (an experiment can combine several drugs) and the resultant fre-quencies of hyperpigmented and nonhyperpigmented phenotypes. Thesearchwas constrainedwith a set of 11 links, representing known regulatoryinteractions between signaling products, and additional 17 links, represent-ing the interactions between pharmacological drugs and signaling products.No parameters were constrained in these predefined links, except thewell-known activator or inhibitor character of the interactions between drugsand signaling products. The algorithm took 36 hours in a computer clusterwith 22CPUs to reverse-engineer themodel shown in Fig. 5A. As shown inthe reversed-engineered model, the algorithm inferred, as necessary, 3 ad-ditional uncharacterized products (labeled a, b, and c in Fig. 5A) and 24regulatory links between products (green links in Fig. 5A), as well as a totalof 145 numerical parameters (see text S3 for the complete system of equa-tions and parameters).

SUPPLEMENTARY MATERIALSwww.sciencesignaling.org/cgi/content/full/8/397/ra99/DC1Text S1. Model implementation, simulation, and evaluation.Text S2. Method for reverse-engineering a stochastic model of hyperpigmentation.Text S3. System of equations and kinetic parameters for the reverse-engineered model.Fig. S1. Hierarchical clustering of the differentially regulated transcripts.TableS1. A list of the 45 transcripts differentially expressedby stage 15 in response to ivermectin.TableS2.A list of the517 transcriptsdifferentially expressedbystage45 in response to ivermectin.Table S3. Enriched GO affected in stage 45 embryos.Table S4. Cancer-related genes in humans of the homologs of the genes differentiallyexpressed in stage 45 Xenopus tadpole embryos after depolarizing ivermectin treatment.






Dow

nloaded from

Table S5. Reference genes and primers for RT-qPCR.Data S1. Differentially expressed transcripts in early and late embryos and their associa-tion with disease or cellular process.Data S2. Results of the training set and validation set experiments and the performance ofthe model for each.

REFERENCES AND NOTES1. C. D. McCaig, B. Song, A. M. Rajnicek, Electrical dimensions in cell science. J. Cell

Sci. 122, 4267–4276 (2009).2. N. Prevarskaya, R. Skryma, Y. Shuba, Ion channels and the hallmarks of cancer.

Trends Mol. Med. 16, 107–121 (2010).3. S. Sundelacruz, M. Levin, D. L. Kaplan, Role of membrane potential in the regulation

of cell proliferation and differentiation. Stem Cell Rev. 5, 231–246 (2009).4. M. H. You, M. S. Song, S. K. Lee, P. D. Ryu, S. Y. Lee, D. Y. Kim, Voltage-gated K+

channels in adipogenic differentiation of bone marrow-derived human mesenchymalstem cells. Acta Pharmacol. Sin. 34, 129–136 (2013).

5. J. Y. Lan, C. Williams, M. Levin, L. D. Black III, Depolarization of cellular restingmembrane potential promotes neonatal cardiomyocyte proliferation in vitro. CellMol. Bioeng. 7, 432–445 (2014).

6. D. Blackiston, D. S. Adams, J. M. Lemire, M. Lobikin, M. Levin, Transmembranepotential of GlyCl-expressing instructor cells induces a neoplastic-like conversionof melanocytes via a serotonergic pathway. Dis. Model Mech. 4, 67–85 (2011).

7. F. Ding, G. Zhang, L. Liu, L. Jiang, R. Wang, Y. Zheng, G. Wang, M. Xie, Y. Duan,Involvement of cationic channels in proliferation and migration of human mesenchymalstem cells. Tissue Cell 44, 358–364 (2012).

8. A. Tseng, M. Levin, Cracking the bioelectric code: Probing endogenous ioniccontrols of pattern formation. Commun. Integr. Biol. 6, e22595 (2013).

9. M. Levin, Molecular bioelectricity in developmental biology: New tools and recentdiscoveries: Control of cell behavior and pattern formation by transmembranepotential gradients. Bioessays 34, 205–217 (2012).

10. M. Levin, Bioelectric mechanisms in regeneration: Unique aspects and futureperspectives. Semin. Cell Dev. Biol. 20, 543–556 (2009).

11. M. Levin, Molecular bioelectricity: How endogenous voltage potentials control cellbehavior and instruct pattern regulation in vivo.Mol. Biol. Cell 25, 3835–3850 (2014).

12. D. M. Thompson, A. N. Koppes, J. G. Hardy, C. E. Schmidt, Electrical stimuli in the centralnervous system microenvironment. Annu. Rev. Biomed. Eng. 16, 397–430 (2014).

13. B. Chernet, M. Levin, Bioelectric signaling in cancer. J. Clin. Exp. Oncol., in press.14. M. Yang, W. J. Brackenbury, Membrane potential and cancer progression. Front.

Physiol. 4, 185 (2013).15. A. Becchetti, Ion channels and transporters in cancer. 1. Ion channels and cell pro-

liferation in cancer. Am. J. Physiol. Cell Physiol. 301, C255–C265 (2011).16. A. Arcangeli, Ion channels and transporters in cancer. 3. Ion channels in the tumor cell-

microenvironment cross talk. Am. J. Physiol. Cell Physiol. 301, C762–C771 (2011).17. B. L. Than, J. A. Goos, A. L. Sarver, M. G. O’Sullivan, A. Rod, T. K. Starr, R. J. Fijneman,

G. A. Meijer, L. Zhao, Y. Zhang, D. A. Largaespada, P. M. Scott, R. T. Cormier, The roleof KCNQ1 in mouse and human gastrointestinal cancers. Oncogene 33, 3861–3868(2013).

18. W. J. Brackenbury, L. L. Isom, Voltage-gated Na+ channels: Potential for b subunitsas therapeutic targets. Expert Opin. Ther. Targets 12, 1191–1203 (2008).

19. W. Stühmer, F. Alves, F. Hartung, M. Zientkowska, L. A. Pardo, Potassium channelsas tumour markers. FEBS Lett. 580, 2850–2852 (2006).

20. A. Felipe, R. Vicente, N. Villalonga, M. Roura-Ferrer, R. Martínez-Mármol, L. Solé,J. C. Ferreres, E. Condom, Potassium channels: New targets in cancer therapy.Cancer Detect. Prev. 30, 375–385 (2006).

21. A. Arcangeli, S. Pillozzi, A. Becchetti, Targeting ion channels in leukemias: A newchallenge for treatment. Curr. Med. Chem. 19, 683–696 (2012).

22. A. Arcangeli, O. Crociani, E. Lastraioli, A. Masi, S. Pillozzi, A. Becchetti, Targetingion channels in cancer: A novel frontier in antineoplastic therapy. Curr. Med. Chem.16, 66–93 (2009).

23. R. Ouwerkerk, M. A. Jacobs, K. J. Macura, A. C. Wolff, V. Stearns, S. D. Mezban,N. F. Khouri, D. A. Bluemke, P. A. Bottomley, Elevated tissue sodium concentration inmalignant breast lesions detected with non-invasive 23Na MRI. Breast Cancer Res.Treat. 106, 151–160 (2007).

24. B. T. Chernet, M. Levin, Transmembrane voltage potential is an essential cellularparameter for the detection and control of tumor development in a Xenopus model.Dis. Model. Mech. 6, 595–607 (2013).

25. C. D. Cone Jr., The role of the surface electrical transmembrane potential in normaland malignant mitogenesis. Ann. N. Y. Acad. Sci. 238, 420–435 (1974).

26. H. S. Burr, L. C. Strong, G. M. Smith, Bio-electric correlates of methylcolantherene-induced tumors in mice. Yale J. Biol. Med. 10, 539–544 (1938).

27. J. Morokuma, D. Blackiston, D. S. Adams, G. Seebohm, B. Trimmer, and M. Levin,Modulation of potassium channel function confers a hyperproliferative invasive pheno-type on embryonic stem cells. Proc. Natl. Acad. Sci. U.S.A. 105, 16608–16613 (2008).

www

28. D. S. Micalizzi, H. L. Ford, Epithelial–mesenchymal transition in development andcancer. Future Oncol. 5, 1129–1143 (2009).

29. C. L. Chaffer, E. W. Thompson, E. D. Williams, Mesenchymal to epithelial transitionin development and disease. Cells Tissues Organs 185, 7–19 (2007).

30. M. Lobikin, B. Chernet, D. Lobo, M. Levin, Resting potential, oncogene-induced tu-morigenesis, and metastasis: The bioelectric basis of cancer in vivo. Phys. Biol. 9,065002 (2012).

31. J. Hannon, D. Hoyer, Molecular biology of 5-HT receptors. Behav. Brain Res. 195,198–213 (2008).

32. C. I. Rodríguez, V. Setaluri, Cyclic AMP (cAMP) signaling in melanocytes and mel-anoma. Arch. Biochem. Biophys. 563, 22–27 (2014).

33. Y. Aoki, N. Saint-Germain, M. Gyda, E. Magner-Fink, Y. H. Lee, C. Credidio,J. P. Saint-Jeannet, Sox10 regulates the development of neural crest-derived melano-cytes in Xenopus. Dev. Biol. 259, 19–33 (2003).

34. M. Agnarsdóttir, L. Sooman, A. Bolander, S. Strömberg, E. Rexhepaj, M. Bergqvist,F. Ponten, W. Gallagher, J. Lennartsson, S. Ekman, M. Uhlen, H. Hedstrand, SOX10expression in superficial spreading and nodular malignant melanomas. MelanomaRes. 20, 468–478 (2010).

35. S. A. Graf, C. Busch, A. K. Bosserhoff, R. Besch, C. Berking, SOX10 promotes mel-anoma cell invasion by regulating melanoma inhibitory activity. J. Invest. Dermatol.134, 2212–2220 (2014).

36. O. Shakhova D. Zingg, S. M. Schaefer, L. Hari, G. Civenni, J. Blunschi, S. Claudinot,M. Okoniewski, F. Beermann, D. Mihic-Probst, H. Moch, M. Wegner, R. Dummer,Y. Barrandon, P. Cinelli, L. Sommer, Sox10 promotes the formation and mainte-nance of giant congenital naevi and melanoma. Nat. Cell Biol. 14, 882–890(2012).

37. E. Casas, J. Kim, A. Bendesky, L. Ohno-Machado, C. J. Wolfe, J. Yang, Snail2 is anessential mediator of Twist1-induced epithelial mesenchymal transition and metas-tasis. Cancer Res. 71, 245–254 (2011).

38. A. S. Dobroff, H. Wang, V. O. Melnikova, G. J. Villares, M. Zigler, L. Huang, M. Bar-Eli,Silencing cAMP-response element-binding protein (CREB) identifies CYR61 as a tumorsuppressor gene in melanoma. J. Biol. Chem. 284, 26194–26206 (2009).

39. D. Lobo, W. S. Beane, M. Levin, Modeling planarian regeneration: A primer for reverse-engineering the worm. PLOS Comput. Biol. 8, e1002481 (2012).

40. P. François, V. Hakim, Design of genetic networks with specified functions by evo-lution in silico. Proc. Natl. Acad. Sci. U.S.A. 101, 580–585 (2004).

41. J. Bongard, H. Lipson, Automated reverse engineering of nonlinear dynamicalsystems. Proc. Natl. Acad. Sci. U.S.A. 104, 9943–9948 (2007).

42. T. J. Perkins, J. Jaeger, J. Reinitz, L. Glass, Reverse engineering the gap genenetwork of Drosophila melanogaster. PLOS Comput. Biol. 2, e51 (2006).

43. A. Crombach, K. R. Wotton, D. Cicin-Sain, M. Ashyraliyev, J. Jaeger, Efficient reverse-engineering of a developmental gene regulatory network. PLOS Comput. Biol. 8,e1002589 (2012).

44. D. Lobo, M. Levin, Inferring regulatory networks from experimental morphologicalphenotypes: A computational method reverse-engineers planarian regeneration.PLOS Comput. Biol. 11, e1004295 (2015).

45. P. D. Nieuwkoop, J. Faber, Normal Table of Xenopus laevis (Daudin) (North-HollandPublishing Company, Amsterdam, ed. 2, 1967).

46. D. J. Blackiston, G. M. Anderson, N. Rahman, C. Bieck, M. Levin, A novel method forinducing nerve growth via modulation of host resting potential: Gap junction-mediatedand serotonergic signaling mechanisms. Neurotherapeutics 12, 170–184 (2014).

47. D. S. Adams, M. Levin, Endogenous voltage gradients as mediators of cell-cell com-munication: Strategies for investigating bioelectrical signals during pattern formation.Cell Tissue Res. 352, 95–122 (2013).

48. A. J. Shaywitz, M. E. Greenberg, CREB: A stimulus-induced transcription factoractivated by a diverse array of extracellular signals. Annu. Rev. Biochem. 68,821–861 (1999).

49. D. Lubelski, T. A. Ponzio, H. Gainer, Effects of A-CREB, a dominant negative inhib-itor of CREB, on the expression of c-fos and other immediate early genes in the ratSON during hyperosmotic stimulation in vivo. Brain Res. 1429, 18–28 (2012).

50. A. Barco, J. M. Alarcon, E. R. Kandel, Expression of constitutively active CREB proteinfacilitates the late phase of long-term potentiation by enhancing synaptic capture. Cell108, 689–703 (2002).

51. M. C. Isoldi, I. Provencio, A. M. Castrucci, Light modulates the melanophore response toa-MSH in Xenopus laevis: An analysis of the signal transduction crosstalk mechanismsinvolved. Gen. Comp. Endocrinol. 165, 104–110 (2010).

52. J. Shi, C. Severson, J. Yang, D. Wedlich, M. W. Klymkowsky, Snail2 controlsmesodermal BMP/Wnt induction of neural crest. Development 138, 3135–3145(2011).

53. E. Eden, R. Navon, I. Steinfeld, D. Lipson, Z. Yakhini, GOrilla: A tool for discoveryand visualization of enriched GO terms in ranked gene lists. BMC Bioinformatics 10,48 (2009).

54. A. Nikitin, S. Egorov, N. Daraselia, I. Mazo, Pathway studio—The analysis and nav-igation of molecular networks. Bioinformatics 19, 2155–2157 (2003).






Dow

nloaded from

55. B. Mayr, M. Montminy, Transcriptional regulation by the phosphorylation-dependentfactor CREB. Nat. Rev. Mol. Cell Biol. 2, 599–609 (2001).

56. A. Slominski, Neuroendocrine activity of the melanocyte. Exp. Dermatol. 18, 760–763(2009).

57. B. G. Jenks, A. H. Kidane, W. J. Scheenen, E. W. Roubos, Plasticity in the melanotropeneuroendocrine interface of Xenopus laevis. Neuroendocrinology 85, 177–185 (2007).

58. B. Lutz, W. Schmid, C. Niehrs, G. Schütz, Essential role of CREB family proteinsduring Xenopus embryogenesis. Mech. Dev. 88, 55–56 (1999).

59. S. Legewie, N. Blüthgen, H. Herzel, Quantitative analysis of ultrasensitive re-sponses. FEBS J. 272, 4071–4079 (2005).

60. J. H. Holland, Adaptation in Natural and Artificial Systems: An Introductory Analysiswith Applications to Biology, Control, and Artificial Intelligence (Michigan Univ.Press, Ann Arbor, MI, 1975).

61. L. Hou, W. J. Pavan, Transcriptional and signaling regulation in neural crest stemcell-derivedmelanocyte development: Do all roads lead toMitf?Cell Res. 18, 1163–1176(2008).

62. R. N. Kelsh, Sorting out Sox10 functions in neural crest development. Bioessays 28,788–798 (2006).

63. M. Wegner, Secrets to a healthy Sox life: Lessons for melanocytes. Pigment CellRes. 18, 74–85 (2005).

64. R. M. Bakos, T. Maier, R. Besch, D. S. Mestel, T. Ruzicka, R. A. Sturm, C. Berking,Nestin and SOX9 and SOX10 transcription factors are coexpressed in melanoma.Exp. Dermatol. 19, e89–e94 (2010).

65. H. Krude, H. Biebermann, W. Luck, R. Horn, G. Brabant, A. Grüters, Severe early-onset obesity, adrenal insufficiency and red hair pigmentation caused by POMC mu-tations in humans. Nat. Genet. 19, 155–157 (1998).

66. L. Hao, X. Zhao, B. Zhang, C. Li, C. Wang, Positive expression of pro-opiomelanocortin(POMC) is a novel independent poor prognostic marker in surgically resected non-smallcell lung cancer. Tumour Biol. 36, 1811–1817 (2014).

67. A. Slominski, POMC gene expression in mouse and hamster melanoma cells. FEBSLett. 291, 165–168 (1991).

68. H. E. Tsai, G. S. Liu, M. L. Kung, L. F. Liu, J. C.Wu, C. H. Tang, C. H. Huang, S. C. Chen,H. C. Lam, C. S. Wu, M. H. Tai, Downregulation of hepatoma-derived growth factorcontributes to retarded lungmetastasis via inhibition of epithelial–mesenchymal transitionby systemic POMC gene delivery in melanoma. Mol. Cancer Ther. 12, 1016–1025(2013).

69. J. M. Farah Jr., J. F. Bishop, H. Q. Nguyen, T. L. O’Donohue, The effects of pro-opiomelanocortin peptides on cyclic AMP and tyrosinase in melanoma cells. Peptides 7,437–441 (1986).

70. R. Cui, H. R. Widlund, E. Feige, J. Y. Lin, D. L. Wilensky, V. E. Igras, J. D’Orazio,C. Y. Fung, C. F. Schanbacher, S. R. Granter, D. E. Fisher, Central role of p53 inthe suntan response and pathologic hyperpigmentation. Cell 128, 853–864 (2007).

71. G. S. Wu, H. R. Luo, C. Dong, C. Mastronardi, J. Licinio, M. L. Wong, Sequencepolymorphisms of MC1R gene and their association with depression and anti-depressant response. Psychiatr. Genet. 21, 14–18 (2011).

72. M. H. Vaarala, K. Porvari, A. Kyllönen, O. Lukkarinen, P. Vihko, The TMPRSS2 geneencoding transmembrane serine protease is overexpressed in a majority of prostatecancer patients: Detection of mutated TMPRSS2 form in a case of aggressive dis-ease. Int. J. Cancer 94, 705–710 (2001).

73. L. M. Pedersen, P. G. Sørensen, Increased urinary albumin excretion rate in breastcancer patients. Acta Oncol. 39, 145–149 (2000).

74. C. G. Lis, J. F. Grutsch, P. G. Vashi, C. A. Lammersfeld, Is serum albumin anindependent predictor of survival in patients with breast cancer? JPEN J. Parenter.Enteral. Nutr. 27, 10–15 (2003).

75. P. Gómez, M. E. Beltrán, M. Rábago, Immunoelectrophoretic demonstration of albuminin breast cancer. Arch. Invest. Med. 14, 241–245 (1983).

76. D. Dubnau, R. Losick, Bistability in bacteria. Mol. Microbiol. 61, 564–572 (2006).77. Y. Yang, D. Atasoy, H. H. Su, S. M. Sternson, Hunger states switch a flip-flop

memory circuit via a synaptic AMPK-dependent positive feedback loop. Cell 146,992–1003 (2011).

78. X. J. Tian, H. Zhang, J. Xing, Coupled reversible and irreversible bistable switchesunderlying TGFb-induced epithelial to mesenchymal transition. Biophys. J. 105,1079–1089 (2013).

79. R. J. Johnston Jr., S. Chang, J. F. Etchberger, C. O. Ortiz, O. Hobert, MicroRNAsacting in a double-negative feedback loop to control a neuronal cell fate decision.Proc. Natl. Acad. Sci. U.S.A. 102, 12449–12454 (2005).

80. W. Xiong, J. E. Ferrell Jr., A positive-feedback-based bistable ‘memory module’ thatgoverns a cell fate decision. Nature 426, 460–465 (2003).

81. T. Shiraishi, S. Matsuyama, H. Kitano, Large-scale analysis of network bistability forhuman cancers. PLOS Comput. Biol. 6, e1000851 (2010).

82. J. E. Ferrell Jr., Self-perpetuating states in signal transduction: Positive feedback,double-negative feedback and bistability. Curr. Opin. Cell Biol. 14, 140–148 (2002).

www

83. M. Levin, T. Thorlin, K. R. Robinson, T. Nogi, M. Mercola, Asymmetries in H+/K+-ATPase and cell membrane potentials comprise a very early step in left-right patterning. Cell 111, 77–89 (2002).

84. D.S.Adams et al., Early,H+-V-ATPase-dependent proton flux is necessary for consistentleft-right patterning of non-mammalian vertebrates.Development 133, 1657–1671 (2006).

85. T. Fukumoto, R. Blakely, M. Levin, Serotonin transporter function is an early step inleft-right patterning in chick and frog embryos. Dev. Neurosci. 27, 349–363 (2005).

86. S. Wang, X. Yu, T. Zhang, X. Zhang, Z. Zhang, Y. Chen, Chick Pcl2 regulates theleft-right asymmetry by repressing Shh expression in Hensen’s node. Development131, 4381–4391 (2004).

87. S. M. Pagán-Westphal, C. J. Tabin, The transfer of left-right positional informationduring chick embryogenesis. Cell 93, 25–35 (1998).

88. N. W. Bellono, E. Oancea, UV light phototransduction depolarizes human melanocytes.Channels 7, 243–248 (2013).

89. S. Hernández-Díaz, M. Levin, Alteration of bioelectrically-controlled processes in theembryo: A teratogenicmechanism for anticonvulsants.Reprod. Toxicol.47, 111–114 (2014).

90. E. J. Meredith, M. J. Holder, A. Chamba, A. Challa, A. Drake-Lee, C. M. Bunce,M. T. Drayson, G. Pilkington, R. D. Blakely, M. J. Dyer, N. M. Barnes, J. Gordon, Theserotonin transporter (SLC6A4) is present in B-cell clones of diverse malignant origin:Probing a potential antitumor target for psychotropics. FASEB J. 19, 1187–1189 (2005).

91. H. Sive, R. M.Grainger, R. M. Harland, Early Development of Xenopus laevis: ALaboratory Manual (Cold Spring Harbor Laboratory Press, New York, 2000).

92. P. D. Nieuwkoop, J. Faber, Normal Table of Xenopus laevis (Daudin): A Systemat-ical & Chronological Survey of the Development from the Fertilized Egg till the End ofthe Metamorphosis (North-Holland Publishing Company, Amsterdam, 1967).

93. A. Keren, A. Keren-Politansky, E. Bengal, A p38 MAPK–CREB pathway functions topattern mesoderm in Xenopus. Dev. Biol. 322, 86–94 (2008).

94. J. Barhanin, F. Lesage, E. Guillemare, M. Fink, M. Lazdunski, G. Romey, KVLQT1and lsK (minK) proteins associate to form the IKs cardiac potassium current. Nature384, 78–80 (1996).

95. R. M. Harland, In situ hybridization: An improved whole-mount method for Xenopusembryos. Methods Cell Biol. 36, 685–695 (1991).

96. M. Lobikin, G. Wang, J. Xu, Y. W. Hsieh, C. F. Chuang, J. M. Lemire, M. Levin, Early,nonciliary role for microtubule proteins in left-right patterning is conserved acrosskingdoms. Proc. Natl. Acad. Sci. U.S.A. 109, 12586–12591 (2012).

97. A. Untergasser, H. Nijveen, X. Rao, T. Bisseling, R. Geurts, J. A. Leunissen, Primer3Plus,an enhanced web interface to Primer3. Nucleic Acids Res. 35, W71–W74 (2007).

98. B. Christen, V. Robles, M. Raya, I. Paramonov, J. C. Izpisúa Belmonte, Regenerationand reprogramming compared. BMC Biol. 8, 5 (2010).

99. X. Liang, C. J. Martyniuk, G. Cheng, J. Zha, Z. Wang, Pyruvate carboxylase as asensitive protein biomarker for exogenous steroid chemicals. Environ. Pollut. 189,184–193 (2014).

100. L. Jostins, J. Jaeger, Reverse engineering a gene network using an asynchronousparallel evolution strategy. BMC Syst. Biol. 4, 17 (2010).

101. T.Bäck,EvolutionaryAlgorithms inTheory andPractice: EvolutionStrategies, EvolutionaryProgramming, Genetic Algorithms (Oxford Univ. Press, New York, 1996), p. xii, 314.

Acknowledgments: We thank A. Allen, R. Lubonja, and E. Switzer for the general labora-tory assistance and frog husbandry, J. Lmire for molecular biology assistance, S. Doughty forhigh-performance computation support, D. Marshall for assistance with statistical analyses,and the members of the Levin laboratory and the bioelectricity community for many usefuldiscussions. We thank V. Pai, C. Fields, M. McVey, S. Fuchs, and J.-F. Pare for the com-ments on a draft of the manuscript. Funding: Computation used a cluster computer awardedby Silicon Mechanics and the Campus Champion Allocation for Tufts University TG-TRA130003 at the Extreme Science and Engineering Discovery Environment, which issupported by NSF grant ACI-1053575. This work was supported by the G. Harold and LeilaY. Mathers Charitable Foundation. Author contributions:M. Lobikin performed inhibitor andmRNA injection experiments. M. Levin and M. Lobikin planned the experiments and interpreteddata. D.J.B. performed extirpation surgeries. D.L. performed computational modeling,software development, and in silico data analysis. E.T. performedRT-qPCR. C.J.M. performedbioinformatics analysis. All co-authors wrote the paper together. Competing interests: Theauthors declare that they have no competing interests. Data and materials availability: Alldata have been deposited to the National Center for Biotechnology Information Gene Expres-sion Omnibus database (accession no. GSE70834, platform GPL10756).

Submitted 27 May 2015Accepted 16 September 2015Final Publication 6 October 201510.1126/scisignal.aac6609Citation: M. Lobikin, D. Lobo, D. J. Blackiston, C. J. Martyniuk, E. Tkachenko, M. Levin,Serotonergic regulation of melanocyte conversion: A bioelectrically regulated network forstochastic all-or-none hyperpigmentation. Sci. Signal. 8, ra99 (2015).



stochastic all-or-none hyperpigmentationSerotonergic regulation of melanocyte conversion: A bioelectrically regulated network for

Maria Lobikin, Daniel Lobo, Douglas J. Blackiston, Christopher J. Martyniuk, Elizabeth Tkachenko and Michael Levin

DOI: 10.1126/scisignal.aac6609 (397), ra99.8Sci. Signal.

process can occur.melanocyte conversion process, and they used computational approaches to explain how this all-or-none, stochastic

. unraveled the molecular signaling pathway and physiological circuit that mediates thiset alhyperpigmented. Lobikin darkly colored tadpoles through a stochastic all-or-none process; the embryos are either normally pigmented orproliferation, altered melanocyte cell shape, and abnormal migration of melanocytes into multiple tissues, which results in regulated by changes in membrane potential. Forced depolarization of instructor cells can result in excessive melanocytedistribution, and shape of melanocytes are determined by a subpopulation of cells called ''instructor cells,'' which are

Melanocytes play key physiological functions; one of the easiest to see is pigmentation. In frogs, the number,Driving melanocyte proliferation and invasion

ARTICLE TOOLS http://stke.sciencemag.org/content/8/397/ra99

MATERIALSSUPPLEMENTARY http://stke.sciencemag.org/content/suppl/2015/10/02/8.397.ra99.DC1

CONTENTRELATED

http://stke.sciencemag.org/content/sigtrans/10/472/eaan2959.fullhttp://science.sciencemag.org/content/sci/325/5939/471.fullhttp://stke.sciencemag.org/content/sigtrans/8/402/ec328.abstracthttp://stke.sciencemag.org/content/sigtrans/8/400/ec310.abstracthttp://science.sciencemag.org/content/sci/224/4653/1111.full.pdfhttp://stke.sciencemag.org/content/sigtrans/2002/150/tw339.abstracthttp://stke.sciencemag.org/content/sigtrans/2/66/ra16.fullhttp://stke.sciencemag.org/content/sigtrans/4/192/tr13.fullhttp://stke.sciencemag.org/content/sigtrans/8/397/fs17.full

REFERENCES

http://stke.sciencemag.org/content/8/397/ra99#BIBLThis article cites 94 articles, 15 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.Science SignalingYork Avenue NW, Washington, DC 20005. The title (ISSN 1937-9145) is published by the American Association for the Advancement of Science, 1200 NewScience Signaling

Copyright © 2015, American Association for the Advancement of Science



Dow

nloaded from

http://stke.sciencemag.org/content/8/397/ra99

http://stke.sciencemag.org/content/suppl/2015/10/02/8.397.ra99.DC1

http://stke.sciencemag.org/content/sigtrans/8/397/fs17.full

http://stke.sciencemag.org/content/sigtrans/4/192/tr13.full

http://stke.sciencemag.org/content/sigtrans/2/66/ra16.full

http://stke.sciencemag.org/content/sigtrans/2002/150/tw339.abstract

http://science.sciencemag.org/content/sci/224/4653/1111.full.pdf

http://stke.sciencemag.org/content/sigtrans/8/400/ec310.abstract

http://stke.sciencemag.org/content/sigtrans/8/402/ec328.abstract

http://science.sciencemag.org/content/sci/325/5939/471.full

http://stke.sciencemag.org/content/sigtrans/10/472/eaan2959.full

http://stke.sciencemag.org/content/8/397/ra99#BIBL

http://www.sciencemag.org/help/reprints-and-permissions

http://www.sciencemag.org/about/terms-service


serotonergic regulation of melanocyte conversion: a ...containing the pineal gland, the pituitary,...

Documents