A retrovirus-based protein complementation assayscreen reveals functional AKT1-binding partnersZhiyong Ding*, Jiyong Liang*, Yiling Lu*, Qinghua Yu*, Zhou Songyang†, Shiaw-Yih Lin*, and Gordon B. Mills*‡

*Department of Molecular Therapeutics, University of Texas M. D. Anderson Cancer Center, 1515 Holcombe Boulevard, Box 950, Houston, TX 77030;and †Department of Biochemistry and Molecular Biology, Baylor College of Medicine, One Baylor Plaza, Houston, TX 77030

Communicated by Louis Siminovitch, University of Toronto, Toronto, ON, Canada, August 17, 2006 (received for review April 24, 2006)

We developed a retrovirus-based protein-fragment complementa-tion assay (RePCA) screen to identify protein–protein interactionsin mammalian cells. In RePCA, bait protein is fused to one fragmentof a rationally dissected fluorescent protein, such as GFP, intenselyfluorescent protein, or red fluorescent protein. The second, com-plementary fragment of the fluorescent protein is fused to anendogenous protein by in-frame exon traps in the enhancedretroviral mutagen vector. An interaction between bait and hostprotein (prey) places the two parts of the fluorescent molecule inproximity, resulting in reconstitution of fluorescence. By usingRePCA, we identified a series of 24 potential interaction partnersor substrates of the serine�threonine protein kinase AKT1. Weconfirm that �-actinin 4 (ACTN4) interacts physically and function-ally with AKT1. siRNA-mediated ACTN4 silencing down-regulatesAKT phosphorylation, blocks AKT translocation to the membrane,increases p27Kip1 levels, and inhibits cell proliferation. Thus, ACTN4is a critical regulator of AKT1 localization and function.

�-actinin 4 � AKT1 � protein–protein interactions � screen

Presently, novel protein–protein interactions are identified byusing yeast two-hybrid (Y2H) systems (1), coimmunoprecipi-tation (Co-IP) and mass spectrometry (2), or protein libraries (3).These approaches, however, do not efficiently identify proteininteractions on the cytoskeleton or membrane, either because of thelocation of the interaction (Y2H) or difficulties in the Co-IP ofcytoskeletal or membrane proteins. Furthermore, conventionalY2H approaches yield false-positive signals with transcription fac-tors precluding screening. The purpose of this study was to developa readily applicable, subcellular localization-, library-, and cell-independent method for the identification of protein–protein in-teractions, including cytoskeletal and membrane proteins, in mam-malian cells and apply it to the well characterized AKTprotooncogene.

The retrovirus-based protein-fragment complementation assay(RePCA) utilizes the technical advantages of two emerging tech-nologies, the enhanced retroviral mutagen (ERM) vector and theprotein-fragment complementation assay (PCA). The ERM vectorfunctions as an exon trap that efficiently activates and tags endog-enous genes (4). The ERM vector in all three ORFs can be appliedto any mammalian cell without the effort and bias of cell-specificlibraries. Furthermore, the ERM vector utilizes the endogenoussplicing machinery of the target cell, allowing the evaluation ofvariations in natural splice variants. In PCA, a reporter protein, suchas a monomeric enzyme or a fluorescent protein (GFP or a variantthereof), is rationally dissected into two fragments that will notreconstitute spontaneously (5–7). When each fragment of thefluorescent protein is fused to one of a pair of interacting proteinpartners, the subsequent protein binding places the fragments inproximity, creating a functional complex that restores fluorescencelevels to near that of the parental molecule (8). Recently, Remy andMichnick (9) identified an AKT1 partner with a PCA-based cDNAlibrary screen using GFP as a reporter. However, the need togenerate cell-specific cDNA libraries and a complex approach tothe identification of candidates renders this approach technicallydemanding and labor-intensive. RePCA combines the power of

ERM (4) with PCA (8) to provide a facile, sensitive approach forthe identification of context-dependent protein–protein interac-tions in mammalian cells, allowing native protein folding andposttranslational modifications.

ResultsRePCA Screen Design. The intensely fluorescent protein (IFP), alsoknown as Venus (10), was dissected into two fragments, an IFPN-terminal portion (IFPN) and IFP C-terminal portion (IFPC), atresidue 158 (11). As shown in Fig. 1A, IFPC was inserted into theERM vector, followed by a splice donor, to construct the RePCAvector. A Tet-responsive promoter obviates potential toxicity of thefusion protein. After infection of the target cells, the retrovirusundergoes reverse transcription and integration into the host ge-nome. If the integration occurs upstream or inside a host gene, thesplice donor in the RePCA vector generates a fusion transcript withIFPC linked in-frame to downstream host exons. The propensity ofthe ERM retrovirus to integrate near the start of transcriptionalunits increases the chance that a full-length or near-full-lengthfusion protein will be created (12).

A host cell line expressing the tetracycline-regulated transac-tivator tTA or reverse tTA to enable the regulated expression ofthe prey from the Tet-responsive promoter is generated bytransfection to express the IFPN-Bait fusion protein (Fig. 1B).Bait fusion protein-expressing cells are infected with the RePCAretrovirus in all three frames to generate in-frame IFPC-endog-enous fusion proteins. Cells in which the retrovirus does notgenerate an in-frame fusion protein or wherein IFPC is not fusedto a binding partner of the bait do not fluoresce. Only cellscontaining IFPC fused in-frame to an interaction partner of thebait f luoresce. The fluorescent cells are cloned, and the targetgenes are identified by RT-PCR with primers contained in theERM vector. The resulting fluorescent clones can be directlyused to characterize the formation, localization, and function-ality of candidate interactions, providing powerful reagents formechanistic exploration without the need for recloning.

Proof-of-Concept Screen for AKT1 Interaction Partners. We choseAKT1 (also known as PKB-�), which plays a central role in cellmetabolism, survival, growth, and tumorigenesis (13, 14), as a baitfor a proof-of-concept RePCA screen. IFPN-AKT1 HeLa Tet-oncells with or without transient transfection of IFPC were notfluorescent, confirming that the fragments do not fluoresce and donot spontaneously associate (data not shown). Transfection of the

Author contributions: Z.D., J.L., Y.L., Q.Y., Z.S., and G.B.M. designed research; Z.D., J.L., Y.L.,and Q.Y. performed research; Z.S. contributed new reagents�analytic tools; Z.D., J.L., Y.L.,Q.Y., S.-Y.L., and G.B.M. analyzed data; and Z.D., J.L., Y.L., S.-Y.L., and G.B.M. wrote thepaper.

The authors declare no conflict of interest.

Abbreviations: PCA, protein-fragment complementation assay; RePCA, retrovirus-basedPCA; ERM, enhanced retroviral mutagen; Y2H: yeast two-hybrid; IFP, intensely fluorescentprotein; IFPN, N-terminal portion of IFP; IFPC, C-terminal portion of IFP; ACTN4, �-actinin4; PDK1, phosphoinositide-dependent kinase 1; PI3K, phosphatidylinositol 3 kinase; Co-IP,coimmunoprecipitation.

‡To whom correspondence should be addressed. E-mail: gmills@mdanderson.org.

© 2006 by The National Academy of Sciences of the USA

15014–15019 � PNAS � October 10, 2006 � vol. 103 � no. 41 www.pnas.org�cgi�doi�10.1073�pnas.0606917103

Dow

nloa

ded

by g

uest

on

June

22,

202

1

known AKT partner, phosphoinositide-dependent kinase 1(PDK1)-IFPC (15), into IFPN-AKT1 HeLa Tet-on cells resulted influorescence, which was enriched in the cell membrane and atpoints of cell–cell contact (Fig. 2A). The membrane localization ofthe PDK1-IFPC:IFPN-AKT1 complex was completely blocked byinhibition of phosphatidylinositol 3 kinase (PI3K) with LY294002,as expected from the known localization of AKT1 and PDK1 (16,17). Thus, linking each part of IFP to interacting proteins AKT1 andPDK1 reconstitutes IFP fluorescence with appropriate subcellularlocalization.

IFPN-AKT1 HeLa Tet-on cells were separately infected withRePCA-IFPC retroviral vectors in all three reading frames. Afterselection with puromycin and induction with doxycycline, singlefluorescent cells were isolated by sorting. Expanded clones dis-played fluorescence with different patterns, i.e., homogenous,cytoplasmic, nuclear, or membrane, suggesting differential local-ization of AKT1 and specific binding partners (see Fig. 2 B–D andTable 1). We chose 70 fluorescent clones for identification of targetgenes (10, 30, and 30 clones for reading frame 1, 2, and 3,respectively), all of which yielded fusion transcripts. In terms ofefficiency, using frame 2 as an example, 2 � 107 cells were sorted,with the 384 most highly fluorescent cells placed in single wells.Doxycycline was withdrawn to reduce the expression of proteinsthat could have adverse effects on cell survival and growth. Of the

384 cells sorted, 120 cells (�30%) formed colonies, with 55 coloniesshowing fluorescence after doxycycline induction (�14%). Wechose 30 clones based on fluorescence intensity and patterns andidentified 11 independent candidate AKT1 partners (Table 1).Thus, from 2 � 107 cells, 11 candidates were identified. This pilotscreen was not taken to saturation; therefore, it is likely thatadditional AKT1-binding partners would be identified in large-scale screens. Twenty-four independent candidates were identifiedfrom all three ORFs (Table 1). Some candidates were present asmultiple clones [i.e., BCL2-antagonist of cell death (BAD) wasidentified three times (Table 1)] and were likely derived from asingle infected parental cell that divided before cell sorting, becausesequencing demonstrated identical fusion transcripts. Limiting cellpropagation before sorting fluorescent cells could increase thetarget diversity identified in the screen. IFPC fusion proteins similarto calculated sizes were detected in all clones (see Fig. 5, which ispublished as supporting information on the PNAS web site, forselected clones) except for AHNAK2, which has an expectedmolecular weight of �600 kDa, precluding detection by Westernblots. The identification of AHNAK2 demonstrates the power ofthe ERM vector, which, through the use of endogenous splicingmachinery, is not restricted in terms of size of insert. In 14 of 24clones, the IFPC fragment was fused to a full-length or near-full-length protein (Table 1).

Fig. 2. Pilot screen for AKT1-interaction part-ners. (A) AKT1 interaction with PDK1 demon-strated by PCA. IFPN-AKT1 HeLa Tet-on cells sta-bly transfected to coexpress PDK1-IFPC yieldedhighly fluorescent cells with enhanced fluores-cence at the cell membrane and points of cell–cellcontact. Inhibition of PI3K with LY294002 (20 �M)for 3 hr abrogated membrane localization. (B)Selected RePCA clones with different fluorescentpatterns. Expanded clones were assessed for flu-orescence after doxycycline induction. The targetgenes activated in each clone are shown undereach image. (C) SSB clone (Table 1, no. 20) show-ing nuclear fluorescence. (D) SLC3A2 (Table 1, no.6) clone showing membrane fluorescence.

Fig. 1. Schematic diagram of the RePCA screen. (A) Generation of IFPC fusion with endogenous proteins in mammalian cells. A RePCA vector is constructedby inserting IFPC into the ERM retroviral vector, followed by a splice donor, in the U3 region of the 3� LTR. A Tet-responsive promoter controls the expressionof IFPC. RePCA retroviruses are generated in packaging cells. After the infection of target cells, integration upstream or inside a host gene may allow thegeneration of in-frame IFPC fusions with a host protein (see Results for details). (B) Procedures of the RePCA screen. Bait is fused to IFPN. A host cell line, preferablycontaining the Tet regulatory complex to enable the expression from a Tet-responsive promoter, is transfected to stably express IFPN-Bait. The cell populationis infected with RePCA viruses to create endogenous protein fusions with IFPC. Fusion of IFPC in-frame to a protein that binds the bait reconstitutes the IFPmolecule, restoring fluorescence. Fluorescent cells are cloned by cell sorting or other approaches and expanded, and target genes are identified by RT-PCR.

Ding et al. PNAS � October 10, 2006 � vol. 103 � no. 41 � 15015

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

June

22,

202

1

Among the targets, BAD is a well characterized AKT substrateand interaction partner (18). Myosin light chain (atrial�embryonicalkali) (MYL4) is part of the myosin II complex, a known AKT-binding partner (19). AHA1, Annexin1, and a Lamin A�C isoform(LaminB1) have been confirmed as AKT-binding partners byCo-IP and mass spectroscopy (J. Downward, personal communi-cation). We also identified two potential AKT substrates becauseheme oxygenase 2 (HMOX2) and C14ORF78 (AHNAK2) areisoforms of HMOX1 and AHNAK1, which are known AKTsubstrates (20, 21). Additional candidates have strong links to AKT(Table 1). For example, CD98 regulates AKT phosphorylation andactivation (22). Both CD98 and AKT interact with integrin-�1 (22,23). Moesin and AKT interact with the hamartin:tuberin complex(24–26). The identification of known and likely AKT partners�substrates validates the screening strategy.

�-Actinin 4 (ACTN4) Is a Functional Partner of AKT1. �-Actininactin-binding proteins, including ACTN4, can bind the p85 regu-latory subunit of PI3K and translocate out of the membrane afterinhibition of PI3K (27–29). In addition, ACTN4 contributes to theprognosis of breast cancer (29) and has been implicated in tumordevelopment, invasion, and metastases (30, 31), making it a can-

didate for further characterization. The fluorescence of the ACTN4clone (IC1-05) obtained from the screen was markedly increased bydoxycycline; therefore, formation of the fluorescent complex wascontrolled by Tet-responsive expression of the IFPC fusion protein(Fig. 3A). The fluorescent IFPN-AKT1::IFPC-ACTN4 complex inclone IC1-05 was located throughout the cytoplasm in serum-starved cells (Fig. 3B). Serum induced the translocation of theAKT1::ACTN4 complex to the membrane, in particular cellularruffles, reminiscent of the PDK1::AKT1 complex (Fig. 2A), and tothe nuclear periphery (Fig. 3B). Compatible with the observationthat ACTN4 translocates out of the membrane upon inhibition ofPI3K (29), serum-dependent translocation of the AKT1::ACTN4complex to the cell membrane was blocked by inhibition of PI3K,indicating that membrane localization depended on the productionof 3-phosphorylated membrane phosphatidylinositols. Expressionof an exogenous IFPC-ACTN4wt in IFPN-AKT1 HeLa Tet-oncells resulted in fluorescence, which was enriched at the leadingedge of cells growing in serum (Fig. 3C), confirming that theapparent interaction between ACTN4 and AKT1 was not due toclonal variation. Expression of a IFPC-ACTN4�310–665 deletionconstruct resulted in homogeneous fluorescence in the cytoplasmwith enhanced fluorescence in the nucleus but completely failed to

Table 1. Interaction partners of AKT1 identified in RePCA screen

No. Protein nameGene

symbol

Vectorreadingframe

Insertionsite, aa

Expected sizeof the fusionprotein, kDa

No. ofclonesfor thetarget

Predominantfluorescencelocalization

Accessionno.

1 BCL2-antagonist of celldeath*

BAD 2 63 23 3 Homogenous NP�116784

2 Myosin light chain(atrial�embryonicalkali)*

MYL4 3 �12 34 1 Cytoplasm NP�001002841

3 Heme oxygenase 2* HMOX2 2 �14 47 2 Cytoplasm NP�0021254 C14ORF78* AHNAK2 2 167 613 1 Cytoplasm,

Leading edgeXP�290629

5 Activator of heat shock90-kDa proteinATPase homolog 1*

AHA1 3 28 45 2 Homogenous NP�036243

6 CD98* SLC3A2 2 38 80 1 Membrane NP�0010126797 Moesin* MSN 1 5 74 2 Cytoplasm NP�0024358 Ribosomal protein L22* RPL22 1 5 25 1 Cytoplasm NP�0009749 Reticulon 4*† RTN4 2 12 32 3 Cytoplasm NP�00893910 GTPase-activating

RANGAP domain-like3*

GARNL3 2 897 22 8 Homogenous NP�115669

11 Lamin A�C* LMNA 3 120 68 1 Nucleus NP�73382212 �-Actinin 4* ACTN4 1 53 105 2 Cytoplasm,

Leading edgeNP�004915

13 Sterol O-acyltransferase‡ SOAT1 2 �3 72 1 Cytoplasm NP�00309214 Peroxiredoxin 1‡ PRDX1 2 �4 34 5 Cytoplasm NP�85904715 Phosphorybosylaminoi

midazolecarboxylase‡

PAICS 2 6 57 2 Cytoplasm,Leading edge

NP�006443

16 Annexin I‡ ANXA1 2 �5 50 1 Homogenous NP�00069117 Protease inhibitor 6‡ PI6 3 �4 53 3 Homogenous NP�00455918 ARCHAIN 1‡ ARCN1 3 218 43 1 Homogenous NP�00164619 Proteolipid protein 2‡ PLP2 3 �31 31 8 Membrane NP�00265920 Autoantigen La‡ SSB 3 186 36 3 Nucleus NP�00313321 Zinc finger protein 185‡ ZNF185 3 232 35 5 Homogenous NP�00908122 F-actin-capping protein

� subunit‡CAPZB 1 2 41 5 Cytoplasm NP�004921

23 Protein translationfactor SUI1 homolog‡

Sui1iso1 2 11 22 3 Homogenous XP�497726

24 Pyruvate kinase, muscle‡ PKM2 3 �93 80 6 Cytoplasm NP�872271

Accession nos. and gene symbols are from GenBank.*Known partners�substrates or potential partners�substrates with known links to AKT.†RTN4 has five splice variants, A–E, with the same C-terminal sequence. The insertion site for RTN4 was defined according to the RTN4 variant C sequence.‡Potential AKT partners�substrates without obvious known links to AKT.

15016 � www.pnas.org�cgi�doi�10.1073�pnas.0606917103 Ding et al.

Dow

nloa

ded

by g

uest

on

June

22,

202

1

mediate localization of the fluorescent complex to the leading edgeof cells growing in serum (Fig. 3C). In contrast, coexpression ofIFPN-AKT1 and IFPC-ACTN4�310–911 did not result in detect-able fluorescence (Fig. 3C). Thus, residues 665–911 of ACTN4,which contain two EF motifs, are required for the interaction withAKT1, whereas residues 310–665 of ACTN4 are critical for thelocalization of the AKT1::ACTN4 complex to the leading edge ofcells. Strikingly, the PH domain of AKT1 (32) was not sufficient totranslocate the AKT1::ACTN4 complex to the cell membrane,suggesting that ACTN4 contributes to AKT1 localization. How-ever, given that LY294002 blocked translocation (Fig. 3B), thetranslocation of the AKT1::ACTN4 complex to the membranedepends on the production of 3-phosphorylated membrane phos-phatidylinositols. Therefore, the screening approach not only pro-vides insight into the binding partners of AKT1, but the resultantcells can also provide important functional information related tothe localization and formation of the protein complexes.

In clone IC1-05, IFPC-ACTN4 was readily immunoprecipitatedby anti-AKT1 antibodies and detected by immunoblotting with apolyclonal anti-GFP antibody, which binds both halves of IFP (Fig.3D). Equal amounts of IFPC-ACTN4 and IFPC-AKT1 werepresent in the immunoprecipitated complex, compatible with theformation of a dimer. In parental HeLa Tet-on cells, ACTN4 couldbe immunoprecipitated by anti-AKT1 antibodies after in vivocross-linking (Fig. 3E; see also Supporting Materials and Methods,which is published as supporting information on the PNAS website), demonstrating an association of endogenous AKT1 withACTN4. We were unable to detect a stable association betweenendogenous AKT1 and ACTN4 by Co-IP in the absence ofcross-linking (data not shown), potentially because the conditionsrequired to efficiently release ACTN4 from the cytoskeleton dis-rupted interactions between AKT1 and ACTN4 or because theinteraction between AKT1 and ACTN4 was transient or indirect.

ACTN4 Silencing Inhibits AKT Translocation, Phosphorylation, Signal-ing, and Cell Proliferation. The membrane localization of AKT playsa pivotal role in AKT activation (13, 14). The distinctive pattern ofACTN4::AKT1 complex localization suggested that ACTN4 couldcontribute to AKT translocation and, thus, activation. In HeLaTet-on cells as well as in IOSE80- hTERT (hTERT, humantelomerase reverse transcriptase) cells not expressing exogenousAKT or ACTN4 IFP fusion proteins (Fig. 4A), an siRNA pooltargeting ACTN4 induced a concordant reduction in ACTN4protein expression and AKT phosphorylation. Strikingly, ACTN4silencing increased levels of the cyclin-dependent kinase inhibitorp27Kip1 protein (Fig. 4A), a known downstream target of AKT (33).Additional siRNA constructs inhibited AKT phosphorylation, withACTN4 knockdown and pAKT phosphorylation being concordant(Fig. 6, which is published as supporting information on the PNASweb site). Insulin induced translocation of AKT1 fused to afull-length GFP protein (in parental HeLa cells not expressing AKTor ACTN4 fusion proteins) to the cell membrane, which wasblocked by ACTN4 siRNA knockdown, confirming a role forACTN4 in the translocation of AKT1 to the membrane (Fig. 4B).The effects of ACTN4 knockdown on the phosphorylation ofHA-AKT1 were bypassed by membrane-targeted myristylatedAKT1 (Fig. 4C), indicating that ACTN4 knockdown preventstranslocation of AKT1 to the cell membrane. ACTN4 knockdowndid not markedly alter F-actin in HeLa cells, suggesting that theeffects of ACTN4 siRNA were not secondary to the disruption ofthe cellular cytoskeleton (Fig. 4D). Finally, knockdown of ACTN4significantly inhibited HeLa cell proliferation (Fig. 4E), with dif-ferent siRNAs demonstrating similar activity on ACTN4 knock-down, decrease in AKT phosphorylation (Fig. 6), and cell prolif-eration. Thus, ACTN4 plays an essential role in AKT translocation,activation, signaling, and function, which demonstrates the ability ofRePCA to uncover functional protein–protein interactions.

Fig. 3. AKT1 interaction with ACTN4. (A) The fluo-rescence intensity of clone IC1-05 (ACTN4) is Tet-responsive. No IFPN-AKT1 HeLa Tet-on cells were in theM1 quadrant. Cells had a mean fluorescence level of3.2. IC1-05 cells exhibited low-level fluorescence with-out doxycycline induction, likely because of leaky ex-pression from Tet-responsive promoters (8.6% of cellsin M1 quadrant). Cells had a mean fluorescence levelof 46.5. After incubation with 2 �g�ml doxycycline for48 hr, the mean fluorescence of IC1-05 cells increasedto 112.9 (62% of cells in the M1 quadrant). (B) Trans-location of IFPN-AKT1::IFPC-ACTN4 complex upon se-rum stimulation in clone IC1-05. Serum-starved cellsshow predominantly cytoplasmic fluorescence. Afterserum (10%) stimulation for 60 min, the fluorescentcomplex translocated to the leading edge of cells andthe periphery of the nucleus in �90% of cells. Inhibi-tion of PI3K with a 3-hr treatment of LY294002 (10 �M)abrogated serum-induced membrane localization. Atleast 100 cells were examined from different fields foreach sample. (C) Confirmation of ACTN4::AKT1 inter-action by PCA and identification of the residues inACTN4 required for interaction with AKT1 and local-ization of the complex. Coexpression of IFPN-AKT1with IFPC-ACTN4wt yielded fluorescence enriched atthe leading edge of the cell (ruffles). Coexpression ofIFPN-AKT1 with IFPC-ACTN4�310–665 yielded cyto-plasmic and enhanced nuclear fluorescence. Coexpres-sion of IFPN-AKT1 with IFPC-ACTN4�310–911 did notresult in fluorescence. A schematic under each set of images shows the domains in wild-type or deletion mutants of ACTN4. CH, calponin homology domain; SPEC,spectrin repeats; EF, EF-hand, calcium-binding motif. (D) Association of IFPN-AKT1 with IFPC-ACTN4 in clone IC1-05. Clone IC1-05 cells expressing IFPN-AKT1 andIFPC-ACTN4 were lysed in RIPA buffer. Co-IP was performed with anti-AKT1 and Western blotting with anti-GFP. Lane 1, Co-IP with normal IgG; lane 2, Co-IP withanti-AKT1; lane 3, total lysate. (E) Association of AKT1 with ACTN4 in parental HeLa Tet-on cells. HeLa Tet-on cells were lysed in RIPA buffer, with in vivocross-linking with BASED. Co-IP was performed with anti-AKT1 and Western blotting with anti-ACTN4. Lane 1, Co-IP with normal IgG; lane 2, Co-IP withanti-AKT1; lane 3, blank; lane 4, total lysate.

Ding et al. PNAS � October 10, 2006 � vol. 103 � no. 41 � 15017

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

June

22,

202

1

DiscussionThe identification of known and likely AKT1-binding partners andsubstrates (Table 1) validated RePCA as being able to identifyprotein–protein interactions and potentially transient interactionsbetween enzyme and substrate in mammalian cells. A previouslyuncharacterized physical and functional interaction was identifiedbetween ACTN4 and AKT1. Functional association was demon-strated by siRNA-mediated ACTN4 silencing, which inhibits AKTtranslocation and phosphorylation, resulting in up-regulation ofp27Kip1 protein and a decrease in cell proliferation. An interactionbetween ACTN4 and AKT may underlie the pathophysiology offocal segmental glomerulosclerosis, which is caused by ACTN4mutations (34), suggesting novel therapeutic approaches. ACTN4also associates with focal adhesions, tight junctions, and adherensjunctions through interactions with the cytoskeleton and the tight-junction protein MAGI-1 (35), linking the PI3K�AKT pathway tomotility and invasion.

RePCA has a number of potential advantages. (i) The abilityto perform screens in a homologous mammalian cell environ-ment allows native protein folding and posttranslational modi-fications. (ii) RePCA can be used to comparatively analyzemultiple different cell lines or genetic backgrounds, avoiding therelated bias and difficulties in generating cell-specific cDNAlibraries. The host range of the retrovirus can be extended bypseudotyping with vesicular stomatitis virus G glycoprotein (36).(iii) RePCA can identify context-dependent interactions underdifferent activation conditions or genetic manipulations or withspecific drugs. (iv) The derivation of the ERM vector from

Moloney murine leukemia virus (12) results in preferentialintegration near the start of transcriptional units, generating ahigh frequency of full-length or near-full-length fusion tran-scripts. Furthermore, because the ERM vector uses nativesplicing machinery, the approach can identify splicing-specificinteractions and the identification of binding partners is notlimited by size. (v) The Tet-responsive promoter allows high-level expression of endogenous targets during screening fol-lowed by repression of potentially toxic endogenous targets.Regulated expression of the endogenous target by doxycyclinealso limits false-positive interactions. (vi) The lack of backgroundfluorescence combined with the high level of f luorescence whenthe PCA fragments are brought into proximity by interactingproteins makes the approach applicable to high-throughputscreening. (vii) The resulting clones provide reagents for studiesof the function and localization of protein complexes, suggestingpotential functional consequences of the interactions. Theseclones can also be used in high-content drug, genomic, orchemical genomic screens aimed at preventing the formation ofcomplexes or blocking the translocation of the complex toparticular subcellular compartments. (viii) The reconstitutedbarrel structure of GFP and, by analogy, IFP is relatively stable(37). Thus, RePCA has the potential to stabilize or trap transientinteractions, such as enzyme–substrate interactions, or to stabi-lize low-affinity interactions, allowing the identification of com-ponents of signaling pathways and networks not discoverable byother approaches. The relative ease and applicability of RePCAto high-throughput analysis allow sequential indentification of

Fig. 4. siRNA-mediated ACTN4 silencing alters AKTsignaling and cell proliferation. (A) The effects ofACTN4 siRNA knockdown on AKT signaling. HeLaTet-on and human ovarian surface epithelial cellsIOSE80(hTERT) were transfected with ACTN4 siRNA ora nontargeting siRNA (NT-siRNA) pool (Dharmacon).Twenty-four hours after transfection, cells were se-rum-starved for 24 hr and stimulated with 5% FBS for3 hr before lysis. Lysates (50 �g per lane) were resolvedby 8% SDS�PAGE for the detection of ACTN4 and 12%for other proteins. �-actin immunoblotting showedequivalent loading and specificity of the siRNA. Scan-ning densitometric values of Western blots from threeindependent experiments in HeLa Tet-on cells(mean � SE) were obtained with NIH IMAGE 1.63.1software and are presented as densitometric values oftarget siRNA divided by control (nontargeting siRNA).(B) ACTN4 silencing blocks insulin-induced AKT1 trans-location to HeLa cell membranes. HeLa cells were sta-bly transfected to express AKT1-GFP. After serum star-vation for 18 hr, insulin stimulation (20 �g�ml) for 10min induced translocation of AKT1-GFP to the cellmembrane in �90% of AKT1-GFP HeLa cells in thepresence or absence of nontargeting siRNA. In con-trast, �5% of cells transfected with ACTN4 siRNA dem-onstrated insulin-induced translocation of AKT1-GFPto the cell membrane. At least 100 cells were examinedfrom different fields for each sample. (C) Inhibition ofAKT1 phosphorylation by ACTN4 siRNA is bypassed bymyristylated AKT1. HeLa cells were transfected with anontargeting siRNA or ACTN4 siRNA pool. After 24 hr,the cells were transfected to express Myr-HA-AKT1 orHA-AKT1. Forty-eight hours after siRNA transfection,cells were serum-starved for 24 hr and stimulated with10% FBS and 75 ng�ml IGF-1 for 10 min before lysis.pAKT1 (473) levels are presented as values relative tonontargeting siRNA. (D) ACTN4 knockdown does notalter the actin cytoskeleton. HeLa cells were trans-fected with nontargeting siRNA or ACTN4 siRNA. Actincytoskeleton was stained with tetramethylrhodamine B isothiocyanate-labeled phalloidin 48 or 72 hr after transfection. Representing images are shown. (E) Theeffects of ACTN4 knockdown on cell proliferation. Proliferation of HeLa cells transfected with siRNA was assessed by 3,(4,5-dimethylthiazol-2-yl)2,5-diphenyltetrazolium bromide assay kit (Sigma) 96 hr after transfection.

15018 � www.pnas.org�cgi�doi�10.1073�pnas.0606917103 Ding et al.

Dow

nloa

ded

by g

uest

on

June

22,

202

1

interacting partners and the creation of pathways and networksbased on protein–protein interactions. (ix) RePCA can identifyinteractions in multiple different intracellular compartments,particularly cytoskeletal, membrane, and transcription-factorinteractions , that are difficult to discover by using a Y2H or massspectrometry approach. (x) Molecular interactions are detecteddirectly, not through secondary events, such as transcriptionactivation.

The RePCA approach has a number of potential limitations.(i) Intron-less genes cannot be captured by the ERM vector.Fortunately, these genes are relatively uncommon. (ii) Thescreen depends on a high-titer virus preparation for targeting asufficient number of genes. (iii) Genes not accessible for viralintegration will not be targeted. (iv) The fluorescent proteinfragment may interfere with the formation of protein–proteininteractions. (v) The fluorescent protein fragment may interferewith membrane insertion of type I integral membrane proteinsor secreted proteins. (vi) The ability to trap enzyme substrate orweak interactions as well as other sources of binding may resultin false-positive results. However, the use of both N- andC-terminal orientations for the bait as well as varying the lengthof linkers between the engineered GFP fragments and thebait�prey is likely to alleviate many of these concerns.

Materials and MethodsCell Lines and Plasmids. HeLa Tet-on cells were from BD Clontech(Palo Alto, CA). 293T�17 was from American Type CultureCollection (Manassas, VA). TAg and human telomerase immor-talized normal ovarian epithelial cells IOSE80(hTERT) (38) werefrom N. Auersperg (University of British Columbia, Vancouver,BC, Canada). Plasmids VYF102 (IFPC vector), 11117-Y101 (ex-pressing IFPN-AKT1), and 21622-Y108 (expressing PDK1-IFPC)were from Odyssey Thera, Inc. (San Ramon, CA) (11). PlasmidPS1941 encoding AKT1-GFP was from Bioimage (Soeborg, Den-mark). Plasmids pcGP and pVSVG were from Xiao-Feng Qin(M. D. Anderson Cancer Center). ERM vectors are described inref. 4. RePCA vectors for each of the three reading frames wereconstructed by inserting a PacI�AscI fragment containing the IFPCcoding sequence into the ERM vectors digested with PacI�AscI (fordetails, see Supporting Materials and Methods). Full-length ACTN4cDNA was from Origene Technologies, Inc. (Rockville, MD). The

human AKT1 gene was cloned from OVCAR3 cells by RT-PCR.Construction of plasmids expressing IFPC-ACTN4wt, IFPC-ACTN4�310–665, IFPC-ACTN4�310–911, and Myr-HA-AKT1are described in Supporting Materials and Methods.

RePCA Screen Procedure. HeLa Tet-on cells were stably transfectedwith plasmid 11117-Y101 to express IFPN-AKT1. IFPN-AKT1HeLa Tet-on cells were infected in exponential growth phase.Preparation of RePCA retrovirus and infection are described in ref.39, with additional details in Supporting Materials and Methods.Infected cells were selected with 0.5 �g�ml puromycin (BD Clon-tech) for 5 days. In the last 2 days of selection, 2 �g�ml ofdoxycycline (BD Clontech) was added to induce the expression ofIFPC fusions from Tet-responsive promoters. Fluorescent cellswere sorted individually into 96-well plates. Doxycycline was with-drawn during recovery to reduce the expression of proteins thatcould have adverse effects on cell survival and growth (for addi-tional details, see Supporting Materials and Methods).

Identification of Target Genes. Total RNA was extracted fromexpanded clones by using RNeasy Mini kits (Qiagen, Valencia,CA). Reverse transcription was performed with a random primerRT-1 (5�-GCAAATACGACTCACTATAGGGATCCNNNN-(GC)ACG-3�; n AGCT) (4), or a PolyT primer RT-1T (5�-GCAAATACGACTCACTATAGGGATCCTTTTTTTTT-TTTTTTT-3�) using a SuperScript III kit (Invitrogen, Carlsbad,CA). The 5� end of the RT-1 primer contains the T7 primersequence. The cDNA was PCR-amplified with a specific IFPCprimer (IFPCR, 5�-ACTTCAAGATCCGCCACAACATCGAG-3�) and the T7 primer (T7-2, 5�-GCAAATACGACTCACTAT-AGGGATC-3�) by using AccuTaq DNA polymerase (Invitrogen).Gel-purified PCR products were directly sequenced, with theresulting sequences used to search GenBank human nonredundantand expressed-sequence-tag databases by using BLAST.

RNAi and Cell Growth Assay. A pool of four siRNAs targeting humanACTN4 and a nontargeting siRNA pool were from Dharmacon(Lafayette, CO). Two single siRNAs targeting human ACTN4 werefrom Ambion (Austin, TX). RNAi silencing was performed ac-cording to the manufacturer’s protocol. The effects of siRNAknockdown on cell growth was assessed by a 3,(4,5-dimethylthiazol-2-yl)2,5-diphenyltetrazolium bromide assay (40).

1. Fields S, Song O (1989) Nature 340:245–246.2. Lin D, Tabb DL, Yates, JR, III (2003) Biochim Biophys Acta 1646:1–10.3. Turk BE, Cantley LC (2003) Curr Opin Chem Biol 7:84–90.4. Liu D, Yang X, Yang D, Songyang Z (2000) Oncogene 19:5964–5972.5. Johnsson N, Varshavsky A (1994) Proc Natl Acad Sci USA 91:10340–10344.6. Ghosh I, Hamilton AD, Regan L (2000) J Am Chem Soc 122:5658–5659.7. Hu CD, Chinenov Y, Kerppola TK (2002) Mol Cell 9:789–798.8. Michnick SW, Remy I, Campbell-Valois FX, Vallee-Belisle A, Pelletier JN (2000)

Methods Enzymol 328:208–230.9. Remy I, Michnick SW (2004) Mol Cell Biol 24:1493–1504.

10. Nagai T, Ibata K, Park ES, Kubota M, Mikoshiba K, Miyawaki A (2002) Nat Biotechnol20:87–90.

11. Yu H, West M, Keon BH, Bilter GK, Owens S, Lamerdin J, Westwick JK (2003) AssayDrug Dev Technol 1:811–822.

12. Wu X, Li Y, Crise B, Burgess SM (2003) Science 300:1749–1751.13. Cantley LC (2002) Science 296:1655–1657.14. Brazil DP, Yang ZZ, Hemmings BA (2004) Trends Biochem Sci 29:233–242.15. Alessi DR, James SR, Downes CP, Holmes AB, Gaffney PR, Reese CB, Cohen P

(1997) Curr Biol 7:261–269.16. Andjelkovic M, Alessi DR, Meier R, Fernandez A, Lamb NJ, Frech M, Cron P,

Cohen P, Lucocq JM, Hemmings BA (1997) J Biol Chem 272:31515–31524.17. Anderson KE, Coadwell J, Stephens LR, Hawkins PT (1998) Curr Biol 8:684–691.18. Datta SR, Dudek H, Tao X, Masters S, Fu H, Gotoh Y, Greenberg ME (1997) Cell

91:231–241.19. Tanaka M, Konishi H, Touhara K, Sakane F, Hirata M, Ono Y, Kikkawa U (1999)

Biochem Biophys Res Commun 255:169–174.20. Salinas M, Wang J, Rosa de Sagarra M, Martin D, Rojo AI, Martin-Perez J, Ortiz

de Montellano PR, Cuadrado A (2004) FEBS Lett 578:90–94.21. Sussman J, Stokoe D, Ossina N, Shtivelman E (2001) J Cell Biol 154:1019–1030.22. Feral CC, Nishiya N, Fenczik CA, Stuhlmann H, Slepak M, Ginsberg MH (2005) Proc

Natl Acad Sci USA 102:355–360.23. Wu D, Thakore CU, Wescott GG, McCubrey JA, Terrian DM (2004) Oncogene 23:8659–

8672.

24. van Slegtenhorst M, Nellist M, Nagelkerken B, Cheadle J, Snell R, van den OuwelandA, Reuser A, Sampson J, Halley D, van der Sluijs P (1998) Hum Mol Genet7:1053–1057.

25. Lamb RF, Roy C, Diefenbach TJ, Vinters HV, Johnson MW, Jay DG, Hall A (2000)Nat Cell Biol 2:281–287.

26. Dan HC, Sun M, Yang L, Feldman RI, Sui XM, Ou CC, Nellist M, Yeung RS, HalleyDJ, Nicosia SV, Pledger WJ, Cheng JQ (2002) J Biol Chem 277:35364–35370.

27. Fukami K, Furuhashi K, Inagaki M, Endo T, Hatano S, Takenawa T (1992) Nature359:150–152.

28. Shibasaki F, Fukami K, Fukui Y, Takenawa T (1994) Biochem J 302:551–557.29. Honda K, Yamada T, Endo R, Ino Y, Gotoh M, Tsuda H, Yamada Y, Chiba H,

Hirohashi S (1998) J Cell Biol 140:1383–1393.30. Menez J, Le Maux Chansac B, Dorothee G, Vergnon I, Jalil A, Carlier MF, Chouaib

S, Mami-Chouaib F (2004) Oncogene 23:2630–2639.31. Honda K, Yamada T, Hayashida Y, Idogawa M, Sato S, Hasegawa F, Ino Y, Ono M,

Hirohashi S (2005) Gastroenterology 128:51–62.32. Stephens L, Anderson K, Stokoe D, Erdjument-Bromage H, Painter GF, Holmes AB,

Gaffney PR, Reese CB, McCormick F, Tempst P, et al. (1998) Science 279:710–714.33. Liang J, Slingerland JM (2003) Cell Cycle 2:339–345.34. Kaplan JM, Kim SH, North KN, Rennke H, Correia LA, Tong HQ, Mathis BJ,

Rodriguez-Perez JC, Allen PG, Beggs AH, Pollak MR (2000) Nat Genet 24:251–256.35. Patrie KM, Drescher AJ, Welihinda A, Mundel P, Margolis B (2002) J Biol Chem

277:30183–30190.36. Burns JC, Friedmann T, Driever W, Burrascano M, Yee JK (1993) Proc Natl Acad

Sci USA 90:8033–8037.37. Magliery TJ, Wilson CG, Pan W, Mishler D, Ghosh I, Hamilton AD, Regan L (2005)

J Am Chem Soc 127:146–157.38. Auersperg N, Maines-Bandiera SL, Dyck HG, Kruk PA (1994) Lab Invest 71:510–

518.39. Soneoka Y, Cannon PM, Ramsdale EE, Griffiths JC, Romano G, Kingsman SM,

Kingsman AJ (1995) Nucleic Acids Res 23:628–633.40. Hansen MB, Nielsen SE, Berg K (1989) J Immunol Methods 119:203–210.

Ding et al. PNAS � October 10, 2006 � vol. 103 � no. 41 � 15019

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

June

22,

202

1