Research ResourceCell Biology

Proximity biotinylation provides insight into the molecular composition of focal adhesions at the nanometer scale

See allHide authors and affiliations

Science Signaling  14 Jun 2016:
Vol. 9, Issue 432, pp. rs4
DOI: 10.1126/scisignal.aaf3572

Focusing on focal adhesions

Focal adhesions are protein complexes that link cells to the surrounding extracellular matrix, and their disassembly and reassembly are required during cell migration. The proteins in focal adhesions could be targeted to prevent cancer cell metastasis, but the composition of these complexes (which is generally referred to as the adhesome) has been controversial. Dong et al. used a label that targeted only proteins in close proximity to paxillin and β integrins, key components of focal adhesions. Mass spectrometry on the labeled proteins revealed that focal adhesions contained fewer proteins than previously suspected and identified several previously unknown components. After analysis of the function of these proteins in focal adhesions, they could be targeted to block metastasis.


Focal adhesions are protein complexes that link metazoan cells to the extracellular matrix through the integrin family of transmembrane proteins. Integrins recruit many proteins to these complexes, referred to as the “adhesome.” We used proximity-dependent biotinylation (BioID) in U2OS osteosarcoma cells to label proteins within 15 to 25 nm of paxillin, a cytoplasmic focal adhesion protein, and kindlin-2, which directly binds β integrins. Using mass spectrometry analysis of the biotinylated proteins, we identified 27 known adhesome proteins and 8 previously unknown components close to paxillin. However, only seven of these proteins interacted directly with paxillin, one of which was the adaptor protein Kank2. The proteins in proximity to β integrin included 15 of the adhesion proteins identified in the paxillin BioID data set. BioID also correctly established kindlin-2 as a cell-cell junction protein. By focusing on this smaller data set, new partners for kindlin-2 were found, namely, the endocytosis-promoting proteins liprin β1 and EFR3A, but, contrary to previous reports, not the filamin-binding protein migfilin. A model adhesome based on both data sets suggests that focal adhesions contain fewer components than previously suspected and that paxillin lies away from the plasma membrane. These data not only illustrate the power of using BioID and stable isotope–labeled mass spectrometry to define macromolecular complexes but also enable the correct identification of therapeutic targets within the adhesome.


Focal adhesions (FAs) are labile intracellular protein complexes that link metazoan cells to the extracellular matrix through integrins. These FAs both link and transduce forces from the internal actin cytoskeleton to the extracellular environment and vice versa. Their assembly requires the activity of Rho guanosine triphosphatase (GTPases) such as RhoA and Cdc42 (1), which promote actomyosin II contraction through ROCK and MRCK (2). Integrin cytoplasmic domains are generally quite small, with the β1, β2, β5, and β7 subunits binding no more than two to three partners (3). These direct partners, in turn, bind various cytoskeletal adaptor proteins, protein kinases, and signaling proteins that make up these macromolecular complexes. Some of these proteins are FA-specific [for example, paxillin, FA kinase (FAK), and talin], whereas others play roles in structures such as cell-cell junctions and actomyosin fibers (4). Live and super-resolution light imaging of tagged FA proteins indicated that each might occupy a particular “zone” relative to the plasma membrane; for example, super-resolution microscopy indicates that FAK occupies a region proximal to integrins, whereas vinculin lies 50 to 100 nm from the plasma membrane (5, 6).

The “adhesome” is a collective term for FA-localized proteins such as talin (7) and paxillin (8) and more than 150 other proteins (9, 10). The composition of the adhesome is sensitive to the actomyosin contractile activity (11, 12). Integrin engagement is required to initiate newly formed nascent adhesions enriched for proteins such as β3 integrin, talin, and paxillin (13). Unbiased proteomic approaches have been used to attempt to tabulate a more complete adhesome (11, 1416). The lability of FAs can be overcome in part by reversible chemical cross-linking (17), but most of the high-confidence “FA proteins” have not been replicated between the different studies (9). The recent “consensus integrin adhesome,” which comprises 60 proteins based on analyses of the most reliable FA proteomic data, nonetheless, includes histone H1 and several RNA binding proteins (10).

With the goal of identifying a small set of FA proteins by in situ labeling rather than through biochemical purification, we have combined biotin labeling identification (BioID) (18) with in-cell stable isotope labeling (SILAC) (19). The BioID strategy has been applied to analyze several complexes, including the inner nuclear membrane (18), the centrosome (20), and the nuclear pore complex (21). We have validated this new method using the FA protein paxillin, which is not enriched in other cellular compartments. Our probe [denoted green fluorescent protein (GFP)–BirA*–paxillin] generates reactive biotinoyl-5′-AMP (adenosine 5´-monophosphate) (22) to label primary amine groups within a radius of ~15 to 20 nm (21). This method, as applied to FA proteins, showed that 90% of proteins with a SILAC ratio >3 were reproducibly detected in biological replicates. Combining these data with those obtained with GFP-BirA*–tagged kindlin-2, which binds directly to integrins (23, 24), was highly informative, although kindlin-2 was also found to be prominently located at cell-cell junctions. Because this method generates relatively modest numbers of validated FA proteins, we were able to carry out rigorous protein-protein interaction assays in mammalian cells covering most protein types in the data sets. The results from these assays revealed a much smaller and credible number of partners for both paxillin and kindlin-2 than those found in the current literature.


Using GFP-BirA*-paxillin to label a subset of the adhesome

FAs are plasma membrane–proximal structures of undefined core composition (25, 26). The FA complex is quite labile after cell lysis, which necessitates chemical cross-linking to allow biochemical purification (16) or rapid fractionation (11). The in situ labeling of paxillin-proximal fusion proteins with biotin by BirA* (BioID) was anticipated to overcome these problems because in situ labeling obviates the need to retain protein-protein interaction during sample isolation. Paxillin is an FA protein with a substantial cytosolic pool (27). We generated U2OS lines expressing GFP-BirA*-paxillin (Fig. 1A and fig. S1). Using GFP-BirA*-paxillin (line 10), which expresses 1:1 of the transgene relative to endogenous paxillin, we carried out SILAC using a GFP-BirA* control line. Purified proteins were visualized by 10% SDS–polyacrylamide gel electrophoresis (PAGE) and Coomassie brilliant blue (CBB) staining, and regions corresponding to proteins >28 kD (red box) were excised and analyzed (Fig. 1B) (28). The tandem mass spectrometry (MS/MS) analysis routinely detected >400 proteins with high confidence, but most were nonspecific based on their SILAC ratio (Fig. 1C). Proteins with a SILAC heavy/light (H/L) ratio >2.5 (vertical red line) were considered to be selectively biotin-labeled in GFP-BirA*-paxillin cells. Most of these (orange) form part of the annotated adhesome (29) and include the key FA proteins uncovered to date (namely, FAK, talin, tensin, vinculin, p130Cas, and zyxin).

Fig. 1 Principles and workflow for SILAC-enriched analysis of paxillin-proximal proteins.

(A) Schematic of the GFP-BirA* and GFP-BirA*-paxillin (pax) constructs and their generation of biotinyl-5′-AMP, which labels both nonspecific (NS) and FA targets upon biotin addition to the culture medium. (B) Summary of workflow used to identify BirA*-paxillin–proximal proteins through BioID. The typical recovery of proteins in the stained polyacrylamide gel is shown, from which bands were excised for MS/MS analysis. Strep., streptavidin; Tx100, Triton X-100. (C) Scatter plot showing SILAC ratio and peptide coverage of the paxillin BioID data set in data file S1. The bulk of proteins are not SILAC-enriched (fall to left of the orange line); those on the right (SILAC >3) and identified as adhesome components (29) are labeled in pink.

We carried out two independent SILAC-BioID MS experiments in U2OS cells using GFP-BirA*-paxillin, which showed unanticipated reproducibility of the MS data in biological replicates compared with previous studies (Table 1 and data file S1) (11, 30). Variation in median SILAC values for highly enriched proteins (with >10 SILAC H/L ratio) is anticipated because of background noise in the “light” channel. The SILAC BioID analysis generated 35 proteins (Table 1, which does not include proteins represented by <4 peptides) with ~90% overlap between the two sets. The radius of BioID labeling in situ is estimated at 15 nm from the protein of interest (21), which represents a volume equivalent to the mammalian ribosome. The extended nature of the N terminus of paxillin could increase this volume to ~25 nm, which would encompass a larger protein data set, but one that lies within 100 nm estimated distance from integrin to the F-actin layer (31).

Table 1 A summary of combined data from two independent experiments using BirA*-paxillin to analyze proximal proteins by SILAC BioID.

Proteins are ordered by median SILAC heavy/light (H/L) ratio based on the higher value in either experiment (blue); “NI” indicates the protein that was not identified in the data set. In the last column, “FA” indicates proteins that have been previously localized to FAs as annotated in the adhesome (; “?” indicates proteins whose cellular localization has not been tested in this context. Proteins were selected on the basis of informative peptide number >3 and SILAC ratio (H/L) value >2.5. Proteins with a SILAC H/L ratio >10 are highlighted in red, 5 to 10 in pink, and 3 to 5 in yellow. The primary data can be found in data file S1.

View this table:

The most abundant paxillin-associated protein (based on peptide coverage) was talin-1, a key protein in FA assembly (32). Talin-1 is thought to span the entire FA from the plasma membrane, where it binds integrin, to the distal F-actin layer (33). We also detected kindlin-2, which is a membrane-proximal protein that also binds β1 integrin. These observations support the notion that BirA*-paxillin labels proximal proteins predominantly in the context of FAs. Paxillin dynamically exchanges between cytoplasmic pools (34) and the nucleus (35), but we did not detect nuclear proteins. The relative intensity-based absolute quantification (iBAQ)–derived abundances of the 36 proteins of interest from three adherent cell lines of different origins (U2OS, A549, and HeLa) are presented in fig. S3A. The deduced cellular amounts of these proteins varied over about three orders of magnitude. For the paxillin BioID data set, our unadjusted average heavy-peptide intensities without iBAQ normalization (36) fell within a 10-fold range (fig. S3B) and are calculated relative to paxillin.

Identification of previously unidentified adhesome components

We classified the paxillin BioID data set according to putative or known functions. Of the 35 proteins (Table 1), 27 have been identified as part of a previously published adhesome (29), a compiled list of 144 publication-validated FA components (Fig. 2A). The U2OS paxillin-proximal proteins not identified in the adhesome list are on the left side (Fig. 2A, red). Proteins that were highly enriched (rows 1 to 20 with SILAC ratios >10) are marked in red (Table 1), in which five of these proteins are not annotated adhesome proteins, namely, Kank2 (KN motif and ankyrin repeat protein 2), RN-Tre, RASnGAP, Pragmin, and Odin. It is notable that GIT, FAK, and PTP-PEST are established binding partners of paxillin (37). The ADP-ribosylation factor GTPase-activating proteins (ArfGAPs) GIT1 and GIT2 are recruited to FAs through the paxillin LD4 motif (38), and the RacGEF β-PIX is a partner of GIT (39). We did not detect any RNA or DNA binding proteins or those involved with protein folding and translation, which are typically recovered in purified FA isolates (29). Thus, our analysis contrasts with the proposed “consensus adhesome” based on various MS studies (10).

Fig. 2 The relationship between paxillin-proximal proteins and the adhesome.

(A) Proteins identified by BirA*-paxillin SILAC-BioID and adhesome proteins categorized according to their biochemical functions (25). Proteins in red were identified in the paxillin BioID data set (Table 1). The boxes to the left feature proteins not in the database at (B) STRING analysis (102) of proteins identified in the BirA*-paxillin SILAC-BioID data set (Table 1) with the addition of the protein VASP. The output was restricted to experimentally validated protein-protein interactions. To visually simplify the network, we removed less-studied isoforms (tensin-3, GIT2, and talin-2). Broader lines indicate higher confidence for each pairwise interactions; dotted lines mark the edges of the clusters. The data set was subject to MCL, which segregates the IPP complex with kindlin-2. The integrin-binding partners talin-1 and kindlin-2 are poorly connected in this network because few partners have been identified.

Unsupervised STRING analysis (40) of this data set was used to assess their connectivity based on previously validated protein-protein interactions. This network visualization (Fig. 2B) confirms that paxillin and FAK are connected directly to multiple components and through the small adaptors Crk and Nck. To improve coverage, we included VASP (six peptides detected with SILAC H/L ratio of 1.7; data file S1), which links various FA components to F-actin (41). The Markov cluster algorithm (MCL) segregates the paxillin-FAK cluster (red) from the IPP (ILK-parvin-PINCH) complex and a VASP-associated set (blue).

Proteins in the vicinity of paxillin that are not identified in the current adhesome (Fig. 2A, left) represent potentially new FA components. These include the F-actin–binding protein utrophin (UTRN), with a single report of its immunolocalization to FAs (42), and liprins that are typically found in puncta adjacent to FAs (43). The presence of RAPH1 (also known as lamellipodin) within the adhesion complex is consistent with its interaction with Rap1 and talin (33). Small heterotrimeric guanine nucleotide–binding proteins (G proteins) of the Rho and Rab families play a major role in establishing FAs in addition to Rap1. Several regulators of small GTPases found proximal to paxillin included RN-Tre, GIT1, ASAP1, β-PIX, Bcar3, and RASnGAP (marked in blue). The adaptor Odin [ankyrin repeat and SAM domain–containing 1A (ANKS1A)] which interacts with other SAM domain proteins, notably, EphA2 (44) and its C-terminal PTB domain, can bind various proteins including talin-2 and RASnGAP (45). RN-Tre (USP6NL) is not annotated in the adhesome but is an FA-enriched RabGAP (46). The labeling of α1-catenin (CTTNA1), but no other adherens junction proteins, suggests a new role for this actin-binding protein (but GFP–α1-catenin is not FA-enriched; fig. S2). Cells lacking α-catenin exhibit impaired cell migration and lamellipodial dynamics (47).

Pragmin (SGK223) is an FA protein that is related to pseudopodium-enriched kinase PEAK1 (48); both are likely to be pseudokinases with minimal activity (49). EphA2 is a receptor tyrosine kinase that is present only in adherent cells and was the only transmembrane protein detected (50). Although many different Ser/Thr kinases affect FA structure (51), only PAK2 was detected in the vicinity of paxillin, where it is recruited by PIX (39). In summary, taking account of isoform and orthologs in our data set, a network of only ~32 different protein species lie within ~25 nm of the N terminus of paxillin. This paxillin-proximal list contains only one membrane-bound protein (EphA2), indicating that, in this cellular context, paxillin does not lie close to the plasma membrane.

A comparison of GFP-paxillin affinity purification MS to BioID

GFP-containing fusion proteins can be efficiently recovered using recombinant single-chain anti-GFP coupled to agarose (GFP-Trap), providing us the opportunity to compare both techniques using the same cell line (fig. S4A). Using SILAC-labeled GFP-BirA*-paxillin (heavy) and GFP-BirA* (light) lysates and a standard MS workflow (fig. S4A), we scored agarose-bound proteins according to their SILAC enrichment (table S1). Those marked in red are bona fide partners, including GIT1, FAK, PTP-PEST, and the IPP complex. The SILAC enrichment of ribosome subunits (RPL13, RPL19, RPL35, and RPL39), nuclear proteins (histone H4, lamin B1, and lamin B2), and RNA binding proteins (HNRNPM) are typical CRAPome contaminants (52), accounting for ~50% of those proteins identified. A different published paxillin “interactome” using Flag-paxillin affinity purification MS (AP-MS) (53) yields different sets of contaminants (fig. S4B). These results confirm that AP-MS of GFP-paxillin after cell lysis yielded considerable amounts of SILAC-enriched contaminants that are not resolved through stringent control(s). Cellular in situ labeling and protein identification through SILAC implementation resolved this problem and demonstrates that paxillin was not proximal to noncytoskeletal components such as metabolic enzymes or RNA binding proteins.

Assessing direct binding partners of paxillin

Affinity copurification from cells remains the method of choice to assess binary protein-protein interactions and is particularly useful for side-by-side analyses. To mitigate the lack of quantitative information from Western blots, we elected to measure the “bait” by CBB staining, which has a lower detection limit of ~100 ng per lane on polyvinylidene difluoride (PVDF) membranes. Multiple GFP fusion proteins were successfully recovered at similar amounts (~100 to 300 ng) (Fig. 3A). A minority of fusion proteins (namely, PEAK1, ANKS1A, and EphA2) were too insoluble and were not tested. In our implementation, GFP-tagged proteins were coexpressed with a limiting amount of paxillin [3:1 DNA ratio of GFP-bait to hemagglutinin (HA)–paxillin] to increase stringency (Fig. 3A).

Fig. 3 Binary protein-protein interactions between paxillin and other local FA proteins.

(A) Paxillin interaction with various GFP-tagged proteins was assessed in cotransfected COS7 cells. GFP-Trap A pulldowns were detected by CBB staining and immunoblotted with an antibody directed against paxillin (anti-paxillin). The red stars indicate the full-length “bait” proteins. The Western blot shows the amount of paxillin in total detergent soluble lysates (5% of input) and the GFP-pulldown (10% of pulldown) with the same exposure. The image is representative of three separate experiments. TCL, total cell lysate. (B) Venn diagram illustrating public database entries for paxillin-binding proteins. The BioGRID 3.4 web interface ( was set to select only binding partners detected in low-throughput analyses. The Molecular INTeraction database is more stringently curated (103). The proteins that directly bind paxillin are marked in red. PPI, protein-protein interaction. (C) Summary of paxillin protein-protein interaction data and localization of respective GFP-tagged constructs of 30 proteins identified by paxillin BioID to FA. Proteins that are insoluble or those with closely related isoforms were not tested (NT). The FA localization of our set of GFP-tagged proteins is shown in fig. S2.

Several proteins with high SILAC H/L ratio such as talin-1, vinculin, and p130Cas did not appreciably interact with paxillin. In this side-by-side protein-protein interaction analysis, we found seven classes of protein that bind paxillin, with PINCH, Kank2, and LPP representing new paxillin-associated proteins. Kindlin-2 has been proposed to be an important partner for paxillin (54), but the interaction of kindlin-2 with paxillin was relatively weak and likely indirect (Fig. 3A). It was notable that 5 of 10 proteins with highest SILAC ratios (Table 1) can directly interact with paxillin. Thus, SILAC enrichment, regardless of absolute protein abundance, did indeed correlate to “nearest neighbors” in situ. The BioGRID 3.4 annotation contains >80 different protein-protein interactions for human paxillin, but most have been described in only one publication. Even for paxillin protein-protein interactions found in more than one study (Fig. 3B), there was, at best, moderate overlap with the curated Molecular INTeraction database (55) and our validated paxillin protein-protein interaction network (Fig. 3B, red). The FAK-paxillin interaction exhibits a dissociation constant (Kd) in the range of 1 to 9 μM (56), but parvin-α weakly binds paxillin (20 to 100 μM) (57) and did not copurify with paxillin in this assay (Fig. 3C). Our analysis suggests that PINCH interacted directly with paxillin.

An analysis of membrane-proximal FA proteins using kindlin-2

Nascent integrin-containing adhesions require recruitment of talin and kindlin, which can simultaneously bind to β-integrin cytoplasmic domains (33). Rap1-mediated talin activation promotes integrin activation (58). Kindlin-2 binds β1 or β3 integrin at a distal NPxY motif either simultaneously or before talin binding (23, 33). Kindlin-2 is a relatively abundant FA protein (fig. S3), and immunostaining of cells with paxillin and kindlin-2 antibodies indicates that both proteins are similarly enriched at both early and late phases of cell attachment (fig. S5). In addition, kindlin-2 tends to accumulate in the lamella (arrows). We generated several GFP–BirA*–kindlin-2 cell lines for proximity labeling in situ (table S2 and data file S1). A substantial number of transmembrane or N-myristoylated proteins and FA proteins were identified. Kindlin-2–mediated integrin activation and FA targeting require binding to the ILK complex (59). We detected β1 integrin in only one of the data set (seven peptides; SILAC ratio 4.0), likely because bound kindlin and talin together obscure the short β-integrin cytoplasmic domain. Talin-1 was the most abundantly labeled protein with >100 SILAC informative peptide pairs. The presence of many validated FA proteins suggests that a considerable portion of GFP–kindlin-2 exists in complex with β-integrin, ILK, and talin-1. Other SILAC-enriched proteins that interact with β-integrin cytoplasmic tails through PTB domain interactions are Numb (60) and tensin (61), although these proteins bind more strongly to β3 or β5 integrin than to β1 integrin. FA proteins proximal to paxillin (Fig. 2) comprised about one-third of those identified by kindlin-2 BioID (Fig. 4A). Among these, Kank2 and RN-Tre had the highest SILAC ratios in both sets. The low number of peptides for the ILK complex probably reflects poor biotin labeling or high lability (turnover) for proteins of average abundance in U2OS cells (fig. S3A).

Fig. 4 Kindlin associates with both FAs and cell-cell junctions.

(A) Summary of kindlin-2–proximal proteins identified in U2OS cells (table S2) and their overlap with the paxillin BioID set (right side). BioID-enriched proteins detected in tight junctions (69) are highlighted in blue. SLC3A2 (also known as CD98) forms a complex with several transmembrane transport proteins (64). Proteins in brown are not junctional-enriched and may be proximal to integrin in FAs. (B and C) Localization of kindlin-2 in FAs and cell-cell junctions revealed by confocal imaging. MDCK cells were stained for indirect immunofluorescence using mouse antibodies specific for kindlin-2 or β-catenin and rabbit antibodies specific for afadin or ILK. Scale bars, 10 μm. All cells in the field show the same phenotype. The images are representative of multiple fields from two separate experiments.

Paxillin, talin-1, Cdc98hc, and band 4.1G showed >40% sequence coverage in the kindlin-2 BioID set. There were >10 transmembrane proteins including EphA2, but no other receptor tyrosine kinase was detected. CD98hc (SLC3A2) is a single-pass type II transmembrane protein and heterodimeric subunit for multiple transporters (62) and is clearly an abundant membrane protein that binds to β1 integrins (63). This finding gave us confidence that integrin-associated membrane proteins were labeled by BirA*–kindlin-2. The CD98hc subunit can function as a transmembrane chaperone for β1 integrin, and its abundance is increased in various cancers (64). CD98 heterodimers form disulfide bonds between the CD98h membrane-proximal section of the extracellular domain and several “light chains” (various transporters), of which our kindlin-2 BioID set included SLC7A5, SLC6A8, SLC30A1, and SLC1A5. CD98 is associated with lipid rafts (65), as are the two plasma membrane palmitoyl transferases identified, termed DHHC-5 and DHHC-8.

Proximity of kindlin-2 to cell-cell junction proteins

The kindlin-2–proximal BioID list (table S2) included many cell-cell junction proteins. Endogenous kindlin-2 localized to junctions in both U2OS and epithelial MDCK cells (Fig. 4B). In human keratinocytes, kindlin-2 is proposed to bind to β1 integrin at cell-cell junctions (66, 67). Kindlin-2 loss can also affect the formation of cell-cell junctions (68). Does kindlin-2 recruit other FA proteins such as ILK to these junctions? We did not detect ILK with kindlin-2 at such cell-cell junctions (Fig. 4C). It was notable that α-catenin was in the paxillin BioID list but not in the kindlin-2 list, suggesting that α-catenin labeling by BirA-paxillin was unrelated to the junctional localization of kindlin-2 and β1 integrin. The presence of kindlin-2 and integrins in the vicinity of tight junctions has also been indicated in another BioID-based MS study using claudin-4 or occludin probes (69) or by comparing with E-cadherin–associated proteins (70). In both cases, kindlin-2 was ranked 38th (normalized proteomics and MS/observable peptide number values of 10.6 and 13.2); other proteins prominent in tight junctions are marked in table S2 (right columns). The kindlin-2–proximal proteins that were also junctional are marked in blue (Fig. 4A). Proteins that interact with small G proteins (table S2) include RN-Tre, RASnGAP, and lamellipodin. Three GTPase effectors are noted: IRSp53, Borg5, and WAVE-3 (WASF2). Our analysis failed to identify small G proteins themselves because of an imposed size cutoff of <28 kD (Fig. 1B).

Identification of Kank2, EFR3A, and liprin as kindlin-2 binding partners

Kindlin-2 is associated with integrin and the IPP complex and may act as an adaptor for other FA proteins (33). We therefore tested a panel of nontransmembrane proteins for their ability to interact with kindlin-2 (Fig. 5A). Our positive controls α-parvin and ILK bound to kindlin-2; however, we could not detect an interaction of kindlin-2 with migfilin. Zyxin family proteins were not identified as proximal to kindlin-2. However, we did identify a robust biochemical interaction of kindlin-2 with both liprin and EFR3A, proteins that promote endocytosis. Liprin can promote integrin endocytosis (43), whereas the EFR3A fly ortholog (rolling blackout) is essential for local endocytosis at synapses (71). EFR3A is a conserved palmitoylated peripheral membrane protein, which recruits PI4KIIIα to the plasma membrane and, thereby, maintains PtdIns(4,5)P2 (PIP2) concentrations (72). In kindlin-2 (±) endothelial cells, cell surface amounts of CD39 and CD73 are increased two- to threefold, which has been linked to alter trafficking at the plasma membrane (73). The interaction of kindlin with both integrins and EFR3A could serve to maintain local PIP2 concentrations. Moreover, the lack of the PIP2-binding protein band 4.1R prevents cell adhesion (74). Thus, kindlin-2, a binding partner for β1 integrin, is associated with both FA and cell-cell junctions, where recruitment of EFR3A and/or liprins likely plays an important role in local endocytosis.

Fig. 5 Identification of additional kindlin-2–binding proteins.

(A) HA–kindlin-2 interaction with various GFP-tagged proteins as indicated (excluding transmembrane proteins) was assessed. GFP-Trap A pulldowns (PDs) were detected with Coomassie brilliant blue (CBB) staining and probed with antibody specific for kindlin-2. The red stars indicate the full-length “bait” proteins. The Western blot shows the amount of kindlin-2 in total detergent soluble lysates (5% of input) and the GFP-pulldown (10% of pulldown) with the same exposure. The image is representative of three separate experiments. (B) Schematic of the domain organization of Kank2 and the constructs generated to test localization. The KN motif is conserved across all human Kank isoforms and not found in other proteins. (C) Localization of various GFP-tagged constructs shown in (B). Typical images of live U2OS cells transiently expressing mApple-paxillin and GFP-Kank2 constructs are shown. More than 90% of transfected cells imaged showed this phenotype (n = 10 cells). The images are representatives of multiple fields from two separate experiments. Scale bar, 10 μm.

FA localization of Kank2 through a conserved KN motif

The cytoskeletal adaptor Kank2 has not been previously linked to FA function. Kank2 was enriched in both paxillin and kindlin-2 BioID (Table 1 and fig. S2) and, probably, is an abundant component of the adhesome. The N terminus of Kank2 contains a region that is conserved among all Kank isoforms (KN motif) required for FA localization (Fig. 5B), which does not require the ankyrin repeat region. Kank2 is clearly an FA-enriched protein (Fig. 5C); however, no protein-protein interaction analyses previously linked it to any adhesions. In support of our data, VAB-19, the Caenorhabditis elegans Kank ortholog, is enriched at muscle attachment sites (75, 76). Here, we identified 1–135 as the region of Kank2 that is sufficient for FA engagement (Fig. 5C).

Using BioID to map the localization of FA components

On the basis of our stringent protein-protein interaction criteria and published data on biologically relevant neighbors of paxillin and kindlin-2, we constructed a local network of interactions (Fig. 6A). Previously characterized protein-protein interactions are marked with light blue lines, whereas the interactions that we identified are marked in dark blue. The figure indicates that Kank2 may bridge kindlin-2 with paxillin. However, it is likely that talin also recruits paxillin through other interactions, for example, FAK. The link between EFR3A and PI4KIIIα can occur through two alternate adaptors, TTC7 or TMEM150A (77).

Fig. 6 Local protein-protein interactions involving paxillin and kindlin-2 identified in this study.

(A) Summary of the protein-protein interactions identified in this study, which link kindlin-2 and paxillin to their local partners. The dark blue lines indicate new interactions not previously reported, and the light blue lines are well-validated interactions (multiple reports), where appropriate membrane-associated proteins are placed in proximity to the plasma membrane. The trimeric IPP and GIT-PIX-PAK2 complexes are treated as single complexes. (B) The consensus integrin adhesome comprising 60 proteins as recently defined (10), which were scored in at least five of seven MS/MS data sets obtained from purified FAs. The overlap of paxillin BioID with integrin-associated material from U2OS cells (10) is marked [+]. Actinin- or F-actin–associated proteins are marked in red (and notably absent from our paxillin BioID list); many of these proteins can be found on both actin stress fibers and FAs (as well as cell-cell junctions). Proteins that are also in the curated adhesome list ( are marked.

We interrogated the 60 proteins in a consensus integrin adhesome (10) to estimate the coverage of our paxillin BioID data set (Fig. 6B). The authors noted that their experimental protocol (10) yielded 42 of these FAs proteins using U2OS cells (Fig. 6B, marked +), of which 16 of the proteins in close proximity to paxillin were represented. Thus, ~50% of the proteins we identified are not routinely identified by conventional FA cross-linking and biochemical purification (10). Of note are 12 proteins that are involved with F-actin or actomyosin interaction (marked in red), a pool of proteins such as α-actinin that we deduce cannot be labeled by BirA*-paxillin (because of the distance from paxillin to the F-actin layer). Most of these actin-associated proteins are not exclusive to FAs and stabilize the FA-actin interface (12).

Combining our two BioID data sets allowed us to build a model reflecting the composition and localization of proteins in a typical U2OS FA (Fig. 7). We excluded proteins that are primarily associated with cell-cell junctions but are aware that there may be considerable overlap between these two compartments, given their close association with lipid rafts. Those proteins in close proximity to kindlin-2 (but not paxillin) that have a role in FAs are shown and include MCAM, which, like integrins, interacts with the cellular matrix. We are cognisant that the overall composition of FAs can change and that certain classes of F-actin–binding proteins such as α-actinin (as mentioned above) are FA-enriched. Overall, our analysis was consistent with the binding of BirA–kindlin-2 to integrins in the presence of talin-1, a lynchpin for FA assembly by activated integrins (78).

Fig. 7 Suggested localization of FA proteins in FAs in U2OS cells based on BioID data.

The schematic shows the protein-protein interactions we detected between paxillin or kindlin-2 and other FA proteins and that have been selected published literature. To simplify the drawing, we have added multiple integrin dimers. Kindlin-2 can interact with the integrin in the presence of talin (32) and likely with Cd98h. Proteins thought to interact directly with integrin are in pink. Transmembrane or peripheral membrane proteins are in blue.

Our data conflict with the most widely accepted models of FA protein localization based on super-resolution microcopy (5). The model we present (Fig. 7) places FAK away from the plasma membrane because it is not labeled by BirA–kindlin-2. An alternative explanation is that kindlin-2 competes with FAK for integrin binding. Because BirA*-paxillin fails to substantially label any membrane-bound proteins except EphA2, paxillin is assigned to an intermediate zone (with FAK). Such proteins that are detected by paxillin BioID (but not by kindlin-2 BioID) are shown in yellow, whereas those in close proximity to kindlin-2 are in green. This model is self-consistent; for example, regulatory complexes that interact directly or indirectly with paxillin such as GIT, FAK, Crk, and p130Cas fall into the “intermediate” category. Integrin binding partners such as tensin, talin, and kindlin-2 are detected in close proximity to paxillin. The recruitment of paxillin to FAs through multiple protein interactions (Fig. 6A) will be the subject of a follow-up study. The nature of the integrin-proximal signaling layer is of particular interest and requires further investigation.

Adhesome proteins and mechanosensitivity

Integrin complexes containing kindlin are the first components to assemble in nascent adhesions (perhaps from preassembled complexes). The association of talin with the integrin-kindlin complex is actomyosin II–dependent, with a suggested stoichiometry of one talin molecule linking two integrin-kindlin complexes (79). Subsequently, LIM domain–containing proteins such as those of the zyxin family become particularly enriched in FAs under tension (12). We characterized paxillin-proximal proteins for their blebbistatin sensitivity using an existing data set (16). Proteins that were depleted from FAs by treatment with blebbistatin were TRIP6, zyxin, VASP, LPP, paxillin, vinculin, LIMD1, p130Cas, FAK, and α-parvin (table S3, marked in blue). Thus, of the membrane-proximal FA proteins identified by kindlin-2 BioID, only the IPP complex exhibits mechanosensitivity.


BioID analysis as the basis for establishing protein-protein interaction networks

An important question is whether the BioID protocol, as reflected in the data presented here, reflects in situ proximity or is biased by the cellular abundance of the various proteins. The cellular protein abundant can be inferred from quantitative proteomics of total protein extracted from cell lines (80) (fig. S3A). The iBAQ algorithm uses quantified MS/MS and normalizes the summed peptide intensities against those theoretically observable for each protein. The relative iBAQ-derived abundances of the FA proteins in U2OS, A549, and HeLa cell lines indicate quite different amounts of FA proteins. It is important to note that proteins identified with high SILAC ratios in the paxillin BioID were not the most abundant FA proteins in U2OS cells. Overall, 33 of 36 proteins are found at measurable amounts in these three cell lines, suggesting that these proteins are common components of FAs.

The basis for the disposition of a minority of FA proteins is understood at the molecular level. For example, kindlin and talin bind directly to certain β-integrin cytoplasmic domains and recruit ILK and vinculin, respectively (81, 82). For most other FA components, multiple protein interactions are thought to contribute to their localization (83). Paxillin requires the double zinc-finger LIM (Lin11, Isl-1, and Mec-3) domains for targeting (38, 84), but no single FA protein is responsible for paxillin localization (85). The biochemical data suggest that paxillin is potentially anchored through multiple interactions with PTP-PEST, FAK, Kank2, and PINCH in the context of U2OS cells. Rigorous biochemical analyses to test interactions among all 32 distinct FA proteins will allow us to infer their interrelationships. Existing protein-protein interaction data sets are contaminated and, thus, of limited use (for example, BioGRID indicates >80 paxillin binders). Our protein-protein interaction data also matched poorly with pairwise associations assessed by fluorescent cross-correlation spectroscopy (FCCS) in cells (86). The FCCS data from pairs of fluorescent tagged proteins in situ (using α-actinin, α-parvin, p130Cas, CSK, FAK, ILK, paxillin, PINCH, talin, tensin, VASP, vinculin, and zyxin) suggested that paxillin directly interacts with α-parvin, FAK, α-actinin, p130Cas, VASP, and vinculin, which does not match our protein-protein interaction data.

On the disposition of proteins within FAs

The BioID method is dependent on both the presence of and access to lysines on target proteins by the AMP-biotin, which will lead to some variability in the extent of protein labeling (21). Our experiments demonstrate the benefits of combining BioID with SILAC analysis, which can accommodate these variations using an appropriate control cell line and accurately probe the environment within the labile FA complex (Fig. 2A and Table 1). In the case of paxillin, among those with the highest SILAC ratio are proteins that directly bind paxillin, namely, parvin, GIT1, liprin, FAK, PTP-PEST, and Kank2 (Fig. 6A).

The nanoscale architecture of integrin adhesion complexes is largely unknown. Electron microscopic (EM) observations of fibroblast FAs and three-dimensional structural reconstruction using cryo-electron tomography have suggested small objects with (x-y) diameters of 25 ± 5 nm, termed “FA-related particles,” which occupy <5% of total FA area (87). Light-based super-resolution methods (5) confirm previous EM studies that an F-actin–rich layer sits 80 to 100 nm above the integrin-enriched zone (31). The median disposition of tagged paxillin appears ~45 nm above the z plane by super-resolution (88), whereas vinculin lies ~70 nm away. A plasma membrane marker tdEos-CAAX in this experimental setup appears ~25 nm above the z plane (88). This is not inconsistent with our model (Fig. 7), which has paxillin and multiple other FA proteins (in yellow) occupying an intermediate region perhaps 25 nm from the lipid bilayer. This layer has mechanosensitive proteins such as FAK and vinculin (shown in yellow), which would preclude a direct interaction between FAK and integrins. Conversely, the distal region (not accessible to BirA-paxillin) would represent an actin-rich zone ~70 to 100 nm from the plasma membrane and not labeled by BirA-paxillin, comprising F-actin and associated proteins such as α-actinin. Larger proteins such as talin and tensin (>150 kD) likely penetrate all three regions. Consistent with this model, the position of the F-actin layer is affected by the engineered distance between the talin “head” and its C-terminal F-actin–binding domain (89). One surprise is that the Arf6GAP GIT1 and the RacGEF PIX (39, 90) are not apparently in the vicinity of their membrane-bound small G proteins targets. There are no RhoGEFs detected in proximity to kindin-2, although a number of Cdc42 and Rac1 effectors such as Borg5, IRSp53, and WAVE-2 are observed.

New FA components identified in this study

Of the proteins that are highly enriched in our analysis (highlighted in red in Table 1), seven have not been previously annotated as FA proteins (α-catenin, Odin, Kank2, Pragmin, RN-Tre, RASnGAP, and UTRN). Here, we validated Kank2 as a bona fide FA component and found that RN-Tre (46) and UTRN (42) were FA-enriched, but that α-catenin was not (fig. S2). Follow-up studies will test the localization of the remaining three. Inspection of the RASnGAP primary sequence indicates that the protein contains a conserved C-terminal cysteine, which is either palmitoylated to allow plasma membrane localization or forms part of a conserved PDZ-binding motif. Two similar Ras GAP proteins termed SynGAP and DAB2IP are thought to regulate Rap1 in addition to Ras (91). The ability of RASnGAP to act as a tumor suppressor in certain breast cancers (92) could relate to its effects on integrin signaling though modulation of Rap1.

FA-enriched protein kinases

Pragmin is somewhat related to PEAK1 and both are likely pseudokinases because they have minimal activity in vitro (49). These proteins therefore act as scaffolds; Pragmin is an Rnd2 effector (93). Pragmin (but not PEAK) contains an EPIYA motif identified in various pathogen proteins that inhibit the C-terminal Src kinase CSK (94). Thus, Pragmin could activate the nonreceptor tyrosine kinase Src (or Yes), which, in turn, is required for full FAK activation. PTP-PEST binds directly to paxillin and keeps FAK activity in check (95). EphA2 was the only transmembrane tyrosine kinase we found in FAs, which is interesting, given that it is present only in cells that are adherent (50). ILK is another pseudokinase that forms a trimeric IPP complex that localizes to FAs through interaction with kindlin and/or paxillin (96). We conclude that PAK2 (the ubiquitous group I PAK) is the primary resident Ser/Thr kinase in FAs. We have previously shown that paxillin is not a good PAK1 substrate (35), but GIT and PIX, which likely forms a large hexomeric complex at this site (97), are phosphorylated by PAK1 in vitro (98).

In summary, our study suggests that the integrin adhesome network is more compact than has been suggested by previous studies. We detected 32 of these proteins in other adherent cells (fig. S3A). It seems reasonable to assume that this experimentally defined set of paxillin-proximal proteins will be largely common to other cell lines. Presently, we are missing an estimated 10 to 15 more distal FA proteins that are associated with the overlying F-actin network. The identity of proteins proximal to the integrin cytoplasmic domains is suggested by our analysis (Fig. 7) but requires further validation. These findings do not rule out (the as yet unidentified) physical connections between FAs and “noncanonical” adhesome components such as ribosomes and microtubules (99).


Plasmids and reagents

The SILAC medium and dialyzed fetal bovine serum were from Thermo Scientific. The heavy l-lysine (U13C6; U15N2 K8) and l-arginine (U13C6; U15N4 R10) were from Cambridge Isotope Laboratories. The light l-lysine (K0) and l-arginine (R0) were from Sigma-Aldrich. High-capacity NeutrAvidin agarose resin (catalog #29204) is from Thermo Scientific, and GFP-Trap A agarose is from ChromoTek GmbH.

A humanized sequence corresponding to BirA-R118G (18) was cloned into pXJ-GFP vector (39) by polymerase chain reaction to create pXJ-GFP-BirA* backbone. Full length of human paxillin open reading frame (ORF; U14588) was cloned into pXJ-GFP-BirA* vector using 5′ Hind III and 3′ Kpn I restriction sites to create pXJ-GFP-BirA*-paxillin (fig. S1A). Full length of mouse kindlin-2 ORF (BCO33436) was cloned into pXJ-GFP-BirA* vector using 5′ Bam HI and 3′ Not I restriction sites to create pXJ–GFP–BirA*–kindlin-2. pBABE-puro was a gift from W. Hong. Other mammalian ORFs were derived from the following publicly available plasmids: USP6NL (HsCD00081860), TNS3 (HsCD00342488), Kank2 (MmCD00312322), Ptpn12 (MmCD00319983), CTNNA1 (HsCD00040121), LPP (MmCD 00313629), BCAR1 (HsCD00295584), BCAR3 (MmCD318547), Cdc42EP1 (HsCD00042021), CDCA3 (HsCD00367923), EFR3A (HsCD00295205), EPB41L5 (HsCD00296963), ZDHHC5 (HsCD00377209), PEAK1 (HsCD00399144), and NUMB (HsCD00365821). Full-length Kank2 complementary DNA (cDNA) clone was amplified from human cDNA prepared from HeLa cell mRNA. All ORFs were cloned into pXJ-GFP vector, and the ORF sequence was confirmed by sequencing.

Cell culture and generation of stable cell lines

The human osteosarcoma cell line U2OS and monkey fibroblast COS-7 line were cultured in high-glucose Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum. U2OS cells were cotranfected with respective GFP-BirA* constructs and pBABE-puro at a 50:1 ratio using Lipofectamine 2000 reagent according to the manual provided by the supplier. Twenty-four hours after transfection, cells were trypsinized, diluted at ratio 1:10, and replated. Stable cell line selection with puromycin (0.7 μg/ml) began 24 hours later. Upon colony formation, individual clones were isolated and screened by cell imaging and Western blot.

NeutrAvidin affinity capture protocol

Established stable cell lines were adapted in either heavy (H) or control light (L) isotopic labeled culture medium for 14 days (28). Puromycin was removed from cell culture during this period. Cells at ~90% confluence were incubated with 100 μM biotin for 5 hours and then washed three times with phosphate-buffered saline (PBS) (5 min, room temperature). Each 95-mm dish was harvested in buffer A [0.5 ml of 100 mM KCl, 25 mM tris-HCl (pH 8.0), 0.5% Triton X-100, 0.5% deoxycholate (DOC), 5 μM EDTA, and protease inhibitor cocktail; Roche]. The cell lysate was transferred to 1.5-ml tube and triturated 10 times, and nuclei were pelleted by centrifugation (20 s at 16,000g in Eppendorf centrifuge 5424). The supernatant was saved, and SDS was added to 0.2%. After sonication (20 s), the centrifugation was repeated for 10 min at 4°C. The clarified cell lysates from heavy and light (control) cells were pooled and added to NeutrAvidin Sepharose (40 μl of slurry), which was incubated with rolling at 4°C overnight. The beads were then washed (twice for 5 min at 25°C) with 1 ml of wash buffer B [25 mM tris (pH 8.0) and 1% SDS + protease inhibitor cocktail] and then twice (10 min) at 4°C in wash buffer C [25 mM tris (pH 7.4), 0.5% DOC, 0.5% Triton X-100, 100 mM KCl, and protease inhibitor cocktail]. Bound proteins were eluted by heating in 80 μl of 1× lithium dodecyl sulfate (LDS) sample buffer (Novex NuPAGE), which was removed, and then 40 μl of distilled H2O was added to rinse the beads. The combined eluate volume was reduced to 50 μl under vacuum, and 40 μl (per lane) was subjected to SDS-PAGE (10% gel). After colloidal blue (Novex, Invitrogen) staining, multiple bands were processed for trypsin digestion and MS (28).

SILAC analysis of GFP-paxillin–associated proteins

U2OS cell lines expressing GFP-BirA*-paxillin or GFP-BirA* were adapted in either heavy (H) or light (L) isotopic culture medium for 14 days. Each 95-mm dish was harvested in lysis buffer [0.5 ml of 100 mM KCl, 25 mM tris-HCl (pH 8.0), 0.5% Triton X-100, 0.5% DOC, 5 mM EDTA, and protease inhibitor cocktail]. The cell lysate was transferred to a 1.5-ml tube and triturated 10 times, and nuclei were pelleted by centrifugation (20 s at 16,000g in Eppendorf centrifuge 5424). After sonication (20 s) of the cell lysate, the centrifugation was repeated for 10 min at 4°C. The “heavy” and “light” cell lysates, separately, were incubated (8°C) with GFP-Trap A beads (20 μl of slurry; ChromoTek) in lysis buffer for 4 hours with rolling. The beads were washed three times [25 mM tris (pH 7.4), 0.5% DOC, 0.5% Triton X-100, 100 mM KCl, and protease inhibitor cocktail]. Bound proteins were eluted by adding 80 μl of 1× LDS buffer [+50 mM dithiothreitol (DTT); Novex NuPAGE] to the beads and placing in 100°C block for 5 min. The volume was then reduced to 40 μl under vacuum. Half of the material (20 μl per lane) was analyzed by SDS-PAGE (10% gel). After staining, the bands were processed for trypsin digestion and MS (28).

Protein identification by SILAC mass spectrometry

Stained polyacrylamide gel pieces (2 mm) were extracted and digested with trypsin under standard conditions (100), and each slice was subjected to nano-liquid chromatography–MS/MS using Orbitrap or Orbitrap XL (Thermo Fisher). Survey full-scan MS spectra [mass/charge ratio (m/z) 310 to 1400] were acquired with a resolution of r =60,000 at m/z 400, an automatic gain control target of 1 × 106, and a maximum injection time of 500 ms. The 10 most intense peptide ions in each survey scan with an ion intensity of >2000 counts and a charge state ≥2 were isolated sequentially to a target value of 1 × 104 and fragmented in the linear ion trap by collisionally induced dissociation using a normalized collision energy of 35%. A dynamic exclusion was applied using a maximum exclusion list of 500 with one repeat count and exclusion duration of 30 s.

Raw data were processed by MaxQuant software (v1.3.0.5) involving the built-in Andromeda search engine (101). The search was performed against the Human database. Database searches were performed with tryptic specificity, allowing maximum two missed cleavages and two labeled amino acids as well as an initial mass tolerance of 6 ppm (parts per million) for precursor ions and 0.5 dalton for fragment ions. Cysteine carbamidomethylation was searched as a fixed modification, and N-acetylation and oxidized methionine were searched as variable modifications. Labeled arginine and lysine were specified. False discovery rates were set to 0.01 for both protein and peptide. Proteins were considered identified when supported by minimum ratio count 2 with a minimum length of seven amino acids. The SILAC ratios (median) for each point are derived from at least five peptides, with peptide confidence >95%.

Immunofluorescence, coimmunoprecipitation, and Western blotting

Various GFP-tagged proteins were transiently expressed with mCherry-paxillin in U2OS cells; after 16 hours, the cells were replated on fibronectin (10 μg/ml)–precoated glass coverslip for live cell imaging. Stable U2OS cell lines were routinely cultured without antibiotics and plated on fibronectin-coated glass coverslips for microscopy. Cellular biotinylation was carried out by incubating cells in 100 μM biotin for 5 hours. Cells were fixed in 4% paraformaldehyde before mounting. Biotinylated proteins were detected using Alexa Fluor 546–conjugated streptavidin (1: 1000; Invitrogen). For other antibody staining, rabbit or mouse antibodies were added to the coverslips and incubated for 2 hours at 25°C at indicated dilution after blocking with 10% goat serum at 25°C for 10min. After washing twice with 0.1% Triton X-100 for 10 min at 25°C, Alexa Fluor 488– or Alexa Fluor 546–conjugated antibodies recognizing rabbit or mouse immunoglobulins (1: 500; Invitrogen) were then added for 1 hour at 25°C. After washing twice with 0.1% Triton X-100 for 10 min at 25°C, coverslip was mounted in PermaFluor Mountant (Thermo Scientific). Primary antibodies used in this study included mouse antibodies specific for kindlin-2 (1:200; Millipore), vinculin (1:500; Sigma), or β-catenin (1:200; Abcam) and rabbit antibodies specific for paxillin (1:500; Epitomics) or afadin (1:200; Sigma). Images were acquired using Olympus BX61 microscope equipped with CoolSnap HQ cold CCD camera or Olympus FluoView confocal system.

For coimmunoprecipitation experiments, COS cells (60-mm dish) were transiently cotransfected with 0.5 μg of GFP-tagged constructs as indicated and 0.15 μg of HA-paxillin or HA–kindlin-2. After 16 hours, the cells were lysed in 0.65 ml of lysis buffer [25 mM tris (pH 7.4), 100 mM KCl, 5 mM MgCl2, 0.5% Triton X-100, 5% glycerol, 5 mM DTT, and protease inhibitor cocktail]. After sonication and centrifugation, the supernatant was saved and used for immunoprecipitation. GFP-Trap A agarose beads (10 μl of slurry; ChromoTek) and 0.6 ml of lysate were incubated with rolling at 4°C for 2 hours. The beads were then washed with lysis buffer (three times for 5 min). The bound proteins were eluted in 60 μl of 2× SDS sample buffer. Of this, 5 μl of the elute was separated by SDS-PAGE and transferred to a PVDF membrane. Immunoblotting was performed with horseradish peroxidase (HRP)–conjugated antibody specific for GFP (1:10,000; Abcam), rabbit antibody specific for paxillin (1:6000; Epitomics), or mouse antibody specific for kindlin-2 (1:2000; Millipore). The secondary antibodies were HRP-conjugated antibodies specific for rabbit or mouse immunoglobulin (1:10,000; Dako). The PVDF membrane was incubated with Immobilon enhanced chemiluminescence solution (Millipore) according to the manufacturer’s instruction before it was exposed to an x-ray film. To detect biotinylated proteins, membranes were blocked in 2.5% bovine serum albumin in PBS with 0.4% Triton X-100 and incubated in the same buffer with HRP-conjugated streptavidin (1:10,000; Cell Signaling).


Fig. S1. Characterization of U2OS cell lines stably expressing GFP-BirA*-paxillin.

Fig. S2. Intracellular localization of some GFP-tagged proteins that were identified through GFP-BirA*-paxillin SILAC BioID.

Fig. S3. The SILAC H/L ratio is not related to its protein abundance in the cell.

Fig. S4. Identification of paxillin-associated proteins by AP-MS.

Fig. S5. Colocalization of paxillin and kindlin-2 in U2OS cells.

Table S1. SILAC-enriched paxillin-associated proteins identified by AP-MS.

Table S2. List of proteins identified by GFP–BirA*–kindlin-2 SILAC BioID.

Table S3. Summary of reported blebbistatin sensitivity of paxillin-proximal proteins based on MS analysis of fibroblast proteins.

Data file S1. Unfiltered data sets of protein identified by paxillin and kindlin-2 BioID.


Funding: This work is supported by the Agency for Science Technology and Research (A*STAR), Singapore. Author contributions: E.M. and J.-M.D. conceived the study and wrote the manuscript with input from all authors. E.M., J.-M.D., and B.B. designed experiments. J.-M.D., T.L., and F.P.-L.T. generated cDNA plasmid constructs and carried out mammalian cell culture experiments. H.L.-F.S. and J.G. carried out MS/MS data collection and analysis. J.-M.D. collected and processed live cell and immunofluorescence imaging of mammalian cell expressing various GFP-tagged constructs. Competing interests: The authors declare that they have no competing interests. Data and materials availability: MS data were deposited in ProteomeXchange ( through the PRIDE partner repository with the primary accession identifier PXD004180.
View Abstract

Stay Connected to Science Signaling

Navigate This Article