Assembly of the β4-Integrin Interactome Based on Proximal Biotinylation in the Presence and Absence of Heterodimerization*

This study characterized the β4-integrin interacting proteome using BioID proximity-dependent biotinylation in epithelial MDCK cells. The analysis identified several novel type II hemidesmosome (HD)-associated proteins and revealed potential connecting protein modules that could orchestrate the observed coordinated coassembly of HDs and focal adhesions (FAs). Curiously, unlike the formation of HDs, the assembly of β4-interactome did not depend on α6β4-heterodimerization. Graphical Abstract Highlights This study reports the first proteomic characterization of a type II hemidesmosomal complex. This study characterizes the interactome of β4-integrin in the presence and absence of α6-integrin in a simple epithelial cell model. The assembly of the β4-integrin interacting complex was largely independent of α6-integrin expression. Integrin-mediated laminin adhesions mediate epithelial cell anchorage to basement membranes and are critical regulators of epithelial cell polarity. Integrins assemble large multiprotein complexes that link to the cytoskeleton and convey signals into the cells. Comprehensive proteomic analyses of actin network-linked focal adhesions (FA) have been performed, but the molecular composition of intermediate filament-linked hemidesmosomes (HD) remains incompletely characterized. Here we have used proximity-dependent biotin identification (BioID) technology to label and characterize the interactome of epithelia-specific β4-integrin that, as α6β4-heterodimer, forms the core of HDs. The analysis identified ∼150 proteins that were specifically labeled by BirA-tagged integrin-β4. In addition to known HDs proteins, the interactome revealed proteins that may indirectly link integrin-β4 to actin-connected protein complexes, such as FAs and dystrophin/dystroglycan complexes. The specificity of the screening approach was validated by confirming the HD localization of two candidate β4-interacting proteins, utrophin (UTRN) and ELKS/Rab6-interacting/CAST family member 1 (ERC1). Interestingly, although establishment of functional HDs depends on the formation of α6β4-heterodimers, the assembly of β4-interactome was not strictly dependent on α6-integrin expression. Our survey to the HD interactome sets a precedent for future studies and provides novel insight into the mechanisms of HD assembly and function of the β4-integrin.


In Brief
This study characterized the ␤4integrin interacting proteome using BioID proximity-dependent biotinylation in epithelial MDCK cells. The analysis identified several novel type II hemidesmosome (HD)-associated proteins and revealed potential connecting protein modules that could orchestrate the observed coordinated coassembly of HDs and focal adhesions (FAs). Curiously, unlike the formation of HDs, the assembly of ␤4-interactome did not depend on ␣6␤4heterodimerization.
Integrins are a large family of ␣␤-heterodimeric extracellular matrix (ECM) 1 receptors that recognize short peptide motifs found in many ECM proteins, including laminins (4). Integrins do not possess intrinsic enzymatic activity. Instead, they cluster together and interact with cytoplasmic effectors that in turn recruit numerous additional components to form large multiprotein complexes, collectively termed as integrin-associated complexes (5)(6)(7). ␤1and ␤4-integrins form distinct adhesion complexes, focal adhesions (FA) and hemidesmosomes (HD), respectively, but they both bind to laminin and might thus synergistically contribute to laminin adhesion and signaling. ␤1-integrins form actin-linked FAs and convey many critical functions in epithelial cells (8 -10). Unlike ␤1integrin, ␤4-integrin pairs only with ␣6-subunit and contains an unusually large cytoplasmic tail that links to intermediate filament network at HDs. Two types of HDs exist, type I HDs are highly organized structures formed at the basal layer of stratified epithelium, such as in the skin epidermis, where they link the epidermal-dermal layers, providing mechanical strength and durability (11). Simple epithelial cells contain type II HDs that are less complex and much less studied (12). Although HDs are generally considered for their adhesive function, ␤4-integrins have also been implicated in the regulation of laminin-triggered polarity signals in epithelial cells (13)(14)(15).
Although a number of comprehensive proteomics-based studies have been performed in fibroblasts and lymphoblasts to characterize the components of FAs, little attention has been put on the composition of HDs in epithelial cells. Here we have characterized the ␤4-interactome in Madin Darby Canine Kidney (MDCK) epithelial cells using biotin ligase (BirA)-based BioID proximity labeling technique (16). In MDCK cells, ␣6␤4-integrins form adhesions that resemble type II HDs. The MDCK-HDs colocalized with basal laminin patches, presumably demarcating sites of cell-driven laminin-assembly. Efficient formation of MDCK-HDs required both ␣6and h␤4ϩ␣6KO cells; BirA-GFP cells and myrGFP-BirA cells) were analyzed by LC-MS/MS (supplemental Tables S1, S2, and S4). Proteins that were enriched at least 3-fold in h␤4-BirA or BirA-h␤4 cells when compared with control cells were determined as components of ␤4-interactome. The ␤4-interactome was also validated by comparing it with BirA-GFP-and myr-GFP-BirA-interactomes (supplemental Table S2). Colocalization analysis was done using samples from at least three independent experiments. The number of images for each assay is indicated in figure legends. The test of normality was performed using Shapiro-Wilk test. Single comparisons were tested for significance with studentЈs two-tailed unpaired t test and multiple comparisons with one-way analysis of variance using TukeyЈs post hoc test. Analysis of bitmaps of segmented objects (Ͼ100 pixels ϳ1 m 2 ) in different cell lines was performed on samples from 3-5 independent experiments as indicated in the figure legends. Statistical significance was tested with one-way analysis of variance using TukeyЈs or Games-HowellЈs post hoc test.
Immunofluorescence Staining and Microscopy-Cells were seeded at 3 ϫ 10 4 /cm on 11 mm Ø acid washed glass coverslips in 24-wells for confocal microscopy or on CELLview™ glass bottom dishes (Greiner, Bionordika, Helsinki, Finland) for TIRF microscopy and cultured for 6 days. Cells were fixed with 4% PFA in PBSϩ/ϩ (PBS with 0.5 mM MgCl 2 and 0.9 mM CaCl 2 ) for 15 min at room temperature. Immunofluorescence staining was performed as previously described (17). Confocal images were acquired with the Zeiss LSM 780 laser scanning confocal microscope using 40ϫ Plan-Apochromat objective (NA ϭ 1.4) and TIRF images were acquired with the Zeiss Cell Observer spinning disc confocal equipped with Hamamatsu camera (EMCCD) using the alpha Plan-Apochromat 63x oil objective (NA ϭ 1.46). Image acquisition software was ZEN (black edition, LSM 780; blue edition Cell Observer; Carl Zeiss Oy, Vantaa, Finland).
Image Analysis-Colocalization in TIRF images was assessed with the PearsonЈs correlation coefficient measured with the Colocalization Threshold plugin in FIJI using Costes method auto threshold determination and excluding zero intensity pixels. Segmentation of adhesions from TIRF images was performed with the Squassh plugin developed for FIJI (18).

SDS-PAGE and Western
Blotting-BirA biotinylation products were separated on 4 -20% Mini-PROTEAN® TGX™ gels (Bio-Rad Laboratories, Helsinki, Finland), and other proteins of interest on 6 -7.5% SDS-PAGE gels. Western blotting was done overnight at ϩ4°C at 20V in 20% ethanol 0.025 M Tris 0.192 M glycine onto nitrocellulose membranes (PerkinElmer, Turku, Finland). Immunolabeling and detection was performed as previously described (17). Labeling with peroxidase-conjugated streptavidin (to visualize surface biotinylated integrins or BirA biotinylation products) was done for 1 h at room temperature. BirA biotinylation products were also visualized directly by colloidal Coomassie staining (21).
Biotin Ligation and Sample Preparation for LC-MS/MS-Cells were seeded as 4.5 ϫ 10 4 /cm on a 10 cm Ø dish (Western blotting) or three 15 cm Ø dishes (LC-MS/MS) and cultured for 24 h in the presence of 50 M biotin. Cells were washed three times with cold PBSϩ/ϩ, scraped into 50 ml falcon tubes and centrifuged at 2000 rpm for 5 min at ϩ4°C. Cell pellets were snap frozen with liquid nitrogen. Cells were lysed and biotinylated proteins purified from cell lysates using Strep-Tactin Sepharose beads (IBA Lifesciences, Thermo Fischer Scientific, Helsinki, Finland) as described in (25). For LC-MS/MS samples were prepared as follows: cysteine bonds were reduced with 5 mM Tris (2-carboxyethyl)phosphine (TCEP) (Sigma-Aldrich) for 20 min at 37°C and alkylated with 10 mM iodoacetamide (Fluka, Thermo Fischer Scientific, Helsinki, Finland; Sigma-Aldrich) for 20 min at room temperature in the dark. A total of 1 g of Sequencing Grade Modified Trypsin (Promega, Madison, Wisconsin) was added and samples digested overnight at 37°C. Samples were quenched with 10% trifluoroacetic acid (TFA) and purified with C-18 Micro SpinColumns (Nest Group Inc., Thermo Fischer Scientific, Helsinki, Finland) eluting the samples to 0.1% TFA in 50% acetonitrile (ACN). Samples were dried by vacuum concentration and peptides were reconstituted in 30 l buffer A (0.1% TFA and 1% ACN in LC-MS grade water) and vortexed thoroughly.
LC-MS/MS Analysis-LC-MS/MS analysis was performed on an Orbitrap Elite ETD hybrid mass spectrometer using the Xcalibur version 2.2 SP 1.48 coupled to EASY-nLCII-system (all from Thermo Fisher Scientific) via a nanoelectrospray ion source. 6 l and 5 l of peptides were loaded from Strep-Tag and BioID-samples, respectively. Samples were separated using a two-column setup consisting of a C18 trap column (EASY-Column™ 2 cm x 100 m, 5 m, 120 Å, Thermo Fisher Scientific), followed by C18 analytical column (EASY-Column™ 10 cm x 75 m, 3 m, 120 Å, Thermo Fisher Scientific). Peptides were eluted from the analytical column with a 60min linear gradient from 5 to 35% buffer B (buffer A: 0.1% FA, 0.01% TFA in 1% acetonitrile; buffer B: 0.1% FA, 0.01% TFA in 98% acetonitrile). This was followed by 5 min 80% buffer B, 1 min 100% buffer B followed by 9 min column wash with 100% buffer B at a constant flow rate of 300 nl/min. Analysis was performed in data-dependent acquisition mode where a high resolution (60,000) FTMS full scan (m/z 300 -1700) was followed by top20 CID-MS2 scans (energy 35) in ion trap. Max-imum fill time allowed for the FTMS was 200 ms (Full AGC target 1,000,000) and 200 ms for the ion trap (MSn AGC target 50,000). Precursor ions with more than 500 ion counts were allowed for MSn. To enable the high resolution in FTMS scan preview mode was used. The MS data, thermo.raw files, spectral libraries (msf-files), and converted mgf-formats for all the above runs are available in a publicly accessible PeptideAtlas raw data repository (http://www.peptideatlas. org/PASS/PASS01198) with deposit ID:PASS01198.
LC-MS/MS Data Analysis-SEQUEST search algorithm in Proteome Discoverer TM software (Version 1.4.1.14, Thermo Fisher Scientific) was used for peak extraction and protein identification from the acquired MS2 spectral data files (Thermo.RAW) using the dog reference proteome (taxonomy Canis lupus familiaris) database (28793 entries, http://www.uniprot.org/, version 2015-09). The decoy database was the reverse of the target database. All data were reported based on 95% confidence for protein identification, as determined by the false discovery rate (FDR) Յ5%. Allowed error tolerances were 15 ppm and 0.8 Da for the precursor and fragment ions, respectively. Database searches were limited to fully tryptic peptides allowing one missed cleavage, and carbamidomethyl ϩ57,021 Da (C) of cysteine residue was set as fixed, and oxidation of methionine ϩ15,995 Da (M) as dynamic modifications. For peptide identification FDR was set to Ͻ0.05. For label-free quantification, SCs for each protein in each sample was extracted and used in relative quantification of protein abundance changes.
Protein Identification with MALDI-TOF Mass Spectrometry-MDCK cells growing on a 10 cm TC-dish were lysed with RIPA buffer and immunoprecipitated as described above using protein G Dynabeads® and ␤4-integrin antibodies. Immunoprecipitated ␤4-integrin complexes were washed three times with RIPA buffer, dissolved into SDS sample buffer and separated using SDS-PAGE and proteins were visualized by colloidal Coomassie staining (21). Three bands (Ͼ200, ϳ150 and ϳ120kD) were visible on the gel. The gel pieces were excised and de-stained by three 5 min incubations with 50 mM ammoniumbicarbonate containing 40% acetonitrile. The gel pieces were then reduced with 20 mM DTT, followed by alkylation with 40 mM iodoacetamide each for 30 min at room temperature. The gel pieces were then washed once with de-staining buffer and twice with 40 mM ammoniumbicarbonate with 9% acetonitrile, followed by an overnight trypsin-treatment (Sigma-Aldrich, proteomics grade) at 37°C. 0.5 l of the supernatant was applied to the mass spectrometers sample plate (anchor chip (800 -384), Bruker Nordic AB, Solna, Sweden) and allowed to dry, followed by addition of 0.5 l of the matrix solution (0.8 mg/ml ␣-cyano-4-hydroxycinnamic acid in 85% acetonitrile containing 0.1% trifluoroacetic acid and 1 mM NH 4 H 2 PO 4 ). Mass spectra were recorded with a UltrafleXtreme MALDI TOF/TOF mass spectrometer in automatic mode, which first measured MS spectra using external calibration (Bruker peptide calibration standard). For the MS spectra the measuring algorithm selected up to 10 ions for MS/MS interrogation. Flex analysis as part of Bruker compass 1.3 and BioTools 3.2 were used for the peaklist generation. Other Mammalia reference sequence database (2,247,961 entries, version NCBInr_2013-01) was searched using the Mascot search engine (Version 2.4.1, Matrix science) applying combined MS and MS/MS data to maximize identification confidence to identify multiple proteins. Database searches were limited to fully tryptic peptides (C-terminal to R and K, if next residue is anything but P) allowing one missed cleavage, carbamidomethyl ϩ57,021 Da (C) of cysteine residue was set as fixed and oxidation of methionine ϩ15,995 Da (M) as dynamic modification. Typical search conditions were 20 ppm mass tolerance for MS and 0.7 Da for MS/MS data, no species restriction.
Retrovirus-mediated Gene Knockdown-DGKD cells were generated by infection of MDCK cells with retroviruses encoding for shRNA constructs and selection of infected cells with puromycin as previously described (13). Two targeting sequences were selected and KD efficiency was determined by qPCR as previously described (supplemental Table S6 -S7) (13). Knockdown efficiency of DG was assessed with the ⌬⌬CT method using glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and TATA binding protein (TBP) as housekeeping genes for normalization (supplemental Table S7).
Lentivirus-mediated Gene Knockout-Gene editing with Cas9 and sgRNA expressing lentivirus (LentiCRISPR) was achieved as previously described (26). Two separate target sequences from constitutive early exons were selected for each gene (supplemental Table S8). Target sequences with no off-target sites with less than three mismatches in the Canis lupus familiaris genome were selected based on FASTA similarity search tool (EMBL-EBI, Hinxton, UK). gRNA oligos with BsmBI (New England Biolabs) overhangs were subcloned into lentiCRISPRv1 or v2 and used for lentivirus preparation. To produce lentiviruses, 70 -80% confluent 293T-D10 on CellBind® 10 cm Ø tissue culture dishes (Corning) were cotransfected with lentiCRISPR, pPAX2 and VSVg plasmids using Lipofectamine® 2000 reagent (Thermo Fisher Scientific) in Opti-MEM™ (Thermo Fisher Scientific). For infection, viral supernatant was either used directly (Lenti-CRISPRv2), or concentrated 100ϫ and then used as 1/10 dilution (LentiCRISPRv1) as described before (27). Subconfluent MDCK cells, seeded at a density of 6 ϫ 10 4 /24-well the previous day, were infected for a period of 24 h, expanded for another 24 h without virus and then trypsinized, reseeded and cultured for 24 h in the presence of 6 g/ml puromycin to select transduced cells. Clonal cell lines were established from the puromycin resistant population and analyzed by western immunoblotting and sequencing the targeted genomic DNA region. To generate double ␣6/␤4-KO and ␤1/␤4-KO cells, clonal ␤4KO cells were transduced with ␣6-targeting lentiCRISPR vector as described above, expanded and negatively selected by fluorescence activated cell sorting (FACS) based on live staining with anti-␣6 (GoH3) or anti-␤1 (AIIB2) antibodies. Antibody staining for FACS was performed as previously described (28). Alexa Fluor 488-negative cell population was collected with the BD FACSAria™ flow cytometer (BD Biosciences). For amplification of the targeted genomic locus, genomic DNA was extracted using the QuickExtract TM DNA Extraction Solution (Epicentre, Lucigen, Immuno Diagnostic Oy, Hä meenlinna, Finland) and amplified with the Phusion® High-Fidelity DNA polymerase (New England Biolabs) using exon-flanking primers, listed in (supplemental Table S9. PCR products were either directly sequenced or first subcloned into pJET2.1/blunt cloning vector using the CloneJET™ PCR Cloning Kit (Thermo Fisher Scientific). Sequencing results are listed in (supplemental Table S10).
Heatmaps-For heatmap visualizations, hierarchical clustering of specifically enriched proteins based on normalized average SC values was performed with Cluster 3.0 software using Euclidean distance as the similarity metric and single linkage as the clustering method. Normalized average SC values of the clustered genes were visualized with the Matrix2png program (35).

␣6␤4-integrins in MDCK Cells Colocalize with Laminins in
Basal Adhesion Patches That Are Distinct from Focal Adhesions-To better characterize the HDs formed in simple epithelial cells, we analyzed the adhesions formed by ␣6␤4-integrin in MDCK cells by coimmunofluorescence staining and total internal reflection fluorescence (TIRF) -microscopy. ␣6and ␤4-subunits colocalized in parallel elongated patches at the basal membrane (see yellow arrowhead in Fig. 1A). Although ␣6-integrin can also form a heterodimer with ␤1integrin, a stronger colocalization was seen with ␤4-subunit (Fig. 1A, 1B and 1H). MDCK cells express laminin-332 (LN-332) and -511 (LN-511), both of which are ligands of ␣6␤4integrin (11,36). In confluent MDCK cells, ␤4-integrin showed stronger colocalization with LN-511 (Fig. 1C, 1D and 1H). MDCK cells downregulate LN-332 synthesis in confluent conditions and produce relatively more LN-511 (36,37). Therefore, the stronger colocalization with LN-511 can be expected. To confirm that the ␣6␤4-integrin staining demarcated HDs rather than FAs, we showed that ␣6␤4-positive patches colocalized with a HD marker plectin ( Fig. 1E and 1H) but were mutually exclusive for a FA marker talin as well as actin stress fibers ( Fig. 1F-1H, yellow and white arrowheads indicate representative areas of mutual exclusion). Thus, in MDCK cells ␣6␤4-integrins form structures resembling type II HDs described in breast epithelial cells (12). Because of the lack of working antibodies against markers that would allow us to distinguish between type I and II HDs in canine cells, we will refer to them as MDCK-HDs.
Generation and Characterization of MDCK Cell Lines Expressing BirA-tagged ␤4-Integrin Constructs-To investigate the molecular composition of MDCK-HDs, we labeled integrin ␤4 proximal proteins in live cells by using proximity-dependent biotin identification (BioID) technology as described previously (23). In the presence of exogenous biotin, a humanized version of Escherichia coli -derived biotin ligase (BirA) converts biotin into highly reactive biotinoyl-5Ј-AMP that will readily react with primary amines in the immediate vicinity of the BirA. When fused to a protein of interest, BirA can be used to specifically biotinylate proteins that are proximal (within ϳ10 -30 nm) to it and its fusion partner. BioID method is particularly suitable method to detect also weaker and more transient interactions that would be difficult to preserve in traditional pull-down assays. Here, BirA was fused to either the C-(h␤4-BirA) or N terminus (BirA-h␤4) of human integrin ␤4 (h␤4), and these constructs (supplemental Fig. S1) were introduced into MDCK cells by stable transfection as confirmed by Western blotting (Fig. 2A). Cell surface expression of both fusion constructs was confirmed by surface biotinylation assay (Fig. 2B). ␤4-myc, h␤4-BirA, and to lesser extent, BirA-h␤4 fusion proteins colocalized with endogenous ␣6integrin at MDCK-HDs ( Fig. 2C and 2D). The BirA-h␤4 was detected as two bands, a full-length form (Fig 2A and 2B, blue arrowheads) and a truncated form (yellow arrowheads in Fig.  2A and 2B). It was noted, that the truncated form was particularly prominent in the surface-expressed pool of BirA-h␤4 (Fig. 2B). Myc-tagged h␤4 and h␤4-BirA, but not BirA-h␤4, coprecipitated a ϳ120kDa protein. Peptide mass fingerprinting analysis of a similarly migrating band from ␤4-integrin immunoprecipitations identified this protein as ␣6-integrin (supplemental Table S5). These data suggest that only h␤4-BirA formed a stable heterodimer with ␣6-subunit (Fig. 2B). Distinct patterns of BirA-biotinylated proteins were observed in h␤4-BirA and BirA-h␤4 expressing cells (Fig. 2E) and could be efficiently precipitated with streptavidin beads in the presence of biotin (Fig. 2F). In h␤4-BirA expressing cells, the biotinylated proteins colocalized with ␣6-integrin within MDCK-HDs thereby confirming that h␤4-BirA, with the cytoplasmic proximity ligation activity, efficiently biotinylated proteins at HDs (Fig. 2G and 2H). Some BirA-h␤4 could be detected at the basal domain (Fig. 2C), but the staining intensity was weaker (Fig. 2C). Biotinylated targets in BirA-h␤4 expressing cells were also seen at the basal surface although they displayed a more diffuse staining pattern that only partially localized to ␣6-positive patches ( Fig. 2G and 2H).
Identification of the ␤4-integrin Interactome-Biotinylated proteins were Strep-Tactin-precipitated and subjected to LC-MS/MS for identification and label-free quantification based on spectral counting. Specific enrichment of proteins in h␤4-BirA and BirA-h␤4 samples relative to the BirA-negative control was visualized by volcano blots (Fig. 4A and 4B). Total of 90 and 61 proteins were more than 3-fold enriched in h␤4-BirA and BirA-h␤4, respectively ( Fig. 4C and supplemental Table S1). For validation of the filtering efficiency (identification of high-confidence interactors) N-terminally BirA-tagged GFP (BirA-GFP) and myristoylated C-terminally BirA-tagged GFP (myr-GFP-BirA) were used as additional controls (supplemental Fig. S2D-S2F). All the pre-filtered ␤4-interactome proteins were significantly enriched also when compared with BirA-GFP-and myrGFP-BirA interactomes, suggesting that unspecific labeling is not a concern in our BioID experiment (supplemental Table S2, supplemental Fig. S2E-S2F). This was further corroborated by analyzing the ␤4-interactome against the Contaminant Repository for Affinity Purification (CRAPome) database that contains lists of most common contaminants found in negative controls of mainly affinity purification-based MS-analyses (38). Only limited overlap was  noted and the pre-filtered ␤4-interacting proteins were rarer and less abundant in CRAPome data sets compared with those removed by filtering (supplemental Fig. S2A-S2B). Finally, comparative analysis of the pre-filtered ␤4-interactome with other BioID-based interactomes of several cell-cell junctional proteins strongly suggests that the biotinylated proteins in h␤4-BirA-and BirA-h␤4-expressing cells represent a specific set of proteins that interact with ␤4-integrin (supplemental Fig. S2C) (39 -41).
Only 5 proteins were shared between h␤4-BirA and BirA-h␤4, which is consistent with the biotin ligase activities being restricted to different cellular compartments. Enriched proteins were ranked based on total spectral count (SC) values and SC values normalized to the number of available lysine residues, which is a factor, along with protein copy number and proximity that defines labeling efficiency ( Fig. 4D and 4E, supplemental Table S1). It is likely that large cytosolic proteins with multiple available biotinylation sites are overrepresented when compared with targets containing fewer available lysines, such as the short cytoplasmic tail of ␣6-integrin. Human integrin ␤4 was ranked in the top 3 in both h␤4-BirA and BirA-h␤4 samples, suggesting that proximity translates into high ranking. In h␤4-BirA samples, the other top ranking proteins, based on total SC, were large junctional scaffold proteins AHNAK, Utrophin (UTRN) and periplakin (PPL). However, when normalized SCs were used, ␣6-integrin and integral membrane proteins, kinectin 1 (KTN1) and cytoskeleton associated protein 4 (CKAP4), ranked in the top (Fig. 4D). For BirA-h␤4, the top-ranking proteins, based on both total and normalized SCs, were ER-resident proteins involved in protein folding and glycosylation (Fig. 4E). The well-known HD component collagen XVII/BP180/BPAG2 (COL17A1) was moderately ranked in h␤4-BirA interactome ( Fig. 4A and 4D) whereas two other HD components laminin chains ␣3 and ␤3 were moderately ranked (LAMA3 and LAMB3 in Fig. 4B and 4E) in BirA-h␤4 samples. Curiously, plectin was identified from all analyzed samples, including negative controls. Thus, it was not enriched in ␤4-interactome although we did show clear colocalization between plectin and ␣6-integrin (Fig. 1E,  1H).
Consistent with the subcellular localization of the biotin ligase domain, proteins enriched in h␤4-BirA samples were mostly cytosolic, whereas those enriched in BirA-h␤4, were proteins destined to the secretory pathway (Fig. 4F). Moreover, h␤4-BirA interactome was enriched with sequence features and domains associated with cytosolic proteins, whereas BirA-h␤4 proximal interactors where enriched with motifs found in secretory proteins (Fig. 4G). These results suggest that labeling is restricted to the appropriate subcellular compartment. About 8 -9% of the proximal interactors in h␤4-BirA and BirA-h␤4 belonged to the literature-curated adhesome (42), whereas, as expected, BirA-h␤4 labeled more proteins belonging to the literature-curated matrisome than h␤4-BirA (Fig. 4H) (43). Sixty percent of h␤4-BirA-and 42% of BirA-h␤4-enriched proteins belonged to the meta-adhesome, which is a collection of proteins identified from isolated FAs by mass spectrometry (Fig. 4I) (6). Two percent (h␤4-BirA) and 5% (BirA-h␤4) belonged to the consensus adhesome that represents the core components common to FAs isolated from various sources (Fig. 4I) (6).
Bioinformatic Analysis of MDCK-HD Associated ␤4-interactome-Although the core HD components have been studied in detail in keratinocytes (44 -46), the interaction landscape remains relatively unexplored in simple epithelial cells. Here we performed biotinylation of ␤4-interacting proteins in steady state conditions, and thus the ␤4-interactome is expected to contain components interacting with both maturing ␤4-complexes during their biosynthetic transport and with the mature ␣6␤4-integrin at the cell surface and in endocytic compartments. The enrichment of GO terms within the interactome was visualized using REVIGO ( Fig. 5A and 5B) (30). h␤4-BirA-labeled proteins fell under terms such as polymeric cytoskeletal fiber, cell-cell junction, coated vesicle membrane and perinuclear region of cytoplasm (Fig. 5A). ER lumen was the most overrepresented term among BirA-h␤4-enriched proteins, but terms for secreted components such as extracellular matrix and cell surface, as well as focal adhesion, also emerged (Fig. 5B). Thus, the analysis of subcellular localization of proximity-labeled proteins suggests that both h␤4-BirA A) and BirA-h␤4 (red asterisks in B) was confirmed by Western blotting using integrin ␤4antibodies. Integrin ␣6 expression levels were also determined and ␤-tubulin blotting was used as a loading control. C, Colocalization of endogenous ␣6-integrin (red) with exogenous myc-tagged h␤4-BirA (green, upper panels) or BirA-h␤4 (green, lower panels) was analyzed by TIRF microscopy. Scale bars ϭ 10 m.

FIG. 4. Interactomes of C-and N-terminally tagged integrin-␤4 identified and quantified by LC-MS/MS. Volcano blots of h␤4-BirA (A)
and BirA-h␤4 (B) labeled proteins whose SC is Ͼ 4 in the respective samples. C, Venn diagrams of all the proteins identified in h␤4-BirA, BirA-h␤4 and negative control samples (Identified proteins), and of proteins that were specifically enriched in h␤4-BirA (A) and BirA-h␤4 (B) samples (FC Ͼ 3, p value Ͻ 0.05 (two-tailed unpaired t test) indicated by dotted lines). D-E, Enriched proteins were ranked by total SC or by counts per the number of lysines available for biotinylation for both h␤4-BirA (D) or BirA-h␤4 (E). Only cytosolic or luminal sequences of the longest canine protein product (UniProt) were used for the counting of lysines in h␤4-BirA and BirA-h␤4 -enriched proteins, respectively. If membrane topology data was not available, domains were estimated based on alignment to the corresponding human sequence with known domain information. F, Distribution of h␤4-BirA and BirA-h␤4 biotinylated proteins between cytosolic compartment and secretory pathway based on UniProt entry information. G, Enrichment of non-redundant sequence features and protein domains between h␤4-BirA and BirA-h␤4 biotinylated proteins analyzed by DAVID. H, Occurrence of literature-curated adhesome (42) and matrisome components (63) and I, components identified from purified adhesions by mass spectrometry (6) within the interactomes of integrin ␤4. and BirA-h␤4 mature, although BirA-h␤4 matures with only low efficiency.
To facilitate protein-protein interaction analysis of potential HD-associated proteins, we classified proteins into groups based on their annotated subcellular localizations (Fig. 5C). The HD localized proteins were manually curated based on the literature and their UniProt entries and then assigned into 6 different groups according to their subcellular compartment ( Fig. 5E and (supplemental Table S3). Proteins that localized to the nucleus, ER, Golgi or mitochondria were classified into a biosynthesis group and proteins, for which there was no information available concerning their subcellular localization, were put into a group named unknown (Fig. 5E). In agreement with the observed defective maturation and HD targeting of BirA-h␤4, SC-based analysis revealed that more than 60% of proteins in the BirA-h␤4 complex belonged to the biosynthesis-related group whereas this group constituted less than 30% of total SCs in h␤4-BirA interactome (Fig. 5E). Members of the biosynthesis group were involved in folding, ER modifications, translation, post-translational modifications and vesicular transport (Fig. 5G and supplemental Table S3).  The second largest and abundant BirA-h␤4-associated group after the biosynthesis group was the extracellular protein group that included several laminin chains that were also shown to colocalize with ␣6␤4-integrin (Fig. 1C, 1D, 1H and 5E). In h␤4-BirA samples, plasma membrane and cell junction groups containing the highest-ranking proteins were most prominent in total SCs (Fig. 5E). Based on UniProt keywords, the junctional proteins found in putative MDCK-HDs were designated as components of FAs, HDs, desmosomes (DS), tight junctions (TJ) and adherens junctions (AJ) (Fig. 5F). This is not unexpected as proteins can be shared between different types of junctions and adhesions. However, whether these proteins truly localize to HDs, needs to be validated case-bycase. Components mediating cell-ECM interactions, especially those involved in laminin-adhesion, are likely to be very closely positioned to HDs. The cytoskeletal proteins included several actin and microtubule-interacting proteins with regulatory or structural functions. Most of the keratins that were identified by mass spectrometry mapped to the human protein sequence and were therefore not considered specific.
To focus the PPI network analysis on proteins that localize to MDCK-HDs we decided to exclude the biosynthesis group from the analysis. The remaining putative HD-associated proteins were mapped to the human interactome and repre- FIG. 7. Basal targeting of ERC1 is independent of integrin ␣6␤4 expression. A-B, TIRF images showing coimmunostaining of integrins ␤4 (A) and ␣6 (B) (red) with ERC1 (green). Colocalized pixels are shown as bitmaps (yellow). C, Colocalization of ␣6 and ␤4 integrins with ERC1 measured by Pearson's correlation coefficient (n ϭ 12-15 images from three experiments). D, TIRF images and corresponding bitmaps of segmented ERC1 objects (Ͼ100 pixels) in control, ␣6KO and ␤4KO cells. E, Quantification of ERC1 objects (n ϭ 15-18 images from three experiments). F-G, TIRF images showing coimmunostaining of integrin ␤4 (F) and vinculin (VCL, G, (green) with ERC1 (red). Colocalized pixels are shown as bitmaps (yellow). H, Colocalization of ␤4 integrin and VCL with ERC1 measured by Pearson's correlation coefficient (n ϭ 10 images from two experiments). Statistical significance was tested with one-way analysis of variance using TukeyЈs or Games-HowellЈs (ERC1 in E) post hoc test (not significant). Scale bars ϭ 10 m.
sented as protein-protein interaction networks (Fig. 5D). h␤4-BirA was better represented overall, but BirA-h␤4 contributed to the extracellular region and cell junction parts of the network. Proteins involved in the formation of laminin-based adhesions and known to interact with the intermediate filament network formed the most notable subnetwork that also included ␣6␤4-integrin (Fig. 5D). Chaperonin Containing TCP1 (CCT) complex that assist folding and stabilization of diverse group of newly-made cytosolic proteins was also linked with this network (Fig. 5D) (47). Interestingly, three smaller nodes were found that represented actin-binding proteins and one node with proteins that associate with dystroglycan (DSG)-mediated laminin adhesions that in turn are known to connect to the actin cytoskeleton. More than one third of the proteins were not incorporated into any PPInetwork in the analysis (Fig. 5D). Profiling of the MDCK-HD candidate proteins based on their functions highlighted the presence of membrane-and cytoskeleton-associated adaptors and regulators, receptors and ECM proteins (Fig. 5G and supplemental Table S3).
Similar to UTRN, ERC1 was found to colocalize with both integrin ␤4 and ␣6 in mature HDs in confluent MDCK cells (Fig. 7A-7C). ERC1, together with two other novel ␤4-interactome candidates, PHLDB1 and PHLDB2, participate in a protein complex containing KANK2 that has been show to activate ␤1-integrins but to reduce force transmission across FAs (51,52). Interestingly, deleting integrin ␣6 or ␤4 expression had no significant effect on the basal localization of ERC1 ( Fig. 7D and 7E). In subconfluent cells, ERC1 displayed partial colocalization with both HD-forming integrin ␤4 and an FA component vinculin (VCL, Fig. 7F-7H). Therefore, ERC1 ap- pears to be recruited to the membrane independently of ␣6␤4 integrins and could, in principle, be necessary for the recruitment of integrin-␣6␤4 to nascent laminin adhesions prior to HD assembly.
UTRN and ERC1 Are Both Dispensable for HD Formation in MDCK Cells-To study the possible role of ERC1 and UTRN in biogenesis of MDCK-HDs we knocked out their expression by using the CRISPR/Cas9 technology. Efficient depletion of UTRN was demonstrated by Western blotting (Fig. 8C). Significant down-regulation of ERC1 protein was also evident in all five different ERC1-KO MDCK cell lines (Fig. 8D). UTRN-KO and ERC1-KO MDCK cells did not display any significant defect in their ability to assemble HDs as judged by formation of integrin-␣6 and -␤4 positive foci (Fig. 8A, 8B, and 8E). These results indicate that neither UTRN nor ERC1 is essential for the HD-assembly.
Proximal Interactions of Integrin ␤4 Are Independent of the Formation of ␣6␤4-Heterodimer-Formation of an ␣␤-heterodimer is considered a prerequisite for the endoplasmic reticulum (ER)-exit and subsequent surface delivery of integrins (53). In order to study how ␣6␤4-heterodimerization affects the composition of ␤4-interactome, we generated MDCK cells expressing either h␤4-BirA or BirA-h␤4 fusion protein but lacking expression of endogenous ␤4and ␣6integrins (␣6/␤4-dKO) ( Fig. 9A and 9B). In principle, lack of ␣6-subunit should block maturation of both h␤4-BirA and BirA-h␤4. Biotinylated samples were collected from h␤4-BirA and BirA-h␤4 cells in the presence (control), or absence (␣6/ ␤4-dKO) of ␣6-integrin. Samples from cells without BirAexpression were used as a negative control. It is expected that depletion of integrin-␣6 leads to ER retention of both ␤4-BirA and BirA-␤4 fusion constructs, thereby preventing the association of proteins interacting with the mature integrin ␤4 and instead preferentially revealing interactions associated with the biosynthetic trafficking and folding of integrin ␤4. Increased SC were indeed observed for a few ER-resident ␤4-interacting proteins, such as Protein Disulfide Isomerase A5 (PDIA5), PDIA6, Signal Recognition Particle Receptor (SRPR) and FK506 Binding Protein 10 (FKBP10) peptidylprolyl cis/trans isomerase in the ␤4-interactome of ␣6KO cells when compared with controls (supplemental Table S4). To our surprise, however, only four proteins (including the depleted ␣6-integrin) displayed significantly reduced interaction with integrin-␤4 in ␣6KO cells (Fig. 9C and 9D). Moreover, the SC of the labeled proteins between ␣6KO and control samples suggested no major change in the abundance of the vast majority of ␤4-interacting proteins ( Fig. 9C and 9E). This suggests that the proximal interactions of ␤4-integrin remain largely unchanged despite the absence of ␣6-integrin.
Several ␤4-interacting proteins involved in the formation and regulation of actin-linked FAs displayed increased SCs in ␣6KO cells. When the ␤4-interacting proteins with significantly differing SCs in ␣6KO cells were mapped to the human interactome and represented as protein-protein interaction networks, two nodes were obtained, one centering on ␣V␤3integrin and another representing an actin-linked junctional module (Fig. 9F). In addition, integrin-␣3, ERC1, KANK2 and KTN1, all of which have been implicated in FA regulation, were among the proteins preferentially interacting with integrin-␤4 in the absence of ␣6-subunit (Fig. 9E, supplemental Table S4). DISCUSSION Recent advancement in mass spectrometry analyses has enabled comprehensive proteomic characterization of various protein complexes, including cellular adhesions. The HD composition has been previously interrogated only by using traditional methods (11). Here we used BirA-based proximity biotinylation (BioID) technology to analyze the interactome of ␤4-integrin, a core component of laminin-adhering HDs. The BioID approach does not rely upon preservation of the protein-protein interactions throughout the adhesion purification procedure. This is particularly advantageous for the study of cell adhesion complexes that are notoriously difficult to preserve (7). It was noted, however, that fusing the BirA domain to the N terminus of ␤4-integrin affected its delivery to HD-like basal patches. Although reduced basal targeting likely limits efficient labeling of ␤4-interacting proteins at HDs, the BirA-␤4-integrin construct revealed proteins that interact with ␤4integrin in the ER and may be critical for its biosynthetic trafficking. Less than 5% of the proteins identified as potential ␤4-integrin interacting components of HDs are shared with the previously reported core consensus adhesome (6). The core consensus adhesome has been established by combining results from selected proteomic analyses of FAs isolated using mechanical and biochemical techniques (7). The limited overlap seen between the core consensus adhesome and ␤4-BioID-interactome is not surprising given that FAs and HDs are distinct complexes. Moreover, as the source cell type and the adhesion purification methods are different, drawing direct comparisons is problematic. Indeed, when Dong and colleagues employed the BioID-technology to characterize the interactomes of kindlin-2 and paxillin, two key compo- nents of FAs, they found that only 22% of their interactomes were included in the core consensus adhesome (54). Curiously, only one third of the proteins labeled in paxillin-and kindlin-2-BirA fusion protein expressing cells were common to both constructs suggesting that, when properly optimized, BioID labeling is strictly limited to proteins in the immediate vicinity of the BirA domain (54).
In our BioID-based ␤4-integrin interactome, we did identify some components previously associated with FAs, which may also mediate laminin interactions. Among the FA components were two ␤1-integrin binding proteins tensin-3 and a recently described talin activating protein KANK2 whose interaction with integrin-␤4 was enhanced in ␣6KO cells (52,55,56). The ␣6␤4-integrin staining did not overlap with the most intensive talin staining (Fig. 1F). However, the strongly talinpositive FAs we observed linked with stress fibers may not represent laminin-binding adhesions at all. Indeed, we observed partial colocalization of ␤1-integrins with ␣6-integrins in MDCK-HDs that were mutually exclusive for FA markers (Fig. 1B, 1F, 1G). Despite limited overlap, FA and HD patterns flanked each other tightly. Thus, it is expected that some molecular connections exist between these two structures and in the absence of functional HDs, such as in ␣6KO cells, remaining HD components may be relocated to FAs (Fig. 9F).
We also investigated a couple of the highly ranked hits, ERC1 and UTRN, for their potential roles in laminin adhesions in more detail. ERC1 was recently reported to regulate FA turnover in a complex with Liprin-␣1 and LL5 (57). The LL5 complex in turn associates with integrin-mediated laminin adhesions in mammary epithelial cells, where it supposedly plays a role in the capture microtubules (58). Both LL5 isoforms, LL5␣ (PHLDB1) and LL5␤ (PHLDB2) were also identified in our ␤4-interactome. UTRN on the other hand is a homolog of dystrophin and is thought to link DG-mediated laminin adhesions to the actin cytoskeleton in non-muscle cells (48). We verified the localization of both ERC1 and UTRN at MDCK-HDs, but neither was an essential HD component as MDCK-HDs did form in UTRNKO and ERC1KO cells. It is thus more likely, that these proteins are accessory components rather than essential structural components. Our data show that in polarized cells ERC1 preferentially associates with HDs but in subconfluent cells it appears to be linked to both FAs and HDs. A potential role for ERC1 in orchestrating coordinated assembly of FAs and HDs is an interesting topic for further studies. The core structural components, at least in type I HDs, have been listed and their interactions precisely defined (11). From these core components, our screen picked up collagen XVII, laminin chains ␣3 and ␤3 and integrin ␣6. The absence of the other components may be related to the structural differences seen between type I and II HDs and thus further investigation is necessary.
The HD-like laminin patches we observed in MDCK cells were not only dependent on ␣6␤4-integrins, but also on DG expression. Moreover, the role of ␤1-integrins in laminin as-sembly in MDCK cells has been previously suggested (3). Therefore, it is possible that these patches that we introduce as MDCK-HDs have a more integrated nature, containing several closely-associated laminin-binding receptor complexes. This could be one possible explanation as to why we identified components known to reside in both FAs and the DG complex in our integrin ␤4 interactome. Indeed, we confirmed that the recruitment of UTRN to laminin adhesions was dependent on both ␣6␤4-integrins and DG. Clearly, further scrutiny and comparative analyses of the coexpressed laminin-binding receptor complexes is needed to resolve their individual contributions to laminin adhesion and signaling.
The most surprising finding of our study was that deletion of ␣6-integrin expression did not significantly affect the assembly of ␤4-integrin interactome. It is likely that ␤4-integrins alone are not able to form robust laminin adhesions. Indeed, efficient assembly of HD-associated laminin patches in the ECM was dependent on ␣6␤4-integrin heterodimer expression, which agrees with mouse studies demonstrating that ␣6-deletion leads to loss of functional HDs and results in lethal fragility in epithelial tissues (59). However, ␤4-integrin, that does not form a heterodimer with ␣6-integrin, could still interact with multiple cytoplasmic effectors and associate with the intermediate filament network (60). Previous data have shown that integrin ␤4 is expressed in excess relative to ␣6-integrin (61). The potential relevance of ␤4-integrin complexes that do not contain ␣6-subunit merits further investigation. Interestingly, mutations in the integrin ␤4 gene contribute to epidermolysis bullosa, a genetic skin blistering disease caused by defective HDs, much more frequently than those in the integrin ␣6 gene (62).
In conclusion, we provide here the first comprehensive characterization of ␤4-integrin interactome in simple epithelium. We demonstrate that ␤4-integrin can associate with most of its proximal interactors, such as UTRN and ERC1, two novel ␤4-associated proteins, independently of ␣6-integrin expression. This interactome serves as valuable resource and as such provides interesting insight into the molecular characteristics of the assembly of HDs in simple epithelia.
Acknowledgments-We thank Riitta Jokela for overall expert technical assistance, Jaana Trä skelin for expert technical assistance at Biocenter Oulu Virus Core Laboratory, Dr. Veli-Pekka Ronkainen for expert assistance in microscopy at Biocenter Oulu Tissue Imaging Center and Dr. Ulrich Bergmann for expert assistance in MALDI/TOF analysis at Biocenter Oulu Mass spectrometry Core Laboratory.

DATA AVAILABILITY
The MS data, thermo.raw files, spectral libraries (msf-files), and converted mgf-formats for all the above runs are available in a publicly accessible PeptideAtlas raw data repository (http:// www.peptideatlas.org/PASS/PASS01198) with deposite ID: PASS01198.