Cell lysate preparation

Stable isotope labeling with amino acids in cell culture (SILAC)-adapted cell lines previously established from mammary gland epithelial cells from ANXA1-heterozygous and ANXA1-deficient mice were used in this study [21]. The SILAC-adapted “heavy-labeled” and “light-labeled” cells were lysed with 4% SDS (w/v), 100 mM Tris/HCl pH 7.6 and 100 mM DTT at room temperature. Protein was quantified using the RCDC assay kit (Bio-Rad) prior to equal mixing of identically concentrated light and heavy SILAC-labeled ANXA1^+/- and ANXA1^-/- lysates.

Reduction, alkylation and digestion

Filter-aided sample preparation (FASP) was performed as described with slight modifications [23]. In short, the sample was mixed with 5 mL UA buffer (8 M urea in 0.1 M Tris/HCl pH 8.5) and poly(ethylene) glycol (PEG) (with a final PEG concentration of 0.5%) in a 15-mL 30-kDa Amicon filter (Millipore) followed by centrifugation at 3850 × g for 45 minutes at room temperature. The retentate was rinsed with 4 mL of UA and followed by 2 mL of UA: 2 mL of 50 mM iodacetamide in UA was then added followed by incubation for 20 minutes in the dark and then by centrifugation to remove the iodoacetamide. Samples were washed one more time with 2 mL of UA, followed by three washes with 2 mL 40 mM ammonium bicarbonate. Sequencing-grade trypsin (Promega) in 40 mM ammonium bicarbonate was added. The trypsin to protein ratio was 1:100 (w/w). After overnight digestion at 37 °C the peptide solution was collected by centrifugation. The amount of peptide obtained was measured using Nanodrop (Thermo Scientific).

Phosphopeptide enrichment

Phosphopeptides were enriched from the tryptic peptides as described with slight modifications [23]. Briefly, around 6 mg of FASP-digested tryptic peptides were diluted four times with 1 M glycolic acid/80% acetonitrile (v/v)/2% trifluoroacetic acid (TFA) (v/v) and added to 13 mg of Titansphere TiO₂ 10 μm (GL Sciences, Inc., Japan). After incubating the mixture for 10 minutes at room temperature, the sample was centrifuged at 4000 g for 3 s. The supernatant was collected and the above process was repeated with another portion of the beads. The beads were washed four times with 1 M glycolic acid/80% acetonitrile (v/v)/2% TFA (v/v) solution and four times with 80% acetonitrile (v/v)/0.1% TFA (v/v) solution after transferring to a 200-μL pipet tip plugged with two layers of C8 empore disks (3 M Empore). Finally, the phosphopeptides were eluted using 500 uL 40% acetonitrile (v/v) containing 15% NH4OH (m/v) followed by vacuum-concentration to ∼ 40 μL.

Anion exchange fractionation of phosphopeptides

Phosphopeptides were fractionated as described, with slight modifications [24]. Briefly, phosphopeptides were diluted five times in Britton & Robinson buffer composed of 20 mM CH₃COOH, 20 mM H₃PO₄, 20 mM H₃BO₃ and NaOH, pH 11. The sample was then fractionated on a micropipette tip with six layers of 3 M Empore Anion Exchange disks (Varian, 1214-5012) stacked into it. Peptides were loaded at pH 11 and fractions were subsequently eluted with buffer solutions of pH 8, 6, 5, 4 and 3, respectively into their respective C18 stage tips.

Mass spectrometry (MS)

Full-scan MS spectra corresponding to m/z 300–1400 were acquired using Orbitrap or Orbitrap XL (Thermo Fisher) for all samples with a resolution of r = 60,000 at m/z 400, an AGC target of 1e6, and a maximum injection time of 500 ms. For each scan, the top ten most intense peptide ions with intensity threshold >2000 and charge state ≥2 were isolated sequentially to a target value of 1e4, and the precursor ions were fragmented by collision-induced dissociation in a linear ion trap using normalized collision energy of 35%. Dynamic exclusion was applied for 30 s using a maximum exclusion list of 500.

Identification and quantification of peptides and proteins

Data were processed using MaxQuant (Version 1.3.0.5) [25, 26] against uniprot 2013-01 mouse database containing 262 commonly observed contaminants. Raw mass spectrometry data corresponding to our previously published proteome profiling carried out under a similar experimental set up was also included for analysis. Database searches were performed allowing a maximum of two missed cleavages upon trypsin digestion, two labeled amino acids, and an initial mass tolerance of 6 ppm for precursor ions and 0.5 Da for fragment ions. Carbamidomethylation of cysteine was searched as a fixed modification, and N-acetylation, oxidized methionine and phosphorylated serine/threonine/tyrosine were searched as variable modifications. Peptide and protein quantification of SILAC pairs was performed using default settings at a maximum false discovery rate (FDR) of 0.01. Only those proteins supported by at least one unique peptide with a minimum length of six amino acids were considered to be identified.

Differential expression analysis

For differential expression analysis only those phosphosites that can be localized with a localization probability >0.75 were considered. The SILAC ratios of the corresponding peptides were log-transformed, and each replicate was median-centered and scaled for median absolute deviation (MAD). Only those phosphopeptides greater than the robust z score threshold of 2 or less than -2 (approximately equating to at least 1.8-fold change in this dataset) reliably in three out of four experiments were considered to be regulated. Functional enrichment for the regulated phosphoproteins was obtained as Gene Ontology (GO) terms corresponding to biological process, cellular component, and molecular function using the Database for Annotation, Visualization and Integrated Discovery (DAVID) functional annotation tool [27]. Only those categories containing at least three genes were considered for analysis. The GO molecular function categories that were significantly enriched with a p value <0.01 in either of the increased or decreased phosphorylation groups were visualized as a heatmap.

Kinase mapping and enrichment

The upstream kinases of the regulated phosphosites were obtained by mapping the site-specific information against a comprehensive knowledge base of experimentally derived kinase-substrate relationships (KSR) available from KSR-LIVE [28]. The integrated database combines KSRs from multiple repositories such as PhosphoSitePlus, PhosphoELM, PhosphoPOINT and the Human Protein Reference Database (HPRD). The derived upstream kinases along with the regulated phosphoproteins were analyzed using the search tool for the retrieval of interacting proteins (STRING) to explore their functional connectivity [29]. Next, the network was filtered to retain only the high-confidence interactions. This filtered network was finally combined with kinase mapping information to build the ANXA1-specific signaling network. The network was rendered in Cytoscape [30].

Upstream kinase prediction was also carried out using the NetworKIN (version 3.0) algorithm that uses kinase consensus motif along with contextual modeling for predicting KSRs [31]. The mouse phosphopeptides were mapped to their corresponding human homologs and the respective upstream kinases were derived for all regulated phosphosites. Then kinase activity estimation was performed based on average abundance of the regulated phosphopeptide in the predicted kinase substrate set [32]. An activity score was computed for each kinase and the statistical significance of the enrichment was tested using the z statistic. The z score was converted to a p value and was FDR-adjusted in R environment.

Comparison of proteome and phosphoproteome datasets

From the proteome data, those proteins that were 2.5 MAD away from the median (roughly equating to twofold change) in three out of four experiments were considered as regulated. For comparative analysis, proteins that displayed change in abundance at both phosphoproteome and proteome level, and those that displayed changes only in the phosphoproteome level were grouped into four distinct clusters. Functional enrichment for the cellular component was carried out for each of the clusters and the significantly enriched clusters (p value <0.05) in at least one of the groups were retained for hierarchical clustering and visualization as a heatmap. To perform gene-set enrichment analysis a pre-ranked list based on average fold change derived (ANXA1^-/-/ANXA1^+/- ratio) from both proteome and phosphoproteome data was created. For genes with multiple phosphopeptides, fold changes corresponding to the most regulated phosphopeptides (highest or lowest fold change) were used for ranking. For the others that did not meet the previously set robust z score cutoff, those phosphopeptides with the highest ratio were used. Gene-set enrichment analysis (GSEA) was performed using all curated mouse GO biological processes and pathways individually for both the proteome and phosphoproteome data [33]. Statistical significance of the enrichment was assessed by permutation analysis, and mutually overlapping gene-set clusters with a p value <0.01 were visualized as a network using the Enrichment Map plugin as implemented in Cytoscape [34].

Network analysis

Expression-driven transcription factor prediction analysis was carried out for all differentially regulated proteins in the proteome dataset using the Transcription Factor database (TRANSFAC) [35]. The analysis was performed against a background that included predictions for all proteins in our dataset and the significance of the enrichment was calculated using the one-sided Fisher’s exact test. The predicted transcription factors that were significantly enriched (p value <0.05) were visualized using the Treemap package from R. For construction of the interaction network we used protein-protein interactions curated in the InWeb_InBioMap database that has 2.8 times more interactions than other comparable resources and is also functionally superior [36]. The identified phosphoproteins along with the predicted transcription factors were mapped onto this network, and later filtered to include only edges connecting altered phosphoproteins or significant transcription factors or both. The phospho-network contained a total of 282 nodes and 779 edges. The network components were arranged into modules based on their functional categorization (such as GTPase or kinase), pathways or cellular localizations for better visualization. For automated cluster enrichment, the GLay community clustering algorithm was implemented and the clusters were extracted for visualization [37].

Quantitative phosphoproteomics of ANXA1-deficient murine mammary gland cells

Our previous quantitative profiling of protein abundances in ANXA1^+/− and ANXA1^−/− murine mammary gland epithelial cells implicated a role for ANXA1 in breast cancer initiation and progression [21]. To further elucidate ANXA1-dependent signaling events, we performed SILAC-based quantitative phosphoproteome analysis of ANXA1^+/− and ANXA1^−/− murine mammary cells as depicted in Fig.1a and described in “Methods”. The collective analysis from biological duplicates of each forward (heavy ANXA1^-/- cells versus light ANXA1^+/- cells) and reverse experiments (heavy ANXA1^+/- cells versus light ANXA1^-/- cells) identified a total of 11,235 unique phosphopeptides carrying 8110 unique phosphorylation sites on 2626 proteins (Additional file 1: Table S1 and Additional file 2: Figure S1A). Of these, a majority (83%) was among the class I phosphorylation sites with high localization probability (≥0.75), and only 6465 sites (~80%) overlapped with previously annotated phosphorylation sites in the PhosphoSitePlus database (Additional file 2: Figure S1B). The total phosphosites identified in the individual biological replicates was comparable, and the replicates demonstrated good reproducibility and quantification accuracy as is evident from their correlation (Fig. 1b and Additional file 2: Figure S1C). While a majority of the phosphopeptides was singly phosphorylated (87%), about 11% displayed two phosphorylation sites per peptide and the rest harbored at least three sites.

Using a robust z score cutoff of 2 (median absolute deviation 2 above or below the median ratio roughly equating to 1.8-fold change) we identified 362 phosphosites on 224 proteins with increased phosphorylation and 220 sites on 153 proteins with decreased phosphorylation in three out of four experiments (Fig. 1c, Additional file 3: Figure S2 and Additional file 4: Table S2). Apart from these altered sites, 189 of these phosphoproteins also harbored additional sites that displayed no change in phosphorylation abundance, suggesting that these site-specific changes plausibly relay ANXA1-dependent signals in breast epithelial cells. Most of the regulated sites belonged to serine/threonine phosphorylation with only two tyrosine sites being regulated in Tensin-like C1 domain-containing phosphatase (TENC1) and cyclin-dependent kinase 1 (CDK1) (Additional file 3: Figure S2B). A few proteins including supervillin (SVIL), BAT2 domain-containing protein 1 (PRRC2C), MAP7 domain-containing 1 (MAP7D1) and leucine rich repeats and calponin homology domain containing 3 (LRCH3), however, displayed differential phosphorylation with both increased and decreased phosphorylation site abundance. These results clearly indicate that ANXA1 is extensively involved in protein phosphorylation signaling in breast epithelial cells.

Comparison of this phosphoproteome dataset to our previously reported proteome profiling (Additional file 5: Figure S3 and Additional file 6: Table S3) revealed that though 77% of the identified phosphorylation sites had a corresponding protein measure, a majority of the changes arose from site-specific differential phosphorylation and not due to changes in protein abundance. Interestingly, one of the proteins belonging to the antioxidant family, peroxiredoxin-6 (PRDX6), showed opposite regulation with elevated protein abundance and reduced phosphorylation at Thr44 spanning the catalytic center, suggesting possible enzymatic activity modulation by ANXA1.

Functional portrait of ANXA1-regulated phosphoproteome

While the function of ANXA1 in mediating anti-inflammatory response is well-documented, a wide spectrum of cellular mechanisms regulated by ANXA1 has only been realized in recent years. In our study, we uncovered site-specific changes on several proteins including kinases and receptors reflecting the impact of ANXA1 on signaling pathways in mammary epithelial cells. Several mitogen-activated protein kinase (MAPK)-associated kinases such as MKNK2, MAP3K8 and RPS6KA3 and microtubule regulatory kinases such as DCLK1, MAST2 and MARK3 are among those with induced phosphorylation. This is interesting as some of these kinases have a profound influence on cell migration activities of pathways including WNT and NF-κB signaling. Enhanced phosphorylation of serine/threonine kinase receptor TGFBR2 on multiple serine sites at Ser573, Ser576 and Ser578 was also observed in ANXA1^-/- cells. This agrees with previous reports of ANXA1 regulating TGFβ signaling possibly via modulating receptor kinase activity [19]. Conversely, kinases involved in inflammatory response such as EIF2AK2 and RIPK1 displayed reduced phosphorylation. Decreased phosphorylations were also observed on kinases involved in cell adhesion and migration. For instance, the non-receptor tyrosine kinase ABL2 associated with actin filament organization and cytoskeletal remodeling displayed reduced phosphorylation on two serine sites flanking the F-actin binding region [38, 39]. For the receptor tyrosine kinase EPHA2, multiple sites displayed decreased phosphorylations including that on the functional site Ser898 that is implicated in promoting cell migration [40]. Among all the sites modulated in the absence of ANXA1, only a few are known regulatory sites (Additional file 7: Table S4), and the functional significance of the remaining are yet to be explored.

Next, we looked into ontology-based enrichment for those modulated phosphoproteins (Fig. 2a). It became evident that a majority of the upregulated phosphoproteins were involved in cell-cell adhesion and cytoskeletal organization, particularly actin-filament-based processes. Accordingly, those components that govern cell shape and movement such as actomyosin structure, focal adhesions and stress fibers were selectively enriched (Additional file 8: Table S5). This is consistent with reports linking ANXA1 and cell migration, and underpins the role of ANXA1 in adhesion complexes. We also observed site-specific upregulation on proteins associated with the Rho family of GTPases that work as molecular switches in cell organization and movement. Specifically, several GTPase-activating protein (GAP) such as ARHGAP21, ARHGAP28 and ARHGAP29 that function as RhoA-GAP, and Rac-gunanine nucleotide exchange factor (GEF) DOCK7 displayed increase in phosphosite abundance. Our data also revealed an interesting link between ANXA1 and the WNT signaling pathway with several members including receptor LRP6, crosslinking factor MACF1 and antagonist dickkopf-3 (DKK3) showing increased phosphorylation in ANXA1-deficient mammary epithelial cells. LRP6 constitutes the WNT-ligand-induced co-receptor complex and confers active WNT signaling. We observed a reproducible increase in phosphorylation of two sites, Ser1490 and Thr1493, which is crucial for receptor activation and signal transduction [41], establishing that ANXA1 level may directly correlate with WNT pathway output. In concordance with these functions, the proteins with elevated site abundance also showed enhanced cadherin and β-catenin binding in comparison with those with reduced phosphorylation (Fig. 2b).

Proteins with downregulated sites in ANXA1-deficient mammary epithelial cells displayed preferential enrichment for microtubule-based processes and ATP-dependent chromatin remodeling. A few phosphoproteins with reduced site abundance were also enriched for actin cytoskeletal processes and localized to focal adhesions indicative of a complex network of phosphorylation-mediated events that establish cell movement. While the role of ANXA1 in regulating the actin cytoskeleton has been reported earlier, our dataset exposes its possible novel associations in influencing microtubule-filament-based processes. We uncovered reduced phosphosite abundance on microtubule motor proteins including kinesin and dynein, and on other microtubule-associated proteins such as MAP2 and CLIP2. It is worthy to note that one of the key proteins that influence microtubule stability, stathmin (STMN1) was found with increased phosphosite abundance at the functional site Ser16 that inhibits microtubule depolymerization in ANXA1-deficient cells, highlighting the pervasive role of ANXA1 in regulating microtubule dynamics [42].

Next, we asked how the functions of those proteins that changed in total or site abundance compared with each other in ANXA1-deficient mammary cells (Additional file 3: Figure S2A, Additional file 9: Figure S4 and Additional file 10: Table S6). For this purpose, we first analyzed the cell component distribution of protein subsets that were regulated at both the proteome and phosphoproteome level or only at the phosphoproteome level by performing combined hierarchical clustering (Additional file 11: Table S7). We observed distinct clusters specific to each subset (Fig. 3a). For example, a subset of proteins that localized to the exterior plasma membrane and ruffle structures displayed elevated protein and phosphosite abundance. The proteins that exclusively displayed elevated phosphorylation with no change at protein level traversed a wide range of cellular components, particularly spanning all the junctional and focal adhesion components. Phosphoproteins were specifically enriched at the leading edge membrane that constitutes the lamellipodium, stress fiber, actomyosin and contractile fibers, revealing extensive phosphosignaling and rewiring of the cell adhesion machinery upon loss of ANXA1. Apart from these, proteins that localized to endosome membrane displayed only phosphosite upregulation. On the other hand, proteins that constitute the microtubule cytoskeleton, spindle and kinesin complex displayed only reduced phosphosite abundance with no protein level changes. Only a small subset of proteins showed decrease in both phosphorylation and corresponding protein level, and these localized to chromatin remodeling components.

In addition, by performing GSEA of canonical biological pathways and processes [33] individually for both the ranked proteome and phosphosite abundance, we showed distinct functions fine-tuned by the protein subsets (Fig. 3b). From the enrichment map it is evident that processes representing immune response, nucleotide metabolic process and cell cycle regulation are majorly represented by downregulated proteins. Interestingly, the proteins associated with cell migration, integrin-mediated adhesion, extracellular matrix (matrisome), stress and estrogen response are enriched in the subset with upregulation of both protein and phosphosite abundance. Proteins involved in calcium ion homeostasis also displayed similar trends underscoring the crucial role of ANXA1 in regulating calcium dynamics and associated signaling. Again, this analysis also highlighted that stress fiber and actin filament assembly along with exocytosis events are exclusively driven by proteins with increased phosphorylation.

Altogether, the current dataset suggests that ANXA1 is crucially involved in reshaping the cytoskeleton and modulates cell adhesion or migratory properties by fine-tuning both protein and phosphorylation levels that work in concert with each other.

Kinases regulated by ANXA1 in mammary epithelial cells

We next mapped site-specific information of the phosphoproteins against a comprehensive knowledge base of experimentally derived kinase-substrate relationships (KSR). Of all the regulated sites, upstream kinase was experimentally curated only for 56 phosphosites on 45 proteins. The interconnectivity of the kinome network (Fig. 4a) was obtained through mapping of protein-protein interactions. With many substrates of MAPK1 showing increased phosphorylation in the network, it is evident that MAPK1 displays elevated activity, and similarly, AKT1 shows reduced activity upon ANXA1 knockout. Whether such modulations are a direct effect of ANXA1 in breast epithelial cells or arise from feedback regulations between the MAPK/ERK and PI3K/AKT pathways is still not known [43–45]. While for these and some other kinases such as PAK1 and Aurora B such activity prediction was possible from the network, for a few others including PKA and CDK the substrates displayed both increased and decreased phosphorylation. This suggests that the activities of these kinases may not be turned on/off entirely, but possibly rewired wherein some substrates are selectively phosphorylated while others are not.

Though the kinome network highlighted KSRs from diverse kinase families, it is not representative of the entire signaling network as many of our regulated phosphosites did not have experimentally curated upstream kinase. Hence, we additionally predicted KSRs for all regulated phosphoproteins using the NetworKIN algorithm that assigns kinases based on linear consensus motifs and network proximity [31]. We performed kinase-substrate enrichment analysis (KSEA) to derive significantly modulated kinase along with their predicted activity scores (Fig. 4b). This analysis mirrored some of the findings from the previous kinome network including activation of MAPK1, CAMK2, CK2A, and highlighted other kinases that could be perturbed in the ANXA1 signaling network. We observed CDK family kinases CDK1 and CDK3 to be activated upon loss of ANXA1, and this along with enhanced MAPK1/3 activity possibly constitutes a pro-proliferative signal. Strikingly, kinases such as PAK1, ROCK1 and MRCK were found to be activated upon ANXA1 knockout. These kinases play important roles in determining cell properties during migration, and thus their activation focuses on the rewiring of adhesion network upon loss of ANXA1. Also, a subset of the kinases was predicted to be deactivated in ANXA1-deficient mammary cells, and many of these were associated with cellular stress and cytokine response. These included kinases IKKα and IKKβ that are activators of NF-κB transcription factor. Notably, ANXA1 has been previously reported to constitutively activate NF-κB in breast cancer cells via the IKK complex [46], and our phosphoproteome data also predicts such a positive correlation between ANXA1 and NF-κB activity status.

Mechanistic network of ANXA1-regulated phosphoproteome

To map the ANXA1-modulated network to its entirety, it is necessary to understand where the downstream signaling impinges and which transcription factors are affected as a consequence. We used changes in protein abundance measures from our previous study to predict transcription factor activation status in ANXA1-deficient mammary cells. Using one-sided Fisher’s exact test, we derived a set of 28 transcription factors (p value <0.05) that were potentially regulated in the absence of ANXA1 (Fig. 5a). We found that the targets of cellular stress-associated transcription factors FOS and JUN were specifically enriched among the downregulated proteins. This aligns with our kinase analysis, which predicted reduced activity for kinase MAP2K4 that lies upstream of JUN in the same pathway. Transcriptional targets of TEA domain family (TEAD) transcription factors that are associated with YAP1/TAZ from the Hippo signaling pathway [47], also showed specific enrichment among the downregulated proteins. Such a link between ANXA1 and the Hippo signaling pathway has not been established previously to our knowledge. Other functional partners of YAP1 including RUNX2 also showed negative enrichment [48]. Alternatively, the activities of myogenic regulatory factor (MYF) and serum response factor (SRF) were found to negatively correlate with ANXA1 abundance. These factors are crucial for actin-cytoskeletal organization and focal adhesion assembly [49], and thus correlate with enrichment observed based on phosphoproteome data. While some of the kinases and transcription factors we predicted through analyses of our phosphoproteome and proteome dataset have previously been associated with ANXA1, several novel associations and pathways became evident through our in-depth investigation.

We next combined the predicted transcription factors along with the regulated phosphoproteins, and assessed the functional connectivity among them. We curated protein-protein interactions from InWeb_InBioMap [36], a large data compendium for high quality protein-protein interaction network, and built an interaction network consisting of 282 proteins including 258 differentially regulated phosphoproteins. A contextual view of the network is shown in Fig. 5b with proteins grouped into distinct clusters based on their overrepresented pathways/processes. We observed that many regulated phosphoproteins displayed direct interactions with several predicted transcription factors. A majority of these interactions were mediated by transcriptional regulators that were differentially phosphorylated in ANXA1-deficient cells. The precise nature of these interactions, however, and whether the phosphorylation status may have any influence on them is not known for most of them.

The network revealed that a majority of the proteins in focal adhesion, stress fibers and ruffled membranes displayed increased phosphorylation and as expected also displayed high connectivity among one another (Additional file 12: Table S8). In fact a cluster assessment based on GLay community clustering also highlighted cell adhesion as one of the top enriched clusters in the network highlighting the influence of ANXA1 in governing adhesion dynamics (Additional file 13: Figure S5) [37]. Projections that form on the leading edge, the lamellipodium, constituted proteins that displayed reduced phosphorylation. We found the adaptor protein talin (TLN2) that is crucial for focal adhesion initiation with elevated phosphorylation at Thr1843 upon ANXA1 loss. Talin establishes core cell extracellular matrix (ECM) links by associating and activating integrin, and simultaneously binding to actin and other cytoskeletal proteins [50]. Adhesion life cycle encompasses recruitment of various proteins at different stages, and only a fraction of adhesions survive to become large stable focal adhesion. Finally, disassembly of these adhesions occurs to enable cell spreading and forward movement. This disassembly and destabilization can be microtubule-mediated, wherein microtubules either stimulate ABL2 to inhibit RhoA, which is crucial for adhesion maturation and stress fiber formation or simply their depolymerized state can induce adhesion instability [51–53]. We observed many phosphorylation changes on microtubule-associated proteins including the increased phosphorylation on stathmin that inhibits microtubule depolymerization/catastrophe [42]. This along with reduced phosphorylation on ABL2 kinase possibly points to defects in microtubule-mediated adhesion disassembly, thus favoring enhanced adhesion and less migratory capacity of ANXA1-deficient mammary epithelial cells.

Apart from these major structural components, other connected proteins mainly belonged to cell signaling and cell-cycle-related processes (Additional file 12: Table S8). It is evident from the network that many of these regulated proteins also share interactions with components in focal adhesion and stress fibers underscoring complex interplay among ANXA1-regulated network. Notably, we observed Rictor, which is a component of mTOR complex 2 with reduced phosphorylation on Thr1592. mTORC2 is a regulator of cytoskeletal dynamics and stimulates Rho GTPases and F-actin stress fibers [54]. The fact that many of the regulated phosphoproteins converge on these structural assemblies emphasizes the notion that ANXA1 primarily governs cell adhesion and migratory properties in mammary epithelial cells.