Mapping of the hSOX10 Proximal Protein Interactome in Human Melanoma

Abstract

The transcription factor SOX10 is a central regulator of melanoma, influencing tumor initiation, progression, phenotypic plasticity, and therapeutic resistance, yet the protein–protein interactions underlying its function remain poorly defined. To address this, we conducted the first dedicated, comprehensive mapping of the human SOX10 (hSOX10) proximal protein interactome using miniTurbo (mT) proximity-dependent biotinylation coupled with mass spectrometry in A375 melanoma cells. Stable lines expressing N- or C-terminal mT-tagged hSOX10 fusion proteins at near-endogenous levels enabled the unbiased capture of proximal proteins in a native cellular context, identifying 847 melanoma-enriched candidate hSOX10 interactors. Stringent statistical filtering, contaminant frequency profiling, and subcellular localization context refined this to 180 high-confidence candidates, including known hSOX10 partners and previously unidentified candidates. Integration of orthogonal biological relevance criteria (functional enrichment and network context, transcriptomic coexpression with hSOX10, and genomic co-occurrence in melanoma) further refined the dataset to 124 biologically relevant candidates enriched for transcriptional regulators, cofactors, chromatin-modifying complexes, and associated pathways. These proteins were stratified using an evidence-based prioritization framework incorporating transcriptomic, genomic, and chromatin-based context without additional exclusion. Collectively, this work provides a high-confidence resource for the hSOX10 proximal protein interactome in melanoma and a framework for generating testable hypotheses regarding hSOX10-associated regulatory networks, melanoma biology, and therapeutic vulnerabilities.

Introduction

Melanoma is an aggressive skin cancer originating from melanocytes, the pigment-producing cells of the skin. It poses a major public health challenge due to its high metastatic potential, increasing global incidence, and frequent resistance to even advanced therapies. , Despite accounting for a minority of skin cancer diagnoses, melanoma is responsible for the majority of skin cancer-related deaths and is characterized by pronounced phenotypic heterogeneity, cellular plasticity, and the capacity to adapt to therapeutic pressures. − Although targeted therapies (e.g., BRAF, MEK, and checkpoint inhibitors) have significantly improved outcomes for subsets of patients, durable responses are often limited by acquired resistance and treatment-related toxicities, underscoring the need for deeper mechanistic insights into melanoma biology and novel therapeutic strategies. −

Melanoma cells, like their normal melanocyte counterparts, originate from the embryonic neural crest (NC), a transient, multipotent cell population that gives rise to diverse lineages during development. , Melanoma initiation, progression, and plasticity are closely linked to transcriptional programs inherited from the NC gene regulatory network (GRN), with SRY-Box Transcription Factor 10 (SOX10) serving as a central lineage-defining transcriptional regulator. SOX10 is essential for melanocyte development and maintenance and functions as a major dependency factor in melanoma. ,,, In melanoma, SOX10 governs tumor growth, proliferation, and phenotypic plasticity, including transitions between proliferative and invasive states. These properties contribute significantly to melanoma aggressiveness and therapeutic resistance. , In patient tumors, SOX10 has been implicated in the modulation of resistance pathways, in part through effects on TGFβ-EGFR signaling and immune checkpoint regulation, further highlighting its pivotal role in melanoma pathogenesis. ,,

Despite its well-established role as a lineage-defining transcription factor (TF), the precise mechanisms through which SOX10 executes its diverse biological functions remain incompletely understood. SOX10 operates within the broader NC GRN through interactions with TFs, cofactors, chromatin regulators, and signaling pathways that collectively control lineage specification, transcriptional output, and melanoma formation and maintenance. ,,, Such protein–protein interactions (PPIs), including both direct physical contacts and indirect associations within larger complexes, are essential for cellular functions such as transcriptional regulation, chromatin remodeling, and cellular adaptation. − Thus, defining the protein microenvironment in which SOX10 functionsits “proximity interactome”is a critical step toward elucidating the regulatory logic underlying SOX10-driven melanoma biology. ,,

Traditional methods for mapping PPIs, such as coimmunoprecipitation, are limited in their ability to capture the transient, low-affinity, and context-dependent interactions characteristic of TFs like SOX10. ,,− Proximity-dependent labeling strategies, including miniTurbo (mT)-based biotinylation, overcome many of these limitations by enabling the labeling of proteins within a defined spatial radius of a bait protein in intact cells. This approach facilitates unbiased identification of candidate protein interactors that may participate in direct or indirect PPIs under physiologic conditions. ,,− Although proximity labeling does not distinguish direct binding from spatial proximity, it provides a powerful, discovery-oriented framework for mapping TF-associated protein environments.

To date, SOX10 has only been included as one of many TFs in a single large-scale multiplexed proximity proteomics study. However, this prior effort was not designed to systematically or comprehensively define the SOX10 proximal protein interactome, and a dedicated map of SOX10-proximal proteins has thus remained lacking.

In this study, we leverage mT-based proximity labeling coupled with mass spectrometry (MS) to generate the first dedicated, comprehensive map of the human SOX10 (hSOX10) proximal protein interactome. ,,, We employed both N- and C-terminal mT-tagged hSOX10 fusion proteins expressed at near-physiologic levels in A375 melanoma cells, followed by streptavidin-mediated enrichment and MS-based identification of candidate proximal protein interactors. A375 cells represent a widely used and well-characterized model of cutaneous melanoma, the most common melanoma subtype, and exhibit hSOX10 dependency along with the BRAFV600E driver mutation frequently observed in patient tumors. , This model enables integration of the hSOX10 proximal protein interactome with existing genomic, transcriptomic, and epigenomic datasets and provides a representative framework for future comparative studies across melanoma subtypes.

Using a structured, integrative analytical workflow, we first applied stringent statistical filtering, contaminant frequency profiling, and subcellular localization context to define a high-confidence set of candidate hSOX10 interactors. We then incorporated orthogonal biological relevance criteria (functional enrichment and network context, transcriptomic coexpression with hSOX10, and genomic co-occurrence in melanoma) to refine this dataset to 124 biologically relevant candidates. These retained proteins were subsequently stratified using an evidence-based prioritization framework that integrates transcriptomic, genomic, and chromatin-based contexts, without further exclusion. Collectively, this work establishes a foundational, high-confidence, hSOX10-focused proximal protein interactome resource that enables hypothesis generation and provides a framework for future mechanistic studies aimed at dissecting hSOX10-driven regulatory networks underlying melanoma plasticity, progression, and therapeutic resistance.

Methods

Cell Lines and Culture Conditions

The human cutaneous melanoma cell line A375 (ATCC, CRL-1619), which harbors the BRAFV600E driver mutation, was used to generate stable lines expressing mT-tagged fusion proteins. Three constructs were evaluated: N-terminal mT-tagged hSOX10 (mT-hSOX10), C-terminal mT-tagged hSOX10 (hSOX10-mT), and mT containing a nuclear localization signal (mT-NLS) as a control. A375 cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM; Corning, MT10013CV) supplemented with 10% fetal bovine serum (FBS; Sigma-Aldrich, F2442) and 1% penicillin-streptomycin (Gibco, 15140122). Puromycin selection was performed in DMEM containing 10% FBS, 1% penicillin-streptomycin, and 1 μg/mL puromycin (Fisher Scientific, NC0721599). For proximity labeling assays, stable fusion protein cell lines were treated in DMEM with 5% FBS, supplemented with 50 μM biotin. HEK293T/17 cells (ATCC, CRL-11268) were used for lentivirus production and maintained in DMEM with 10% FBS and 1% penicillin-streptomycin, except during transfection, when antibiotic-free medium was used. All cultures were maintained at 37 °C in a humidified incubator containing 5% CO₂.

Cloning of hSOX10 Fusion Constructs and NLS Control

Both mT-hSOX10 and hSOX10-mT fusion constructs were generated using the Gateway recombinational cloning system. The hSOX10 coding sequence was obtained from pDONR221-hSOX10 (Addgene #24749). Two versions of this plasmid were utilized: one containing the native stop codon and a second in which the stop codon was removed by Q5 site-directed mutagenesis (New England Biolabs). Recombination into destination vectors was performed using LR Clonase II Plus enzyme mix (Invitrogen, 12538120) following established protocols. Briefly, the pLenti663-UBC-mT-hSOX10 (mT-hSOX10) construct was generated by multisite Gateway cloning to fuse mT to the N-terminus of hSOX10 (25 ng mT Entry clone, 25 ng UBC Entry clone, 40 ng hSOX10, 100 ng pDest663 destination, and LR Clonase II Plus enzyme mix). The pLenti667-UBC-hSOX10-mT (hSOX10-mT) construct was generated by single-site Gateway cloning to fuse mT to the C-terminus of hSOX10 lacking a stop codon (50 ng hSOX10 without a stop codon, 100 ng pLenti667-UBC-GW-mT destination, and LR Clonase II Plus enzyme mix). Constructs were transformed into E. coli Stbl3 chemically competent cells (Invitrogen, C737303) and validated by restriction digestion, Sanger sequencing, and whole-plasmid sequencing (Plasmidsaurus). Sanger sequencing primers included M13R, an in-house primer for the UBC promoter (TGAAGCTCCGGTTTTGAACT), and in-house primers for both the N- (CATTTCCTCCCTCTGCTTCC) and C-terminal (AATAGACCCGTGAAGCTGATC) mT fusion regions. The mT-NLS control construct was previously generated by and obtained from the Major laboratory (Washington University School of Medicine in St. Louis, St. Louis, MO).

Lentiviral Production

Lentiviral particles were produced in HEK293T/17 cells using calcium phosphate-mediated transfection. Cells (4.4 × 10⁶) were plated in 10 cm dishes 24 h prior to transfection. Transfer plasmids (8 μg; mT-hSOX10, hSOX10-mT, or mT-NLS), psPAX2 packaging plasmid (6 μg; Addgene #12260), and pMD2.G envelope plasmid (2 μg; Addgene #12259) were combined with 2.5 M CaCl₂ (Sigma-Aldrich, C4901) and 2× HEPES-buffered saline (HBS; 280 mM NaCl, 1.5 mM Na₂HPO₄·H₂O, 50 mM HEPES, pH 7.10) to form DNA precipitates, which were added dropwise to cells. Medium was replaced after a 24-h incubation at 37 °C, and viral supernatants were collected 48 h later, filtered through 0.45-μm PVDF membranes, aliquoted, and stored at −80 °C.

Generation of Stable Cell Lines

A375 cells (5.0 × 10⁵) were plated per 10 cm dish, allowed to adhere for 48–72 h, and transduced with unconcentrated lentiviral supernatants in the presence of Polybrene (10.66 μg/mL; Sigma-Aldrich, TR-10030G). Lentiviral dosages of 0.5 mL, 1 mL, 2 mL, 3 mL, 4 mL, and 5 mL were tested to determine the optimal dosage. Following 24 h of transduction, the medium was replaced, and puromycin selection (1 μg/mL) was initiated at 48 h post-infection. Stable populations were expanded in selection medium, cryopreserved in 90% FBS/10% DMSO, and maintained under selection for subsequent experiments.

Proximity-Dependent Biotinylation and Protein Extraction

A375 cells stably expressing hSOX10-mT, mT-hSOX10, or mT-NLS were grown to 85–90% confluence in 15 cm dishes. For each construct, three biological replicates were prepared, each comprising seven plates. Proximity labeling was induced by treatment with 50 μM biotin for 90 min at 37 °C. Cells were then washed twice with ice-cold PBS, scraped on ice, pooled by replicate into prechilled conical tubes, and pelleted by centrifugation at 4 °C. Cell pellets were lysed in RIPA buffer (50 mM HEPES pH 8.0, 150 mM NaCl, 2 mM EDTA, 0.1% SDS, 1% Triton X-100, 0.2% sodium deoxycholate, 10% glycerol) supplemented with 1× Halt protease inhibitor cocktail (100×; Thermo Scientific, 1861279), 1× N-ethylmaleimide (freshly prepared from a 1 M ethanol stock; Thermo Scientific, 23030), and benzonase (1:5000 dilution; Sigma, E1014–25KU). Lysates were passed through a 21-gauge needle five times to shear membranes, rocked at 4 °C for 15 min, clarified by centrifugation at 4 °C, and supernatants were stored at −80 °C.

Streptavidin Affinity Enrichment

Protein concentrations were determined using the Pierce BCA Protein Assay Kit according to the manufacturer’s instructions. For each replicate, 20 mg of total protein was prepared in a final volume of 8 mL RIPA lysis buffer. Streptavidin-conjugated Sepharose beads (Cytiva, GE17–5113–01) were prewashed three times in RIPA lysis buffer and centrifuged at 400g for 2 min at 4 °C after each wash. After washing, lysates were incubated with prewashed beads (30 μL each) at 4 °C overnight with gentle rotation. Beads were pelleted at 400g for 2 min at 4 °C, and unbound supernatant was discarded. Beads were washed sequentially with 1 mL of increasingly stringent wash buffers to remove nonspecific binders: two washes with 2% SDS in water (WB1), one wash with a high-salt buffer (WB2; 500 mM NaCl, 0.1% deoxycholate, 1% Triton X-100, 1 mM EDTA, 50 mM HEPES pH 7.5), one wash with lithium chloride buffer (WB3; 250 mM LiCl, 0.5% Triton X-100, 0.5% deoxycholate, 1 mM EDTA, 50 mM HEPES pH 8.1), one wash with a low-salt buffer (WB4; 150 mM NaCl, 50 mM HEPES pH 7.4), and three washes with 50 mM ammonium bicarbonate (ABC buffer; WB5). After each wash, beads were resuspended, inverted gently ten times, and centrifuged at 400g for 2 min at room temperature (RT). Beads were stored at 4 °C or processed immediately for downstream analyses.

Sample Preparation for Mass Spectrometry

Proteins bound to beads were digested on-bead by resuspending in 100 μL of 50 mM ABC buffer and incubating overnight at 37 °C with 1 μg trypsin (Promega V5113) and 0.5 mAU Lys-C (Wako Chemicals, 129–02541). An additional 0.5 μg of trypsin/Lys-C was added the next day, and digestion continued for 2 h. Peptides were collected by centrifugation, pooled with bead washes, and passed through BioPureSPN columns (Nest Group, 10 μm frit, C100500) prewashed with 50 mM ABC buffer. Filtered peptides were acidified to 2% formic acid (FA) and dried in a SpeedVac. Dried peptides were resuspended in 25 μL of 0.1% FA and 2% acetonitrile in water. Peptide concentrations were measured with a fluorometric BCA assay, and samples were normalized to 1 μg of peptide per 10 μL injection volume. Final samples were transferred to autosampler vials for MS.

Western Blotting

Protein lysates were analyzed by Western blotting to validate bait expression, biotinylation, and selected interactors. Protein concentrations were determined using the Pierce BCA Protein Assay Kit, and all samples were normalized to the lowest concentration for consistent loading. Proteins were separated on 4–12% Criterion XT Bis-Tris Protein Gels (Bio-Rad, 3450124; 18-well format, 30 μL loads per well). Protein samples were mixed with 4× LDS loading buffer (Invitrogen, NP0007) and lysis buffer to a final volume of 19 μL, and 1 μL of 20× DTT (1M, Cell Signaling, 7016) was added immediately before loading. Samples were heated at 70 °C for 5–10 min, cooled, centrifuged briefly, vortexed, and loaded alongside Precision Plus Protein Kaleidoscope Prestained Protein Standards (Bio-Rad, 1610395).

Electrophoresis was performed in 1× MOPS Running Buffer (Invitrogen, NP0001) at 80 V for 20–30 min, followed by 120 V for 60–90 min. Proteins were transferred onto nitrocellulose membranes (Bio-Rad, 1704271) according to the manufacturer’s protocol for midi gels. Membranes were washed with 1× TBS and blocked in 5% BSA solution for 1 h at RT, then incubated overnight at 4 °C while rocking in one of the following primary antibodies diluted in blocking buffer: hSOX10 (1:1000, Abcam, ab264405), hSOX10 (1:400, Abcam, ab227680), GAPDH (1:1000, Cell Signaling, 14C10), Anti-V5 antibody (Invitrogen, R960–25), YAP1 (1:1000, Cell Signaling, 4912S), ZNF609 (1:1000, SigmaMillipore, HPA040742), TRIM24 (1:2000, Bethyl, A300–815A-T), MAML1 (1:1000, Cell Signaling, 4608S), STAT3 (1:1000, Cell Signaling, 79D7 #4904), and KMT2A (1:1000, Cell Signaling, 14197). Membranes were washed four times for 5–10 min each with 1× TBS-T and incubated for 1 h at RT with secondary antibodies in blocking buffer containing 0.1% Tween-20. Secondary antibodies included IRDye 680RD Streptavidin (1:5000, LI-COR, 925–68079), IRDye 800CW Donkey anti-Mouse (1:10,000, LI-COR, 926–32212), and IRDye 680RD Goat anti-Rabbit IgG (1:10,000, LI-COR, 926–68071). Membranes were washed four times for 10–20 min each in 1× TBS-T, and dual-channel fluorescence imaging (700 and 800 nm) was performed using a LI-COR Odyssey Imaging System.

Mass Spectrometry Data Acquisition

Trypsinized peptides were loaded onto a μPAC^TM Trapping column (Thermo Scientific, COL-TRPNANO16G1B2) and separated on 50 cm μPAC^TM Neo HPLC column (Thermo Scientific, COL-NANO050NEOB). Chromatography was performed with the following parameters: initial 2.8 min gradient from 2% to 10% buffer B at 0.750 μL/min, followed by an increase to 12% buffer B at 4.5 min and a reduced flow rate of 300 nL/min. A 57.7 min gradient from 12% to 40% buffer B at 300 nL/min was then applied, followed by a ramp to 90% buffer B over 0.3 min and subsequent column wash and re-equilibration. MS analysis was performed on an Orbitrap Eclipse (Thermo Fisher Scientific) in data-dependent acquisition mode. MS1 scans were acquired in the Orbitrap at 240k resolution, with a 250% normalized automated gain control (AGC) target, automatic maximum injection time, and 375–2000 m/z scan range. MS2 targets (charge states 2–7) were filtered with a dynamic exclusion of 60 s and accumulated using a 0.7 m/z quadrupole isolation window. MS2 scans were acquired in the ion trap at turbo scan rate following higher-energy collisional dissociation at 35% normalized collision energy, with a 100% normalized AGC target and 35 ms maximum injection time.

MaxQuant Data Search

Raw MS data files were analyzed for protein identification and label-free quantification (LFQ) using MaxQuant (version 2.5.1.0) with default Orbitrap settings and minor modifications. Datasets for N- and C-terminal mT-tagged hSOX10 constructs and the mT-NLS control were searched against the human UniProt reference proteome database (UP000005640_9606, updated 2024–02–08, https://www.uniprot.org/proteomes/UP000005640) plus a list of likely contaminants (e.g., streptavidin, trypsin, albumin, mT constructs with different subcellular localizations) and the MaxQuant contaminants list. Variable modifications included methionine oxidation (+15.994914 Da) and protein N-terminal acetylation (+42.010564 Da), with a maximum of five modifications per peptide. Digestion was set to Trypsin/P with a maximum of four missed cleavages and a minimum allowed peptide length of seven amino acids. Both unique and razor peptides were used for protein quantification, and the minimum LFQ ratio count was set to 2. Fast LFQ with classic normalization was used, with minimum and average numbers of neighbors set to 3 and 6, respectively. For samples within a group, “match between runs” (match time window of 0.4 min; alignment time window 20 min) and “second peptides” were enabled. The raw MS proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) with the dataset identifier PXD067868.

Perseus Data Analysis

Data filtering and statistical analysis were performed using Perseus (version 2.0.11.0) following an initial preprocessing step in Microsoft Excel. The MaxQuant “proteinGroups.txt” file was first curated in Excel to remove proteins annotated as “Only identified by site,” “Reverse,” or “Potential contaminant,” as well as the mT bait protein, proteins identified by only a single unique peptide, and proteins detected exclusively in the mT-NLS control samples. The curated dataset was exported as a tab-delimited (.txt) file and imported into Perseus for downstream analysis. Within Perseus, raw protein intensity values were log₂-transformed, and the individual columns were grouped by bait. Proteins lacking quantified or valid values across all three replicates for at least one experimental group were removed. The remaining missing values were imputed using the “Replace missing values from normal distribution” function with a width of 0.3 and a downshift of 1.8 from the standard deviation for the total matrix. Principal component analysis (PCA) was performed to assess clustering of mT-hSOX10, hSOX10-mT, and mT-NLS replicates. To identify proteins enriched in mT-hSOX10 and hSOX10-mT samples relative to mT-NLS, two-sided unpaired Student’s t tests were performed for each hSOX10 fusion construct versus the mT-NLS control, with multiple testing correction using a permutation-based false discovery rate (FDR) approach (250 permutations; FDR = 0.05). Proteins were considered significantly enriched if they met all of the following criteria: (i) absolute log₂ fold change (FC) (hSOX10 vs NLS) ≥ 2.0 or ≤ −2.0, (ii) adjusted p-value (q-value) ≤ 0.05, and (iii) identification by at least two unique peptides in the hSOX10 mT bait samples. Volcano plots were generated using log₂ FC (x-axis) and -log₁₀(p-value) (y-axis) to visualize enriched proteins.

Protein Structure Prediction and Visualization

AlphaFold 3 was used to assess the structural effects of mT fusion. For each construct, prediction Model 0, the top-ranked prediction, was selected for downstream visualization and interpretation. For hSOX10 alone, an additional model (Model 4) was also examined and included for comparison, as Models 0 and 4 differed modestly in the relative spatial arrangement of key domains. Structural models were visualized using the built-in AlphaFold 3 interface. Functional regions of hSOX10, including the dimerization domain (DIM), high-mobility group (HMG) DNA-binding domain, and transactivation domains (TAM/TAC), were manually annotated and color-coded. Measurements of predicted protein dimensions were obtained within AlphaFold 3 to estimate overall protein sizes and spatial positioning of the mT tag relative to hSOX10 domains. Model confidence metrics are provided in Figures S1–S4.

CRAPome Contaminant Frequency and Subcellular Localization Analysis

Contaminant frequency analysis was performed using the Contaminant Repository for Affinity Purification (AP) (CRAPome; https://www.crapome.org). The top 207 shared candidates identified from the initial proximity enrichment analysis were queried individually against the CRAPome database. For each protein, CRAPome-derived metrics (percentage of experiments in which the protein was detected, average spectral count, and maximum spectral count) across unrelated AP-MS experiments were extracted. Subcellular localization annotations were evaluated using the Human Protein Atlas (HPA; http://www.proteinatlas.org). For each candidate protein, subcellular localization was first obtained from the Protein Expression and Localization section, which integrates antibody-based immunofluorescence data across multiple human cell lines. Proteins explicitly annotated as being nuclear-localized were classified accordingly. The Protein Function section of HPA (derived from UniProt-curated functional evidence) was then systemically reviewed for all candidates to ensure consistency with reported localization annotations. In a limited number of cases where nuclear localization status was not explicitly reported in the Protein Expression and Localization section but strong evidence of nuclear function was documented (e.g., reported roles in transcriptional regulation, chromatin association, and DNA-binding activity), proteins were classified as nuclear-associated.

CRAPome-derived contaminant frequency metrics were integrated with proximity labeling enrichment values and HPA-based localization annotations to facilitate prioritization of high-confidence candidate interactors. Visualization of average log₂ FC versus CRAPome detection frequency was performed using GraphPad Prism. Minor visual refinements (e.g., coloring and labeling) were performed using Adobe Illustrator.

Functional Enrichment and Network-Based Annotation of hSOX10-Proximal Proteins

Gene ontology (GO) Molecular Function enrichment analysis was conducted using the STRING database, with the top 180 prioritized candidate proteins as input. STRING visualization tools were used to identify and highlight enriched molecular functions. To complement STRING-based GO Molecular Function analysis, GO Biological Process and Reactome pathway enrichment analyses were performed using the ClueGO plugin (version 2.5.10) within Cytoscape. ClueGO analysis parameters and output statistics are provided in Tables S1 and 2.

STRING was also used to identify candidate interactors belonging to known transcriptional regulatory and chromatin remodeling complexes. Proteins were categorized into complexes or functional groups based on GO Molecular Function and Cellular Component annotations. For the purposes of these analyses, when proteins were associated with multiple functional categories (e.g., transcriptional coregulator and transcriptional repressor), manual assignment was made to the most specific or biologically informative category. Similarly, proteins belonging to well-defined complexes were prioritized for complex-based grouping over more general classifications.

Previously reported hSOX10 interactors were identified using the BioGRID interaction database (https://thebiogrid.org), which curates experimentally supported PPIs from the published literature. Proteins annotated in BioGRID as interacting with hSOX10 based on prior biochemical, affinity-based, or interaction-focused studies were designated as known hSOX10 interactors and used for reference in the downstream network annotation and visualization.

Network visualization of functional groups and candidate interactors was performed with Cytoscape. Candidate interactors were manually organized into functional- and complex-based groupings and ordered within groups based on average log₂ FC derived from proximity labeling enrichment analysis. Heat maps of log₂ protein intensities across constructs were generated with Microsoft Excel. Minor visual refinements, including grouping outlines and labeling, were performed using Adobe Illustrator.

Gene Expression Correlation Analysis

Gene expression correlation analysis was performed using Gene Expression Profiling Interactive Analysis 2 (GEPIA2; http://gepia2.cancer-pku.cn). Pearson correlation coefficients (R) and associated p-values were calculated using The Cancer Genome Atlas (TCGA) Skin Cutaneous Melanoma (SKCM) tumor samples. Correlation analyses were performed individually for each candidate relative to hSOX10 expression. For visualization and downstream analysis, GEPIA2-reported p-values of 0 (reflecting values below the platform’s numerical precision) were set to 1 × 10^–16 to enable log transformation. Correlation plots were generated using GraphPad Prism, and final figure formatting (e.g., colors, annotations, labeling) was performed using Adobe Illustrator.

Genomic Co-occurrence and Mutation Frequency Analysis

Genomic co-occurrence and mutual exclusivity analyses were performed using cBioPortal for Cancer Genomics (https://www.cbioportal.org). Analyses were conducted using publicly available cutaneous melanoma datasets aggregated within cBioPortal. For each candidate gene, the log₂ odds ratio (OR) and corresponding FDR-adjusted q-value were calculated using Fisher’s exact test to evaluate statistically significant co-occurrence or mutual exclusivity with hSOX10 alterations across tumor samples. For numerical consistency and visualization, cBioPortal-reported q-values listed as <0.001 were set to 0.001, and log₂ ORs reported as >3 or <−3 were capped at 3 or −3, respectively. Mutational frequencies for candidate genes were also extracted from cBioPortal cutaneous melanoma cohorts and included somatic single-nucleotide variants, small insertions/deletions, and copy number alterations. Aggregate mutation frequencies were incorporated into downstream analyses, while detailed per-sample mutation annotations for all analyzed genes are provided in Table S3. Plots were generated in GraphPad Prism (genomic co-occurrence) or Microsoft Excel (mutational frequency), and colors, annotations, and labeling were adjusted in Adobe Illustrator.

ChIP-seq-Based hSOX10 Target Gene Annotation

ChIP-seq-based target gene annotation for hSOX10 was performed using ChIP-Atlas (https://chip-atlas.org), with hSOX10 specified as the query TF. Target genes were identified based on MACS2 peak enrichment within ±10 kb of transcription start sites. Publicly available hSOX10 ChIP-seq datasets included one cutaneous melanoma dataset (501Mel; GSM1517752), two uveal melanoma datasets (92–1; GSM7729129 and GSM7729130), one soft tissue sarcoma dataset (DTC1; GSM5720706), and three additional soft tissue sarcoma datasets (SU-CCS-1; GSM5720714, GSM5720717, and GSM5720720). For each candidate gene, ChIP-Atlas-reported binding scores were extracted for individual datasets as well as for the ChIP-Atlas-provided average hSOX10 binding score (Table S4). Average binding scores were also calculated separately for uveal melanoma datasets and nonmelanoma datasets. ChIP-Atlas binding scores and associated annotations were integrated into network visualizations using Cytoscape. Minor figure formatting and annotation adjustments were performed using Adobe Illustrator.

Results

To our knowledge, this study represents the first dedicated, comprehensive mapping of the hSOX10 proximal protein interactome. We employed mT-based proximity biotinylation to label proteins within an estimated 10–35 nm radius ,− of hSOX10, followed by streptavidin enrichment and MS to identify candidate hSOX10 interactors in A375 melanoma cells (Figure ). This experimental workflow was coupled with a multistage computational framework integrating technical and statistical filtering, biological relevance assessment, and evidence-based prioritization to generate a high-confidence, hypothesis-generating map of the hSOX10 proximal protein interactome in melanoma.

1.

Experimental and computational workflow for proximity-based mapping and integrative prioritization of the hSOX10 proximal protein interactome in melanoma. The top panel (Experimental) depicts generation of stable melanoma cell lines expressing N- or C-terminal mT-tagged hSOX10 fusion proteins, induction of proximity-dependent biotinylation following exogenous biotin addition, total protein extraction, streptavidin-based enrichment of biotinylated proximal proteins, and identification by MS. The bottom panel (Computational) summarizes the multistage refinement and prioritization strategy applied to MS data, including Phase 1 Technical and Statistical Filtering, Phase 2 Biological Relevance Filtering using orthogonal functional, transcriptomic, and genomic datasets, and Phase 3 Integrative Prioritization based on cumulative evidence supporting hSOX10 association. Together, this workflow generates a high-confidence, hypothesis-generating map of the hSOX10 proximal protein interactome in human melanoma. Abbreviations: hSOX10, human SOX10; MS, mass spectrometry; mT, miniTurbo; NLS, nuclear localization signal; nm, nanometer; UBC, ubiquitin C promoter.

Generation of hSOX10 Proximity Labeling Constructs and Structural Modeling

Three lentiviral constructs were generated to probe the hSOX10 proximity interactome: mT-hSOX10, hSOX10-mT, and an mT-NLS nuclear control (Figure A). Both N- and C-terminal mT-tagged hSOX10 fusion constructs were designed to capture complementary protein microenvironments surrounding hSOX10 and to increase confidence in proximal protein interactor identification, while the mT-NLS construct controlled for nonspecific nuclear biotinylation. All constructs included a V5 epitope tag and were expressed under a constitutive UBC promoter to achieve near-endogenous expression levels.

2.

Design and validation of mT-tagged hSOX10 proximity labeling constructs. (A) Schematic representation of the mT-tagged constructs used for proximity biotinylation, including N-terminal mT-tagged hSOX10 (mT-hSOX10), C-terminal mT-tagged hSOX10 (hSOX10-mT), and nuclear-localized mT-NLS control constructs, all driven by the UBC promoter and containing a V5 epitope tag. (B) Domain architecture of hSOX10 highlighting key functional regions, including the DIM, HMG, TAM, and TAC domains, with amino acid residue numbering indicated. (C) Predicted structural models of untagged hSOX10, mT-tagged hSOX10 fusion proteins, and mT alone generated using AlphaFold 3, illustrating the relative positioning of the mT tag with respect to annotated hSOX10 domains. Color coding corresponds to hSOX10 domains and mT, as indicated in the legend. Scale bar shown. (D) Immunoblot analysis confirming successful expression of the indicated constructs following viral transduction, with detection of endogenous hSOX10 (unmodified, ∼55 kDa; phosphorylated, ∼70 kDa), hSOX10 fusion proteins (∼80 kDa), V5-tagged mT control fusion proteins (∼32 kDa), and GAPDH as a loading control. (E) Streptavidin-based immunoblot detection of biotinylated proteins following biotin supplementation, demonstrating robust proximity-dependent biotinylation in cells expressing mT-hSOX10 (lanes 9–10) and hSOX10-mT (lanes 14–15) compared with WT (lanes 2–3, 7–8, and 12–13) and mT-NLS controls (lanes 4–5). (F) Time-course experiments demonstrating progressive labeling with increasing biotin exposure times, with maximal signal achieved at 90 min. Abbreviations: DIM, dimerization domain; HMG, high-mobility group DNA-binding domain; hSOX10, human SOX10; kDa, kilodaltons; mT, miniTurbo; NLS, nuclear localization signal; TAC, transcriptional activation domain; TAM, transactivation domain; UBC, ubiquitin C promoter.

To evaluate whether mT fusion alters hSOX10 structural organization (Figure B), protein structure predictions were generated using AlphaFold 3 for untagged hSOX10, mT-hSOX10, hSOX10-mT, and mT alone (Figure C). For each construct, model 0 (top-ranked prediction) was selected for visualization and comparative analysis. For hSOX10 alone, model 4 was additionally examined, as models 0 and 4 exhibited modest differences in the relative spatial positioning of the DIM and HMG DNA-binding domains. While both models preserved the folded core domains of hSOX10 and displayed similar overall confidence metrics, model 4 more closely resembles the relative domain organization observed in the fusion protein models. Importantly, these differences largely involved regions predicted with low or very low confidence.

Across all models, the core functional domains of hSOX10 (DIM, HMG, TAM, and TAC domains) remained intact, with variability primarily confined to intrinsically disordered regions exhibiting low or very low predicted local distance difference test (pLDDT) scores (low: 70 > pLDDT > 50; very low: pLDDT < 50). AlphaFold 3-based measurements suggested an overall fusion protein diameter of approximately 5–15 nm, depending on measurement orientation and inclusion of disordered regions (Figures S1–S4). Given the effective labeling radius of mT (10–35 nm ,− ), both tagging orientations are expected to access overlapping yet partially distinct protein neighborhoods, supporting their complementary use for proximal protein interactome mapping. Together with extensive prior validation of mT-based proximity labeling across diverse biological systems, − ,− these predictions are consistent with overall preservation of hSOX10 structural integrity in both fusion constructs.

Stable Expression of Fusion Proteins and Optimization of Proximity Labeling

Western blot analysis confirmed stable expression of all constructs in A375 cells (Figure D). Lentiviral titration revealed that increasing viral volume did not substantially enhance fusion protein levels. Accordingly, 3 mL of virus was used for hSOX10 fusion protein constructs and 2 mL for mT-NLS. Endogenous hSOX10 (∼55 kDa unmodified; ∼70 kDa phosphorylated) was detected at comparable levels across all cell lines, indicating that fusion protein expression did not markedly alter endogenous hSOX10 abundance. Distinct ∼80 kDa bands corresponding to the hSOX10 fusion proteins were present only in mT-hSOX10 (lanes 7–9) and hSOX10-mT (lanes 10–13) samples, while the ∼32 kDa mT-NLS protein was observed only in control lanes (lanes 3–6). Within each hSOX10 fusion protein group, bands were uniform in size and intensity across replicates. V5 blotting confirmed robust expression of both fusion proteins, with a stronger signal in hSOX10-mT samples, suggesting higher expression of the C-terminal construct. Differences between hSOX10- and V5-detected signal intensities, particularly for hSOX10-mT, are likely attributable to antibody-specific detection properties.

Functional activity of the mT tag was confirmed by streptavidin-based detection following biotin supplementation (Figure E). Both hSOX10 fusion constructs produced robust and distinct biotinylation profiles compared with the mT-NLS control, aside from expected endogenous biotinylated mitochondrial carboxylases at ∼75 kDa and ∼25 kDa. − Time-course experiments demonstrated progressive labeling with increasing biotin exposure times, with maximal signal achieved at 90 min (Figure F). A 90 min biotin incubation time was therefore selected for all subsequent experiments.

Identification of the hSOX10 Proximal Protein Interactome by Mass Spectrometry and Initial Technical and Statistical Refining Steps

Streptavidin-enriched, on-bead-digested proteins from mT-hSOX10, hSOX10-mT, and mT-NLS cells (three biological replicates each) were analyzed by MS and processed using MaxQuant and Perseus. PCA of log₂-transformed protein intensities revealed tight clustering of biological replicates and clear separation of hSOX10 fusion samples from mT-NLS controls, indicating robust construct-specific proximity labeling (Figure A). Peptide recovery was markedly higher in hSOX10 fusion samples than controls, with an average of 1212 ± 187.45 peptides per replicate in mT-NLS samples, 2465 ± 226.69 in mT-hSOX10 samples, and 5066 ± 430.50 in hSOX10-mT samples (Figure B). These differences are consistent with higher expression of the C-terminal hSOX10 fusion construct and correspondingly increased detection of hSOX10-proximal proteins relative to the nuclear control.

3.

Multistep technical and statistical filtering of candidate hSOX10 interactors. (A) Principal component analysis of MS data demonstrating distinct clustering of mT-hSOX10, hSOX10-mT, and mT-NLS control samples. (B) Number of peptides detected per sample, highlighting increased proteome coverage in hSOX10 proximity labeling samples relative to controls. (C) Volcano plots depicting log₂ FC versus -log₁₀(p-value) for proteins enriched in mT-hSOX10 (top) and hSOX10-mT (bottom) samples compared with mT-NLS controls. Dashed lines indicate statistical thresholds. Boxed numbers indicate the total number of significantly enriched proteins per fusion orientation, with known hSOX10 interactors highlighted in green. (D) Overlap of significantly enriched proteins identified by N-terminal and C-terminal hSOX10 proximity labeling, yielding 207 shared candidate hSOX10 interactor proteins. (E) Distribution of average log₂ FC, unique peptide counts, and MS/MS spectral counts for the top enriched candidates shared between both fusion orientations. (F) Integration of proximity labeling enrichment with contaminant frequency profiling and subcellular localization annotation. Average log₂ FC is plotted against CRAPome detection frequency for the 207 shared candidates. Proteins retained for downstream analyses included both known hSOX10 interactors (green) and additional high-confidence candidates exhibiting low contaminant frequencies (<40%) and nuclear localization (purple). Proteins with high CRAPome detection frequencies (>40%; gray) or lacking nuclear localization (pink) were excluded as likely background or off-target signals, resulting in a refined set of 180 high-confidence candidates. Abbreviations: CRAPome, Contaminant Repository for Affinity Purification; FC, fold change; hSOX10, human SOX10; MS, mass spectrometry; mT, miniTurbo; NLS, nuclear localization signal; PCA, principal component analysis. ^a Although 209 shared proteins were initially identified across both fusion orientations, two grouped protein entries (HOXB3; HOXA3; HOXD3 and STMN1; STMN2) were excluded due to ambiguity arising from shared peptide assignments that prevented confident gene-level resolution by MS. ^b AHNAK was excluded from panel E to improve figure readability due to its outlier values (average log₂ FC = 2.65; 122 unique peptides; MS/MS count = 684).

Following removal of contaminants, reverse hits, proteins identified only by site, the mT bait, single-unique-peptide identifications, and proteins detected exclusively in mT-NLS samples, our raw dataset was refined to 1019 candidates. Two-sample t tests comparing each hSOX10 fusion protein to the mT-NLS control were then performed with permutation-based FDR control. Proteins were considered significantly enriched if they exhibited absolute log₂ FC ≥ 2.0 or ≤ −2.0, q-value ≤ 0.05, and ≥ 2 unique peptides in hSOX10 samples (Figure C). Using these criteria, 847 proteins were enriched in at least one hSOX10 fusion dataset (Table S5). Of these, 209 proteins were shared between both N- and C-terminal fusions, with 4 unique to mT-hSOX10 and the remainder (n = 634) detected only in hSOX10-mT samples (Figure D). Two grouped protein entries (HOXB3; HOXA3; HOXD3 and STMN1;STMN2) were excluded from downstream gene-level analyses, yielding 207 high-confidence, shared, and gene-resolved candidates.

Across these 207 candidates (excluding the outlier AHNAK), average log₂ FC values ranged from 2.3 to 10.5 (median 5.3), with unique peptide counts ranging from 2 to 53 (median 11) and MS/MS counts from 5 to 459 (median 40), indicating strong and reproducible enrichment (Figure E). Although AHNAK (average log₂ FC 2.65, 122 unique peptides, and 684 MS/MS spectra) was excluded from panel E visuals, it was retained in the dataset for downstream analyses.

Contaminant Frequency and Subcellular Localization Filtering Refine High-Confidence hSOX10 Protein Interactor Candidates

To further refine the shared candidate dataset, contaminant frequency profiling and subcellular localization annotation were evaluated for the 207 gene-resolved candidates (Figure F). CRAPome analysis revealed that the majority of candidates were infrequently detected across unrelated AP-MS datasets. All previously reported hSOX10 interactors (n = 23) exhibited CRAPome detection frequencies below 20%. Across the dataset, 77.3% of candidates were detected in fewer than 20% of CRAPome experiments, and 91.3% in fewer than 40%. Only 8.7% displayed high detection frequencies (>40%), consistent with common contaminants, and were removed from the dataset for downstream analyses.

Subcellular localization was assessed by using annotations from the Human Protein Atlas. The majority of retained candidates were annotated as nuclear. For proteins lacking explicit localization annotations, established nuclear functions (e.g., transcriptional regulation or chromatin remodeling) supported classification as nuclear-associated. Proteins lacking evidence of nuclear localization or function were removed from the dataset for downstream analyses. Together, integration of these filters yielded a refined set of 180 candidate hSOX10 protein interactors (Table S5).

Functional Enrichment Highlights Transcriptional and Chromatin-Regulatory Programs among the Top 180 hSOX10-Proximal Candidate Protein Interactors

To establish biological context and support downstream biological relevance filtering (Phase 2; Figure ), functional enrichment analyses were performed on the 180 retained hSOX10 protein interactor candidates. GO Molecular Function enrichment analysis using STRING revealed strong overrepresentation of transcription-associated functions, including transcription coregulator activity, transcription coactivator activity, nuclear receptor binding, DNA-binding TF, and chromatin binding (FDR range: 1.0 × 10^–51 to 1 × 10^–12; Figure A). Complementary ClueGO analyses of GO Biological Process and Reactome pathways identified enrichment for chromatin organization, transcriptional regulation, SUMOylation and SUMO-dependent transcriptional control, DNA damage and TP53-associated signaling, regulation of cell state transitions, transcriptional regulation of white adipocyte differentiation, and heme signaling (Figures S5–S6 and Tables S1−S2).

4.

Functional organization and validation of transcriptional and chromatin-associated proteins within the top 180 candidate hSOX10 interactors. (A) GO Molecular Function enrichment analysis (STRING) of the top 180 high-confidence hSOX10-proximal candidate protein interactors, revealing strong enrichment for transcriptional coregulator activity, DNA-binding transcription factor-associated functions, and chromatin binding. (B) Heat map of log₂ protein intensities across mT-hSOX10, hSOX10-mT, and mT-NLS control samples for a functionally connected subset of candidates within the top 180, including components of major chromatin remodeling complexes or transcriptional regulatory pathways. (C) Network representation of functionally connected proteins within the top 180 candidate interactors, highlighting known hSOX10 interactors and novel candidates, complex membership, and prey–prey interactions. Nodes are grouped by GO Molecular Function annotations, and only candidates linked by shared functional roles, known interactions, or complex membership are shown. (D) Western blot validation of selected candidate proteins following streptavidin-based enrichment of biotinylated proteins, demonstrating enrichment in hSOX10 proximity labeling conditions relative to mT-NLS controls. Abbreviations: FC, fold change; GO, Gene Ontology; hSOX10, human SOX10; MS, mass spectrometry; mT, miniTurbo; NLS, nuclear localization signal; STRING, Search Tool for the Retrieval of Interacting Genes/Proteins; TF, transcription factor.

These enrichment patterns provided the basis for Biological Relevance Filtering Criterion 1, defined as membership in transcriptional regulatory complexes, chromatin-associated functional groups, or functionally enriched pathways represented within the proximity labeling dataset. To illustrate how these global enrichment patterns map to individual proteins, we identified a subset of 64 candidates (including hSOX10) that contribute to these enriched functional categories, transcriptional or chromatin-associated complexes, and/or prey–prey interactions identified within the dataset. These candidates display consistent enrichment across both hSOX10 fusion constructs relative to the mT-NLS control (Figure B).

Network visualization of this subset revealed extensive functional connectivity and coordinated enrichment, consistent with assembly into related regulatory modules (Figure C). A subset of candidates within this network was further supported by biochemical validation: streptavidin-enriched Western blotting confirmed enrichment of proteins such as YAP1, MAML1, TRIM24, and KMT2A in hSOX10 fusion samples relative to controls (Figure D).

Transcriptomic and Genomic Integration Identifies hSOX10-Associated Regulatory Candidates in Melanoma

To further refine biological relevance beyond functional enrichment, transcriptomic and genomic data were integrated for the top 180 hSOX10-proximal candidate protein interactors, corresponding to Biological Relevance Filtering Criteria 2 and 3 for Phase 2 of the analytical framework (Figure ).

Transcriptomic coexpression analysis (Biological Relevance Filtering Criterion 2) was performed using GEPIA2 to evaluate the relationship between candidate protein interactor gene expression and hSOX10 expression across TCGA cutaneous melanoma samples (Figure A). Candidates were stratified based on the strength and direction of Pearson correlation coefficients, enabling prioritization of proteins exhibiting strong positive coexpression with hSOX10 (e.g., ARID1A, SMARCA4, BRD4, EP300, and MAML1) as potential components of hSOX10-associated transcriptional programs (Figure A). Thirteen previously reported hSOX10-associated proteins were enriched among positively correlated candidates, supporting the biological plausibility of this criterion.

5.

Integrative transcriptomic and genomic analyses of the top 180 candidate hSOX10 protein interactors in melanoma. (A) GEPIA2 gene expression correlation analysis of the top 180 high-confidence candidate hSOX10 protein interactors across TCGA cutaneous melanoma samples. Pearson correlation coefficients (R) and corresponding significance values are shown relative to hSOX10 gene expression. Candidates are stratified into tiers based on the strength and direction of coexpression, with positively correlated candidates prioritized as biologically relevant to hSOX10-associated transcriptional programs. (B) cBioPortal-based genomic co-occurrence and mutual exclusivity analysis of the same candidate set, plotting log₂ ORs versus -log­(q-values) to identify genes whose genomic alterations significantly co-occur with hSOX10 alterations in melanoma. Candidates enriched for co-occurrence are highlighted as high-confidence, potentially cooperative regulators. (C) cBioPortal-reported mutation frequencies for the top 180 candidate hSOX10 protein interactors across melanoma datasets, illustrating the distribution of recurrent genomic alterations within the candidate set and highlighting the most frequently altered genes. Previously reported hSOX10 protein interactors are outlined and labeled in panels A and B. Abbreviations: GEPIA2, Gene Expression Profiling Interactive Analysis 2; MS, mass spectrometry; R, Pearson correlation coefficient; TCGA, The Cancer Genome Atlas.

Genomic co-occurrence and mutual exclusivity analyses using cBioPortal (Biological Relevance Filtering Criterion 3) were conducted to assess patterns of genomic alteration among the same candidate sets. Log₂ ORs and corresponding q-values were used to identify genes whose alterations significantly co-occur with hSOX10 alterations across melanoma datasets, highlighting candidates that may participate in shared or cooperative oncogenic pathways, including KMT2D, BRD4, SMARCA4, EP300, and NCOA family members (Figure B). Mutation frequency distributions across the 180 candidates illustrated the prevalence of recurrent genomic alterations among hSOX10-proximal proteins and highlighted higher mutational frequencies in a subset of proteins including ZFHX4, MECOM, KMT2D, ARID1A, and ARID2 (13–25%) (Figure C).

Integrated Biological Relevance Filtering and Evidence-Based Prioritization of the hSOX10 Proximal Protein Interactome

To systematically refine the hSOX10 proximal protein interactome and identify candidates most strongly supported by orthogonal lines of biological evidence, the Phase 2 Biological Relevance Filtering framework introduced in Figure was applied to the 180 high-confidence hSOX10 candidate protein interactors. Candidates were retained if they satisfied at least one of the following biological relevance filtering criteria derived from the preceding analyses: (i) functional organization within transcriptional or chromatin-associated groups or prey–prey interaction networks identified in the proximity labeling dataset (Figure B–C), (ii) strong transcriptional coexpression with hSOX10 across TCGA cutaneous melanoma samples (Figure A; Tiers 1 or 2), or (iii) high-confidence genomic co-occurrence with hSOX10 based on cBioPortal analyses (Figure B). Application of these criteria yielded a final refined set of 124 biologically relevant candidate protein interactors, including hSOX10, which were carried forward without further exclusion (Table S5).

These 124 retained candidates were visualized as an integrated hSOX10-centered network incorporating proximity labeling enrichment, functional and network context, transcriptomic coexpression, genomic co-occurrence, melanoma mutation frequency, and chromatin-based evidence (Figure ). As an additional contextual layer, publicly available ChIP-Atlas datasets were queried to assess evidence of hSOX10 genomic occupancy across aggregated pan-cancer and melanoma-specific ChIP-seq experiments. All 124 candidates exhibited evidence of hSOX10 genomic occupancy in aggregated pan-cancer datasets, with average binding scores ranging from 10.4 to 1124.1 (mean 161.1). While the single human cutaneous melanoma dataset identified relatively few targets, uveal melanoma datasets supported hSOX10 binding for 113 of 124 candidates.

6.

Integrated network representation of the final biologically relevant hSOX10-proximal candidate protein interactor set (n = 124, including hSOX10), generated by applying orthogonal transcriptomic, genomic, and network-based criteria to the top 180 high-confidence candidates. Candidates were retained if they met at least one of the following biological relevance filtering criteria: (i) membership in transcriptional regulatory complexes or functional groups enriched within the proximity labeling dataset, including transcriptional coregulators and chromatin-associated complexes, and/or enrichment for prey–prey interactions; (ii) strong transcriptional coexpression with hSOX10 based on GEPIA2 gene expression correlation analysis (Tier 1 or Tier 2 candidates); or (iii) high-confidence genomic co-occurrence with hSOX10 based on cBioPortal mutual exclusivity analysis. Functionally connected candidates belonging to known complexes or regulatory groups are shown in the upper network, whereas the remaining retained candidates lacking functional connectivity within the 124-candidate set are displayed separately as individual nodes (lower network). Node size reflects melanoma mutation frequency, and node color denotes evidence of hSOX10 genomic binding derived from pan-cancer ChIP-Atlas analyses. Node outline color indicates melanoma-specific identification of hSOX10 target genes based on ChIP-Atlas analyses, distinguishing targets identified in both uveal and epidermal datasets (burgundy), only uveal melanoma datasets (blue), or not detected in any of the available melanoma datasets (gray). Edge colors indicate previously reported hSOX10 interactions (green), novel candidate associations (blue), and known prey–prey interactions (gray), while edge styles denote support from gene coexpression (dashed) or genomic co-occurrence (dotted) analyses, or both (mixed dashed and dotted). Abbreviations: hSOX10, human SOX10; GEPIA2, Gene Expression Profiling Interactive Analysis 2; MACS2, Model-based Analysis of ChIP-Seq 2.

Several chromatin regulators and transcriptional cofactors, including ARID1A, ARID1B, SMARCA4, BRD4, EP300, KMT2D, and CIC, exhibited convergence across multiple dimensions of evidence, highlighting them as particularly well-supported candidate protein interactors. To systematically capture this convergence across the entire refined dataset, a Phase 3 Integrative Prioritization step was applied to the final 124 candidates without additional filtering (Figure A). Each candidate was evaluated against five orthogonal criteria derived from earlier analyses: (i) functional or network-based support within the proximity labeling dataset, (ii) transcriptional coexpression with hSOX10, (iii) genomic co-occurrence with hSOX10, (iv) evidence of hSOX10 genomic binding in aggregated ChIP-Atlas datasets, and (v) evidence of hSOX10 binding in melanoma-derived ChIP-seq datasets (Figure B). The majority of candidates were supported by multiple independent criteria. Over half satisfied three criteria, with an additional 21.8% meeting four and 8.1% meeting all five criteria. Visualization of this evidence-based stratification highlights the heterogeneity in support across the retained candidates and provides a transparent framework for interpreting the relative confidence within the final hSOX10 proximal protein interactome (Figure C).

7.

Evidence-based stratification of the final hSOX10 proximal protein interactome. (A) Schematic overview of the multiphase analytical workflow used to derive the final hSOX10 proximal protein interactome, highlighting three major phases. Phase 1 (Technical and Statistical Filtering) and Phase 2 (Biological Relevance Filtering) comprise five sequential filtering steps that progressively refine the initial proteomic dataset to 124 high-confidence, nuclear-enriched, hSOX10-associated candidates. Phase 3 (Integrative Prioritization) applies an evidence-based stratification step to the final 124 retained candidates without further exclusion. (B) Summary of the five orthogonal prioritization criteria used in Phase 3 to stratify the 124 retained candidate hSOX10 protein interactors. Bar plots show the percentage of candidates meeting each individual criterion (left panel) as well as the distribution of candidates by the total number of criteria satisfied (right panel). (C) Treemap representation of the 124 candidate hSOX10 protein interactors grouped by the number of prioritization criteria met (1–5). Within each group, individual proteins are listed, and the specific criteria satisfied are indicated in the corner of each region. Abbreviations: GEPIA2, Gene Expression Profiling Interactive Analysis 2; hSOX10, human SOX10.

Together, these analyses define a rigorously filtered yet inclusive hSOX10 proximal protein interactome that integrates proteomic, transcriptomic, genomic, and chromatin-based contexts. Rather than asserting direct interactions or regulatory mechanisms, this resource provides a structured, evidence-weighted foundation for prioritizing candidates for future functional and mechanistic studies.

Discussion

This study represents the first dedicated, comprehensive characterization of the hSOX10 proximal protein interactome, leveraging mT-based proximity biotinylation coupled to MS. By profiling the protein microenvironment surrounding hSOX10 in A375 melanoma cells, we identified 847 melanoma-enriched candidate protein interactors. Through stringent statistical filtering, contaminant frequency profiling, and subcellular localization annotation, this dataset was refined to 180 high-confidence, nuclear-enriched candidates. Integration of transcriptomic coexpression, genomic co-occurrence, and hSOX10 ChIP-seq binding evidence further converged on a prioritized network of 124 candidate protein interactors. Together, these analyses define a structured, high-confidence resource that captures the regulatory protein neighborhoods within which hSOX10 operates in melanoma.

Although SOX10 is not among the most frequently mutated genes in melanoma, extensive evidence indicates that its dysregulation occurs predominantly through nonmutational mechanisms, including altered expression levels, enhancer rewiring, post-translational modification, and context-dependent changes in cofactor engagement. ,,,− SOX10 expression is tightly linked to melanoma lineage identity and proliferative states, and its modulation promotes dynamic transitions toward invasive, therapy-resistant phenotypes. These observations underscore the importance of regulatory context, rather than genetic alteration alone, in shaping SOX10 activity in melanoma and provide a strong rationale for defining its proximal protein interactome.

Proximity-dependent labeling identifies proteins within a defined spatial radius of the bait protein and does not distinguish direct physical interactions from indirect, transient, or complex-mediated associations. Accordingly, the proximal associations reported here should be interpreted as hypothesis-generating rather than as evidence of direct binding or causality. This approach is particularly well-suited for TFs such as hSOX10, whose regulatory activity is mediated by dynamic, context-dependent protein assemblies that are often difficult to capture by traditional AP-based interaction assays. ,,, In this context, the hSOX10 proximal protein interactome provides a biologically relevant view of the regulatory environments that support hSOX10-driven transcriptional programs in melanoma.

hSOX10 Operates within Chromatin- and Enhancer-Centered Regulatory Environments

A dominant feature of the hSOX10 proximal protein interactome is the strong enrichment of transcriptional regulators, chromatin remodeling complexes, and enhancer-associated cofactors. These findings align with hSOX10’s established role as a lineage-defining TF that governs melanocyte identity, melanoma proliferation, and phenotypic plasticity. ,,, Rather than acting in isolation, TFs typically function within highly dynamic, multiprotein neighborhoods that integrate chromatin accessibility, enhancer architecture, and upstream signaling. − Large-scale proximity proteomics efforts demonstrate that such environments are often dominated by indirect or transient interactions that reflect functional regulatory context rather than stable binary complexes.

The composition of the hSOX10 proximal protein landscape described here supports a model in which hSOX10 operates within chromatin-centered regulatory hubs that coordinate enhancer activity and transcriptional output. The convergence of chromatin remodelers, transcriptional coregulators, and architectural factors suggests that hSOX10-associated regulatory programs are shaped by higher-order chromatin organization and dynamic cofactor engagement, features that are central to melanoma cell state plasticity and adaptive behavior.

Chromatin-Remodeling and Enhancer-Regulatory Complexes as Core Components of the hSOX10 Proximal Protein Interactome

Among the most prominent features of the hSOX10 proximal protein interactome is the strong enrichment of SWI/SNF chromatin remodeling complexes. Multiple SWI/SNF subunits were identified among the top 180 high-confidence candidate protein interactors and remained highly represented within the final 124-candidate network, including both known hSOX10 partners (e.g., ARID1A, ARID1A, and ARID2) and novel candidates (e.g., SMARCA4, SMARCC1, SMARCC2, SMARCE1, BICRA/GLTSCR1, and SMARCD3). SWI/SNF complexes play vital roles in regulating chromatin accessibility, enhancer activation, and lineage specification, and their dysregulation is common in melanoma, where it influences transcriptional programs, tumor progression, and therapeutic response. − The robust representation of SWI/SNF components within the hSOX10 proximal landscape is therefore consistent with prior evidence that hSOX10 cooperates with chromatin remodelers to regulate melanocyte and melanoma gene expression.

In parallel, components of the MLL/COMPASS family, particularly members of the MLL3/4 (KMT2C/KMT2D) complex, also emerged as high-confidence proximal protein interactors, including KMT2A, KMT2C, and KMT2D. MLL3/4 complexes are key regulators of enhancer-associated histone modifications and are required for establishing and maintaining active enhancer states. Loss-of-function alterations in MLL3/4 components can reshape enhancer landscapes and promote transcriptional reprogramming in multiple malignancies. − Their proximity to hSOX10 suggests potential convergence at lineage-defining enhancer elements and supports a model in which hSOX10 operates within enhancer-regulatory hubs that integrate nucleosome remodeling and histone modification.

Lineage Transcription Factor Crosstalk and Chromatin Architecture

Beyond chromatin remodeling machinery, the hSOX10 proximal protein interactome highlights TFs and architectural regulators that may modulate hSOX10-driven lineage programs. Notably, SOX5 emerged as a prioritized candidate supported by multiple independent lines of evidence. SOX5 is a member of the SOX family with established roles in NC development and has been shown to antagonize SOX10-dependent transcriptional programs, including repression of MITF expression in melanoma models. − Its proximity to hSOX10 suggests potential crosstalk between SOX family members within shared regulatory environments, providing a plausible mechanism for fine-tuning lineage identity and phenotype switching in melanoma.

NIPBL, the cohesin loader required for establishing chromatin loops and enhancer-promoter communication, was also identified as a prioritized candidate. Recent studies have demonstrated that TFs can recruit NIPBL to open chromatin regions to facilitate cohesin loading and three-dimensional genome organization. − The proximity of NIPBL to hSOX10 supports a model in which hSOX10-associated regulatory environments may extend beyond local chromatin remodeling to influence higher-order chromatin architecture and enhancer-promoter interactions that govern transcriptional output.

Transcriptional Coregulators and Chromatin Repression within the hSOX10 Landscape

The hSOX10 proximal protein interactome also includes multiple transcriptional coregulators that balance activating and repressive chromatin states. EP300, a lysine acetyltransferase previously shown to stabilize and potentiate SOX10 activity in melanoma, , and NCOR2, a nuclear corepressor implicated in chromatin remodeling and melanoma prognosis, were identified among high-confidence candidates. Genetic variation in NCOR2 has been associated with cutaneous melanoma outcomes, suggesting a role in modulating transcriptional programs relevant to disease progression. − Their presence within the hSOX10 proximal protein interactome is consistent with their previously detected hSOX10 interactions and established roles in transcriptional regulation, though direct mechanistic links remain to be defined.

In addition, ELMSAN1 and WIZ emerged as prioritized candidates. ELMSAN1 is a core component of the MiDAC histone deacetylase complex and functions as a scaffold for HDAC1/2-mediated transcriptional repression. , Its proximity to hSOX10 suggests that hSOX10-associated regulatory environments may incorporate defined deacetylase complexes to dynamically regulate chromatin accessibility and transcriptional activity. WIZ, a recruiter of the G9a/GLP histone methyltransferase complex responsible for H3K9 dimethylation, , was also detected among prioritized candidates, further supporting the presence of both activating and repressive chromatin-modifying activities within hSOX10-proximal regulatory hubs.

Emerging Candidate Regulators

Beyond these established regulators, our integrative prioritization strategy highlighted several transcriptional regulators with strong multidimensional support, including MECOM, BRD4, and ZFHX4. MECOM − and BRD4 − are well-established enhancer-associated regulators implicated in transcriptional plasticity and melanoma progression, and their proximity to hSOX10 likely reflects shared localization within transcriptionally active chromatin environments. Notably, ZFHX4 − emerged as a high-confidence candidate supported by multiple orthogonal criteria and displays the highest mutation frequency among prioritized candidates in melanoma cohorts (25%, cBioPortal). Although its role in melanoma remains poorly defined, ZFHX4 has been linked to transcriptional regulation and immune modulation, warranting further functional investigation.

Resource Value, Limitations, and Future Directions

A central goal of this work was to generate a high-confidence, integrative resource that defines the hSOX10 proximal protein interactome in melanoma and provides a rational framework for prioritizing candidates for future mechanistic studies. By combining proximity labeling with stringent proteomic filtering, contaminant frequency profiling, subcellular localization context, and biological relevance criteria (functional enrichment and network context, transcriptomic coexpression, and genomic co-occurrence), we refined an initial proximity network to a biologically relevant set of 124 candidate hSOX10 protein interactors.

Several limitations must be acknowledged. Analyses were performed in a single, BRAF-mutant SOX10-dependent cutaneous melanoma. While this model is highly relevant, the hSOX10-proximal landscape may vary across melanoma subtypes and contexts. Second, proximity labeling captures spatial proximity rather than direct binding, necessitating an orthogonal biochemical and biophysical validation. In addition, the two hSOX10 fusion constructs differed in expression levels and labeling efficiency, resulting in datasets of unequal size and depth. To mitigate the effects of this, we focused downstream analyses on shared candidates and on stringent statistical filters.

Despite these limitations, the strong convergence of proteomic, transcriptomic, genomic, and chromatin-based evidence supports the robustness of the prioritized network. In summary, this study establishes a high-confidence map of the hSOX10 proximal protein interactome in melanoma and integrates multiple orthogonal data types to prioritize a coherent set of candidate cofactors and pathways. By providing a structured view of the protein neighborhoods within which hSOX10 operates, this work lays a foundation for future studies interrogating chromatin remodelers, transcriptional cofactors, and architectural regulators across diverse melanoma models and perturbation conditions, with the ultimate goal of identifying context-specific vulnerabilities that underlie melanoma plasticity, progression, and treatment resistance.

Conclusions

This study provides the first dedicated, comprehensive map of the hSOX10 proximal protein interactome. Using miniTurbo-based proximity biotinylation coupled with mass spectrometry, we identified a high-confidence network of 124 candidate hSOX10 protein interactors in human melanoma cells. These included both known interaction partners and novel candidates enriched for transcriptional regulators, chromatin remodeling and enhancer-associated complexes, and post-translational regulatory machinery. These findings position hSOX10 within chromatin-centered regulatory neighborhoods that support melanoma cell identity and plasticity. Collectively, this work provides a robust, hypothesis-generating resource to guide future mechanistic studies of hSOX10-dependent transcriptional programs and potential therapeutic vulnerabilities in melanoma and other hSOX10-dependent contexts.

Glossary

Abbreviations

ABC

Ammonium bicarbonate

AGC

Automatic gain control

AP

Affinity purification

ATCC

American Type Culture Collection

BCA

Bicinchoninic acid

ChIP-seq

Chromatin immunoprecipitation followed by sequencing

CRAPome

Contaminant Repository for Affinity Purification

DIM

Dimerization (domain)

DMSO

Dimethyl sulfoxide

DMEM

Dulbecco’s Modified Eagle Medium

DNA

Deoxyribonucleic acid

DTT

Dithiothreitol

EDTA

Ethylenediaminetetraacetic acid

FA

Formic acid

FBS

Fetal bovine serum

FC

Fold change

FDR

False discovery rate

GEPIA2

Gene Expression Profiling Interactive Analysis 2

GO

Gene Ontology

GRN

Gene regulatory network

GW

Gateway (cloning)

HBS

HEPES-buffered saline

HDAC

Histone deacetylase

HMG

High-mobility group (domain)

HPA

Human Protein Atlas

HPLC

High-performance liquid chromatography

kDa

Kilodalton

LDS

Lithium dodecyl sulfate

LFQ

Label-free quantification

M

Molar

MACS2

Model-based Analysis of ChIP-Seq 2

mAU

Milli-Absorbance Unit

mT

miniTurbo (proximity-dependent biotinylation enzyme)

MS

Mass spectrometry

NC

Neural crest

NLS

Nuclear localization signal

nm

Nanometer

OR

Odds ratio

PBS

Phosphate-buffered saline

PCA

Principal component analysis

pLDDT

Predicted local distance difference test

PPI

Protein–protein interaction

PVDF

Polyvinylidene difluoride

R

Pearson correlation coefficient

RIPA

Radioimmunoprecipitation assay (buffer)

RT

Room temperature

SDS

Sodium dodecyl sulfate

SKCM

Skin cutaneous melanoma

SOX10/hSOX10

SRY-Box Transcription Factor 10/human SOX10

STRING

Search Tool for the Retrievel of Interacting Genes/Proteins (database)

SUMO­(ylation)

Small Ubiquitin-like Modifier (conjugation)

TAM/TAC

Transactivation (domains)

TBS

Tris-buffered saline

TBS-T

Tris-buffered saline with Tween-20

TCGA

The Cancer Genome Atlas Program

TF

Transcription Factor

UBC

Ubiquitin C (promoter)

V

Volts

WB

Wash buffer

The raw MS proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) with the dataset identifier PXD067868.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.5c00852.

• AlphaFold 3-predicted structures and associated confidence metrics for hSOX10, mT-hSOX10, hSOX10-mT, and mT (Figures S1–S4); functional enrichment analysis of the hSOX10 proximal protein interactome using Gene Ontology Biological Process annotations (Figure S5); and Reactome pathway enrichment analysis of the hSOX10 proximal protein interactome (Figure S6) (PDF)

• ClueGO Gene Ontology Biological Process enrichment results for the hSOX10 proximal protein interactome (Table S1) (XLSX)

• ClueGO Reactome pathway enrichment results for the hSOX10 proximal protein interactome (Table S2) (XLSX)

• Detailed cBioPortal mutation and alteration annotations for individual candidate interactors (Table S3) (XLSX)

• ChIP-Atlas transcription factor binding and chromatin enrichment data associated with candidate interactors (Table S4) (XLSX)

• Comprehensive annotation and prioritization table summarizing all candidate interactors, including the full set of identified proteins and successive filtered subsets (847, 207, 180, and 124 candidates), along with associated proteomic, transcriptomic, genomic, and functional metadata (Table S5) (XLSX)

C.M.N.S. and C.K.K. developed the research question and overall research plan; D.P.B. and M.B.M. provided guidance and training, experimental protocols, proteomics expertise, and key reagents; C.M.N.S executed and troubleshot most experimental protocols; D.P.B. conducted mass spectrometry sample preparation, data acquisition, and initial computational analyses; C.M.N.S and C.K.K. conducted additional analyses and performed literature reviews; C.M.N.S. and D.P.B. prepared manuscript figures. All authors contributed to manuscript preparation.

Research reported in this publication was supported in part by the National Cancer Institute of the National Institutes of Health under award number R01CA240633. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

The authors declare no competing financial interest.