Sex differences in the cancer proteome

WHAT IS ALREADY KNOWN ON THIS TOPIC Sex influences cancer incidence, progression, treatment response and tumour genomic and transcriptomic profiles, but its impact on the cancer proteome remains poorly understood.WHAT THIS STUDY ADDS This study characterises sex differences across 1590 proteomes from eight cancer types, showing pronounced protein abundance differences in lung adenocarcinoma and more modest differences in other cancers.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings support considering sex as a biological variable in cancer proteomics and highlight the need for larger, sex-balanced proteogenomic cohorts.Introduction Proteins play a central role in cancer biology as drug targets and as diagnostic, prognostic and predictive biomarkers—largely quantified via immunohistochemistry. 1 Proteomics enables high-throughput profiling of protein abundance, structure and modifications, directly quantifying the terminal products of gene expression.2 The proteome is highly dynamic and is shaped by both genetic and environmental factors. Understanding what determines proteomic characteristics is critical for identifying reliable biomarkers for precision oncology and in mechanistic studies.Sex influences gene expression in normal cells. For example, proteomic studies identified thousands of proteins with sex-differential abundance in brain and cerebrospinal fluid, many of which are associated with psychiatric and neurodegenerative conditions.3 4 Blood proteomic studies revealed substantial sex differences in protein abundance associated with cardiovascular disease and obesity.5 6 These findings highlight the importance of considering sex as a factor in proteomic studies, and yet the impact of sex on the cancer proteome remains largely unknown.Indeed, sex has well-documented effects on cancer, with extensive research characterising sex differences at the genomic level. For example, tumours arising in male patients exhibit higher somatic mutation burdens in cancers such as bladder, melanoma, kidney and liver but lower in glioblastoma (GBM).7 Copy number aberrations (CNAs) differ by sex and are associated with clinical phenotypes.8 In colon cancer, genes with sex-differential essentiality have been identified.9 These findings suggest sex-specific somatic mutations may at least partially explain sex differences in clinical presentation, including in disease incidence, progression, survival and treatment response.10–16 Previous pan-cancer analyses using TCGA data further identified sex differences in the transcriptome across multiple cancer types, including lung adenocarcinoma (LUAD),kidney renal clear cell carcinoma (KIRC) and head and neck squamous cell carcinoma (HNSC).17 Integrative studies, including OncoSexome, have highlighted widespread sex differences across cancer transcriptomic, genomic, immune and metabolic landscapes.18 However, the extent to which these well-characterised sex differences at the genomic and transcriptomic levels are translated into variations in protein abundance remains largely unknown.To fill this gap, we analysed 1590 high-resolution mass spectrometry-derived proteomes from 934 cancer patients representing eight cancer types. We quantified sex differences in protein abundance in each cancer type using univariable and multivariable statistical modelling, adjusting for key confounders like age and stage. We identified sex differences in protein abundances in six cancer types, with distinct biological pathways influenced by sex. These findings underscore the importance of understanding the influence of sex on cancer molecular profiles, as it can inform patient stratification and guide the development of more effective, tailored treatments.Results Sex differences in protein abundance We analysed 1590 proteomes across eight cancer types from the Clinical Proteomic Tumour Analysis Consortium (CPTAC). These comprised 916 tumour tissue specimens and 674 normal adjacent tissues (NATs; figure 1A). Cancer types, abbreviations and sample numbers are in online supplemental table 1. The number of samples analysed ranged from 109 in GBM to 314 in liver hepatocellular carcinoma (LIHC).19–26 There were 73±21 male and 34±20 female patients per cancer type (median±median absolute deviation (MAD)). On average, 8890±637 proteins were detected and quantified per patient. We applied a statistical approach modified from an established two-stage analysis workflow to assess sex differences in protein abundances.7 8 27 28 Briefly, we first excluded genes with sex differences in normal tissues, followed by univariable analysis to identify putative sex-differential proteins and finally multivariable analysis to control for confounding variables including age and stage (figure 1B; online supplemental table 1).SP210.1136/bmjonc-2026-001130.supp2Supplementary dataFigure 1Cancer-type variability in proteomic sex differences and hallmark enrichment in lung adenocarcinoma (LUAD). (A) Overview of sex differences in cancer proteome across eight cancer types: LUAD, liver hepatocellular carcinoma (LIHC), kidney renal clear cell carcinoma (KIRC), pancreatic ductal adenocarcinoma (PAAD), lung squamous cell carcinoma (LUSC), colon adenocarcinoma (COAD), head and neck squamous cell carcinoma (HNSC) and glioblastoma (GBM). From left to right, the number of female (pink) and male (blue) patients in each cancer type, number of genes with sex-differential copy number status (red for gain and steel blue for loss), total number of proteins quantified and number of genes with significant protein abundance favouring males (blue) and females (pink; q value < 0.1) in normal adjacent tissues (NATs) and tumours. (B) Statistical workflow for identifying proteins with sex-differential abundance in tumours. (C) Histogram of p values from t-tests comparing protein abundance between sexes in each cancer type. (D) Heatmap showing relative protein abundance of top 35 genes with significant sex differences in protein abundance (q value < 0.01, effect size >0.45 or <−0.45) in LUAD. The top annotation panel indicates patient sex, age and cancer stage. Barplot to the right shows the q values from multivariable linear regression (MLR). (E, F) Top 20 cancer hallmarks enriched among genes with male-biased (E) and female-biased (F) protein abundance in LUAD (q value < 0.05). Dot size indicates the number of genes mapped to each cancer hallmark gene set, while colour denotes the q value.We first analysed sex differences in protein abundance in NATs, finding that very few proteins exhibited significant sex-differential abundance (q value <0.05; figure 1A; online supplemental table 1), with the exception of LUAD and LIHC, where 369 and 352 proteins showed significant sex difference, respectively; all other cancer types had two or fewer. To assess sex differences in protein abundance of autosomal genes in tumours, we first employed a univariable approach across the eight cancer types after excluding the few proteins that exhibited sex-differences in tissue-matched NAT. We observed pronounced rightward skews in p value distributions for KIRC, LIHC, LUAD and pancreatic adenocarcinoma (PAAD), indicating strong sex differences in these cancers (figure 1C; online supplemental figure 1). Moderate but clear rightward skews were observed for HNSC and lung squamous cell carcinoma (LUSC; figure 1C; online supplemental figure 1).SP110.1136/bmjonc-2026-001130.supp1Supplementary dataPutative sex-differential proteins (q value < 0.1, t-test) were further adjusted using multivariable linear regression (MLR) for confounders known to impact gene expression,27 28 including age, race, smoking, body mass index (BMI), tumour stage and grade (online supplemental table 1). In LUAD, 901 proteins remained significant (q value < 0.05, online supplemental table 2), with the top 35 genes (q value < 0.01 and effect size >0.45 or <−0.45) shown in figure 1D. Several of these genes have established roles in tumour biology. Among male-biased proteins, CCNA2 and MKI67 (q value=7.7×10−3 and 1.1×10−3; effect size=0.65 and 0.52) are key regulators of cell cycle progression and proliferation, commonly upregulated in LUAD and associated with poor prognosis.29 30 PRC1 and TPX2 (q value=7.5×10−3 and 3.9×10−4; effect size=0.55 and 0.54) are involved in mitotic spindle organisation and chromosomal stability,31–33 and SERPINB2 (q value=6.0×10−3; effect size=0.47) has been linked to immune modulation and favourable clinical outcomes.34 35 In contrast, several female-biased proteins such as EPDR1, QPRT and ABCC3 (q value=9.4×10−3, 8.1×10−4 and 6.7×10−4; effect size=−0.53, −0.48 and −0.86) have been linked to proliferation, metabolic regulation or drug resistance in lung cancer and other cancer types.36–38 Cancer hallmark enrichment analysis showed genes with significant male-bias are significantly enriched in MYC targets (q value=1.0×10−55), E2F targets (q value=4.2×10−37) and G2M checkpoint (q value=1.1×10−32; figure 1E), while genes with significant female-bias showed enrichment in metabolic and stress-response pathways, including peroxisome, heme metabolism and the p53 pathway (figure 1F).Twenty proteins were identified across LUSC, GBM, PAAD, KIRC and LIHC, with each cohort yielding 1–9 significant proteins (q value < 0.05, figure 2A). No proteins remained significant after MLR adjustment in HNSC and colon adenocarcinoma (COAD) (figure 1A; online supplemental figure 1; online supplemental table 2). Power analyses estimated that at least 187 and 181 samples are needed to detect 20 sex-differential genes in these cohorts based on observed effect sizes ranging from 0.30 to −0.20 in HNSC and 0.13 to −0.12 in COAD (online supplemental figure 2). Thus, while our analysis exploits the largest available proteomics cohorts to date, statistical power to identify sex differences is limited to relatively large effect sizes.Figure 2Association between sex differences in protein abundance, ploidy status, and gene dependency. (A) Boxplots showing genes with significant sex-differential protein abundance (q value < 0.05, multivariable linear regression (MLR)) in lung squamous cell carcinoma (LUSC), glioblastoma (GBM), pancreatic ductal adenocarcinoma (PAAD), kidney renal clear cell carcinoma (KIRC) and liver hepatocellular carcinoma (LIHC). Boxes are coloured by sex (blue for male, pink for female). Barplots to the right show corresponding q value from MLR. (B) Dotmap of genes with significant sex-differential protein abundance and corresponding copy number aberration (CNA) gain or loss in lung adenocarcinoma (LUAD), KIRC, PAAD and LIHC. For each cancer type, the left panel shows MLR results for protein abundance, and the right panel shows the proportion test results for CNAs status. Background shading reflects q values (protein) or p values (CNA), dot size represents effect size (protein) or the difference in the proportion of patients with CNA gain/loss, and dot colour indicates male (light/dark blue) or female (pink/red) bias. Crosses indicate non-significant CNA differences. (C) Volcano plots showing effect sizes and q values for sex difference in protein abundance from tumours derived from patients with LUAD. Effect sizes and q values were calculated using t-test with false discovery rate correction. Light blue and pink indicate a significant bias in protein abundance (q value < 0.1) favouring males and females, respectively. Dark blue and dark pink represent an additional biased CNA for males and females. (D) Boxplots showing gene dependency of sex-differential genes compared with all other genes quantified in LUAD-derived cell lines. The left panel shows cell lines from male patients and the right panel from female patients. Horizontal barplot represents the Mann-Whitney U test p values comparing the two categories of genes. (E) Dot map of Mann-Whitney U test comparing the effect sizes of sex differences in protein abundance between genes with and without sex-differential copy number aberrations (CNAs). Dot colour indicates the direction of effect size (blue for male-biased, red for female-biased), dot size represents effect size of Mann-Whitney U test, and background darkness corresponds to statistical significance (darker indicates lower p values). (F, G) From top to bottom: fold change (FC) in protein abundance (males−females), heatmap of protein relative abundance, sex differences of copy number gain (males−females), proportion value of copy number gain in males (blue) and females (pink) separately, sex differences of copy number loss (males−females) and proportion value of copy number loss in males (blue) and females (pink). Stacked figures are aligned according to the gene coordinates in the human genome.Protein abundance differences in genes with sex-differential CNAs Given the variability of sex differences in protein abundance across cancer types and the known influence of CNAs on gene expression, we next focused on genes with sex-differential ploidy status. Using data from the Pan-Cancer Analysis of Whole Genomes (PCAWG) project, we identified 1939±2100 genes with putative sex-differential CNAs within each of the eight cancer types (proportion test, p value < 0.1 without adjusting for confounders; figure 1A; online supplemental table 3). Given the high correlation among CNAs across the genome, a strict multiple-testing correction was not applied at this stage to avoid excluding biologically relevant signals. We then integrated CNA findings with proteomic data by comparing genes with sex-differential CNAs within each cancer type to those identified through MLR analysis of sex-differential protein abundance, with multiple-testing adjustment. In PAAD, KIRC, LIHC and LUAD, 42 genes with putative sex-differential CNAs also showed significant (q value < 0.1) sex-differential protein abundance (figure 2B). In contrast, no sex-differential CNAs showed significant (q value <0.1) sex-differential protein abundance in COAD, GBM, HNSC or LUSC (online supplemental table 2).Seven genes with sex-differential protein abundance also had putative sex-differential ploidy change in PAAD, KIRC or LIHC, with no overlap across cancer types (figure 2A; online supplemental table 2). RAB42 and RABEP2, two members of the Ras oncogene superfamily, had opposite sex difference in protein abundance and CNA status in KIRC tumours. RAB42 had male-biased protein abundance (q value=8.2×10−5; effect size=1.02) and higher proportion of female patients with CNA loss (p value=0.050; proportion difference (male−female)=−0.12), while RABEP2 had female-biased protein abundance (q value=1.3×10−4; effect size=−0.25) and more male patients with CNA gain (p value=0.05, proportion difference (male−female)=0.15). CAST, encoding calpain and prognostic in pancreatic cancer,39 had elevated protein abundance in males with PAAD (q value=2.0×10−3, effect size=0.14) and higher proportions of CNA gain (p value=0.07, proportion difference (male−female)=0.07) and loss (p value=0.07, proportion difference (male−female)=0.1) in males. CXXC5, a suppressor of the Wnt/β-catenin signalling pathway,40 showed higher protein abundance in females with LIHC (q value=2.7×10−4, effect size=−0.73), but higher male patient proportion with CNA gain (p value=0.08, proportion difference (male−female)=0.11).LUAD tumours had the greatest number of genes with sex-differential protein abundance, however only 35 of the 901 genes showed sex-differential copy number—thus the majority of sex differences in LUAD were not driven by CNAs (figure 2B,C), none of which overlap with genes significant in PAAD, KIRC or LIHC. Six ribosomal proteins (eg, RPS5, RPS7, RPL4, RPL5, RPLP1 and RPL18A) had elevated protein abundance in male LUAD tumours (q values between 5.7×10−3 and 0.033, effect size between 0.11 and 0.42; online supplemental table 2) and varying CNA loss patterns by sex. For example, RPS5 and RPL18A had higher proportion of CNA loss in males (p values=0.013 and 0.080, proportion differences (male−female)=0.34 and 0.43), while RPLP1, RPL4, RPS7 and RPL5 had more CNA losses in females (q values between 0.043 to 0.083, proportion differences (male−female) between −0.25 to −0.34). PFKP, a key regulator of glycolysis and metabolic reprogramming in lung cancer,41 showed higher protein abundance in males (q value=0.023, effect size=0.31) and higher proportion of CNA loss in females (p value=0.011, proportion differences (male−female)=−0.4), highlighting sex-specific differences in metabolic pathways. CLIC6, as mentioned above with increased abundance in females, had more male patients with CNA loss (p value=0.070, proportion differences (male−female)=0.31). These findings suggest distinct biological processes, including protein synthesis, metabolism and immune modulation, influenced by sex in LUAD.Pan-cancer consistency and molecular correlates of sex differences To evaluate the consistency of sex-differential protein abundance across cancer types, we examined the correlation of effect sizes between each pair of cancer types ( online supplemental figure 3). Overall, effect sizes showed weak to moderate correlations (Spearman’s ρ ranging from −0.15 to 0.2). Although sex-differential protein abundance was identified in PAAD, KIRC, LIHC and LUAD, little correlation was observed with other cancer types. GBM and LUSC, which showed the lowest significant sex-differential protein abundance, exhibited the largest positive correlations, suggesting shared regulatory mechanisms influencing sex-differential protein abundance (Spearman’s ρ=0.2, p value=5.1×10−88). These results indicate that sex-differential proteins and the magnitude of differences vary substantially between cancer types.Given the large number of genes with significant sex-differential protein abundance in LUAD, we next investigated whether these differences were mirrored at the RNA level. We analysed transcriptome data from the CPTAC-LUAD cohort and observed concordance between the sex-differential RNA and protein abundance changes across all quantified genes (online supplemental figure 4). Among these, 73 genes showed concordant significant sex differences in both RNA and protein abundance (q value < 0.05), including genes associated with cell cycle regulation, metabolism and immune-related processes (online supplemental figure 5).To investigate whether genes with sex-differential protein abundance exhibit distinct functional importance in cancer cells, we analysed CRISPR gene dependency data from 52 LUAD-derived cell lines. Genes with sex-differential protein abundance showed modestly stronger dependency (lower DepMap Gene Effect values; Mann-Whitney U-test p value < 0.05) compared with other quantified genes in all male-derived and female-derived cell lines (figure 2D). This suggests that genes with sex-differential protein abundance are more likely to contribute to cancer cell fitness and may represent biologically important molecular vulnerabilities in LUAD.To evaluate the clinical relevance of sex-differential proteins, we performed multivariable Cox proportional hazards analyses for proteins with significant sex-differential abundance across the eight cancer types, adjusting for sex, age and tumour stage when available. Although several proteins showed nominal associations with overall survival, none remained significant after multiple-testing correction (online supplemental table 4).To understand the impact of sex differences in genomic alterations on the cancer proteome, we performed integrated analysis with CNA alterations. Previously, we reported sex differences in CNAs in both coding and noncoding regions of the genome at the pan-cancer level and in specific cancer types, including KIRC and LIHC.7 8 We applied the Mann-Whitney U test to compare the effect sizes of sex differences in protein abundance between genes with and without sex-differential CNAs. In LUAD, KIRC, PAAD and LIHC, the cancer types where sex-differential proteins were identified, genes with sex-differential CNA status tended to have lower effect sizes in protein abundance, indicating that their protein levels were generally higher in female tumours (p value from 3.4×10−4 to 0.023; figure 2E). Visualisation of protein abundance differences with CNA gain and loss frequencies across genomic coordinates showed partial concordance between sex-differential CNA patterns and protein abundance differences (figure 2F,G). In contrast, for the other cancer types, no significant enrichment of sex-differential protein abundance was observed among genes with sex-differential CNAs, suggesting that additional regulatory mechanisms contribute to these differences (online supplemental figure 6).Discussion Our analysis of cancer proteomes revealed variable proteomic sex differences across specific cancer types. Pronounced differences were observed in LUAD, while KIRC, PAAD, LIHC, GBM and LUSC showed moderate variability. These may be driven by interactions between sex hormones (eg, oestrogen and androgen), environmental exposures and the tumour microenvironment, which together shape the sex-dependent evolution of cancers. For example, several P450 enzymes, involved in the metabolism of tobacco-derived carcinogens, are regulated by sex hormones and may further contribute to carcinogenesis. 42 Oestrogen and progestin are known to promote angiogenesis in lung cancer,43 altering the tumour microenvironment. X-inactivation escape in certain X-linked genes may also lead to imbalanced expression of autosomal genes between males and females through regulatory mechanisms.44 These factors all contribute to sex differences in protein abundance. Interestingly, the male-dominant lung cancer subtype, LUSC, exhibited minimal sex-related differences in the proteome, suggesting distinct tumour origins and progression pathways in LUAD and LUSC. This highlights the differential influence of endogenous and exogenous factors on tumour evolution and gene expression, leading to varying disease manifestations between sexes.Previous pan-cancer analyses identified extensive sex-associated transcriptomic differences across multiple cancer types, including LUAD, KIRC, LIHC, LUSC and HNSC.17 In contrast, our proteomic analysis revealed substantial sex differences in LUAD, with more modest differences observed in KIRC, LIHC and LUSC. We further observed concordance between sex-associated RNA and protein abundance changes in LUAD. These findings suggest that sex-associated transcriptomic differences are not uniformly translated into proteomic differences across cancer types, and a subset of them are preserved at the proteomic level. In LUAD-derived cell lines, genes with sex-differential protein abundance exhibited stronger CRISPR gene dependency, suggesting functional relevance to cancer cell fitness. However, survival analyses showed no significant associations after multiple-testing correction. Integrated CNA analyses showed only partial concordance between sex-biased genomic alterations and differences in protein abundance, suggesting that additional regulatory mechanisms beyond genomic dosage contribute to sex-associated proteomic differences.Interpreting sex differences in proteomics is intrinsically complex due to clinico-epidemiological variables partially correlated with sex or influenced by gender-associated behaviours. We applied a statistical approach to identify sex-differential proteins, beginning with exclusion of genes showing sex difference in normal tissues, followed by univariable screening in tumours and multivariable regression to adjust for confounders such as age, race, stage, tumour site, smoking and BMI.7 8 The impact of unmeasured factors like alcohol consumption, diet, exercise and environmental exposures is a significant limitation. Ideally, sex differences of tumour molecular signatures would be directly quantified using multivariable models with comparisons of their estimates and CIs. The limited sample sizes of cancer proteogenomic cohorts constrain the ability to derive precise estimates, particularly given these numerous confounders.Indeed, while our study extends our understanding of sex differences from genomics to proteomics, it is fundamentally limited by the number of patients and cancer types analysed. To detect 20 sex-differential proteins with q value < 0.05 in LUSC, at least 211 patients would be required based on a large effect size threshold of >1.86 or <−0.22 (online supplemental figure 2). However, many observed sex differences in protein abundance had modest effect sizes, although they were broadly concordant with transcriptomic differences (online supplemental figure 4). While limited sample size likely reduced statistical power, some sex-associated proteomic differences may also be intrinsically subtle in magnitude. Therefore, these findings should be interpreted cautiously and validated in larger independent cohorts.Additionally, the datasets we used had imbalanced sample sizes, with fewer female patients in most cancers. Future studies should aim for sex-balanced patient cohorts that reflect the sex-specific prevalence of specific cancer types, ensuring unbiased representation and improving the generalisability of findings. Increasing the sample size is also critical for studying sex differences in low-frequency somatic mutations and their impact on the proteome. To improve statistical power, meta-analysis across independent datasets can be leveraged and statistical methods can help adjust for sample imbalance. Integrating clinical and epidemiological data, including hormone levels, environmental exposures and lifestyle factors, will be critical for understanding the drivers of sex-differential proteomics. While collecting comprehensive metadata is challenging, wearable devices could provide continuous, unbiased tracking of lifestyle variables. Additionally, large-scale biobank data can be leveraged to disentangle the interactions between sex, confounders and tumour proteome alterations. Overall, our findings highlight the importance of investigating sex as a key factor in cancer biology and treatment response, paving the way for personalised therapies based on sex-specific molecular profiles.Methods Proteomics and transcriptomics data Protein abundance matrices from eight non-reproductive cancer cohorts (HNSC, LUAD, GBM, KIRC, LUSC, COAD, LIHC and PAAD) were obtained from the CPTAC. 19–26 The processed and normalised protein abundance matrices of tumour and NATs, along with associated clinical information, were downloaded from Proteomic Data Commons (PDC, https://proteomic.datacommons.cancer.gov). Eight samples were excluded from analysis due to failing quality control, either showing high correlation (Pearson’s R >0.9) or being female samples with high Y-chromosomal protein abundance. The demographics (eg, age and weight) and clinical variables (eg, tumour stage) of patients being analysed are in online supplemental table 1. Genes allocated on the sex chromosomes are excluded from analysis as they are sex-biased by nature. RNA-level gene expression data Transcripts Per Million (TPM) from CPTAC’s LUAD cohort was retrieved using the GDCquery function from the R Package TCGAbiolinks (V.3.16), with the parameter project set to ‘CPTAC-3’.Sex difference of CNAs Sex differences of CNAs for the eight cancer types were estimated as previously described. 7 PCAWG consensus CNAs (syn8042988) were categorised into gain, neutral or loss calls per gene. Putative sex-differential genes were identified by comparing the proportions of tumours with gains and losses between sexes using two-tailed proportion tests. No false discovery rate (FDR) correction was applied at this exploratory stage and genes with a p value < 0.1 were selected for further analysis.Sex difference of protein abundance A multistep statistical analysis workflow was used to identify genes with sex-differential protein abundance ( figure 1B). First, t-tests were performed on NAT protein abundance to evaluate sex differences, followed by FDR correction (q value < 0.05). Next, t-tests were applied to tumour samples to assess sex-differential protein abundance, with FDR correction applied (q value < 0.1). Genes meeting this threshold were further adjusted using MLR to control for confounders, including age, tumour stage, race and smoking history,7 8 with FDR correction (online supplemental table 1). MLR results were then annotated with sex-differential CNA status. All statistical tests in this workflow were two-tailed. Cancer hallmark enrichment analysis was performed using the R package hypeR (V.1.10.0)45 and MSigDB gene sets46 47 for genes with q value < 0.1. Mann-Whitney’s test was used to compare the effect sizes of sex differences in protein abundance between genes with and without sex-differential CNAs.Sex differences in RNA abundance in LUAD RNA abundance data (TPM) for all quantified genes in the CPTAC-LUAD cohort were analysed for sex differences. Two-tailed t-tests were performed to compare RNA abundance between male and female tumours for each gene, followed by FDR correction.Power analysis Power analysis was conducted per protein for each cancer type, using the effect size and SE calculated from t-tests, under the assumption of equal standard errors between males and females and equal numbers of male and female patients. To focus on biologically meaningful differences, we restricted the analysis to the top 20% of genes ranked by absolute effect size in each cancer type. The t-value for each protein was calculated as:tij=Effect SizeSE × njwhere n represents the total number of samples with cancer type ianalysed in the study, and j is the sample size of interest. For each sample size j, the t values for all proteins were converted into p values using the degree of freedom j-1, assuming a two-tailed test. These p values were then adjusted to q values using the FDR method. The analysis was iteratively repeated across varying sample sizes to determine the number of proteins identified with significant sex differences (q value < 0.05) at each sample size.Gene dependency analysis Gene dependency data for LUAD-derived cell lines were obtained from the Cancer Dependency Map (DepMap, 26Q1 release). Genes with significant sex-differential protein abundance in LUAD (q value < 0.05, MLR) were compared against all other quantified genes using DepMap CRISPR Gene Effect scores. 48 Mann-Whitney U tests were performed separately within each cell line.Survival analysis Multivariable Cox proportional hazards models were used to evaluate associations between protein abundance and overall survival for genes with significant sex-differential protein abundance across the eight cancer types (q value < 0.05, MLR). Models were adjusted for sex, age and tumour stage when available. P values were adjusted using the FDR method.Statistical reporting Results are reported as median values followed by the MAD, unless otherwise specified. All boxplots show all data points, the median (centre line), upper and lower quartiles (box limits) and whiskers extend to the minimum and maximum values within 1.5 times the IQR. All comparisons were performed on biological replicates, defined as independent patients or tumours as appropriate to each analysis.Data availability Protein abundance matrices were downloaded from PDC with the following accession numbers: PDC000221 (HNSC), 49 PDC000153 (LUAD),50 PDC000204 (GBM),51 PDC000127 (KIRC),52 PDC000234 (LUSC),53 PDC000109 (COAD),54 PDC000198 (LIHC)55 and PDC000270 (PAAD).56 RNA-seq data for LUAD are available on Genomic Data Commons (GDC, Project: CPTAC-3, Primary Site: Lung, Primary Diagnosis: Adenocarcinoma, NOS). Data used in this publication were generated by the National Cancer Institute CPTAC. CRISPR gene effect scores for LUAD-derived cell lines were downloaded from the DepMap portal (26Q1 release).Code availability All statistical analysis and data visualisation were performed in the R statistical environment (V.4.0.2). All visualisations were performed using the BoutrosLab.plotting.general package (V.6.0.3). 57 Data analysis and visualisation scripts are available on request.