We previously applied the Joint and Individual Variance Explained (JIVE) algorithm to decompose male and female glioblastoma transcriptome data into components shared among males and females, and those unique to each sex [26]. This approach identified “sex-specific” gene expression patterns and sex-based molecular subtypes of GBM. While informative, the JIVE approach has limited clinical utility because it requires categorical assignment of gene expression to “male-specific”, “female-specific”, and shared components. “Sex” as a categorical variable has important but limited value when investigating the spectrum of sex and gender differences or attempting to stratify individual patients for sex and gender-informed treatments. Thus, we sought a method for generating individual patient-specific values along an axis that traversed between female and male cancer transcriptional sex and gender “poles”.

We created UMAPs from the transcriptomes of each cancer based on similarities in gene expression. The head and neck squamous carcinoma (HNSC) UMAP, illustrates the process of identifying poles in the data (Fig. 1A). We discovered local areas in the UMAP where samples were primarily male (blue exes) or female (red circles). We quantified the local sex and gender enrichment in the transcriptome-similarity space using the TI values as described in the methods section, and identified the most skewed regions as poles (filled red circle and filled blue square) to detect skewing in cancer transcriptomes.

We derived TI values for all individual cases and median TI values in every cancer population separately (Supplemental Fig. 1). Then, we combined all 7881 adult TI values to create a pan-cancer TI population distribution (Fig. 1B). As can be seen, the female and male values are skewed. Further, cases with TI values below 0.25 are exclusively female while only males have values above 0.75. These cases represent the female and male population poles, respectively. It is also clear from the data that a large fraction of the cancer population possesses TI values between 0.25 and 0.75. We can expect that as female cases approach TI values of 0.5, they represent a changing balance between pole effects that will mechanistically differ from those in male cases approaching the same TI value.

As expected, median TI values for each cancer type positively correlated with their incidence rate ratios (IRR, M:F) as calculated from these datasets. As illustrated in Fig. 1C, esophageal carcinoma (IRR = 4.22) and thyroid carcinoma (IRR = 0.41) exhibit median TI values of (0.65) and (0.41), respectively. Similarly, the other cancers with IRRs of less than 1 (sarcoma, adrenocortical carcinoma, diffuse large B cell lymphoma, thyroid carcinoma) exhibit median TI values of less than 0.50 (Supplemental Table 1). Regression analysis of IRR versus median TI identified a significant correlation between the two (Fig. 1D). Thus, TI value distributions are concordant with sex and gender differences in individual cancer IRRs. Importantly, TI value indicates that many male individuals with IRR < 0.5 cancers exhibit individual TI values that are shifted towards the female pole and that many female individuals with IRR > 0.5 cancers exhibit individual TI values that are shifted towards the male pole. This does not mean that some female cancers are “male-like”, and some male cancers are “female-like.” Instead, it highlights the shortcomings of these classifications and ambiguities that can arise with their use. The TI value simply describes a phenotype like a taller than average female or a shorter than average male.

These data indicate that individual transcriptomic variation across cancer types retains signatures of sex and gender, and suggest that sexual differentiation may have foundational effects on cancer phenotypes. Thus, we next sought to identify the genes and pathways that define the high and low TI poles. We did so by looking for consistency in sex and gender-skewed mechanisms across cancer types using the 7881 adult and in parallel, the 1069 pediatric cases. Those genes with the greatest effect on low and high TI values were identified by performing association analysis between TI and gene expression. Genes that were significantly (FDR < 0.05) associated with high TI were identified as “male-skewed genes”, while those negatively associated with high TI were identified as “female - skewed” (Supplemental Table 2).

Cancer transcriptomes exhibit skewing by sex and gender. (A) UMAP of 566 HNSC transcriptomes clustered by similarity. Male: Female Incidence rate ratio is shown. Male (blue X’s) and female (Red circles) distribute throughout the transcriptional space. Local enrichments for male and female transcriptomes were recognized and quantified to define female (filled red circle) and male (filled blue square) poles of gene expression. TI value is color-coded and confidence in the posterior probability is indicated by symbol size as indicated. (B) Ridge plots of the TI value distributions for Male, Female, and All patients from the PANCAN data (7881 total, 4668 M, 3213 F., M: F IRR = 1.45:1), with vertical lines indicating the 5%, 25%, 75% and 95% quantiles, respectively. (C) Ridge plots for TI population distributions for Esophageal Carcinoma (ESCA, M/F IRR = 4.22) and Thyroid Carcinoma (THCA, M/F IRR = 0.41) illustrate the correlation between IRRs and median TI values for the 26 adult cancers. (D) Regression analysis of IRR vs. Median TI values. Shown is the best fit and 95% confidence intervals. R and p values are shown

Cancer Hallmark Pathway analysis across cancer types indicated that most hallmark pathways [29] exhibit sex-skewing in gene expression and revealed several patterns of male versus female transcriptomic polarization (Fig. 2A and B). Seventeen of the 26 cancer types were enriched for genes involved in oxidative phosphorylation and/or cell cycle regulation at the male pole. Twelve of the 26 cancers were enriched for genes involved in inflammation and immunity at the female pole (Fig. 2A and B). Mesothelioma was the only cancer without evidence of transcriptomic polarization. Interestingly, sarcoma differed from the predominant polarization patterns such that male cases were enriched for inflammation/immunity signatures and female cases for cell cycle regulation. This emphasizes the need to interpret TI values within the context of cancer type and patient sex and gender.

Most Cancers exhibit sex and gender - skewed hallmark pathway activation. (A) Genes with the greatest effect on low (female pole) and high (male pole) TI values were identified. Cancer Hallmark Pathway analysis of pole-associated genes revealed a predominant polarization pattern involving cell cycle regulation and oxidative phosphorylation at the male pole (blue circles) and multiple inflammatory/immunity pathways at the female pole (red circles). Gene counts (count) are symbolized by the size of the circles and False Discovery Rates (FDR) by the saturation of the fill as indicated in the legends. (B) Frequency of pathway skewing is listed in rank order of cases involved (in parentheses) and the ratio of involved female (red text) to male cases (blue text)

These data indicate that varying degrees of sex and gender - correlated gene expression exist across cancer types and that a predominant shared pattern between multiple cancers involves skewed gene expression in cell cycle regulation versus inflammation/immunity pathways. This validates the TI approach as these pathways are known to be strongly sex and gender-biased in action [11,12,13,14]. The replication of these polarization patterns across cancer types provides a measure of cross-validation for the approach. Thus, we conclude that TI value can successfully localize individual cancer cases along axes that traverse between sex and gender poles in targetable mechanisms like cell cycle regulation and immunity/inflammation.

With some exceptions, in utero sexual differentiation results in outwardly recognizable male or female newborns, who differ in growth rates, immunity, metabolism, and disease risks, even prior to puberty and in the absence of circulating sex hormones. Thus, we hypothesized that if sexual differentiation patterns gene expression in cancer, we would also observe TI value sex-skewing in pediatric cancers. We analyzed two pediatric transcriptome datasets: the Gabriella Miller Kids First Pediatric Research Program (Kids First (KF)), which included 209 neuroblastoma patients and the Children’s Brain Tumor Network (CBTN), which included 865 patients comprised of 101 high grade glioma, 105 medulloblastoma, 79 ependymoma, and 214 low grade glioma cases (Table 1). From the KF data, 6330 male - skewed genes and 6089 female - skewed genes were identified (FDR < 0.05, Supplemental Table 3). From the CBTN data, 3063 male - skewed genes and 2062 female - skewed genes were identified (FDR < 0.05, Supplemental Table 4). There were 1126 shared male - skewed genes, and 742 shared female - skewed genes between the two datasets (both with p < 2.2e-16, Supplemental Table 5).

Like the adult cancers, these pediatric cancers exhibited biased distributions of TI values (Fig. 3A, Supplemental Fig. 2). Neuroblastoma (IRR = 1.11) is strongly polarized and again, those cases with high TI values were enriched for cell cycle regulation while those associated with low TI values were enriched for inflammation and immunity (Fig. 3B). The CBTN brain tumor data includes the diverse tumor types common in pediatric neuro-oncology. We focused our analysis on the most common and malignant pediatric brain tumors. Ependymoma (IRR = 1.5) exhibited the strongest polarization. Again, low TI values were enriched for inflammation and immunity and oxidative phosphorylation was strongly correlated with high TI values (Fig. 3C). The most common malignant brain tumor of childhood is medulloblastoma (IRR: 1.8:1) [30]. The strongest association in medulloblastoma was between low TI value and cell cycle regulation, reminiscent of what was observed for adult sarcomas. In pediatric high-grade glioma (IRR ≈ 1), high TI values were strongly correlated with cell cycle regulation, while there were no distinct gene expression patterns associated with low TI value cases. Thus, like adult cancers, pediatric cancers exhibit sex and gender-skewed gene expression that varies in magnitude and involved pathways, with similarities between the pediatric and adult cancers in the associations between cell cyle regulation and oxidative phosphorylation with male cases, versus immunity and inflammation with female cases. Importantly, skewed gene expression and pathway activation are evident even when the incidence ratios of cancer types (e.g., pediatric high-grade glioma) near equivalence. Therefore, individuals with any cancer type can be more extensively phenotyped for personalized approaches to treatment using a TI analysis than without.

Pediatric neural tumors also exhibit sex and gender - skewed gene expression. (A) Ridge plots for neuroblastoma and the three most common malignant brain tumors of childhood (489 total, 259 M, 230 F., M: F IRR = 1.13:1) demonstrating sex -skewed TI population distributions, with vertical lines indicating the 5%, 25%, 75% and 95% quantiles, respectively. (B)(C) Cancer Hallmark Pathway analysis of those genes that exerted the greatest effects on the male and female poles for each cancer. A predominant polarization pattern is identified with inflammatory/immunity pathways associated with the female pole (red circles) and cell cycle regulatory pathways associated with the male pole (blue circles). Gene counts (count) are symbolized by the size of the circles and False Discovery Rates (FDR) by the saturation of the fill as indicated in the legends

Cancer patients with extremes of high and low TI value might be approachable with something akin to sex and gender-specific treatments. However, TI values for most patients lie between the poles. Therefore, we expected that their transcriptomes would exhibit both female- and male- skewed components. We hypothesized that for female cases, translation along the TI axis from < 0.25 to midrange values would involve decreased female - skewed effects and/or increased male - skewed effects. We predicted that the opposite would be true for male cases with midrange TI value compared to those with TI > 0.75. If this proved to be the case, we expected TI could serve as a tool for stratification for sex and gender – informed treatments, even for those of differing sex and gender with identical TI value. To address this hypothesis, we compared the PANCAN transcriptomes of all cases with midrange TI value to those with transcriptomes closer to their respective poles. We then performed pathway analysis to determine which pathways were altered relative to the poles. Several clear patterns of change in different cancer types emerged. Across cancer types, most female cases, exhibited a loss of the inflammatory/immunity signatures (Fig. 4A). Female cases of four cancer types (LIHC, LUAD, COAD, DLBC) exhibited a gain in cell cycle regulatory signature. Male cases of seven cancer types (PRAAD, KIRC, LIHC, BLCA, COAD, LUAD, LUSC), exhibited a clear increase in the inflammation/immunity signature (Fig. 4A). There were mixed patterns of gains and losses of the other “pole-defining” pathways, such as cell cycle regulation and oxidative phosphorylation, across cancer types. In contrast, female sarcoma cases with midrange TI value cases exhibited gains in female - skewed cell cycle regulatory and male - skewed inflammation/immunity patterns of gene expression. Midrange male sarcoma cases exhibited no significant change in these pathways. Finally, several cancer type-specific changes in epithelial-to-mesenchymal transition (EMT) and key intracellular signaling pathways such as MYC, MTORC1, or KRAS, occurred in female and male cases with midrange TI values. Together, these data emphasize the potential of this approach for identifying sex and gender-skewed actions in targetable pathways, even for those with overlapping mid-range TI value.

Mid-range TI values exhibit distinct pathway signatures relative to the sex and gender – defined poles. (A) (Left Panel) Heatmap of pathway activation signatures underlying changes in midrange TI values for female adult cancers relative to their pole. (Right Panel) Heatmap of pathway activation signatures underlying changes in midrange TI values for male adult cancers relative their pole. (B) (Left Panel) Heatmap of pathway activation signatures underlying changes in midrange TI values for pediatric female cancers relative to their pole. (Right Panel) Heatmap of pathway activation signatures underlying changes in midrange TI values for pediatric male cancers relative their pole. For all panels, changes in male (blue circles) and female (red circles) signatures are indicated. Gene counts (count) are symbolized by the size of the circles and False Discovery Rates (FDR) by the saturation of the fill as indicated in the legends. Only cancer types and pathways with significant change are shown

We performed the same analysis in the pediatric datasets. Like the PANCAN analysis, translation away from respective poles to midrange TSI values occurred concomitantly with a shift in pole - defining pathway involvement (Fig. 4B). For increased power, we combined all malignant CBTN brain tumor cases for this analysis. We found that midrange TI value female cases exhibited decreased MYC targets (V1 and V2) and oxidative phosphorylation. Male CBTN cases with midrange TI value, exhibited significant changes in almost all hallmark pathways with gains in both the female - skewed inflammatory signature and loss of the male - skewed cell cycle and oxidative phosphorylation signatures.

In neuroblastoma, female midrange TI cases exhibited a strong acquisition of a male - skewed cell cycle regulatory signature as well as decreased female - skewed inflammation/immunity and metabolism signatures (Fig. 4B). No skewed pathway signature changes were detectable in the male midrange TI cases. Together, these results support the hypothesis that midrange TI value can be associated with different molecular pathway activation profiles in female and male individuals with different cancer diagnoses. Thus, even when similar in transcriptomic phenotype, particular subsets of female and male cancer patients may benefit from sex and gender-informed therapies.