NF-κB and STAT3 signaling uniquely stratify survival in female glioblastoma patients

Jason P Wong; Lorida Llaci; Lloyd Tripp; Myung Sik Jeon; Lihua Yang; Nicholas Reinhold; Nicole M Warrington; Jingqin Luo; Robi D Mitra; Joshua B Rubin

doi:10.1016/j.isci.2026.114761

. 2026 Jan 21;29(2):114761. doi: 10.1016/j.isci.2026.114761

NF-κB and STAT3 signaling uniquely stratify survival in female glioblastoma patients

Jason P Wong ¹, Lorida Llaci ², Lloyd Tripp ^2,³, Myung Sik Jeon ^4,⁵, Lihua Yang ¹, Nicholas Reinhold ¹, Nicole M Warrington ¹, Jingqin Luo ^4,⁵, Robi D Mitra ^2,³, Joshua B Rubin ^1,^6,^7,^8,^∗

PMCID: PMC12914307 PMID: 41716996

Summary

The mechanisms underlying sex differences in glioblastoma (GBM) incidence, treatment response, and survival are not well understood. Increased activation of nuclear factor-kappa B (NF-κB) and signal transducer and activator of transcription 3 (STAT3) signaling is associated with shorter survival in GBM. We looked at the expression of NF-κB- or STAT3-related genes in GBM for evidence of a sex skew in activity. Survival analysis of male and female GBM patients revealed that NF-κB- or STAT3-related gene expression was correlated with shorter survival only in female patients. We further explored mechanisms of this sex effect in an established murine model of sex differences in GBM. Concordant with human data, female murine GBM cells expressed stronger signatures of NF-κB and STAT3 genes and exhibited stronger responses to pathway stimulation and inhibition than their male counterparts. This study illustrates the advantage of sex-stratified data analysis in the development of sex-informed treatments for greater precision in cancer treatments.

Subject areas: health disparity, cancer

Graphical abstract

Highlights

•
Female GBM patients with upregulated NF-κB/STAT3 signaling have shorter survival
•
In murine GBM, NF-κB and STAT3 signaling exhibit sex differences
•
Female murine GBM cells are more sensitive to NF-κB and STAT3 inhibition
•
Sex-informed use of drugs targeting these pathways could benefit female patients

Health disparity; Cancer

Introduction

Inflammatory and immune pathways are known to factor in the genesis of cancers.¹^,²^,³^,⁴^,⁵ Additionally, these pathways are known to exhibit multiple, substantive, and significant sex differences in health and diseases.¹ This important detail has not been consistently studied in human, animal, or cellular cancer research. In multiple published studies, we and others have identified significant sex differences in tumor-associated immune and inflammatory pathway activation.³^,⁴^,⁶^,⁷^,⁸^,⁹ Repeatedly, female cancers exhibit higher expression of immune and inflammatory mediators than male cancers.³^,⁴ This is consistent with known sex differences in inflammation and immunity throughout life and may underlie the male-biased responses to immunotherapy.¹^,¹⁰ Like most other cancers, glioblastoma (GBM) has a higher incidence and is associated with shorter median survival in affected males.¹¹ Multiple cancer hallmark pathways appear to contribute to these sex differences.³^,⁵^,¹² To date, immunotherapy approaches to GBM have been disappointing.¹³ While many factors are likely involved in this lack of efficacy thus far, sex differences in GBM immunity are underexplored.

Here, we investigated sex-biased modulation of signal transducer and activator of transcription 3 (STAT3), a critically important transcription factor in cancers and inflammation.¹⁴^,¹⁵^,¹⁶^,¹⁷ This focus was motivated by published results indicating that inflammatory and immunity signatures were among the most consistently different pathways when comparing male and female patients across cancer types.³^,⁴ This is concordant with tissue-based studies demonstrating that male tumors have greater numbers of exhausted T cells than female tumors,⁷ which is likely the basis for the superior response to immune checkpoint inhibition in male patients.⁷^,¹⁰ Together, the available data indicate that there are important regulators of cancer immunity that are only evident when investigating their sex differences. Thus, we are aided in this study by the parallel analysis of combined male and female, male-only, and female-only datasets.³

Similar to many cancers, female GBM tumors are enriched for inflammatory gene signatures.³ A key pathway in inflammatory signaling is the nuclear factor kappa-B (NF-κB) pathway, which responds to various stimuli such as growth factors, cytokines, damage-associated molecular patterns (DAMPs), and pathogen-associated molecular patterns (PAMPs).¹⁸ Inflammatory pathways can be subverted in cancers, and chronic inflammation is associated with increased cancer progression.¹⁵^,¹⁷^,¹⁸^,¹⁹ Previous reports have noted that upregulation of the NF-κB pathway in GBM leads to poorer responses to radiation¹⁹ and transition to a mesenchymal-like phenotype in cells,²⁰ both of which are correlated with shorter survival.²¹ Activation of the NF-κB pathway can result in the secretion of interleukin-6 (IL-6) and the activation of the STAT3 pathway.¹⁷ STAT3 is known to promote cell survival and epithelial-to-mesenchymal transition (EMT), both of which are known resistance mechanisms.¹⁴^,¹⁵^,²² Several reports also suggest that elevated activation of STAT3 and STAT3-dependent genes are prognostic for shorter survival in GBM.¹⁵^,¹⁷ Interestingly, although female GBM patients have longer median survival, female GBM tumors exhibit an upregulated inflammatory gene signature.³ Therefore, we sought to ask if NF-κB and STAT3 contribute to this female-biased inflammatory gene signature, and whether or not having upregulated activation of these pathways affects survival in GBM patients. Using clinical data and a murine model, we identified multiple consistently skewed elements in a STAT3 regulatory network for activation involving tumor necrosis factor-α (TNF-α), NF-κB, IL-6, and protein kinase A (PKA).

Results

An NF-κB or IL-6/JAK/STAT3 gene signature is prognostic in female GBM patients

We previously reported, in a pan-cancer transcriptomic analysis, that females exhibit upregulated inflammatory and immunity gene signatures.³ Importantly, upregulated NF-κB signaling, which regulates inflammation and immunity, is associated with resistance to radiation treatment in GBM and a more mesenchymal-like phenotype in GBM, with both corresponding to shorter survival.¹⁹^,²⁰ Thus, we examined whether there were any correlations between survival and the expression of genes related to NF-κB signaling. We utilized The Cancer Genome Atlas (TCGA) data downloaded from GlioVis for survival analysis (N = 368; 225 males and 143 females)²³ and performed univariate survival analysis of 38 NF-κB-related genes that are reported to promote radiation resistance (Figures 1A and 1B, and Table S1).¹⁹ We partitioned patients into low or high expression of the gene-of-interest separately in all, male, or female GBM patients, and estimated the overall survival (OS) hazard ratios (HRs) of patients with high versus low expression based on median mRNA expression for each of the 38 genes. In the all-patient analysis, we saw that higher expression of the majority of NF-κB-related genes resulted in an HR greater than one, and thus poor OS, which is consistent with prior clinical reports (Figures 1A and 1C).¹⁹ The greatest significant effects were for 5 genes (CD44, CCL2, BCL2A1, IRF2, and NOD2). The effect was partially mitigated by CSF2, higher expression of which corresponded to better survival. When analyzing the male-only or female-only dataset separately, the OS-prolonging effect of CSF2 was predominantly observed only in the males, but not in the females. Two additional genes (PSMB9 and S100A6) also exhibited significant pro-survival effects in males but an opposite effect in females, and consequently, no effects were observable in the combined analysis. In contrast, ten genes correlated with shorter survival (HRs >1) in the female-only analysis (CD44, CCL2, IL15RA, NQO1, BCL2A1, IL-6, SOD2, IRF2, NFKBIA, and NOD2). This was in striking contrast to the fact that only three genes were found significant in the male-only analysis and all three genes prolonged survival in males. These results indicate that female patients with high expression drove the worse OS HRs seen for the NF-κB genes in the combined population analysis and in previously published analyses.¹⁵^,¹⁹

High expression of NF-κB and IL-6/JAK/STAT3 genes is a biomarker of shorter survival in female patients

Survival analysis performed from TCGA data downloaded from the GlioVis portal (N = 368; 225 males and 143 females).²³

(A) Forest plots of hazard ratios from univariate Cox regression survival analysis in all patients, male patients, or female patients for individual NF-κB mesenchymal genes (Bhat et al., *Cancer Cell*, 2013).¹⁹

(B) Kaplan-Meier survival curves stratified by high and low expression of *IL-6* in all patients, male patients, or female patients.

(C) Hazard ratios from univariate Cox regression survival analysis of NF-κB mesenchymal genes in male, female, or all patients.¹⁹

(D) Hazard ratios from univariate Cox regression survival analysis of IL-6/JAK/STAT3 genes (Human Molecular Signatures Database).²⁴

In (A), significant genes (p < 0.05) are indicated in red; in (B), significance was determined by the log rank test; hazard ratios in (C) and (D) are presented as median ± interquartile range. ∗, p < 0.05; ∗∗, p < 0.01; ∗∗∗, p < 0.001; ∗∗∗∗; p < 0.0001 as determined by Kruskal-Wallis test followed by a Dunn’s post-hoc pairwise comparison with multiplicity corrections.

See also Figure S1 and Tables S1 and S2.

IL-6 is a critical downstream effector of TNF-induced NF-κB signaling.¹⁸ It is through IL-6 binding to its cognate receptors that NF-κB induces a signaling cascade through JAK1 and/or JAK2 and activation of the STAT3 transcription factor.¹⁴^,¹⁷ Kaplan-Meier survival curves for IL-6 (Figure 1B) showed no significant effect of high IL-6 expression on survival when analyzing all patients together. However, when disaggregating male and female patients, we found that IL-6 had slightly higher expression in males than in females (median expression = 4.5 vs. 4.3, p = 0.056) but impacted OS only in females (HR = 1.56, 95% CI = 1.08–2.26, p = 0.017) and not in males (p = 0.52). Females with high IL-6 expression had significantly poor survival than those with low IL-6 expression. These data suggest there is a sex-biased effect of IL-6 on survival. Therefore, we examined if genes related to IL-6/JAK/STAT3 signaling also exhibit sex effects on survival and found a pattern similar to the NF-κB genes (Figure 1D).²⁴

In the combined analysis (Figures 1D and S1, and Table S2), higher IL-6/JAK/STAT3 gene expression correlated with worse prognosis, consistent with previous clinical reports.¹⁵ The analysis of IL-6/JAK/STAT3 gene expression and survival after disaggregating the results by sex yielded discordant results in males and females. In the female-only analysis, patients with high expression of IL-6/JAK/STAT3 genes had higher HRs than those with lower expression for the majority of genes. Like the NF-κB gene analysis, these data indicate that the female population drives the higher HR in the all-patient analysis. Similar to the NF-κB gene set, all significant genes in the female-only analysis (CCR1, CD14, CD44, CSF1, CXCL3, IL13RA1, IL15RA, IL4R, IL-6, IL-7, IRF9, ITGA4, OSMR, PIK3R5, PTPN2, STAT2, TGFB1, TLR2, and TNFRSF1B) correlated with a higher HR. However, this was only the case for 2 of 9 significant genes, IL6ST (HR = 1.77, p = 0.0001) and PDGFC (HR = 1.40, p = 0.02), in the male-only analysis (Table S2), while the remaining 7 significant genes (ACVR1B, BAK1, CBL, CSF2, IL12RB1, IL9R, and INHBE) were pro-survival. Together, these data again indicate that inflammatory signaling impacts survival differently in male and female GBM patients.

An NF-κB gene signature is uniquely enriched in female GBM patients

We previously analyzed TCGA transcriptomic data for male- and female-biased processes on a pan-cancer level.³ Similar to many cancers, GBM tumors exhibited a female bias in hypoxia, inflammatory response, and TNF-α signaling by NF-κB. In an effort to further develop transcriptomic signatures for sex-informed prognostics and treatment stratifications, we utilized male and female GBM patient data from Yanovich-Arad et al. 2021.²⁵ This clinically, well-annotated, single institution database provides some advantages in the transcriptomic analysis of sex-biased effects of these pathways. First, the clinical annotation supports strong survival analysis and the single institution of origin decreases the potential confounding effects of variations in sample handling, processing, and data generation and analysis.²⁶

After filtering for wild-type (WT) IDH1 tumors, 41 of 68 patients were classified as untreated at the time of biopsy and were used for subsequent analysis. Additional quality control (QC) analysis identified three male GBM transcriptomes as outliers in the principle component analysis (PCA) and Euclidean clustering. These three samples were also removed, leaving 38 GBM patients (N = 25 and 13 for males and females, respectively) for further analysis (Table S3), corresponding to a 1.92:1 male-to-female ratio as reflective of the well-known greater incidence of GBM in males. Previously, we demonstrated that male and female pan-cancer transcriptomes exhibit overlapping but distinct population distributions based on Euclidean distances (Figure 2A).³ After deriving Euclidean distances between patient transcriptomes, we similarly observed that male and female transcriptomes from GBM tumors exhibited an overlapping but sex-skewed population distribution, with a large portion of male and female samples skewed left or right, respectively, on the heatmap (Figures 2B and 2C).³^,²⁷

TNF-α signaling by NF-κB uniquely stratifies female GBM patients relative to each other

(A) Combined male and female GBM patient transcriptomes can be examined for sex-skewed gene expression. The same data can also be disaggregated by sex to examine variance within the male-only or female-only transcriptomic phenotypes.

(B–H) Human IDH (isocitrate dehydrogenase)-wt (wild-type) GBM transcriptomes (Yanovich-Arad et al., *Cell Rep*, 2021) of male (N = 25) and female (N = 13) patients with no prior treatment were analyzed.²⁵

(B) PCA plot of all-patient transcriptomes, with sex indicated by blue (male) and red (female) symbols.

(C) K-means clustering of top 1,000 most variable (by variance) genes in the combined dataset. The optimal number of clusters was determined by elbow plot of the averaged dispersion.

(D) Gene enrichment of clusters identified in (C). Sex-skewed clusters identified for males (blue bars) and females (red bars). Upregulation (↑) and downregulation (↓) of genes in the cluster in a sex-biased manner are indicated by arrows next to the cluster number. FDR values are indicated to the right.

(E and F) K-means clustering (E) and gene enrichment of clusters (F) of the male transcriptomes as in 2C and 2D.

(G and H) Parallel analysis of the female transcriptomes by k-means clustering (G) and gene enrichment of clusters (H).

In (D), (F), and (H), Sig. is abbreviated for signaling, panc. is pancreatic, inflamm. is inflammatory, and EMT is epithelial-to-mesenchymal transition. In (E) and (G), sample names are abbreviated from OtB6XXX, where XXX is the unique identifier number (e.g., 179 is OtB6179). See also Figures S2 and S4, Table S3, and Data S1, S2, S3, S4, S5, S6, and S13.

We selected the top 1,000 genes that were the most variable (based on variance) from all samples and performed k-means clustering followed by gene set enrichment analysis (GSEA) to identify sex-biased genes and pathways (Figures 2C and 2D, and Data S1).²⁸ Sex-biased genes and pathways were identified by splitting the heatmap into approximately three even groups (left, middle, and right) based on the clustering dendrogram. A sex-biased cluster had a difference of at least 0.5 in the median Z score (either the left group for males or the right group for females) from all other groups and was found to be significantly different in the Kruskal-Wallis test when comparing the combined Z scores for all genes in the cluster across the other groups. Comparing the median Z score among groups also informed if the cluster exhibited an upregulation or downregulation of the genes within that cluster in a sex-biased manner. There was a male-biased upregulation (left skew) of the genes in clusters 2 and 4 and a female-biased downregulation (right skew) in cluster 1. The emergence of sex-skewed clusters from the combined (ALL) analysis highlights the presence of a sex effect and warrants deeper analysis of its mechanisms.

Male-biased cluster signatures were enriched for EMT, growth (KRAS signaling), and metabolism (glycolysis) pathways (Figure 2D and Data S2). Interestingly, the female-biased cluster was not enriched for any hallmark pathways (Figure 2D and Data S2).²⁹^,³⁰ Clusters 3, 6, and 7 were enriched for inflammatory processes, though a sex-biased skew was not present (Data S2). In our previous pan-cancer report from TCGA data, we identified cell cycle regulation as a male-biased cancer phenotype.³ Additional male-biased pathways are also concordant with the known biology of sex differences.³^,¹²^,³¹^,³²

At the gene level, sex chromosome genes were differentially expressed (referred to as DEGs) in females versus males, as expected (Data S3).³³ Additionally, there was sex-biased autosomal allele enrichment of CCL18 (a macrophage-associated chemokine) and COL1A1 (a collagen gene important for promoting a mesenchymal-like phenotype) in male patients and of EPHA3 (a receptor tyrosine kinase with high expression in GBM mesenchymal-like cells) and SEMA3A (a protein that promotes growth) in female patients (Figure S2A).³⁴^,³⁵^,³⁶^,³⁷ Interestingly, although COL1A1 and EPHA3 are markers tied to a mesenchymal-like phenotype in GBM,³⁵^,³⁶ they are differentially upregulated in males and females, respectively. This highlights that similar pathways can be enriched in both sexes without the underlying DEGs being the same.³

Although we had previously identified inflammatory signatures as a female-biased signature in a pan-cancer report,³ we did not see a female-biased inflammatory signature in our ALL GBM patient analysis. Most reports have focused on sex differences as a combined population analysis between males and females but rarely examined how an individual differs from others of their sex. Survival analysis of NF-κB genes showed large differences between females on a population level, but not among males. This additional personalization of cancer phenotypes is likely important for sex-informed treatment plans as it provides unique insight into how to contextualize the clinical, radiographic, and laboratory data in the framework of variable sex effects. This supports the determination of whether each patient is within or outside the norm for their sex rather than the norm for the whole population.¹² Thus, we next analyzed males and females separately. We applied k-means clustering to the top 1,000 genes by variance in male-only samples, and in parallel, female-only samples. In the male samples (Figures 2E and 2F, and Data S4), cluster 4 had similar signatures (EMT, myogenesis, and hypoxia) to those in cluster 2 in the all-sample analysis (Data S2). Interestingly, an IFN-γ response signature and a pancreas beta cell signature, which were not present within the all-sample analysis, were noted when comparing male samples to each other. This suggested that these pathways exhibit variance within male samples. In the female-only analysis (Figures 2G and 2H, and Data S5), cluster 3 (IFN-γ, IFN-α, and EMT) had similar signatures to those in cluster 7 in the all-sample analysis (Data S2). Importantly, female cluster 2 exhibited a very strong TNF-α signaling by NF-κB gene signature (FDR = 9.44E-16; Figure 2H), which was not present in the all-sample or male-only analysis.

To further confirm that TNF-α/NF-κB signaling varied within female GBM patients, but not within males, we performed GSEA comparing males against other males and females to other females. To generate two groups for GSEA, we divided males or females, each based on their similarity from the dendrogram generated from Euclidean hierarchical clustering (Figure S2B).²⁹^,³⁰ As the first branching point for male samples would generate a group of only two samples, we used the first and second branching points to split the male samples. For the female samples, the first branching point was used. In the male samples, there were no significantly enriched pathways (FDR < 0.05) that could distinguish between the two groups (Figure S2C and Data S6). In the female samples, multiple significantly enriched pathways (including TNF-α signaling by NF-κB) that distinguished between the two female groups were identified (Figures S2B–S2D and Data S6). The other group of female patients showed an enrichment for IFN-α and IFN-γ responses (Figure S2C). These results demonstrate that examining differences within a sex can be just as informative as looking at sex differences between males and females.

Concordance between sex-skewed gene expression in patients and a murine model of GBM

We previously have used an isogenic murine GBM model (Nf1−/− DNp53 transformed astrocytes) to demonstrate concordance in sex differences in human and murine data regarding retinoblastoma (Rb) signaling,³¹^,³⁸ Tp53 functions,³¹^,³⁹^,⁴⁰ bromodomain-containing protein 4 (Brd4) gene regulation,³⁹ glutamine metabolism,³² and senescence in response to irradiation.⁶ Similar to those studies, we generated individual cell lines in which all male or female pups were combined from a single litter to form paired cell lines (denoted as a lot). Lots were derived from five independent litters (biological replication) to capture a spectrum of sex differences in phenotypes, and each cell line was evaluated in triplicate (technical replication). To confirm our patient GBM findings (Figure 1), we conducted transcriptomic analysis (Data S7) and again saw distinct, but overlapping, male and female murine transcriptomic patterns (Figures 3A and S3A). As expected, replicate samples from the same cell line (lot) showed the greatest resemblance to each other and clustered together. Clustering of the 5 different lots demonstrated that we were able to fully capture a spectrum of biological sex differences and that our findings are not due to batch effects. Differential expression of sex chromosome genes confirmed separation of male and female transcriptomes (Data S8).³³ When looking at autosomal alleles, males exhibited enrichment of Sprr1a, insulin-like growth factor 2 mRNA-binding proteins (Igf2bp), Igf2bp1, and Igf2bp3 (Figure S3B). In females, enrichment of Mest, Apoe, and Itga1 was evident.

TNF-α signaling by NF-κB is enriched in female murine transformed astrocytes

Each cell line was harvested multiple times (n = 4) to generate replicates for RNA sequencing. One male and female replicate pair in lots 6 and 12 was dropped from further analysis after quality control (QC).

(A) K-means clustering of the top 1,000 most variable (by variance) genes from murine transformed astrocyte transcriptomes. Lot (6, 7, 8, 11, and 12) refers to male and female littermate cultures. As each lot is derived from independent litters, they serve as (n = 5) biological replication. Sex of each transcriptome, cluster number, and Z score are as indicated. The optimal number of clusters was determined by elbow plot of the averaged dispersion.

(B) Gene enrichment of clusters from astrocyte cell lines in (A). Sex-skewed clusters are identified by blue bars (male) and red bars (female). Upregulation (↑) and downregulation (↓) of genes in the cluster in a sex-biased manner are indicated by arrows next to the cluster number. FDR values are indicated to the right.

(C and D) K-means clustering (C) and gene set enrichment of genes in clusters of interest (D) of the male transcriptomes as in (A) and (B).

(E and F) Parallel analysis of the female transcriptomes by k-means clustering (E) and gene set enrichment of clusters of genes (F).

See also Figures S3 and S4, and Data S7, S8, S9, S10, S11, S12, and S13.

K-means clustering of the top 1,000 most variable genes (based on variance) in the combined analysis (ALL) revealed female sex-biased clusters. Clusters 1 and 5 exhibited upregulation of a subset of genes that was enriched for female transcriptomes (Figures 3A and 3B, and Data S9), while cluster 6 exhibited a downregulation of genes for female transcriptomes. Pathways downregulated in the female-biased cluster included KRAS signaling up. When analyzing the transcriptomes from the male lots only, we saw enrichment of G2M checkpoint, mitotic spindle, and E2F targets pathways, as well as TNF-α signaling by NF-κB (Figures 3C and 3D, and Data S9). Similar to the male human GBM analysis, EMT was a very strong stratifying signature of variance in male phenotypes (Figure 2F). Pathways upregulated in female-biased clusters of the all-sample analysis included EMT, myogenesis, and TNF-α signaling by NF-κB (Figures 3A and 3B, and Data S9). This was similar to female cluster 2 in the female-only human analysis. The analysis of female-only murine transformed astrocyte transcriptomes also identified MYC signaling as a pathway that stratified female murine cells (Figures 3E and 3F and Data S9). EMT, hypoxia, myogenesis, and TNF-α signaling by NF-κB strongly stratified female murine GBM transcriptomes (Figure 3F), which was similarly reflected within the human female-only GBM analysis (Figure 2H).

A consistent finding across the human and murine analyses was the association between female sex and TNF-α signaling by the NF-κB gene signature. In the human female-only analysis, the strength of TNF-α signaling by NF-κB signature varied greatly, with higher NF-κB signaling being associated with worse survival. We also observed this variance in the pathway in the murine analyses. TNF-α signaling by the NF-κB pathway was also identified as the strongest female-enriched pathway in the GSEA for genes differentiating male from female murine transformed astrocytes (Figures S3C and S3D, and Data S10).

To look more closely into the pathways that were uniquely identifiable within the male-only or female-only analysis, but not in the combined analysis, we examined which of the top 1,000 genes from our k-means clustering overlapped or not between the three different analyses. In the human GBM patient analysis (Figure S4A), we found that 545 of 1,534 genes overlapped between all three analyses. EMT, hypoxia, and myogenesis were common to all three analyses (Figure S4B and Data S11). In the male-only analysis, we saw “KRAS signaling down” and “estrogen response early” hallmark gene sets as uniquely enriched pathways. Pathways enriched in the female-only analysis again included the inflammatory/immunity pathways of complement and TNF-α signaling by NF-κB.

When examining the murine transcriptomes, we saw 506 of 1,542 genes overlapping between all three analyses (Figure S4C). However, unlike the patient analysis, TNF-α signaling by the NF-κB gene signature was present in all three analyses (Figure S4D and Data S12). Similar to the human transcriptome analysis, EMT, myogenesis, and hypoxia signatures were common to all three murine analyses. Similar to the male patient transcriptome analysis, “estrogen response early” was enriched in the male-only murine transcriptome analysis. In striking contrast to the female patient analysis, pathways enriched in the female-only murine transcriptome analysis were related to cell cycle (E2F targets, G2M checkpoint, Myc signaling, and mTORC1 signaling).

To further confirm the congruency between the human and murine transcriptomes, we converted human genes to their murine orthologs and determined the overlap between each of the three analyses. In all three analyses, there was a significant overlap between the human and murine transcriptomes (Figure S4E) and the enrichment of EMT, hypoxia, and UV response down (Figure S4F and Data S13). When looking at the overlapping genes between the human and murine female-only analyses, TNF-α signaling by the NF-κB gene signature was uniquely enriched (Figure S4F and Data S13), which once again highlighted its importance in stratifying females.

Sex differences in the NF-κB pathway are dependent on IL-6 and STAT3

The GBM patient data indicated that diagnostic molecular signatures of NF-κB signaling stratified female, but not male, survival. As there are multiple clinically actionable targets in the NF-κB pathway, we sought to identify which of those are required for sex differences in survival. We reasoned that these would be priority targets for a sex-informed treatment. The NF-κB transcription factor functions as a heterodimer that is composed of varying subunits.⁴¹ In the canonical signaling pathway, the most common heterodimer is p65/p50.⁴¹ Using five lots of male and female transformed astrocytes (n = 5 biological replicates) to capture a range of sex-related phenotypes, we observed no significant sex differences in the mean basal expression of phospho-p65 (p-p65) S536, a marker of p65 activation (Figures 4A and S5A), or in mean NF-κB activity in bioluminescent reporter assays (Figure S5B). Interestingly, in both assays, the data points were distributed across distinct ranges of male and female values, with some overlap. This suggested that there may be distinct male and female correlations between p-p65/p65 levels and NF-κB activity.

Murine transformed astrocytes are more responsive to changes in NF-κB signaling

(A) Western blot analysis of phospho-p65 (p-p65) levels in male (blue boxes) and female (red boxes) from 5 cell lots. Shown is a representative blot (n = 3) of independent experiments in each of the 5 lots.

(B) Correlation between p-p65 (S536) abundance and basal NF-κB activity as measured by NF-κB luciferase activity assay. Male: R² = 0.10; r = −0.32; ns. Female: R² = 0.82; r = −0.90; ∗, p < 0.05.

(C) Correlation between the percentage of senescent cells as measured by senescence-associated beta-galactosidase (SABG) in the same lots (from Broestl et al., *Commun Biol*., 2022) and basal NF-κB activity.⁶ Male: R² = 0.03; r = −0.18; ns. Female: R² = 0.74; r = −0.86; ns.

(D) Correlation between TNF-α (10 ng/mL, 4 h)-induced NF-κB activity (treated/untreated in each respective cell line) and basal NF-κB activity. Male: R² = 0.85; female R² = 1.00. Curve fit comparing straight line to one-phase decay (male, ns; female, ∗∗, p < 0.01).

(E) Western blot of NF-κB signaling following 2 h of TNF-α (10 ng/mL) treatment. Replicate lot 6 results serve for normalization between the blots. Shown is a representative blot (n = 3) of independent experiments in each of the 5 lots. Male and female cells are indicated by blue and red boxes, respectively.

(F and G) Correlation between TNF-α-induced p-p65 (male: R² = 0.03, r = −0.18, ns; female: R² = 0.38, r = −0.62, ns; F) or p-IκBα (male: R² = 0.02, r = 0.14, ns; female: R² = 0.48, r = −0.69, ns; G) and basal NF-κB activity.

(H) Correlation between basal NF-κB activity and TNF-α-induced expression changes (ΔΔCT, treated/untreated within each respective cell line) as measured by RT-qPCR of *Ccl*2 (male: R² = 0.25, r = −0.50, ns; female: R² = 0.87, r = 0.93, ∗, p < 0.05), *IL-6* (male: R² = 0.00, r = −0.01, ns; female: R² = 0.57, r = 0.76, ns), *Ccl5* (male: R² = 0.00, r = −0.03, ns; female: R² = 0.68, r = 0.82, ns), and *Nfkbia* (male: R² = 0.09, r = −0.29, ns; female: R² = 0.91, r = 0.96, ∗, p < 0.05).

All data in (A)–(H) are compiled from n = 3 independent experiments (technical replication) in each 5 lots (biological replication), except for the previously published SABG% data in (C), which was determined from 2 technical replicates in n = 5 biological replicates. Pearson correlation coefficients in (B), (C), (F), (G), and (H) are reported for all.

See also Figures S5 and S6.

Separate male and female linear and nonlinear regression analyses allowed us to examine whether there were sex differences in correlation coefficients and slopes. This determined the strength (r, R², p) and direction (positive or negative) of the association, a measure of the sensitivity of NF-κB transcriptional activity to differences in p-p65/p65 levels within each sex. Under basal conditions, p-p65 and NF-κB activity were found to be strongly correlated in females (r = −0.90, R² = 0.82, p = 0.036), but not in males (Figure 4B). Surprisingly, the correlation was negative, possibly suggesting that, under basal conditions, there may be a strong counter-regulatory homeostatic mechanism to prevent excessive transcriptional activity. There was also a trend toward correlation (p = 0.06) between the baseline NF-κB activity and basal levels of senescence in female, but not male (p = 0.78), transformed astrocytes (Figure 4C). Since there was a stronger sensitivity in both female correlation analyses, we next sought to determine whether there were also sex differences in the responsiveness of the NF-κB pathway. Upon treatment with TNF-α, we observed a relationship between basal NF-κB activity and fold change in the induced NF-κB activity (responsiveness) in female cells, and this relationship was best fit by a one-phase decay model (Figure 4D). The lots with lower basal levels of NF-κB activity were the most responsive. The lack of responsiveness to TNF-α in the female lots with high basal NF-κB activity may relate to the negative correlation observed between p-p65/p65 and NF-κB activity measures. While the basal NF-κB activity in the male lots exhibited a wide range of values, none were as low as those in the most-responsive female lots. The male points were best analyzed as a linear fit. The male line was very similar to a linear fit of only female lots with the lowest levels of induced activity. This may indicate that in both male and female cells, there is a threshold of basal NF-κB activity, over which there is no further response to TNF-α. We derived two conclusions from these studies. First, there are sex differences in the relations between p-p65/p65 and NF-κB activity, and second, these differences result in sex differences in TNF-α response.

NF-κB effects are negatively regulated by IκBα. Canonically, NF-κB is sequestered in the cytoplasm by IκBα, which, when phosphorylated (S32) in response to TNF-α treatment, is degraded and releases NF-κB for nuclear action.⁴¹ Thus, we next sought to determine if TNF-α-induced IκBα degradation contributed to sex differences in NF-κB activity. TNF-α treatment induced variable changes in p-p65, as expected, and similarly variable changes in p-IκBα (Figure 4E). In regression analyses, there was a weak correlation in female lots between the fold change in p-p65 (Figure 4F) or p-IκBα levels (Figure 4G) and the basal NF-κB activity. There was no correlation between p-p65 or p-IκBα expressions and the basal NF-κB activity in males (Figures 4F and 4G). When examining the distributions of induced NF-κB activity, fold change in p-p65, and fold change in p-IκBα, we once again saw overlapping, though not significantly different, male and female data distributions (Figures S5C–S5E). We conclude that p-IκBα levels do not underlie sex differences in the NF-κB activity.

To determine whether sex differences in NF-κB reporter activity are correlated with sex differences in NF-κB transcriptional activity, we measured the expression of NF-κB target genes (Ccl2, Ccl5, Nfkbia, and IL-6) 2 h after TNF-α treatment (Figure 4H).¹⁸ Differences in the expression are reported as the ΔΔCT before transformation (by raising to the power of 2) to perform a linear correlation analysis. Therefore, a more negative ΔΔCT would indicate a larger fold change after transformation. There was a significant positive correlation between the fold change in Ccl2 and Nfkbia expressions and the basal NF-κB activity but a moderate (though not significant) correlation between the fold change in Ccl5 (p = 0.09) and IL-6 (p = 0.14) and the basal NF-κB activity in female lots. There was no correlation between any of the genes and basal NF-κB activity in male lots (Figure 4H). With the exception of Ccl5, fold induction of these genes also exhibited overlapping distributions, though female cell activity skewed higher (Figure S5F). The lot 6 females also exhibited a strong induction for some genes (Ccl2 and IL-6) and represented the extreme end of the spectrum for female cells, which is concordant with the clustering in Figure 3E. We conclude that the NF-κB pathway exhibits greater plasticity in TNF-α responses (a wider range of measurements and sensitivity to TNF-α) in female cells than in male cells. We further conclude that the baseline NF-κB activity is more strongly correlated with TNF-α treatment in the female lots. This may be reflective of sex differences in enhancer activity as we have previously reported that female GBM cells, but not male cells, exhibit increased Brd4-bound enhancer activity at Rela (which encodes p65) motifs.³⁹

As the measures for NF-κB signaling exhibited overlapping but distinct distributions with regard to sex, we also examined if there were differences among cells that were identified as having a low or high NF-κB gene signature from the clustering analysis. There were 6 murine (males, n = 4; females, n = 2) GBM cell lines with a low NF-κB gene signature (↓NF-κB), while there were 4 murine (males, n = 1; females, n = 3) GBM cell lines with a high NF-κB gene signature (↑NF-κB). In all metrics (Figures S6A–S6D and S6F) with the exception of p-IκBα levels (Figure S6E) and Ccl5 fold induction (Figure S6F), there were significant differences between the two groups. In some measures, the distribution of the single high male NF-κB gene signature cell line overlapped with or was within the distribution of the four low male NF-κB gene signature cell lines. This contrasts with the distribution seen in female lots, where the low and high NF-κB gene signature cell lines more commonly segregated into two groups. We conclude that female cell lines exhibit a large variation in NF-κB-related measures that are clearly split according to their NF-κB gene signature, whereas male cell lines do not. This is reminiscent of the prognostic significance of NF-κB-related genes split by high and low expression in female GBM patients, but not in males.

Sex differences in the NF-κB and IL-6/JAK/STAT axis has therapeutic implications

The clinical data indicated that NF-κB and IL-6/JAK/STAT3 had prognostic significance only for female GBM patients (Figure 1). In addition, the in vitro data demonstrated that there are sex differences in the basal and TNF-α-induced NF-κB activity and gene expression. Thus, we hypothesized that there would be sex differences in the further downstream pathway effectors IL-6/JAK/STAT3 and that this would render male and female cells differentially sensitive to their inhibition with clinically available agents. At baseline, there was a trend toward greater IL-6 levels in standard cell culture supernatants from female lots (Figure S7A), which coincided with a slight trend toward higher levels of STAT3 activation (p-STAT3 Y705) in female lots (Figures S7B and S7C). Further, there was a significant correlation between IL-6 expression and p-STAT3 Y705 levels in female lots, but not in male lots (Figure 5A). Consistent with the preceding results, these additional data confirm sex differences in activity in clinically actionable targets in the TNF-α/NF-κB pathway.

Female murine transformed astrocytes are more sensitive to regulation and inhibition of STAT3 signaling than male cells

(A) Correlation between IL-6 expression and p-STAT3 (Y705) expression in full serum conditions (n = 5 lots). Male: r = 0.51, R² = 0.26, ns; female: r = 0.96, R² = 0.93, ∗∗, p < 0.01 (Pearson’s correlation).

(B and C) Western blot (B) and quantification (C) of p-STAT3 (Y705) levels in serum-starved conditions following TNF-α treatment (10 ng/mL) with blocking using 10 μg/mL of either α-IgG control or α-IL-6 for 24 h in female cells (n = 3 lots). Western blot representative of n = 3 independent experiments.

(D) Western blot of p-STAT3 (Y705) levels in serum-starved conditions with or without supplemental 10 ng/mL epidermal growth factor (EGF) in male and female cells (n = 5 lots). Replicate lot 6 results serve for normalization between blots.

(E and F) Quantification of p-STAT3 Y705 levels from (D), segregated by sex (n = 5 lots; E) or strength of NF-κB gene signature (n = 6 weak, n = 4 strong; F); 6F is the abbreviation for lot 6 females.

(G) Correlation between p-STAT3 (Y705) levels and IC₅₀ of AZD1480 in serum-starved conditions in murine transformed astrocytes (n = 5 lots per sex with or without supplemental EGF). Male: r = −0.26, R² = 0.25, ns; female: r = −0.79, R² = 0.53, ∗∗, p < 0.01 (Spearman’s correlation).

(H and I) Western blot (H) and quantification (I) of p-STAT3 (Y705) and p-p65 (S536) levels in serum-starved conditions following treatment with vehicle-control (DMSO) or H-89 (10 μM) for 18 h in lot 6 male and female cells. Western blot representative of n = 3 independent experiments. Expression levels normalized to vehicle-treated (DMSO) female cells.

(J) Hazard ratios from univariate survival analysis of wild-type (WT) or altered (alt) *EGFR* tumors in all (*N =* 164 WT, N = 198 alt), male-only (*N =* 94 WT, N = 125 alt), or female-only (*N =* 70 WT, N = 73 alt) patients of IL-6/JAK/STAT3 genes from the Human Molecular Signatures Database.²⁴ Data are presented as the median ± interquartile range.

In (C), (E), (F), and (I), data are presented as the mean ± SD. In (C), (E), (F), (I), and (J), ∗, p < 0.05; ∗∗, p < 0.01; ∗∗∗, p < 0.001; ∗∗∗∗; p < 0.0001; ns, not significant as determined by Tukey’s post hoc pairwise comparison following two-way ANOVA test (C, E, F, and I) or Kruskal-Wallis test followed by a Dunn’s test for post-hoc pairwise comparisons with multiplicity corrections (J). (C) Lot 6: α-IL-6 (∗, p < 0.05); TNF-α (∗∗∗, p < 0.001); interaction (ns, not significant). Lot 8: α-IL-6 (∗, p < 0.05); TNF-α (∗, p < 0.05); and interaction (ns, not significant). Lot 12: α-IL-6 (∗∗∗, p < 0.001); TNF-α (ns, not significant); and interaction (∗∗, p < 0.01) by two-way ANOVA. (E) Sex (ns, not significant); EGF (∗, p < 0.05); and interaction (ns, not significant) by two-way ANOVA. (F) NF-κB (∗, p < 0.05); EGF (∗∗, p < 0.01); and interaction (ns, not significant) by two-way ANOVA. (I) p-p65 (S536): sex (ns, not significant); H-89 (ns, not significant); and interaction (ns, not significant). P-STAT3 (Y705): sex (∗∗∗∗; p < 0.0001); H-89 (∗∗, p < 0.01), and interaction (∗∗, p < 0.01) by two-way ANOVA. Blue and red boxes in (D) and (H) indicate male and female murine transformed astrocytes, respectively. All data in (A)–(I) are compiled from n = 3 independent experiments (technical replication). In (A), (D), (E), (F), and (G), all data are compiled from each of the 5 lots (biological replication).

See also Figure S7 and Table S4.

Because IL-6 and STAT3 can be inhibited with clinically available drugs,¹⁴ we next determined whether sex differences in this pathway would correlate with sex differences in responses to inhibitor activity. Serum and TNF-α can activate the NF-κB pathway and induce IL-6 expression.²⁰ Consistent with this, we found that IL-6 levels were higher in serum-containing conditions than in serum-free conditions (Figures S7A and S7D). There was a slight trend toward higher IL-6 levels in the female lots. To prevent confounding by serum-associated IL-6 or IL-6 induction, we used serum-free conditions to assess IL-6 effects. As only female lots strongly responded to the TNF-α treatment in this system, we determined whether there were effects of IL-6-blocking antibody on TNF-α responses in the female lots (Figures 5B and 5C). TNF-α-induced p-STAT3 in female lots 6 and 12, which had strong NF-κB gene signatures (Figures 3A and 3B), and treatment with the IL-6-blocking antibody significantly reduced p-STAT3 (Figures 5B and 5C). This is in contrast to the lack of response of the IL-6-blocking antibody in lot 8, which had a weak NF-κB gene signature. These data indicate that TNF-α-dependent IL-6 induction activates JAK/STAT3 signaling and that female cells are more sensitive to its inhibition. As female but not male patients exhibited differences in survival based on IL-6 expression levels, these results suggest that clinical targeting of IL-6 signaling may exhibit sex-skewed responses. It will be important to examine data from the open but not yet enrolling clinical trial of the IL-6-blocking antibody tocilizumab (NCT04729959) for any evidence of such a skew.

JAK2, a kinase that activates STAT3, can be pharmacologically inhibited by AZD1480 (Figure S7E).¹⁶ While neurotoxicity of AZD1480 in cancer clinical trials (NCT01112397 and NCT01219543) were not acceptable, a second STAT3 inhibitor WP1066 has been evaluated in phase 1 trial (NCT01904123) and determined to be well tolerated and safe. Though no preliminary outcome results have been reported, the preclinical studies and rationale for testing STAT3 inhibition in cancer patients already exist. A newly opened but not yet recruiting trial (NCT06964815) of silibinin, another STAT3 inhibitor, underscores the importance of establishing appropriate biomarkers of response, which appears to include patient sex and stratification by high versus low activation of the NF-κB/IL-6/STAT3 pathway.

Thus, we next determined whether sex differences in STAT3 activation correlate with sex differences in AZD1480 efficacy in vitro. In serum-free media, STAT3 activation is known to promote cell survival,¹⁴^,¹⁵ and we used cell viability assays to determine AZD1480 IC₅₀ values in each of the male and female lots. The female lots exhibited a wide distribution in IC₅₀ values, while male lots were generally sensitive to AZD1480 (Figure S7F). We probed p-STAT3 levels prior to AZD1480 treatment in the cells grown with or without EGF (Figures 5D and 5E) to determine if there was a correlation between initial p-STAT3 expression and sensitivity to AZD1480. Interestingly, two male lots exhibited elevated p-STAT3 expression relative to female lots upon serum starvation (Figures 5D, 5E, and S7B). There was a weak, nonsignificant (p = 0.13) correlation between p-STAT3 and sensitivity to AZD1480 in the female lots and no correlation between the two in the male lots (Figure S7G) in the cells grown without EGF.

Similar to measures of NF-κB signaling, we saw overlapping but distinct distributions of STAT3-related measures between sexes. Therefore, we examined if having a strong NF-κB gene signature (↑NF-κB)—a female-biased signature—was predictive of STAT3 activity and sensitivity to AZD1480. As expected, cell lines with a strong NF-κB gene signature had higher levels of p-STAT3 than those with a weak NF-κB gene signature in serum conditions (Figures S7B and S7H). Furthermore, p-STAT3 expression remained elevated in serum-starved conditions in cell lines with a strong NF-κB gene signature compared to cell lines with a weak signature (Figure 5F), and this corresponded to the increased sensitivity to AZD1480 (Figure S7I). EGF is known to increase STAT3 phosphorylation, and we next determined whether increasing STAT3 activation overall might improve our sensitivity for detecting AZD1480 effects. Similar to the no-EGF conditions, EGF treatment in female lots resulted in a moderate, but not significant, correlation (p = 0.08) between p-STAT3 expression (Figures 5D and 5E) and sensitivity to AZD1480 (Figure S7J). There was no correlation in the male murine lots. When combining both analyses, there was a significant correlation observed between p-STAT3 expression and sensitivity to AZD1480 in female lots, but not in male lots (Figure 5G). We conclude that the sensitivity to AZD1480 is strongly correlated to p-STAT3 levels in female cell lines, but not in male cell lines. This suggest that females, but not males, may be targetable with STAT3 inhibitors.

Sex differences in modulators of STAT3 activity

STAT3 activity is known to be modulated by other signaling pathways. We reasoned that these pathways might also be targetable in a sex-informed treatment. We previously demonstrated sex differences in cyclic AMP (cAMP) regulation, with female transformed astrocytes having a higher capacity for cAMP synthesis and greater sensitivity toward its inhibition.⁴² Furthermore, female patients with low-grade glioma had a greater incidence of single nucleotide polymorphisms (SNPs) in adenylate cyclase 8 (ADCY8, AC8).⁴² AC8 activity elevates cAMP, which increases PKA activity. PKA activity is known to regulate both p65⁴³ and STAT3 phosphorylation.⁴⁴ Therefore, we wondered if sex differences in NF-κB or STAT3 signaling are affected by sex differences in cAMP and PKA. Using lot 6 as a representative of male and female cells with a weak and strong NF-κB gene signature, respectively, we saw that the treatment with H-89, a PKA inhibitor, resulted in a significant decrease in STAT3 phosphorylation only in female cells (Figures 5H and 5I). Both male and female p-p65 levels were unresponsive to H-89 treatment. These results suggest that STAT3, but not p65, is regulated by PKA in female cells and that targeting cAMP and PKA may be another viable approach to the sex-informed, STAT3-directed treatment.

STAT3 is also a target of the EGF receptor (EGFR) kinase, and EGFR is a known modulator of STAT signaling.⁴⁵^,⁴⁶ In GBM, EGFR can be wild-type (WT), constitutively activated by mutation (EGFRvIII), or amplified.⁴⁶^,⁴⁷^,⁴⁸ Thus, we investigated whether the EGFR status affects STAT3 activation and sensitivity to AZD1480. Interestingly, EGF treatment of the male and female lots resulted in decreased STAT3 phosphorylation (Figures 5E and S7K) and decreased sensitivity to AZD1480 (Figures S7F and S7M), suggesting that EGFR activation can be a resistance mechanism to JAK2 inhibition. Similarly, when comparing cell lines with a strong or weak NF-κB gene signature, differences in p-STAT3 levels and AZD1480 sensitivity were ablated when cells were grown in the presence of EGF (Figures 5F, S7I, and S7N). Therefore, we determined whether the genetic status of EGFR affected STAT3 signaling.⁴⁶^,⁴⁹ We performed the univariate survival analysis of genes in the IL-6/JAK/STAT3 gene set from Figure 1D to compare HRs between patients with WT EGFR and EGFR-altered (alt) status.²⁴ Similar to our previous analysis, HRs were elevated when analyzing all patients with a WT EGFR status, which was driven by the female population rather than the male population (Figure 5J and Table S4). Strikingly, female patients with EGFR-altered tumors resulted in a significant reduction in the HRs, implying reduced or even no effect on survival. This corresponded with a reduction in the HRs from the all-patient analysis, while there was no change in the male-only analysis. These results suggest that IL-6/JAK/STAT3 signaling is prognostic for females with WT EGFR tumors compared to those with EGFR-altered tumors.

Discussion

Prior pan-cancer genomic studies have identified substantial and significant sex differences in inflammation and immunity pathway gene expression.³^,⁴^,¹⁰ To dig deeper into the clinical aspects and molecular underpinnings of these differences, we disaggregated TCGA data by sex and discovered that the TNFα/NF-κB/IL-6 axis had significant prognostic value for female GBM patients, but not for male patients. These data suggest that sex-informed targeting of this pathway could improve clinical response. Further, in a validated murine model of sex differences in GBM, we found that, mechanistically, sex differences in this pathway are dependent upon STAT3 phosphorylation, and STAT3 is likely to be the optimal therapeutic target for sex-informed treatments. These findings are particularly important as there are clinically available agents to block STAT3 activation (phosphorylation), and testing of whether there are sex differences in response to this approach in GBM and other cancers can be readily performed.¹⁴^,¹⁵^,¹⁶^,¹⁷^,²²^,⁵⁰ This effort has been fueled by the demonstrated roles of STAT3 signaling in GBM stem cell activity, tumor progression, and treatment resistance.¹⁵^,¹⁷^,²²^,⁵⁰ More recently, blocking STAT3 activation has been pursued as an adjuvant to immunotherapy.⁵¹^,⁵²^,⁵³

Sex differences in STAT3 signaling are not surprising. STAT3 is a critically important regulator of inflammatory signaling and immune function,¹⁴^,⁵¹ and lifelong sex differences in immunity are pervasive and well documented.¹^,⁵⁴^,⁵⁵ Both male and female patient populations exhibited variance in IL-6/JAK/STAT3 pathway activation. High levels of STAT3 activation have previously been associated with inferior survival in glioma patients.¹⁵ In our analysis of data disaggregated by sex, higher levels of IL-6/JAK/STAT3 pathway activation are associated with inferior survival only in female patients. A similar sex skew in a STAT3 survival correlation was recently reported for hepatocellular carcinoma.⁵⁶ Our findings extend our knowledge of STAT3 effects in GBM in a translationally important manner. They suggest that sex-informed stratification of patients for STAT3-directed therapy could increase the fraction of patients that respond to such treatments and avoid unnecessary toxicity of the treatment in those unlikely to respond. They also identify multiple targets whose inhibition can lower STAT3 phosphorylation in a female-biased manner.

STAT3 phosphorylation is a point of convergence for multiple regulatory pathways.¹⁴^,¹⁶^,¹⁷^,⁴³^,⁴⁵ Therefore, we sought to determine which of these are required for the observed sex differences. In addition to the TNF-α/NF-κB/IL-6 induction of STAT3 phosphorylation, STAT3 can be directly phosphorylated by PKA, and we found that the inhibition of PKA reduced STAT3 phosphorylation only in female cells.⁴⁴ Additionally, EGFR, a major component of GBM biology,⁴⁶^,⁴⁷^,⁴⁹ also induces STAT3 phosphorylation.⁴⁵^,⁴⁷ EGFR is genetically altered (constitutively activated and/or amplified) in more than 60% of GBM cases.⁵⁷ Moreover, patient data suggest that the difference in survival between female patients with low or high expression of STAT3 genes is evident only in tumors with WT EGFR status, suggesting that constitutive EGFR activation abrogates the prognostic interaction between sex and STAT3. We were able to recapitulate this by continuously treating transformed astrocytes with EGF, which resulted in a downregulation of p-STAT3 levels compared with those in untreated cells, and a decrease in sensitivity to the STAT3 inhibitor AZD1480. Therefore, sex differences in STAT3 signaling in GBM may be the greatest when basal signaling is within a normal dynamic range. This may also be relevant to the relationship between high basal NF-κB activity and the lack of response to TNF-α. Since STAT3 signaling is important in determining the efficacy of immunotherapies,⁵¹^,⁵²^,⁵³ these findings also suggest that sex differences in immunomodulatory therapies (either by checkpoint inhibitors or inhibitors targeting NF-κB and/or STAT3) may be greatest in case of GBM patients with WT EGFR tumors and that combinations of EGFR and STAT3 inhibitors might be worth considering. Another potential mediator of sex-biased STAT3 signaling could be secreted phosphoprotein 1 (SPP1). Female GBM patients have increased activation of SPP1,⁵⁸ and SPP1 has been reported to promote STAT3 signaling and a proneural-to-mesenchymal shift in GBM cells.⁵⁹ Future studies identifying to what extent each of these mediators promotes sex differences in STAT3 signaling will help tailor targeted treatment regimens.

These studies benefited from an analytical approach that was designed to detect differences between the sexes and individual variation within a sex. We previously used such an approach in a pan-cancer study that identified a female skew in inflammatory/immunity signatures and a male skew in oxidative phosphorylation and cell cycle signatures.³ Variation in the inflammatory signature uniquely identified male and female subsets that would exhibit the strongest responses to immune checkpoint inhibition in lung cancer. Therefore, when investigating sex differences, one should perform three kinds of analyses: (1) combined male and female data to detect overall sex differences, and disaggregated data to detect (2) variance between males, and similarly (3) variance between females. Indeed, some signatures (such as TNF-α signaling by NF-κB) in the human analyses arose when analyzing variance within a sex rather than comparing sex-skewed differences. This is also reflected in the survival analysis of both NF-κB and STAT3 gene sets, where greater variance in survival was observed in females, but not in males.

In parallel transcriptomic analyses in male and female transformed astrocytes, we found extensive concordance in the patient and murine datasets regarding the sex skew in the NF-κB signaling pathway. The degree of concordance validated the murine model for mechanistic studies. By using male and female littermates from five independent litters (lots), we were able to capture natural heterogeneity in TNF-α/NF-κB/IL-6/JAK/STAT3 phenotypes and identify male and female lots with high versus low levels of basal NF-κB activity. In the female lots, these expression differences defined two groups that differed across many measures. In contrast, no such separation was evident in the male lots where the single high-expressing NF-κB gene signature lot possessed a phenotype that fell within the distribution of the other four low-expressing lots. These results suggest that NF-κB and STAT3 signaling strongly stratify the female lots, but not the male lots, and highlight the value of examining variance within a sex.

When comparing the human and murine datasets, EMT, hypoxia, and myogenesis were also strongly enriched in all three analyses of both datasets. Interestingly, although STAT3, which can regulate EMT,⁶⁰ was a female-biased signal, EMT also strongly stratified males. This result highlights that although the same pathways may be involved in male and female phenotypes, the underlying changes in gene expression can differ.¹²^,³⁹ Further studies of sex differences in all three signaling pathways may be mechanistically and translationally fruitful.

There were some differences in our transcriptomic analyses of the murine transformed astrocytes and patient tumors. TNF-α signaling by the NF-κB gene signature strongly stratified female patients against each other and was evident only in the female-only analysis. In the murine cells, this signature was present in all three analyses, though it was the strongest in the female-only analysis. The inconsistency in the enrichment of NF-κB signature may be due to the absence of activating mutations of Egfr in the murine model. This condition is expected to increase the female effect in the combined analysis. Generation of other common mutational models will be beneficial in (1) recapitulating the full spectrum of sex differences seen in humans, and (2) identifying the pathways that exhibit sex differences under certain genetic backgrounds. Enrichment of pathways identified within the sex-specific analyses in the murine model, but not in the human analyses, may also be due to the heterogeneity in the numbers and types of non-neoplastic cells present in the biopsy.²¹ This could underlie the lack of concordance in the strength of cell cycle signatures in the murine and human datasets. Cell cycle-related signatures may be more enriched in tumor specimens of higher purity. Therefore, appropriate use of murine models can inform us of potential tumor intrinsic pathways that may be masked when analyzing tumors that have low purity.

Finally, the therapeutic value of these analyses and results must be validated in clinical trials. We carefully considered whether to perform in vivo studies and decided that these were not the critical next step in advancing clinical applications of what we have found. Both IL-6-blocking antibodies and STAT3 inhibition have been evaluated in cancer clinical trials, including those for GBM. The next critical steps are to perform retrospective analyses of clinical trial data in which patient sex is reliably annotated as well as new prospective clinical trials using sex-informed design and appropriate statistical models for detecting sex effects.

Limitations of the study

This study is limited in three ways. First, it is a retrospective analysis of patient transcriptomes and survival whose correlation remains to be shown in prospectively acquired data collected for this purpose. Second, preclinical murine models of GBM are frequently not predictive of clinical response, and additional retrospective analyses of data from cancer clinical trials testing NF-κB pathway blockade could further validate these findings. This could be especially important for the optimal design of prospective clinical trials to test whether there will be greater response to NF-κB pathway blockade in a subset of female patients. Finally, while many effects of sex on cancer biology are common across species like mice and humans, their underlying mechanisms may differ. Thus, in inbred strains of laboratory mice, males and females are autosomally isogenic and have identical X chromosomes and mitochondria. All males have the same Y chromosome. Beyond XX versus XY effects, sex differences in laboratory mice primarily arise from variable masculinization of female littermates by the male pups within the shared placenta and amniotic sac. This is not the case for humans whose autosomes, sex chromosomes, and mitochondria exhibit extraordinary inter-individual variations. These species differences in how sex differences arise may limit the predictive value of some findings.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Joshua B. Rubin (rubin_j@wustl.edu).

Materials availability

Materials generated in this study are available upon request from the lead contact with a completed materials transfer agreement. Murine transformed astrocyte lines will be made readily available upon request to the lead contact. No other unique reagents were generated in this study.

Data and code availability

•
Murine RNA-sequencing data from this study have been deposited at the NCBI Gene Expression Omnibus (GEO) database under accession number GEO: GSE295851 and are publicly available as of the date of publication. RNA-sequencing data for human tumors were obtained from Yanovich-Arad et al., Cell Reports, 2021 (DOI: https://doi.org/10.1016/j.celrep.2021.108787; GEO: GSE149009).²⁵
•
This paper does not report original code. Packages used for processing sequencing and survival data are denoted in the STAR Methods.
•
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

Work in the Rubin lab is supported by the National Cancer Institute R01 CA174737-07, P01CA245705, Joshua’s Great Things, St. Louis Children’s Hospital Foundation, Taylor Rozier’s Hope for a Cullen Foundation, Prayers from Marfan Foundation, and the Haubrich and Griffiths family foundations. We thank Fumihiko Urano for providing the NF-κB reporter constructs and the corresponding Renilla control. We thank the following cores at Washington University in St. Louis for help: Genome Technology Access Center (GTAC) at the McDonnell Genome Institute (MGI), the Spike-in Cooperative at the DNA Sequencing Innovation Lab (DSIL) at the Center for Genome Sciences & Systems Biology (CGS&SB), and the High-Throughput Screening Center. We thank the Brain Tumor Center at Washington University in St. Louis for shared equipment and helpful feedback, with special thanks to Yang (Eric) Li, Akanksha Mahajan, and Keerthana Nagesh Prabhu.

Author contributions

Conceptualization, J.P.W. and J.B.R.; formal analysis, J.P.W., L.L., L.T., and J.L.; validation, J.P.W., N.R., and J.B.R.; investigation, J.P.W., L.T., L.Y., N.R., N.M.W., and J.B.R.; methodology, J.P.W. and R.D.M.; data curation, J.P.W., L.L., L.T., and M.S.J.; visualization, J.P.W., L.L., and J.B.R.; resources, J.L., R.D.M., and J.B.R.; software, L.L., L.T., M.S.J., and J.L.; supervision, J.L., R.D.M., and J.B.R.; writing – original draft, J.P.W.; writing – review and editing, J.P.W., L.L., L.T., N.M.W., J.L., R.D.M., and J.B.R.; funding acquisition, J.B.R.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies

control α-IgG1	Biolegend	Cat#400432; RRID:AB_11150772
α-IL6 antibody	Biolegend	Cat#504512; RRID:AB_2814418
p-STAT3 Y705	Cell Signaling Technologies	Cat#9145; RRID:AB_2491009
STAT3	Cell Signaling Technologies	Cat#9139; RRID:AB_331757
p-p65 S536 NF-κB	Cell Signaling Technologies	Cat#3033; RRID:AB_331284
p65 NF-κB	Cell Signaling Technologies	Cat#6956; RRID:AB_10828935
β-actin	Cell Signaling Technologies	Cat#8457; RRID:AB_10950489
p-IκBα S32	Cell Signaling Technologies	Cat#2859; RRID:AB_561111
IκBα	Cell Signaling Technologies	Cat#4814; RRID:AB_390781
p-PKA C T197	Cell Signaling Technologies	Cat#5661; RRID:AB_10707163
PKA C	Cell Signaling Technologies	Cat#5842; RRID:AB_10706172
Anti-rabbit HRP	Cell Signaling Technologies	Cat#7074; RRID:AB_2099233
Anti-mouse HRP	Cell Signaling Technologies	Cat#7076; RRID:AB_330924

Chemicals, peptides, and recombinant proteins

Mouse Recombinant TNF-alpha	StemCell Technologies	Cat#78069
Human Recombinant EGF	StemCell Technologies	Cat#78006.1
AZD1480	Selleckchem	Cat#S2162
H-89 2-HCl	Selleckchem	Cat#S1582
Halt™ Protease and Phosphatase Inhibitor Cocktail (100X)	Thermo Fisher Scientific	Cat#78440
NuPAGE™ Sample Reducing Agent	Invitrogen	Cat#NP0004

Critical commercial assays

LookOut® Mycoplasma PCR Detection Kit	Sigma	Cat#MP0035-1KT
FuGENE® HD Transfection Reagent	Promega	Cat#E2311
Dual-Luciferase® Reporter Assay System	Promega	Cat#E1910
DC protein assay	Bio-Rad	Cat#5000112
Clarity™ Western ECL Substrate	Bio-Rad	Cat#1705060
Cytiva Amersham™ ECL™ Prime	Cytvia	Cat#RPN2232
RNeasy Plus Mini Kit	Qiagen	Cat#74136
QuantiTect Reverse Transcription Kit	Qiagen	Cat#205313
iTaq™ Universal SYBR® Green Supermix	Bio-Rad	Cat#1725121
CellTiter-Glo® Luminescent Cell Viability Assay	Promega	Cat#G7571
Mouse Il-6 ELISA kit	Proteintech	Cat#KE10007

Deposited data

Human GBM RNA-sequencing data	Yanovich-Arad et al.²⁵	DOI: https://doi.org/10.1016/j.celrep.2021.108787; GEO: GSE149009
Murine transformed astrocytes RNA-sequencing data	This paper	GEO: GSE295851
TCGA GBM Cohort	Gliovis (Bowman et al.)²³	DOI: https://doi.org/10.1093/neuonc/now247

Experimental models: Cell lines

Cell line: Nf1-/- DNp53 astrocyte	Sun et al.³¹ and Broestl et al.⁶	–

Oligonucleotides

Primers used for genotyping and gene quantification, see Table S5	Integrated DNA Technologies	–

Software and algorithms

ImageJ v1.53t	Fiji Software	RRID:SCR_003070
Primer-BLAST tool	NIH	RRID:SCR_003095
CutAdapt v2.10	Martin⁶¹	RRID:SCR_011841
STAR v2.7.6a	Dobin et al.⁶²	RRID:SCR_004463
Subread v2.0.2	Liao et al.⁶³	RRID:SCR_009803
iDEP2.0	Ge et al.²⁸	–
DESeq2	Love et al.³³	RRID:SCR_015687
bioDBnet: dbOrtho tool	Mudunuri et al.⁶⁴	–
GeneOverlap	Li Shen	RRID:SCR_018419
Gene enrichment/GSEA v4.3.2	Subramanian et al.²⁹, Mootha et al.³⁰, Howe et al.⁶⁵	RRID:SCR_016863
Survival package	Therneau and Grambsch⁶⁶	RRID:SCR_021137
Survminer package	Kassambara et al.⁶⁷	RRID:SCR_021094
Prism v.10.4.0	GraphPad	RRID:SCR_002798
Adobe Illustrator	Adobe	–
BioRender	BioRender	–

Other

pGL4.32[luc2P/NF-κB-RE/Hygro] NF-κB firefly luciferase plasmid	Kindly provided by Dr. Fumihiko Urano	–
pRL-TK Renilla reporter plasmid	Kindly provided by Dr. Fumihiko Urano	–
Amicon® Ultra Centrifugal Filter, 10 kDa MWCO	Millipore	Cat#UFC801024

Open in a new tab

Experimental model and study participant details

Cell lines

All cell lines were grown in 5% CO₂ at 37°C. The isolation and generation of Nf1-/- DNp53 model glioblastoma (GBM) male and female paired astrocyte cell lines (or lots) were previously generated as reported.⁶^,³¹ Generation of these cell lines was previously performed with mice that were approved by the Institutional Animal Care and Use Committee (IACUC) at Washington University in St. Louis in accordance with the Guide for the Care and Use of Laboratory Animals (NIH). To confirm the sex of each cell line, PCR amplification for the X or Y chromosome paralogs Kdm5c (Jarid1c) and Kdm5d (Jarid1d) (Forward: CTG AAG CTT TTG GCT TTG AG, Reverse: CCA CTG CCA AAT TCT TTG G) or Sry, a Y chromosome gene, (Forward: AGC CCT ACA GCC ACA TGA TA, Reverse: GTC TTG CCT GTA TGT GAT GG) was performed as previously described.⁶^,³² Nf1-/- DNp53 transformed astrocytes were maintained in DMEM/F12 (Gibco) supplemented with 10% FBS (Sigma) and 1% penicillin/streptomycin (Gibco). Cell lines were confirmed and routinely tested to be mycoplasma negative using the LookOut® Mycoplasma PCR Detection Kit (Sigma, MP0035-1KT). Cell lines were not used past 20 passages from collection.

Method details

Cell line treatment

Cells were treated with 10 ng/mL TNF-α (StemCell Technologies, 78069) at the time course indicated in the figure legends. For EGF “treatment”, cells were maintained with 10 ng/mL EGF (StemCell Technologies, 78006.1) in serum-free media over a course of at least three days followed by withdrawal for 24 hours. AZD1480 (S2162) or H-89 2-HCl (S1582) was purchased from Selleckchem, and dissolved in DMSO to make stock concentrations. Cells were treated with AZD1480 for 48 hours to obtain IC50 curves. Il-6 blocking was performed in serum-starved conditions of murine GBM cells using 10 μg/mL of either a control α-IgG1 (Biolegend, 400432, RRID:AB_11150772) or α-Il6 antibody (Biolegend, 504512, RRID:AB_2814418) for 24 hours.

Luciferase activity assay

Cells were plated and then reverse transfected with a pGL4.32[luc2P/NF-κB-RE/Hygro] NF-κB firefly luciferase (Promega) and a constitutive pRL-TK Renilla reporter (kind gifts from Dr. Fumihiko Urano) using FuGENE® HD Transfection Reagent (Promega, E2311) at a 3:1 reagent:DNA ratio. The ratio of NF-κB firefly luciferase to the control Renilla reporter was 10:1. Cells were harvested 72 hours after transfection. If applicable, cells were treated with 10 ng/mL TNF-α (StemCell Technologies, 78069) for the indicated time course. To measure luciferase levels, the Dual-Luciferase® Reporter Assay System (Promega, E1910) was performed according to the manufacturer instructions; cells were lysed for 15 minutes using the passive lysis buffer provided in the kit.

Immunoblot analysis

Cells were washed with cold PBS before being harvested using RIPA buffer (1% NP-40, 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 5 mM EDTA, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with Halt™ Protease and Phosphatase Inhibitor Cocktail (100X) (Thermo Fisher Scientific, 78440). Lysates were passaged through a 23-gauge x ¾” needle and clarified by centrifugation. Protein concentration was determined using the DC protein assay (Bio-Rad, 5000112). Samples were prepared using NuPAGE™ Sample Reducing Agent (Invitrogen, NP0004) and urea loading buffer (5x: 8M urea, 10% w/v SDS, 10 mM 2-mercaptoethanol, 20% v/v glycerol, 0.2M Tris-HCl pH 6.8, 0.05% w/v bromophenol blue). Proteins were resolved on NuPAGE™ Bis-Tris protein gels in NuPAGE™ MOPS SDS Running Buffer (Invitrogen, NP0001) and then transferred onto nitrocellulose membrane (Licor, 926-31092). Primary antibodies purchased from Cell Signaling Technologies include p-STAT3 Y705 (1:1000, 9145, RRID:AB_2491009), STAT3 (1:2500, 9139, RRID:AB_331757), p-p65 S536 NF-κB (1:1000, 3033, RRID:AB_331284), p65 NF-κB (1:2500, 6956, RRID:AB_10828935), β-actin (1:5000, 8457, RRID:AB_10950489), p-IκBα S32 (1:1000, 2859, RRID:AB_561111), IκBα (1:2500, 4814, RRID:AB_390781), p-PKA C T197 (1:1000, 5661, RRID:AB_10707163), and PKA C (1:2000, 5842, RRID:AB_10706172). Anti-rabbit (1:2500, 7074, RRID:AB_2099233) and anti-mouse (1:2500, 7076, RRID:AB_330924) secondary antibodies were purchased from Cell Signaling Technologies. Blots were developed using Clarity™ Western ECL Substrate (Bio-Rad, 1705060) or Cytiva Amersham™ ECL™ Prime (Cytvia, RPN2232). Blots probed for phospho-proteins were stripped using Nitro Stripping Buffer 5x (Licor, 928-40030) and probed with the corresponding total antibody afterwards. Image quantification performed on blots using ImageJ (v1.53t; RRID:SCR_003070).

Quantitative RT-PCR

RNA was isolated from cells using the RNeasy Plus Mini Kit (Qiagen, 74136) according to the manufacturer instructions. β-mercaptoethanol was added to Buffer RLT Plus as recommended by the manufacturer. RNA concentration was quantified using a NanoDrop 1000 spectrophotometer (Thermo Scientific), and 1 μg of RNA was used to synthesize cDNA using the QuantiTect Reverse Transcription Kit (Qiagen, 205313). Quantitative RT-PCR was performed using the iTaq™ Universal SYBR® Green Supermix (Bio-Rad, 1725121) on the CFX Connect Real-Time PCR Detection System (Bio-Rad). Primers were developed using the Integrated DNA Technologies (IDT) PrimerQuest Tool, checked using the NIH Primer-BLAST tool (RRID:SCR_003095), and confirmed by checking amplification and melt curves. β-actin primers were used as described previously.⁶ Primers used to measure gene expression are in Table S5.

CellTiter-Glo® (CTG)

Cell viability was measured using the CellTiter-Glo® Luminescent Cell Viability Assay (Promega, G7571) according to the manufacturer instructions. The assay was performed in Greiner CELLSTAR® 96 well plates (Sigma, M0187-32EA). Cell numbers plated were 500 (in serum conditions) or 1000 (in serum-starved conditions) per well. IC50s were determined 48 hours after treatment with AZD1480.

Il6 ELISA

Il6 in the supernatant was quantified using a mouse Il-6 ELISA kit (Proteintech, KE10007), and performed according to the manufacturer instructions. Briefly, cells were plated overnight, and supernatant was harvested 24 hours later. Supernatant was centrifuged to remove debris, concentrated using an Amicon® Ultra Centrifugal Filter, 10 kDa MWCO (Millipore, UFC801024), and snap frozen for further analysis. Total volume after concentration was noted to perform normalization after quantification by ELISA. Supernatant was diluted using sample diluent PT 1-ef at a 1:2 ratio when performing the ELISA.

Senescence-associated beta-galactosidase (SABG) assay values

SABG data was previously reported in Broestl et al., Commun Biol., 2022 in Figure 3C.⁶ Raw values were obtained and normalized for further linear regression analysis.

BRB RNA-sequencing (BRB-seq) for murine GBM samples

RNA was isolated from cells using the RNeasy Plus Mini Kit (Qiagen, 74136) according to the manufacturer instructions. Each cell line was cultured and harvested four separate times to generate libraries for replication. Library construction was performed in a similar manner as described previously.⁶⁸ Briefly, poly-adenylated transcripts are annealed to primers containing a 16-nucleotide sequence unique to each sample (sample barcode), a unique molecular identifier, and a poly A 3’ end (Integrated DNA Technologies). First-strand cDNA synthesis of barcoded transcripts was performed using Maxima H Minus (Invitrogen) with a template switch oligo (TSO). Hybrid transcript:cDNA was pooled and exonuclease treated (New England Biotech, M0293), followed by subsequent cDNA amplification (KAPA HiFi HotStart). Library construction was performed with 2 ng of amplified cDNA using the Illumina Nextera XT Protocol. Following sequencing, the original FASTQ files were demultiplexed using CutAdapt (v2.10, RRID:SCR_011841) to generate sample-level FASTQ files.⁶¹ Reads were aligned to the mouse genome (GRCm38) using STAR (v2.7.6a, RRID:SCR_004463),⁶² and raw counts were tabulated with Subread (v2.0.2, RRID:SCR_009803).⁶³

Murine GBM RNA-sequencing analysis

Preliminary analyses (filtering, PCA and clustering) was performed using iDEP2.0²⁸ with independent analysis performed afterwards. One replicate of the Lot 6 females was identified as an outlier from the other Lot 6 females and was removed along with the corresponding male replicate from further downstream analysis. One replicate of the Lot 12 males failed quality control (QC) and was removed. The corresponding Lot 12 female replicate was also removed from further downstream analysis. Genes were filtered requiring a minimum count per million (CPM) cutoff of at least 5 and present in at least n = 6 samples for all the three types of analyses: male and female together (all samples), male cell lines only, and female cell lines only. The CPM cutoff of 5 was identified using density plots. Outliers were identified by principal component analysis (PCA) and Euclidean clustering. Transformation was performed using VST transformation.³³ The top 1000 genes having the highest standard deviation was used for k-means clustering with the number of clusters determined using an elbow plot. K-means clustering was performed on data that was centered and scaled to have a mean of zero, a standard deviation of one, and a max z-score of -3 to 3. Gene enrichment was performed for each gene cluster to determine pathways enriched in each cluster.²⁹^,³⁰ DESeq2 analysis (RRID:SCR_015687) was performed using an FDR cutoff of 0.05 with a minimum log2 fold-change of 0.5 to determine differentially expressed genes (DEGs).³³ The top 13 non-sex chromosomal sex genes upregulated in male or female GBM cells was plotted on a volcano plot.⁶⁹ Sex chromosomal genes were referred using the University of California Santa Cruz Genome Browser (RRID:SCR_005780).⁷⁰

Human GBM RNA-sequencing analysis

Human GBM RNA-sequencing data was pulled from Yanovich-Arad et al., Cell Reports, 2021.²⁵ Patients without a negative IDH signature were removed from analysis. Patients were further screened for having no prior treatments (N = 28 male and N = 13 female, all primary tumors). Preliminary analysis (filtering, PCA, and clustering) was performed using iDEP2.0,²⁸ and identified OtB6181, OtB6208, and OtB6220 as outliers, which were removed from further downstream analysis. For gene filtering, a cutoff of 7 for minimum CPM (determined based on density plots) and expression present in at least n = 4 samples was used when analyzing all transcriptomes (male and female together). When analyzing male- or female-specific GBM transcriptomes separately, expression must be present in n = 8 or n = 4 samples, respectively; namely requiring expression in at least one-third of the samples. Rlog transformation was applied due to widely variable library sizes, and subsequent k-means clustering and DESeq2 (RRID:SCR_015687) analysis was performed in a similar manner as in the murine sample analysis (see above).

Gene enrichment (GE)/gene set enrichment analysis (GSEA)

Gene enrichment analysis was performed using the “Investigate” function under the human or mouse collections in the molecular signatures database (RRID:SCR_016863).²⁹^,³⁰ Enrichment for MH: orthology-mapped hallmark gene sets (mouse) or H: hallmark gene sets (human) was determined for genes within different clusters identified by k-means clustering from RNA-sequencing data. Default parameters were used (FDR q-value cutoff of less than 0.05) with the exception of showing the top 20 significant pathways rather than 10. When comparing overlapping gene sets from human and murine datasets, human Ensembl gene IDs were converted into the murine ortholog using the bioDBnet: dbOrtho tool from the National Cancer Institute.⁶⁴ Venn diagram figures were generated using the Molbiotools – multiple list comparator tool. Hypergeometric distribution analysis was conducted to test for the significance of the overlapping gene lists using the GeneOverlap (RRID:SCR_018419) package.

Gene set enrichment analysis (GSEA, v4.3.2) was performed (RRID:SCR_003199) on filtered and normalized (non-transformed) counts obtained using the DESeq2 (RRID:SCR_015687) package³³ to identify gene sets enriched with differential genes. Default parameters were utilized with a phenotype permutation used in analyses with at least 7 samples, otherwise gene set permutation was used. For the male- or female-only human GBM sample analyses, we generated two groups (denoted as cluster 1 and cluster 2) for GSEA by using the dendrogram from the Euclidean hierarchical clustering analysis. As the first branching point for the male samples would generate groups of 2 and 23 samples, which would be insufficient to examine variance across samples, we used the first and second branching point to generate two groups. For the female samples, the first branching point was used. For murine GBM cells, counts were averaged across the replicates (e.g. Lot 6 male replicate 1, 2, and 3) from the normalized count table before running GSEA.

Sex-biased cluster analysis

To identify sex-biased clusters, samples were grouped into approximately three even groups according to the clustering dendrogram. For the human transcriptomes, this resulted in a left (male-biased) group composing of samples from OtB6233 to OtB6232 (n = 12), a middle group (samples OtB6197 to OtB6204, n = 14), and a right (female-biased) group composing of samples OtB6215 to OtB6190 (n = 12). A Kruskal-Wallis test was used to examine if each group was significantly different from each other when comparing the z-scores of all genes within a group in a cluster. A cluster was considered a male-biased or female-biased cluster if the left or right group, respectively, had a z-score median difference greater than 0.5 and was also significantly different to all other groups. If the z-score median was lower for the sex-defining group, the cluster was identified as genes being downregulated in a sex-biased manner, whereas a higher median z-score denoted a cluster containing genes upregulated in a sex-biased manner. If a cluster had both a significantly upregulated and downregulated sex-biased group that met the z-score difference criteria (which was the case for only cluster 4), the cluster was defined based on the upregulated group. A similar analysis was performed for the murine transcriptomes. Since each cell line had multiple replicates, the average z-score was taken prior to testing for significance. The left (male-biased) group contained samples from Lot 8 males to Lot 11 males (n = 4), the middle contained Lot 8 females and Lot 7 males (n = 2), and the right (female-biased) group contained samples Lot 7 females to Lot 12 females (n = 4).

Univariate survival analysis

Kaplan-Meier survival curves were generated on individual gene expression (Hgu133 microarray platform) data of The Cancer Genome Atlas (TCGA) GBM cohort downloaded, together with the corresponding phenotypes, from glioVis: http://gliovis.bioinfo.cnio.es.²³ Only patients with known sex and IDH1 WT tumors were used for downstream analysis. Patients were stratified as high and low using the candidate gene’s median mRNA expression within all patients (for all-sample analysis) or their corresponding sex (for sex-specific analysis). The NF-κB gene list specific to the MES subtype was obtained from Bhat et al., Cancer Cell, 2013 from Table S6.¹⁹ IL6/JAK/STAT3 gene list was pulled from the MSigDB Molecular Signatures database (https://www.gsea-msigdb.org/gsea/msigdb/human/geneset/HALLMARK_IL6_JAK_STAT3_SIGNALING.html) in GSEA (RRID:SCR_016863).²⁹^,³⁰^,⁶⁵ Analysis was performed using the survival (RRID:SCR_021137) and survminer (RRID:SCR_021094) packages in R to determine significance and hazard ratios for each candidate gene for all patients, male-only, and female-only analyses.⁶⁶^,⁶⁷ EGFR status of patients was determined using TCGA.

Quantification and statistical analysis

The number of samples (or lots), replicates, and statistical analysis used are described in the figure legends. Statistical analysis of data was performed using GraphPad Prism v.10.4.0 (RRID:SCR_002798). All tests were two-sided, and significance was defined if p < 0.05. For comparison of two groups, the student’s unpaired t-test was used if the distributions were normal using the D’Agostino-Pearson omnibus (K2) test, under equal variances (tested based on the F test), while unpaired t-test with a Welch’s correction was applied under unequal variance. If data distributions were not normal, Mann-Whitney test was used for two groups and Kruskal-Wallis test was used for multiple groups followed by Dunn’s test for post hoc pairwise comparisons. When comparing more than two groups or multiple factors are modeled, ANOVA was used for normally distributed data to test for the significance of each factor and the interaction of factors, followed by a Tukey’s test for post hoc pairwise comparisons.

A simple linear regression was used, separately for each sex, to model the relationship between two variables (in Figures 4, 5, and S7) with the exception of Figure 4D where a simple linear regression model was fit to the male and a one phase decay model was fit to female with the model assessed and selected based on model goodness of fit based on F test (p < 0.05). For the strength of correlation between two variables (in Figures 4, 5, and S7), Pearson correlation (r) and equivalently coefficient of determination (R²) in simple linear regression, and accompanied p-value, were provided for normally distributed data; otherwise, nonparametric Spearman correlation was used instead.

The HRs of genes in Figures 1, 5, and S1 (high vs. low expression dichotomized by median) were estimated from univariate Cox regression model with HRs compared across patient groups using a Kruskal-Wallis test, followed by a Dunn’s test for post-hoc pairwise comparisons with multiplicity corrections. The Kaplan-Meier method was applied to estimate overall survival (OS) by high and low expression of individual genes, while survival difference was compared by a log-rank test.

Published: January 21, 2026

Footnotes

Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.114761.

Supplemental information

Document S1. Figures S1–S7 and Tables S1–S3 and S5

mmc1.pdf^{(1.7MB, pdf)}

Table S4. Hazard ratios of IL-6/JAK/STAT3 genes when analyzing all (male and female), male-only, or female-only patients with EGFR WT or EGFR-altered status, related to Figure 5

mmc2.xlsx^{(21.2KB, xlsx)}

Data S1. K-means clustering of top 1,000 genes from male and female (ALL) analysis of human GBM transcriptomes, related to Figure 2

mmc3.xlsx^{(467.9KB, xlsx)}

Data S2. Gene enrichment analysis of clusters from all sample, male-only, and female-only analyses of human GBM samples, related to Figure 2

mmc4.xlsx^{(18.6KB, xlsx)}

Data S3. DEGs of male vs. female human GBM samples, related to Figure 2

mmc5.xlsx^{(1.6MB, xlsx)}

Data S4. K-means clustering of top 1,000 genes from male-only analysis of human GBM transcriptomes, related to Figure 2

mmc6.xlsx^{(320.3KB, xlsx)}

Data S5. K-means clustering of top 1,000 genes from female-only analysis of human GBM transcriptomes, related to Figure 2

mmc7.xlsx^{(183.7KB, xlsx)}

Data S6. GSEA of male GBM compared to other male GBM, and similarly in female GBM compared to other female GBM, related to Figures 2 and S2

mmc8.xlsx^{(15.3KB, xlsx)}

Data S7. Raw counts of genes from murine transformed astrocyte transcriptomes, related to Figure 3

mmc9.xlsx^{(6.5MB, xlsx)}

Data S8. DEGs of male vs. female murine transformed astrocyte samples, related to Figure 3

mmc10.xlsx^{(995.7KB, xlsx)}

Data S9. Gene enrichment analysis of clusters from all sample (male and female), male-only, and female-only analyses of murine GBM samples, related to Figure 3

mmc11.xlsx^{(26.3KB, xlsx)}

Data S10. GSEA of murine male transformed astrocytes compared to female transformed astrocytes, related to Figures 3 and S3

mmc12.xlsx^{(13.9KB, xlsx)}

Data S11. Enrichment of genes identified in all three analyses, male-only analysis, or female-only analysis from human GBM transcriptomes, related to Figures 3 and S4

mmc13.xlsx^{(12.5KB, xlsx)}

Data S12. Enrichment of genes identified in all three analyses, male-only analysis, or female-only analysis from murine GBM transcriptomes, related to Figures 3 and S4

mmc14.xlsx^{(12.6KB, xlsx)}

Data S13. Enrichment of overlapping genes from human and murine analyses, related to Figures 2, 3, and S4

mmc15.xlsx^{(11.9KB, xlsx)}

References

1.Klein S.L., Flanagan K.L. Sex differences in immune responses. Nat. Rev. Immunol. 2016;16:626–638. doi: 10.1038/nri.2016.90. [DOI] [PubMed] [Google Scholar]
2.Rubin J.B., Abou-Antoun T., Ippolito J.E., Llaci L., Marquez C.T., Wong J.P., Yang L. Epigenetic developmental mechanisms underlying sex differences in cancer. J. Clin. Investig. 2024;134 doi: 10.1172/JCI180071. [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Yang W., Rubin J.B. Treating sex and gender differences as a continuous variable can improve precision cancer treatments. Biol. Sex Differ. 2024;15:35. doi: 10.1186/s13293-024-00607-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Han J., Yang Y., Li X., Wu J., Sheng Y., Qiu J., Wang Q., Li J., He Y., Cheng L., Zhang Y. Pan-cancer analysis reveals sex-specific signatures in the tumor microenvironment. Mol. Oncol. 2022;16:2153–2173. doi: 10.1002/1878-0261.13203. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Rubin J.B., Lagas J.S., Broestl L., Sponagel J., Rockwell N., Rhee G., Rosen S.F., Chen S., Klein R.S., Imoukhuede P., Luo J. Sex differences in cancer mechanisms. Biol. Sex Differ. 2020;11:17. doi: 10.1186/s13293-020-00291-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Broestl L., Warrington N.M., Grandison L., Abou-Antoun T., Tung O., Shenoy S., Tallman M.M., Rhee G., Yang W., Sponagel J., et al. Gonadal sex patterns p21-induced cellular senescence in mouse and human glioblastoma. Commun. Biol. 2022;5:781. doi: 10.1038/s42003-022-03743-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Lee J., Nicosia M., Hong E.S., Silver D.J., Li C., Bayik D., Watson D.C., Lauko A., Kay K.E., Wang S.Z., et al. Sex-Biased T-cell Exhaustion Drives Differential Immune Responses in Glioblastoma. Cancer Discov. 2023;13:2090–2105. doi: 10.1158/2159-8290.CD-22-0869. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Bayik D., Zhou Y., Park C., Hong C., Vail D., Silver D.J., Lauko A., Roversi G., Watson D.C., Lo A., et al. Myeloid-Derived Suppressor Cell Subsets Drive Glioblastoma Growth in a Sex-Specific Manner. Cancer Discov. 2020;10:1210–1225. doi: 10.1158/2159-8290.CD-19-1355. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Ochocka N., Segit P., Wojnicki K., Cyranowski S., Swatler J., Jacek K., Grajkowska W., Kaminska B. Specialized functions and sexual dimorphism explain the functional diversity of the myeloid populations during glioma progression. Cell Rep. 2023;42 doi: 10.1016/j.celrep.2022.111971. [DOI] [PubMed] [Google Scholar]
10.Ye Y., Jing Y., Li L., Mills G.B., Diao L., Liu H., Han L. Sex-associated molecular differences for cancer immunotherapy. Nat. Commun. 2020;11 doi: 10.1038/s41467-020-15679-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Ostrom Q.T., Cioffi G., Gittleman H., Patil N., Waite K., Kruchko C., Barnholtz-Sloan J.S. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2012-2016. Neuro Oncol. 2019;21:v1–v100. doi: 10.1093/neuonc/noz150. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Yang W., Warrington N.M., Taylor S.J., Whitmire P., Carrasco E., Singleton K.W., Wu N., Lathia J.D., Berens M.E., Kim A.H., et al. Sex differences in GBM revealed by analysis of patient imaging, transcriptome, and survival data. Sci. Transl. Med. 2019;11 doi: 10.1126/scitranslmed.aao5253. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Maccari M., Baek C., Caccese M., Mandruzzato S., Fiorentino A., Internò V., Bosio A., Cerretti G., Padovan M., Idbaih A., Lombardi G. Present and Future of Immunotherapy in Patients With Glioblastoma: Limitations and Opportunities. Oncologist. 2024;29:289–302. doi: 10.1093/oncolo/oyad321. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Johnson D.E., O’Keefe R.A., Grandis J.R. Targeting the IL-6/JAK/STAT3 signalling axis in cancer. Nat. Rev. Clin. Oncol. 2018;15:234–248. doi: 10.1038/nrclinonc.2018.8. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Tan M.S.Y., Sandanaraj E., Chong Y.K., Lim S.W., Koh L.W.H., Ng W.H., Tan N.S., Tan P., Ang B.T., Tang C. A STAT3-based gene signature stratifies glioma patients for targeted therapy. Nat. Commun. 2019;10:3601. doi: 10.1038/s41467-019-11614-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Hedvat M., Huszar D., Herrmann A., Gozgit J.M., Schroeder A., Sheehy A., Buettner R., Proia D., Kowolik C.M., Xin H., et al. The JAK2 Inhibitor AZD1480 Potently Blocks Stat3 Signaling and Oncogenesis in Solid Tumors. Cancer Cell. 2009;16:487–497. doi: 10.1016/j.ccr.2009.10.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.McFarland B.C., Hong S.W., Rajbhandari R., Twitty G.B., Gray G.K., Yu H., Benveniste E.N., Nozell S.E. NF-κB-induced IL-6 ensures STAT3 activation and tumor aggressiveness in glioblastoma. PLoS One. 2013;8 doi: 10.1371/journal.pone.0078728. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Taniguchi K., Karin M. NF-κB, inflammation, immunity and cancer: coming of age. Nat. Rev. Immunol. 2018;18:309–324. doi: 10.1038/nri.2017.142. [DOI] [PubMed] [Google Scholar]
19.Bhat K.P.L., Balasubramaniyan V., Vaillant B., Ezhilarasan R., Hummelink K., Hollingsworth F., Wani K., Heathcock L., James J.D., Goodman L.D., et al. Mesenchymal Differentiation Mediated by NF-κB Promotes Radiation Resistance in Glioblastoma. Cancer Cell. 2013;24:331–346. doi: 10.1016/j.ccr.2013.08.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Schmitt M.J., Company C., Dramaretska Y., Barozzi I., Göhrig A., Kertalli S., Großmann M., Naumann H., Sanchez-Bailon M.P., Hulsman D., et al. Phenotypic Mapping of Pathologic Cross-Talk between Glioblastoma and Innate Immune Cells by Synthetic Genetic Tracing. Cancer Discov. 2021;11:754–777. doi: 10.1158/2159-8290.CD-20-0219. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Wang L., Jung J., Babikir H., Shamardani K., Jain S., Feng X., Gupta N., Rosi S., Chang S., Raleigh D., et al. A single-cell atlas of glioblastoma evolution under therapy reveals cell-intrinsic and cell-extrinsic therapeutic targets. Nat. Cancer. 2022;3:1534–1552. doi: 10.1038/s43018-022-00475-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Loftus A.E.P., Romano M.S., Phuong A.N., McKinnel B.J., Muir M.T., Furqan M., Dawson J.C., Avalle L., Douglas A.T., Mort R.L., et al. An ILK/STAT3 pathway controls glioblastoma stem cell plasticity. Dev. Cell. 2024;59:3197–3212.e7. doi: 10.1016/j.devcel.2024.09.003. [DOI] [PubMed] [Google Scholar]
23.Bowman R.L., Wang Q., Carro A., Verhaak R.G.W., Squatrito M. GlioVis data portal for visualization and analysis of brain tumor expression datasets. Neuro Oncol. 2017;19:139–141. doi: 10.1093/neuonc/now247. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Liberzon A., Birger C., Thorvaldsdóttir H., Ghandi M., Mesirov J.P., Tamayo P. The Molecular Signatures Database Hallmark Gene Set Collection. Cell Syst. 2015;1:417–425. doi: 10.1016/j.cels.2015.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Yanovich-Arad G., Ofek P., Yeini E., Mardamshina M., Danilevsky A., Shomron N., Grossman R., Satchi-Fainaro R., Geiger T. Proteogenomics of glioblastoma associates molecular patterns with survival. Cell Rep. 2021;34 doi: 10.1016/j.celrep.2021.108787. [DOI] [PubMed] [Google Scholar]
26.Fang Z., Cui X. Design and validation issues in RNA-seq experiments. Brief. Bioinform. 2011;12:280–287. doi: 10.1093/bib/bbr004. [DOI] [PubMed] [Google Scholar]
27.Rubin J.B. The spectrum of sex differences in cancer. Trends Cancer. 2022;8:303–315. doi: 10.1016/j.trecan.2022.01.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Ge S.X., Son E.W., Yao R. iDEP: an integrated web application for differential expression and pathway analysis of RNA-Seq data. BMC Bioinf. 2018;19:534. doi: 10.1186/s12859-018-2486-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Subramanian A., Tamayo P., Mootha V.K., Mukherjee S., Ebert B.L., Gillette M.A., Paulovich A., Pomeroy S.L., Golub T.R., Lander E.S., Mesirov J.P. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA. 2005;102:15545–15550. doi: 10.1073/pnas.0506580102. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Mootha V.K., Lindgren C.M., Eriksson K.-F., Subramanian A., Sihag S., Lehar J., Puigserver P., Carlsson E., Ridderstråle M., Laurila E., et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 2003;34:267–273. doi: 10.1038/ng1180. [DOI] [PubMed] [Google Scholar]
31.Sun T., Warrington N.M., Luo J., Brooks M.D., Dahiya S., Snyder S.C., Sengupta R., Rubin J.B. Sexually dimorphic RB inactivation underlies mesenchymal glioblastoma prevalence in males. J. Clin. Investig. 2014;124:4123–4133. doi: 10.1172/JCI71048. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Sponagel J., Jones J.K., Frankfater C., Zhang S., Tung O., Cho K., Tinkum K.L., Gass H., Nunez E., Spitz D.R., et al. Sex differences in brain tumor glutamine metabolism reveal sex-specific vulnerabilities to treatment. Med. 2022;3:792–811.e12. doi: 10.1016/j.medj.2022.08.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Love M.I., Huber W., Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15:550. doi: 10.1186/s13059-014-0550-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Zhu Z., Zhang X., Yu Z., Zhou Y., Zhu S., Zhang Y.H., Lin X.P., Mou Y., Zhang J. Correlation of Tim-3 expression with chemokine levels for predicting the prognosis of patients with glioblastoma. J. Neuroimmunol. 2021;355 doi: 10.1016/j.jneuroim.2021.577575. [DOI] [PubMed] [Google Scholar]
35.Comba A., Faisal S.M., Dunn P.J., Argento A.E., Hollon T.C., Al-Holou W.N., Varela M.L., Zamler D.B., Quass G.L., Apostolides P.F., et al. Spatiotemporal analysis of glioma heterogeneity reveals COL1A1 as an actionable target to disrupt tumor progression. Nat. Commun. 2022;13:3606. doi: 10.1038/s41467-022-31340-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Day B.W., Stringer B.W., Al-Ejeh F., Ting M.J., Wilson J., Ensbey K.S., Jamieson P.R., Bruce Z.C., Lim Y.C., Offenhäuser C., et al. EphA3 Maintains Tumorigenicity and Is a Therapeutic Target in Glioblastoma Multiforme. Cancer Cell. 2013;23:238–248. doi: 10.1016/j.ccr.2013.01.007. [DOI] [PubMed] [Google Scholar]
37.Jeon H.-M., Shin Y.J., Lee J., Chang N., Woo D.-H., Lee W.J., Nguyen D., Kang W., Cho H.J., Yang H., et al. The semaphorin 3A/neuropilin-1 pathway promotes clonogenic growth of glioblastoma via activation of TGF-β signaling. JCI Insight. 2023;8 doi: 10.1172/jci.insight.167049. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Kfoury N., Sun T., Yu K., Rockwell N., Tinkum K.L., Qi Z., Warrington N.M., McDonald P., Roy A., Weir S.J., et al. Cooperative p16 and p21 action protects female astrocytes from transformation. Acta Neuropathol. Commun. 2018;6:12. doi: 10.1186/s40478-018-0513-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Kfoury N., Qi Z., Prager B.C., Wilkinson M.N., Broestl L., Berrett K.C., Moudgil A., Sankararaman S., Chen X., Gertz J., et al. Brd4-bound enhancers drive cell-intrinsic sex differences in glioblastoma. Proc. Natl. Acad. Sci. USA. 2021;118 doi: 10.1073/pnas.2017148118. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Rockwell N.C., Yang W., Warrington N.M., Staller M.V., Griffith M., Griffith O.L., Gurnett C.A., Cohen B.A., Baldridge D., Rubin J.B. Sex- and Mutation-Specific p53 Gain-of-Function Activity in Gliomagenesis. Cancer Res. Commun. 2021;1:148–163. doi: 10.1158/2767-9764.CRC-21-0026. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Hoffmann A., Natoli G., Ghosh G. Transcriptional regulation via the NF-κB signaling module. Oncogene. 2006;25:6706–6716. doi: 10.1038/sj.onc.1209933. [DOI] [PubMed] [Google Scholar]
42.Warrington N.M., Sun T., Luo J., McKinstry R.C., Parkin P.C., Ganzhorn S., Spoljaric D., Albers A.C., Merkelson A., Stewart D.R., et al. The cyclic AMP pathway is a sex-specific modifier of glioma risk in type I neurofibromatosis patients. Cancer Res. 2015;75:16–21. doi: 10.1158/0008-5472.CAN-14-1891. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Zhong H., Voll R.E., Ghosh S. Phosphorylation of NF-kappa B p65 by PKA stimulates transcriptional activity by promoting a novel bivalent interaction with the coactivator CBP/p300. Mol. Cell. 1998;1:661–671. doi: 10.1016/s1097-2765(00)80066-0. [DOI] [PubMed] [Google Scholar]
44.Wang W., Guo C., Zhu P., Lu J., Li W., Liu C., Xie H., Myatt L., Chen Z.-J., Sun K. Phosphorylation of STAT3 mediates the induction of cyclooxygenase-2 by cortisol in the human amnion at parturition. Sci. Signal. 2015;8 doi: 10.1126/scisignal.aac6151. [DOI] [PubMed] [Google Scholar]
45.Chua C.Y., Liu Y., Granberg K.J., Hu L., Haapasalo H., Annala M.J., Cogdell D.E., Verploegen M., Moore L.M., Fuller G.N., et al. IGFBP2 potentiates nuclear EGFR–STAT3 signaling. Oncogene. 2016;35:738–747. doi: 10.1038/onc.2015.131. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Hoogstrate Y., Ghisai S.A., de Wit M., de Heer I., Draaisma K., van Riet J., van de Werken H.J.G., Bours V., Buter J., Vanden Bempt I., et al. The EGFRvIII transcriptome in glioblastoma: A meta-omics analysis. Neuro Oncol. 2022;24:429–441. doi: 10.1093/neuonc/noab231. [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Li X.-P., Guo Z.-Q., Wang B.-F., Zhao M. EGFR alterations in glioblastoma play a role in antitumor immunity regulation. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1236246. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Rabah N., Ait Mohand F.-E., Kravchenko-Balasha N. Understanding Glioblastoma Signaling, Heterogeneity, Invasiveness, and Drug Delivery Barriers. Int. J. Mol. Sci. 2023;24 doi: 10.3390/ijms241814256. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Felsberg J., Hentschel B., Kaulich K., Gramatzki D., Zacher A., Malzkorn B., Kamp M., Sabel M., Simon M., Westphal M., et al. Epidermal Growth Factor Receptor Variant III (EGFRvIII) Positivity in EGFR-Amplified Glioblastomas: Prognostic Role and Comparison between Primary and Recurrent Tumors. Clin. Cancer Res. 2017;23:6846–6855. doi: 10.1158/1078-0432.CCR-17-0890. [DOI] [PubMed] [Google Scholar]
50.Zhang Y., Xiao X., Yang G., Jiang X., Jiao S., Nie Y., Zhang T. STAT3/TGFBI signaling promotes the temozolomide resistance of glioblastoma through upregulating glycolysis by inducing cellular senescence. Cancer Cell Int. 2025;25:127. doi: 10.1186/s12935-025-03770-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Zou S., Tong Q., Liu B., Huang W., Tian Y., Fu X. Targeting STAT3 in Cancer Immunotherapy. Mol. Cancer. 2020;19:145. doi: 10.1186/s12943-020-01258-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Ganesh S., Kim M.J., Lee J., Feng X., Ule K., Mahan A., Krishnan H.S., Wang Z., Anzahaee M.Y., Singhal G., et al. RNAi mediated silencing of STAT3/PD-L1 in tumor-associated immune cells induces robust anti-tumor effects in immunotherapy resistant tumors. Mol. Ther. 2024;32:1895–1916. doi: 10.1016/j.ymthe.2024.03.035. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Witt K., Evans-Axelsson S., Lundqvist A., Johansson M., Bjartell A., Hellsten R. Inhibition of STAT3 augments antitumor efficacy of anti-CTLA-4 treatment against prostate cancer. Cancer Immunol. Immunother. 2021;70:3155–3166. doi: 10.1007/s00262-021-02915-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Dunn S.E., Perry W.A., Klein S.L. Mechanisms and consequences of sex differences in immune responses. Nat. Rev. Nephrol. 2024;20:37–55. doi: 10.1038/s41581-023-00787-w. [DOI] [PubMed] [Google Scholar]
55.Roved J., Westerdahl H., Hasselquist D. Sex differences in immune responses: Hormonal effects, antagonistic selection, and evolutionary consequences. Horm. Behav. 2017;88:95–105. doi: 10.1016/j.yhbeh.2016.11.017. [DOI] [PubMed] [Google Scholar]
56.Huang C.-Y., Tan K.-T., Huang S.-F., Lu Y.-J., Wang Y.-H., Chen S.-J., Tse K.-P. Study of sex-biased differences in genomic profiles in East Asian hepatocellular carcinoma. Discov. Oncol. 2024;15:276. doi: 10.1007/s12672-024-01131-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Heimberger A.B., Hlatky R., Suki D., Yang D., Weinberg J., Gilbert M., Sawaya R., Aldape K. Prognostic effect of epidermal growth factor receptor and EGFRvIII in glioblastoma multiforme patients. Clin. Cancer Res. 2005;11:1462–1466. doi: 10.1158/1078-0432.CCR-04-1737. [DOI] [PubMed] [Google Scholar]
58.Jang B., Yoon D., Lee J.Y., Kim J., Hong J., Koo H., Sa J.K. Integrative multi-omics characterization reveals sex differences in glioblastoma. Biol. Sex Differ. 2024;15:23. doi: 10.1186/s13293-024-00601-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Yu W., Gui S., Peng L., Luo H., Xie J., Xiao J., Yilamu Y., Sun Y., Cai S., Cheng Z., Tao Z. STAT3-controlled CHI3L1/SPP1 positive feedback loop demonstrates the spatial heterogeneity and immune characteristics of glioblastoma. Dev. Cell. 2025;60:1751–1767.e9. doi: 10.1016/j.devcel.2025.01.014. [DOI] [PubMed] [Google Scholar]
60.Lin W.-H., Chang Y.-W., Hong M.-X., Hsu T.-C., Lee K.-C., Lin C., Lee J.-L. STAT3 phosphorylation at Ser727 and Tyr705 differentially regulates the EMT-MET switch and cancer metastasis. Oncogene. 2021;40:791–805. doi: 10.1038/s41388-020-01566-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Martin M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 2011;17:10–12. doi: 10.14806/ej.17.1.200. [DOI] [Google Scholar]
62.Dobin A., Davis C.A., Schlesinger F., Drenkow J., Zaleski C., Jha S., Batut P., Chaisson M., Gingeras T.R. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29:15–21. doi: 10.1093/bioinformatics/bts635. [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Liao Y., Smyth G.K., Shi W. The Subread aligner: fast, accurate and scalable read mapping by seed-and-vote. Nucleic Acids Res. 2013;41 doi: 10.1093/nar/gkt214. [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Mudunuri U., Che A., Yi M., Stephens R.M. bioDBnet: the biological database network. Bioinformatics. 2009;25:555–556. doi: 10.1093/bioinformatics/btn654. [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Howe D.G., Blake J.A., Bradford Y.M., Bult C.J., Calvi B.R., Engel S.R., Kadin J.A., Kaufman T.C., Kishore R., Laulederkind S.J.F., et al. Model organism data evolving in support of translational medicine. Lab Anim. 2018;47:277–289. doi: 10.1038/s41684-018-0150-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Therneau T.M., Grambsch P.M. Springer; 2000. Modeling Survival Data: Extending the Cox Model. [DOI] [Google Scholar]
67.Kassambara A., Kosinski M., Biecek P., Fabian S. 2025. survminer: Drawing Survival Curves using “ggplot2.” Version 0.5.1.https://cran.r-project.org/web/packages/survminer/index.html [Google Scholar]
68.Alpern D., Gardeux V., Russeil J., Mangeat B., Meireles-Filho A.C.A., Breysse R., Hacker D., Deplancke B. BRB-seq: ultra-affordable high-throughput transcriptomics enabled by bulk RNA barcoding and sequencing. Genome Biol. 2019;20:71. doi: 10.1186/s13059-019-1671-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Su S., Law C.W., Ah-Cann C., Asselin-Labat M.-L., Blewitt M.E., Ritchie M.E. Glimma: interactive graphics for gene expression analysis. Bioinformatics. 2017;33:2050–2052. doi: 10.1093/bioinformatics/btx094. [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Perez G., Barber G.P., Benet-Pages A., Casper J., Clawson H., Diekhans M., Fischer C., Gonzalez J.N., Hinrichs A.S., Lee C.M., et al. The UCSC Genome Browser database: 2025 update. Nucleic Acids Res. 2025;53:D1243–D1249. doi: 10.1093/nar/gkae974. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Document S1. Figures S1–S7 and Tables S1–S3 and S5

mmc1.pdf^{(1.7MB, pdf)}

Table S4. Hazard ratios of IL-6/JAK/STAT3 genes when analyzing all (male and female), male-only, or female-only patients with EGFR WT or EGFR-altered status, related to Figure 5

mmc2.xlsx^{(21.2KB, xlsx)}

Data S1. K-means clustering of top 1,000 genes from male and female (ALL) analysis of human GBM transcriptomes, related to Figure 2

mmc3.xlsx^{(467.9KB, xlsx)}

Data S2. Gene enrichment analysis of clusters from all sample, male-only, and female-only analyses of human GBM samples, related to Figure 2

mmc4.xlsx^{(18.6KB, xlsx)}

Data S3. DEGs of male vs. female human GBM samples, related to Figure 2

mmc5.xlsx^{(1.6MB, xlsx)}

Data S4. K-means clustering of top 1,000 genes from male-only analysis of human GBM transcriptomes, related to Figure 2

mmc6.xlsx^{(320.3KB, xlsx)}

Data S5. K-means clustering of top 1,000 genes from female-only analysis of human GBM transcriptomes, related to Figure 2

mmc7.xlsx^{(183.7KB, xlsx)}

Data S6. GSEA of male GBM compared to other male GBM, and similarly in female GBM compared to other female GBM, related to Figures 2 and S2

mmc8.xlsx^{(15.3KB, xlsx)}

Data S7. Raw counts of genes from murine transformed astrocyte transcriptomes, related to Figure 3

mmc9.xlsx^{(6.5MB, xlsx)}

Data S8. DEGs of male vs. female murine transformed astrocyte samples, related to Figure 3

mmc10.xlsx^{(995.7KB, xlsx)}

Data S9. Gene enrichment analysis of clusters from all sample (male and female), male-only, and female-only analyses of murine GBM samples, related to Figure 3

mmc11.xlsx^{(26.3KB, xlsx)}

Data S10. GSEA of murine male transformed astrocytes compared to female transformed astrocytes, related to Figures 3 and S3

mmc12.xlsx^{(13.9KB, xlsx)}

Data S11. Enrichment of genes identified in all three analyses, male-only analysis, or female-only analysis from human GBM transcriptomes, related to Figures 3 and S4

mmc13.xlsx^{(12.5KB, xlsx)}

Data S12. Enrichment of genes identified in all three analyses, male-only analysis, or female-only analysis from murine GBM transcriptomes, related to Figures 3 and S4

mmc14.xlsx^{(12.6KB, xlsx)}

Data S13. Enrichment of overlapping genes from human and murine analyses, related to Figures 2, 3, and S4

mmc15.xlsx^{(11.9KB, xlsx)}

Data Availability Statement

•
Murine RNA-sequencing data from this study have been deposited at the NCBI Gene Expression Omnibus (GEO) database under accession number GEO: GSE295851 and are publicly available as of the date of publication. RNA-sequencing data for human tumors were obtained from Yanovich-Arad et al., Cell Reports, 2021 (DOI: https://doi.org/10.1016/j.celrep.2021.108787; GEO: GSE149009).²⁵
•
This paper does not report original code. Packages used for processing sequencing and survival data are denoted in the STAR Methods.
•
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

[bib1] 1.Klein S.L., Flanagan K.L. Sex differences in immune responses. Nat. Rev. Immunol. 2016;16:626–638. doi: 10.1038/nri.2016.90. [DOI] [PubMed] [Google Scholar]

[bib2] 2.Rubin J.B., Abou-Antoun T., Ippolito J.E., Llaci L., Marquez C.T., Wong J.P., Yang L. Epigenetic developmental mechanisms underlying sex differences in cancer. J. Clin. Investig. 2024;134 doi: 10.1172/JCI180071. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib3] 3.Yang W., Rubin J.B. Treating sex and gender differences as a continuous variable can improve precision cancer treatments. Biol. Sex Differ. 2024;15:35. doi: 10.1186/s13293-024-00607-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib4] 4.Han J., Yang Y., Li X., Wu J., Sheng Y., Qiu J., Wang Q., Li J., He Y., Cheng L., Zhang Y. Pan-cancer analysis reveals sex-specific signatures in the tumor microenvironment. Mol. Oncol. 2022;16:2153–2173. doi: 10.1002/1878-0261.13203. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib5] 5.Rubin J.B., Lagas J.S., Broestl L., Sponagel J., Rockwell N., Rhee G., Rosen S.F., Chen S., Klein R.S., Imoukhuede P., Luo J. Sex differences in cancer mechanisms. Biol. Sex Differ. 2020;11:17. doi: 10.1186/s13293-020-00291-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib6] 6.Broestl L., Warrington N.M., Grandison L., Abou-Antoun T., Tung O., Shenoy S., Tallman M.M., Rhee G., Yang W., Sponagel J., et al. Gonadal sex patterns p21-induced cellular senescence in mouse and human glioblastoma. Commun. Biol. 2022;5:781. doi: 10.1038/s42003-022-03743-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib7] 7.Lee J., Nicosia M., Hong E.S., Silver D.J., Li C., Bayik D., Watson D.C., Lauko A., Kay K.E., Wang S.Z., et al. Sex-Biased T-cell Exhaustion Drives Differential Immune Responses in Glioblastoma. Cancer Discov. 2023;13:2090–2105. doi: 10.1158/2159-8290.CD-22-0869. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] 8.Bayik D., Zhou Y., Park C., Hong C., Vail D., Silver D.J., Lauko A., Roversi G., Watson D.C., Lo A., et al. Myeloid-Derived Suppressor Cell Subsets Drive Glioblastoma Growth in a Sex-Specific Manner. Cancer Discov. 2020;10:1210–1225. doi: 10.1158/2159-8290.CD-19-1355. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib9] 9.Ochocka N., Segit P., Wojnicki K., Cyranowski S., Swatler J., Jacek K., Grajkowska W., Kaminska B. Specialized functions and sexual dimorphism explain the functional diversity of the myeloid populations during glioma progression. Cell Rep. 2023;42 doi: 10.1016/j.celrep.2022.111971. [DOI] [PubMed] [Google Scholar]

[bib10] 10.Ye Y., Jing Y., Li L., Mills G.B., Diao L., Liu H., Han L. Sex-associated molecular differences for cancer immunotherapy. Nat. Commun. 2020;11 doi: 10.1038/s41467-020-15679-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] 11.Ostrom Q.T., Cioffi G., Gittleman H., Patil N., Waite K., Kruchko C., Barnholtz-Sloan J.S. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2012-2016. Neuro Oncol. 2019;21:v1–v100. doi: 10.1093/neuonc/noz150. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib12] 12.Yang W., Warrington N.M., Taylor S.J., Whitmire P., Carrasco E., Singleton K.W., Wu N., Lathia J.D., Berens M.E., Kim A.H., et al. Sex differences in GBM revealed by analysis of patient imaging, transcriptome, and survival data. Sci. Transl. Med. 2019;11 doi: 10.1126/scitranslmed.aao5253. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib13] 13.Maccari M., Baek C., Caccese M., Mandruzzato S., Fiorentino A., Internò V., Bosio A., Cerretti G., Padovan M., Idbaih A., Lombardi G. Present and Future of Immunotherapy in Patients With Glioblastoma: Limitations and Opportunities. Oncologist. 2024;29:289–302. doi: 10.1093/oncolo/oyad321. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib14] 14.Johnson D.E., O’Keefe R.A., Grandis J.R. Targeting the IL-6/JAK/STAT3 signalling axis in cancer. Nat. Rev. Clin. Oncol. 2018;15:234–248. doi: 10.1038/nrclinonc.2018.8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib15] 15.Tan M.S.Y., Sandanaraj E., Chong Y.K., Lim S.W., Koh L.W.H., Ng W.H., Tan N.S., Tan P., Ang B.T., Tang C. A STAT3-based gene signature stratifies glioma patients for targeted therapy. Nat. Commun. 2019;10:3601. doi: 10.1038/s41467-019-11614-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib16] 16.Hedvat M., Huszar D., Herrmann A., Gozgit J.M., Schroeder A., Sheehy A., Buettner R., Proia D., Kowolik C.M., Xin H., et al. The JAK2 Inhibitor AZD1480 Potently Blocks Stat3 Signaling and Oncogenesis in Solid Tumors. Cancer Cell. 2009;16:487–497. doi: 10.1016/j.ccr.2009.10.015. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib17] 17.McFarland B.C., Hong S.W., Rajbhandari R., Twitty G.B., Gray G.K., Yu H., Benveniste E.N., Nozell S.E. NF-κB-induced IL-6 ensures STAT3 activation and tumor aggressiveness in glioblastoma. PLoS One. 2013;8 doi: 10.1371/journal.pone.0078728. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib18] 18.Taniguchi K., Karin M. NF-κB, inflammation, immunity and cancer: coming of age. Nat. Rev. Immunol. 2018;18:309–324. doi: 10.1038/nri.2017.142. [DOI] [PubMed] [Google Scholar]

[bib19] 19.Bhat K.P.L., Balasubramaniyan V., Vaillant B., Ezhilarasan R., Hummelink K., Hollingsworth F., Wani K., Heathcock L., James J.D., Goodman L.D., et al. Mesenchymal Differentiation Mediated by NF-κB Promotes Radiation Resistance in Glioblastoma. Cancer Cell. 2013;24:331–346. doi: 10.1016/j.ccr.2013.08.001. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib20] 20.Schmitt M.J., Company C., Dramaretska Y., Barozzi I., Göhrig A., Kertalli S., Großmann M., Naumann H., Sanchez-Bailon M.P., Hulsman D., et al. Phenotypic Mapping of Pathologic Cross-Talk between Glioblastoma and Innate Immune Cells by Synthetic Genetic Tracing. Cancer Discov. 2021;11:754–777. doi: 10.1158/2159-8290.CD-20-0219. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib21] 21.Wang L., Jung J., Babikir H., Shamardani K., Jain S., Feng X., Gupta N., Rosi S., Chang S., Raleigh D., et al. A single-cell atlas of glioblastoma evolution under therapy reveals cell-intrinsic and cell-extrinsic therapeutic targets. Nat. Cancer. 2022;3:1534–1552. doi: 10.1038/s43018-022-00475-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib22] 22.Loftus A.E.P., Romano M.S., Phuong A.N., McKinnel B.J., Muir M.T., Furqan M., Dawson J.C., Avalle L., Douglas A.T., Mort R.L., et al. An ILK/STAT3 pathway controls glioblastoma stem cell plasticity. Dev. Cell. 2024;59:3197–3212.e7. doi: 10.1016/j.devcel.2024.09.003. [DOI] [PubMed] [Google Scholar]

[bib23] 23.Bowman R.L., Wang Q., Carro A., Verhaak R.G.W., Squatrito M. GlioVis data portal for visualization and analysis of brain tumor expression datasets. Neuro Oncol. 2017;19:139–141. doi: 10.1093/neuonc/now247. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib24] 24.Liberzon A., Birger C., Thorvaldsdóttir H., Ghandi M., Mesirov J.P., Tamayo P. The Molecular Signatures Database Hallmark Gene Set Collection. Cell Syst. 2015;1:417–425. doi: 10.1016/j.cels.2015.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib25] 25.Yanovich-Arad G., Ofek P., Yeini E., Mardamshina M., Danilevsky A., Shomron N., Grossman R., Satchi-Fainaro R., Geiger T. Proteogenomics of glioblastoma associates molecular patterns with survival. Cell Rep. 2021;34 doi: 10.1016/j.celrep.2021.108787. [DOI] [PubMed] [Google Scholar]

[bib26] 26.Fang Z., Cui X. Design and validation issues in RNA-seq experiments. Brief. Bioinform. 2011;12:280–287. doi: 10.1093/bib/bbr004. [DOI] [PubMed] [Google Scholar]

[bib27] 27.Rubin J.B. The spectrum of sex differences in cancer. Trends Cancer. 2022;8:303–315. doi: 10.1016/j.trecan.2022.01.013. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib28] 28.Ge S.X., Son E.W., Yao R. iDEP: an integrated web application for differential expression and pathway analysis of RNA-Seq data. BMC Bioinf. 2018;19:534. doi: 10.1186/s12859-018-2486-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib29] 29.Subramanian A., Tamayo P., Mootha V.K., Mukherjee S., Ebert B.L., Gillette M.A., Paulovich A., Pomeroy S.L., Golub T.R., Lander E.S., Mesirov J.P. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA. 2005;102:15545–15550. doi: 10.1073/pnas.0506580102. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib30] 30.Mootha V.K., Lindgren C.M., Eriksson K.-F., Subramanian A., Sihag S., Lehar J., Puigserver P., Carlsson E., Ridderstråle M., Laurila E., et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 2003;34:267–273. doi: 10.1038/ng1180. [DOI] [PubMed] [Google Scholar]

[bib31] 31.Sun T., Warrington N.M., Luo J., Brooks M.D., Dahiya S., Snyder S.C., Sengupta R., Rubin J.B. Sexually dimorphic RB inactivation underlies mesenchymal glioblastoma prevalence in males. J. Clin. Investig. 2014;124:4123–4133. doi: 10.1172/JCI71048. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib32] 32.Sponagel J., Jones J.K., Frankfater C., Zhang S., Tung O., Cho K., Tinkum K.L., Gass H., Nunez E., Spitz D.R., et al. Sex differences in brain tumor glutamine metabolism reveal sex-specific vulnerabilities to treatment. Med. 2022;3:792–811.e12. doi: 10.1016/j.medj.2022.08.005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib33] 33.Love M.I., Huber W., Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15:550. doi: 10.1186/s13059-014-0550-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib34] 34.Zhu Z., Zhang X., Yu Z., Zhou Y., Zhu S., Zhang Y.H., Lin X.P., Mou Y., Zhang J. Correlation of Tim-3 expression with chemokine levels for predicting the prognosis of patients with glioblastoma. J. Neuroimmunol. 2021;355 doi: 10.1016/j.jneuroim.2021.577575. [DOI] [PubMed] [Google Scholar]

[bib35] 35.Comba A., Faisal S.M., Dunn P.J., Argento A.E., Hollon T.C., Al-Holou W.N., Varela M.L., Zamler D.B., Quass G.L., Apostolides P.F., et al. Spatiotemporal analysis of glioma heterogeneity reveals COL1A1 as an actionable target to disrupt tumor progression. Nat. Commun. 2022;13:3606. doi: 10.1038/s41467-022-31340-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib36] 36.Day B.W., Stringer B.W., Al-Ejeh F., Ting M.J., Wilson J., Ensbey K.S., Jamieson P.R., Bruce Z.C., Lim Y.C., Offenhäuser C., et al. EphA3 Maintains Tumorigenicity and Is a Therapeutic Target in Glioblastoma Multiforme. Cancer Cell. 2013;23:238–248. doi: 10.1016/j.ccr.2013.01.007. [DOI] [PubMed] [Google Scholar]

[bib37] 37.Jeon H.-M., Shin Y.J., Lee J., Chang N., Woo D.-H., Lee W.J., Nguyen D., Kang W., Cho H.J., Yang H., et al. The semaphorin 3A/neuropilin-1 pathway promotes clonogenic growth of glioblastoma via activation of TGF-β signaling. JCI Insight. 2023;8 doi: 10.1172/jci.insight.167049. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib38] 38.Kfoury N., Sun T., Yu K., Rockwell N., Tinkum K.L., Qi Z., Warrington N.M., McDonald P., Roy A., Weir S.J., et al. Cooperative p16 and p21 action protects female astrocytes from transformation. Acta Neuropathol. Commun. 2018;6:12. doi: 10.1186/s40478-018-0513-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib39] 39.Kfoury N., Qi Z., Prager B.C., Wilkinson M.N., Broestl L., Berrett K.C., Moudgil A., Sankararaman S., Chen X., Gertz J., et al. Brd4-bound enhancers drive cell-intrinsic sex differences in glioblastoma. Proc. Natl. Acad. Sci. USA. 2021;118 doi: 10.1073/pnas.2017148118. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib40] 40.Rockwell N.C., Yang W., Warrington N.M., Staller M.V., Griffith M., Griffith O.L., Gurnett C.A., Cohen B.A., Baldridge D., Rubin J.B. Sex- and Mutation-Specific p53 Gain-of-Function Activity in Gliomagenesis. Cancer Res. Commun. 2021;1:148–163. doi: 10.1158/2767-9764.CRC-21-0026. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib41] 41.Hoffmann A., Natoli G., Ghosh G. Transcriptional regulation via the NF-κB signaling module. Oncogene. 2006;25:6706–6716. doi: 10.1038/sj.onc.1209933. [DOI] [PubMed] [Google Scholar]

[bib43] 42.Warrington N.M., Sun T., Luo J., McKinstry R.C., Parkin P.C., Ganzhorn S., Spoljaric D., Albers A.C., Merkelson A., Stewart D.R., et al. The cyclic AMP pathway is a sex-specific modifier of glioma risk in type I neurofibromatosis patients. Cancer Res. 2015;75:16–21. doi: 10.1158/0008-5472.CAN-14-1891. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib44] 43.Zhong H., Voll R.E., Ghosh S. Phosphorylation of NF-kappa B p65 by PKA stimulates transcriptional activity by promoting a novel bivalent interaction with the coactivator CBP/p300. Mol. Cell. 1998;1:661–671. doi: 10.1016/s1097-2765(00)80066-0. [DOI] [PubMed] [Google Scholar]

[bib45] 44.Wang W., Guo C., Zhu P., Lu J., Li W., Liu C., Xie H., Myatt L., Chen Z.-J., Sun K. Phosphorylation of STAT3 mediates the induction of cyclooxygenase-2 by cortisol in the human amnion at parturition. Sci. Signal. 2015;8 doi: 10.1126/scisignal.aac6151. [DOI] [PubMed] [Google Scholar]

[bib46] 45.Chua C.Y., Liu Y., Granberg K.J., Hu L., Haapasalo H., Annala M.J., Cogdell D.E., Verploegen M., Moore L.M., Fuller G.N., et al. IGFBP2 potentiates nuclear EGFR–STAT3 signaling. Oncogene. 2016;35:738–747. doi: 10.1038/onc.2015.131. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib47] 46.Hoogstrate Y., Ghisai S.A., de Wit M., de Heer I., Draaisma K., van Riet J., van de Werken H.J.G., Bours V., Buter J., Vanden Bempt I., et al. The EGFRvIII transcriptome in glioblastoma: A meta-omics analysis. Neuro Oncol. 2022;24:429–441. doi: 10.1093/neuonc/noab231. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib48] 47.Li X.-P., Guo Z.-Q., Wang B.-F., Zhao M. EGFR alterations in glioblastoma play a role in antitumor immunity regulation. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1236246. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib49] 48.Rabah N., Ait Mohand F.-E., Kravchenko-Balasha N. Understanding Glioblastoma Signaling, Heterogeneity, Invasiveness, and Drug Delivery Barriers. Int. J. Mol. Sci. 2023;24 doi: 10.3390/ijms241814256. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib50] 49.Felsberg J., Hentschel B., Kaulich K., Gramatzki D., Zacher A., Malzkorn B., Kamp M., Sabel M., Simon M., Westphal M., et al. Epidermal Growth Factor Receptor Variant III (EGFRvIII) Positivity in EGFR-Amplified Glioblastomas: Prognostic Role and Comparison between Primary and Recurrent Tumors. Clin. Cancer Res. 2017;23:6846–6855. doi: 10.1158/1078-0432.CCR-17-0890. [DOI] [PubMed] [Google Scholar]

[bib51] 50.Zhang Y., Xiao X., Yang G., Jiang X., Jiao S., Nie Y., Zhang T. STAT3/TGFBI signaling promotes the temozolomide resistance of glioblastoma through upregulating glycolysis by inducing cellular senescence. Cancer Cell Int. 2025;25:127. doi: 10.1186/s12935-025-03770-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib52] 51.Zou S., Tong Q., Liu B., Huang W., Tian Y., Fu X. Targeting STAT3 in Cancer Immunotherapy. Mol. Cancer. 2020;19:145. doi: 10.1186/s12943-020-01258-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib53] 52.Ganesh S., Kim M.J., Lee J., Feng X., Ule K., Mahan A., Krishnan H.S., Wang Z., Anzahaee M.Y., Singhal G., et al. RNAi mediated silencing of STAT3/PD-L1 in tumor-associated immune cells induces robust anti-tumor effects in immunotherapy resistant tumors. Mol. Ther. 2024;32:1895–1916. doi: 10.1016/j.ymthe.2024.03.035. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib54] 53.Witt K., Evans-Axelsson S., Lundqvist A., Johansson M., Bjartell A., Hellsten R. Inhibition of STAT3 augments antitumor efficacy of anti-CTLA-4 treatment against prostate cancer. Cancer Immunol. Immunother. 2021;70:3155–3166. doi: 10.1007/s00262-021-02915-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib55] 54.Dunn S.E., Perry W.A., Klein S.L. Mechanisms and consequences of sex differences in immune responses. Nat. Rev. Nephrol. 2024;20:37–55. doi: 10.1038/s41581-023-00787-w. [DOI] [PubMed] [Google Scholar]

[bib56] 55.Roved J., Westerdahl H., Hasselquist D. Sex differences in immune responses: Hormonal effects, antagonistic selection, and evolutionary consequences. Horm. Behav. 2017;88:95–105. doi: 10.1016/j.yhbeh.2016.11.017. [DOI] [PubMed] [Google Scholar]

[bib57] 56.Huang C.-Y., Tan K.-T., Huang S.-F., Lu Y.-J., Wang Y.-H., Chen S.-J., Tse K.-P. Study of sex-biased differences in genomic profiles in East Asian hepatocellular carcinoma. Discov. Oncol. 2024;15:276. doi: 10.1007/s12672-024-01131-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib58] 57.Heimberger A.B., Hlatky R., Suki D., Yang D., Weinberg J., Gilbert M., Sawaya R., Aldape K. Prognostic effect of epidermal growth factor receptor and EGFRvIII in glioblastoma multiforme patients. Clin. Cancer Res. 2005;11:1462–1466. doi: 10.1158/1078-0432.CCR-04-1737. [DOI] [PubMed] [Google Scholar]

[bib59] 58.Jang B., Yoon D., Lee J.Y., Kim J., Hong J., Koo H., Sa J.K. Integrative multi-omics characterization reveals sex differences in glioblastoma. Biol. Sex Differ. 2024;15:23. doi: 10.1186/s13293-024-00601-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib60] 59.Yu W., Gui S., Peng L., Luo H., Xie J., Xiao J., Yilamu Y., Sun Y., Cai S., Cheng Z., Tao Z. STAT3-controlled CHI3L1/SPP1 positive feedback loop demonstrates the spatial heterogeneity and immune characteristics of glioblastoma. Dev. Cell. 2025;60:1751–1767.e9. doi: 10.1016/j.devcel.2025.01.014. [DOI] [PubMed] [Google Scholar]

[bib61] 60.Lin W.-H., Chang Y.-W., Hong M.-X., Hsu T.-C., Lee K.-C., Lin C., Lee J.-L. STAT3 phosphorylation at Ser727 and Tyr705 differentially regulates the EMT-MET switch and cancer metastasis. Oncogene. 2021;40:791–805. doi: 10.1038/s41388-020-01566-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib62] 61.Martin M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 2011;17:10–12. doi: 10.14806/ej.17.1.200. [DOI] [Google Scholar]

[bib63] 62.Dobin A., Davis C.A., Schlesinger F., Drenkow J., Zaleski C., Jha S., Batut P., Chaisson M., Gingeras T.R. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29:15–21. doi: 10.1093/bioinformatics/bts635. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib64] 63.Liao Y., Smyth G.K., Shi W. The Subread aligner: fast, accurate and scalable read mapping by seed-and-vote. Nucleic Acids Res. 2013;41 doi: 10.1093/nar/gkt214. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib65] 64.Mudunuri U., Che A., Yi M., Stephens R.M. bioDBnet: the biological database network. Bioinformatics. 2009;25:555–556. doi: 10.1093/bioinformatics/btn654. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib66] 65.Howe D.G., Blake J.A., Bradford Y.M., Bult C.J., Calvi B.R., Engel S.R., Kadin J.A., Kaufman T.C., Kishore R., Laulederkind S.J.F., et al. Model organism data evolving in support of translational medicine. Lab Anim. 2018;47:277–289. doi: 10.1038/s41684-018-0150-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib67] 66.Therneau T.M., Grambsch P.M. Springer; 2000. Modeling Survival Data: Extending the Cox Model. [DOI] [Google Scholar]

[bib68] 67.Kassambara A., Kosinski M., Biecek P., Fabian S. 2025. survminer: Drawing Survival Curves using “ggplot2.” Version 0.5.1.https://cran.r-project.org/web/packages/survminer/index.html [Google Scholar]

[bib69] 68.Alpern D., Gardeux V., Russeil J., Mangeat B., Meireles-Filho A.C.A., Breysse R., Hacker D., Deplancke B. BRB-seq: ultra-affordable high-throughput transcriptomics enabled by bulk RNA barcoding and sequencing. Genome Biol. 2019;20:71. doi: 10.1186/s13059-019-1671-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib70] 69.Su S., Law C.W., Ah-Cann C., Asselin-Labat M.-L., Blewitt M.E., Ritchie M.E. Glimma: interactive graphics for gene expression analysis. Bioinformatics. 2017;33:2050–2052. doi: 10.1093/bioinformatics/btx094. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib71] 70.Perez G., Barber G.P., Benet-Pages A., Casper J., Clawson H., Diekhans M., Fischer C., Gonzalez J.N., Hinrichs A.S., Lee C.M., et al. The UCSC Genome Browser database: 2025 update. Nucleic Acids Res. 2025;53:D1243–D1249. doi: 10.1093/nar/gkae974. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

NF-κB and STAT3 signaling uniquely stratify survival in female glioblastoma patients

Jason P Wong

Lorida Llaci

Lloyd Tripp

Myung Sik Jeon

Lihua Yang

Nicholas Reinhold

Nicole M Warrington

Jingqin Luo

Robi D Mitra

Joshua B Rubin

Summary

Graphical abstract

Highlights

Introduction

Results

An NF-κB or IL-6/JAK/STAT3 gene signature is prognostic in female GBM patients

Figure 1.

An NF-κB gene signature is uniquely enriched in female GBM patients

Figure 2.

Concordance between sex-skewed gene expression in patients and a murine model of GBM

Figure 3.

Sex differences in the NF-κB pathway are dependent on IL-6 and STAT3

Figure 4.

Sex differences in the NF-κB and IL-6/JAK/STAT axis has therapeutic implications

Figure 5.

Sex differences in modulators of STAT3 activity

Discussion

Limitations of the study

Resource availability

Lead contact

Materials availability

Data and code availability

Acknowledgments

Author contributions

Declaration of interests

STAR★Methods

Key resources table

Experimental model and study participant details

Cell lines

Method details

Cell line treatment

Luciferase activity assay

Immunoblot analysis

Quantitative RT-PCR

CellTiter-Glo® (CTG)

Il6 ELISA

Senescence-associated beta-galactosidase (SABG) assay values

BRB RNA-sequencing (BRB-seq) for murine GBM samples

Murine GBM RNA-sequencing analysis

Human GBM RNA-sequencing analysis

Gene enrichment (GE)/gene set enrichment analysis (GSEA)

Sex-biased cluster analysis

Univariate survival analysis

Quantification and statistical analysis

Footnotes

Supplemental information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases