Patients and samples

To investigate the association between obesity and tumor aggression and size, we performed transcriptional pathway analysis and statistical analysis on data from The Cancer Genome Atlas (TCGA). From the full data set with 10,783, we defined two nested subsets of data. Subset 1 consisted of 7309 patients spanning 23 different cancer types, all annotated with information on age, sex, and pathological tumor stage (T-stage). Subset 2 consisted of 1781 patients, all from Subset 1, who had Body Mass Index (BMI) information available, these patients spanned 10 cancer types (Fig. 1).

Fig. 1

Data cohort. A schematic representation of the full data set from TCGA and Thorsson et al. 2018, and the two subsets that we perform the analysis on. For the full data set we have gene expression data and T-cell receptor diversity information. For Subset 1, which includes 7309 of the patients from the full data set, we have pathological T-stage, age and sex annotations for all patients. For Subset 2, which includes 1781 patients from Subset 1, we have height and weight information for all patients at diagnosis. The figure was created using BioRender

Patients with high BMI more commonly harbor smaller, less invasive tumors

First, we endeavored to investigate if obese individuals in general present with smaller tumors, indicative of a less aggressive phenotype driving early cancer development. To explore this, we used pathological T stage as a proxy for tumor size. T stage is a component of the standardized TNM (Tumor, Node, Metastasis) staging system developed by the American Joint Committee on Cancer (AJCC) and used globally for staging cancers (Edge and Compton 2010). As part of this, T stage describes the size and extent of the primary tumor, and is typically graded as T1-T4. While the exact definition varies by cancer type, T1 tumors are typically smaller, while T4 tumors are larger and may have more extensive growth into local tissue. When we compared the BMI of patients based on T stage, we found that patients with low T stage, particularly T1 tumors, had higher BMI relative to patients with higher stage tumors (Fig. 2A). When we stratified the patients into two groups based on T stage, small tumors (T1, T2) and large tumors (T3, T4), we found that patients with small tumors had a significantly higher BMI (median = 26.4), relative to patients with large tumors (median = 25.8, P = 0.0056) (Fig. 2B). There was no significant difference in BMI by sex in this cohort (female median = 26, male median 26.2, P = 0.68). However, we observed the same pattern within each sex, where patients with small tumors had a significantly higher BMI relative to patients diagnosed with larger tumors (small tumors, female median = 26.4, male median 26.4; large tumors, female median = 25.7, male median 26, P female = 0.045, P male = 0.056, Fig. 2C). When we further stratified patients based on BMI into obese (BMI >  = 30) and non-obese (BMI < 30), we found a significant enrichment of small tumors in patients with obesity (Obese 57.2% vs Non-obese 47.5%, P = 0.000427) (Fig. 2d).

Fig. 2

BMI and pathological T-stage on Subset 2. A Patients BMI stratified by their tumor’s pathological T stage. Colored by tumor size (Small: T1 and T2, Large: T3 and T4). B Patients BMI plotted stratified by their tumor size. C Patients BMI plotted against their tumor size stratified by sex. D Patients are stratified by obesity, BMI >  = 30, and the proportion of small and large tumors are shown for each group

Small tumors in patients with high BMI show unique immune profiles

To investigate if smaller tumors from obese patients may be the result of more aggressive immune activity, we explored the differences in immune cell infiltration between small and large tumors from obese and non-obese patients. Given that the immune system decays with age due to immunosenescence (Pawelec 2018), we further stratified these analyses based on age. We investigated immune infiltration by utilizing the TIL score from Danaher (Danaher et al. 2017), and found that the small tumors of obese patients had a significantly higher level of immune infiltration relative to their non-obese counterparts (P = 0.00025), in younger (< 60 years) patients (Fig. 3A). We observed no differences in immune infiltration within older patients nor between the larger tumors in patients with or without obesity (Fig. 3B).

Fig. 3

Tumor immune infiltration and diversity. A Tumor Infiltrating leukocytes (TIL) score in the younger (< 60 years) patients. The patients are stratified by tumor size and colored by obesity (non-obese: BMI < 30, obese: BMI >  = 30). B TIL score in the older (> = 60 years) patients. The patients are stratified by tumor size and colored by obesity. C Adaptive/innate immune ratio of younger patients. The patients are stratified by tumor size and colored by obesity. The Y-axis is log2-scaled. D Adaptive/innate immune ratio of older patients. The patients are stratified by tumor size and colored by obesity. The Y-axis is log2-scaled. E T-cell receptor (TCR) shannon diversity of younger patients. The patients are stratified by tumor size and colored by obesity. F TCR shannon diversity of older patients. The patients are stratified by tumor size and colored by obesity

Next, we investigated the composition of infiltrating immune cells using the ratio of adaptive to innate immune cells (A/I ratio). We have previously shown that within tumors the A/I ratio is associated with improved survival (Ahrenfeldt et al. 2023). Here, we found that in younger patients with small tumors, obese patients had a higher A/I ratio relative to non-obese patients (P = 0.041) (Fig. 3C). We found no significant differences in the older patients (Fig. 3D).

To investigate the landscape of tumor infiltrating adaptive immune cells, we obtained TCR diversity and richness estimates from the TCGA data, previously published by Thorsson et al. (Thorsson et al. 2018). We found that small tumors exhibited a significantly higher TCR Shannon diversity index in younger patients with obesity relative to younger patients without obesity (P = 0.0067) (Fig. 3E). We found no significant difference in the older cohort with small tumors or between obese and non-obese patients with large tumors, neither in the young nor old cohort (Fig. 3F).

Tumors from obese individuals show distinct pathway expression profiles

Tumor size is strongly prognostic, and is therefore likely associated with a more aggressive biological phenotype. To investigate this, we compared gene expression profiles between small and large tumors across the 7309 samples from 23 cancer types in Subset 1 with T stage annotations, and compared large tumors to small tumors within each cancer type. For this analysis, we summarized gene expression to pathways, gene set variation analysis (GSVA) of the 50 hallmark pathways (Liberzon et al. 2015). All pathways were tested for significant differential expression across all 23 cancer types. In this manner, we observed that 38 showed a significantly different expression between small and large tumors at least once, ranging from 0 to 15 significant pathways per cancer type (Fig S1A). To summarize these results across cancer types, the hallmark pathways were scored as either significantly expressed or not significantly expressed in each cancer type, using an FDR adjusted p-value of 0.1 as cutoff. We then used a binomial test to determine if a hallmark pathway was significantly enriched across multiple cancer types. Here, we found that large tumors have a significantly higher expression of the EPITHELIAL_MESENCHYMAL_TRANSITION, ANGIOGENESIS, and HYPOXIA pathways, all of which have previously been associated with poor outcome and aggressive cancer (Thiery et al. 2009; Oshi et al. 2021; Evans and Koch 2003). Furthermore, we found large tumors to have a significantly higher expression of the GLYCOLYSIS metabolic pathway, whereas small tumors have a significantly higher expression of FATTY_ACID_METABOLISM (Fig. 4A). We also found that proliferative pathways such as MYC_TARGETS_V1 and V2 and G2M_CHECKPOINT were most highly expressed in large tumors, although this was not significant.

Fig. 4

Differences in pathway expression in small and large tumors. A A bar plot showing the pathways where there are more than 5 cancer types with an overexpression in small or large tumors, and the number of cancer types that are significantly overexpressed in either direction. An asterisk, *, marks the pathways where the distribution of cancer types into small or large are significantly different from 50/50, given a binomial distribution. This analysis is performed on Subset 1. B A volcano plot showing the difference of mean (and p-value given a t-test) GSVA values for each pathway between non-obese and obese patients with small tumors. The pathways are colored by their overall process category. This analysis is performed on Subset 2. C A volcano plot showing the difference of mean (and p-value given a t-test) GSVA values for each pathway between non-obese and obese patients with large tumors. The pathways are colored by their overall process category. This analysis is performed on Subset 2

Next, to investigate the impact of obesity in tumor phenotype, we further explored if obesity might impact the observed differences between small and large tumors in Subset 1. By comparing gene expression data between obese and non-obese patients, within small and large tumors separately, we observe lower expression of the proliferative pathways (small tumors: MYC_TARGETS_V1 and V2, large tumors: E2F_TARGETS, MYC_TARGETS_V1 and G2M_CHECKPOINT) and higher expression of immune related pathways (small tumors: IL6_JAK_STAT3_SIGNALING, INFLAMMATORY_RESPONSE, COMPLEMENT and ALLOGRAFT_REJECTION, large tumors: COAGULATION) in both small (Fig. 4B) and large tumors (Fig. 4C) in obese patients.

To investigate the cancer-specific origin of the differential expression, we stratified the analysis on cancer type and found that for small tumors, overexpression of the proliferative pathways in non-obese patients were predominantly driven by liver cancer, esophagus cancer and renal cancer (Fig S1B). Likewise, overexpression of immune pathways in obese patients mostly originated from liver cancer and bladder cancer. In large tumors overexpression of the proliferative pathways in non-obese patients mostly originated from liver cancer and colon cancer, while overexpression of immune pathways in obese patients mostly originated from melanoma and uveal melanoma (Fig S1C).

To investigate if there were differences in gene expression between older and younger patients, we performed the analysis stratified into older and younger patients, as above. We found that when we compared RNA expression from small tumors between younger and older patients, tumors from younger patients had a higher expression of proliferative pathways, such as E2F_TARGETS, G2M_CHECKPOINT and MITOTIC_SPINDLE. Conversely, in small tumors from older patients we found a higher expression of metabolic pathways including XENOBIOTIC_METABOLISM, BILE_ACID_METABOLISM, FATTY_ACID_METABOLISM, HEME_METABOLISM and OXIDATIVE_PHOSPHORYLATION (Fig S2A). When we repeated the analysis in large tumors, we found that younger patients had a higher expression of TGF_BETA_SIGNALING and APICAL_JUNCTION while no pathways had a significantly higher expression in older patients (Fig S2B).

Next, to investigate if there were any significant differences in the expression between the two sexes, we performed the same analysis stratified by sex. For this analysis we excluded sex-specific cancer types, BRCA, CESC, PRAD and TGCT. When we compared small tumors between male and female patients, we found no significant difference (Fig S3A). When we performed the analysis using large tumors, we found that 18 of the 50 pathways are significantly higher expressed in female patients compared to male patients (Fig S3B), these include mainly immune related pathways (INFLAMMATORY_RESPONSE, COMPLEMENT, IL6_JAK3_STAT_SIGNALING, ALLOGRAFT_REJECTION, INTERFERON_GAMMA_RESPONSE and COAGULATION) and signaling pathways (TNFA_SIGNALING_VIA_NKFB, IL2_STAT5_SIGNALING, KRAS_SIGNALING, ESTROGEN_REPONSE_EARLY and ESTROGEN_RESPONSE_LATE).

Genotypic patterns in large vs small tumors

To investigate if the landscape of cancer driver mutations might differ between small and large tumors, we categorized all mutations found within known cancer genes in tumors from Subset 1 into whether they were likely driver mutations or likely passenger mutations. We explored how often individual cancer driver mutations occurred together with other driver mutations within the same tumor. To investigate this, we defined a driver weight score. The driver weight score was determined for each driver mutation, within each tumor, as simply 1/ndriver. We then compared the differences in mean driver weight across genes and cancer types. We found that there were more genes with a significantly higher driver weight in small tumors relative to in large tumors (Fig. 5A). Examples of these are PIK3CA in both BRCA and HNSC, LRP1B in both LUSC and BRCA and TP53 in HNSC and SKCM. However, TP53 also has a higher driver weight in Large MESO tumors. To investigate whether small tumors had a higher driver weight in general, we compared the driver weight of small vs large tumors for each cancer type, where the tumor's driver weight was the same as for each of its driver mutations 1/ndriver. We investigated the mean difference in driver mutations between small and large tumors and found that in three cancer types (BRCA, HNSC and KIRC) large tumors had significantly higher number of driver mutations (Fig. 5B). When we looked at the frequency of specific driver mutations between large and small tumors we only found two significantly enrichment genes (Fig. 5C), HRAS in small BLCA tumors and CDH1 in large BRCA tumors.

Fig. 5

Driver genes and tumor size. A A volcano plot showing the mean difference of driver weight per gene, per cancer type between small and large tumors. The driver weight is 1/number of driver mutations per tumor. The p-value is calculated by a t-test. B A volcano plot showing the mean difference in number of driver mutations per tumor for each cancer type between small and large tumors. The p-value is calculated by a t-test. C A volcano plot showing the odds ratio for an enrichment of certain driver mutations in small or large tumors. Odd ratio and p-value is calculated by fisher’s exact test