3.1 Cholesterol metabolism reprograming is highly heterogeneous across tumor types

Cholesterol metabolism is a complex process involving a dynamic balance between biosynthesis, uptake, efflux, storage, and catabolism [25]. To study this multifactorial process, gene sets of the five pathways involved in cholesterol metabolism were manually curated (Fig. 1a). RNA-seq data from a total of 8832 patients across 19 distinct solid cancer types (Fig. 1b) were obtained from TCGA, and compared with 2249 normal tissue RNA-Seq profiles from GTEx and TCGA. The expression of genes defining the cholesterol-related pathways showed a clear pattern across distinct normal tissues. In contrast, tumor samples showed greater heterogeneity, occasionally clustering distinct cancer types together (Supplementary Fig. S1).

Fig. 1

Overview of the pan-cancer and cholesterol metabolism pathways atlas. a. Schematic representation of cholesterol metabolism pathways, indicating the genes included in each pathway for ssGSEA-based enrichment scoring; b. Anatomical distribution of 26 TCGA cancer projects across 19 tumor sites, with corresponding numbers of tumor and normal samples. ACC, adrenocortical carcinoma; BLCA, bladder carcinoma; BRCA, breast carcinoma; CESC, cervical carcinoma; COAD, colon adenocarcinoma; ESCA, oesophageal carcinoma; GBM, glioblastoma multiforme; KICH, kidney chromophobe carcinoma; KIRC, kidney renal clear cell carcinoma; KIRP, kidney renal papillary carcinoma; LGG, low grade glioma; LIHC, hepatocellular carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; PAAD, pancreatic carcinoma; PRAD, prostate carcinoma; READ, rectal adenocarcinoma; OV, ovarian serous adenocarcinoma; SARC, sarcoma; SKCM, skin melanoma; STAD, stomach carcinoma; TGCT, testicular germ cell tumor; THCA, thyroid carcinoma; UCEC, uterine carcinoma; UCS, uterine carcinosarcoma; UVM, uveal melanoma

To assess the validity of the new gene sets in an organ-specific context, we compared the cholesterol pathway scores across normal tissues. Among analyzed tissues, the liver ranked highest for all cholesterol-related pathways, except for storage, in which the breast ranked highest (Supplementary Fig. S2). However, in tumor samples, although the liver maintained the highest scores, tissue ranking changed markedly (Supplementary Fig. S2).

Due to organ-specific differences in baseline cholesterol metabolism, cholesterol-related scores were compared between tumors and normal tissue by organ (Fig. 2). This comparison showed a similar number of tumors with decreased or increased cholesterol biosynthesis- and efflux- gene expression scores across organs. In contrast, catabolism and storage scores were predominantly downregulated in tumors (80% and 73% respectively) (Supplementary Table S3). Conversely, genes related to the cholesterol uptake were significantly upregulated in about 67% of tumors (Supplementary Table S3). Of note, the opposite trends observed for storage and catabolism were restricted to neuroendocrine tumors (kidney, brain, and thyroid) in storage, and to kidney, brain, and pancreatic tumors in catabolism (pancreas: no differences in storage; thyroid: non-significant increase of catabolism) (Fig. 2). Moreover, tumors with decreased uptake-related gene expression, contrary to the general increasing, included the lung, liver, intestine, and breast.

Fig. 2

Mean scores comparison of normal and tumor samples cholesterol-related pathways across each organ of origin. n refers to the number of samples. p denotes the two-sided p-value obtained from either a Student’s t-test or a Mann–Whitney U test, depending on data distribution, which was assessed using the Shapiro–Wilk test. Information on adjacent normal tissue was not available for uvea and adrenocortical samples

3.2 Cholesterol uptake increase is consistently conserved in tumor samples across cancer types

To determine whether the up- and downregulation of cholesterol-related pathways observed in tumors versus normal tissue was consistent across samples, we evaluated the variability of each cholesterol-related pathway. To accomplish this, the pan-cancer coefficient of variation (SD/mean) and the median of cholesterol-related scores were calculated and tumoral and normal values were compared (Supplementary Fig. S3a). Cholesterol uptake had the highest median increase between normal and tumor samples and displayed the lowest heterogeneity across tumor samples, given by its lowest coefficient of variation. Additionally, consistent with previous observations, median catabolism and storage scores tended to decrease, while biosynthesis and efflux scores tended to increase from normal to tumor samples (Supplementary Fig. S3a). Furthermore, compared with normal, tumor samples exhibited higher coefficients of variation for all pathways, indicating greater heterogeneity in cholesterol metabolism among tumors (Supplementary Fig. S3a). Coefficient of variation versus median score plot, across cholesterol-related pathways, and the tumor-specific coefficients of variation histogram, across distinct TCGA projects, are available in the Supplementary Fig. S3b and c.

3.3 Cholesterol metabolism varies across disease stages and subtypes

Next, we compared cholesterol-related pathways scores across disease stages, histological and molecular subtypes to explain the heterogeneity found in tumors and gain insights into how cholesterol-related pathways impact or are impacted along tumor progression.

Comparison of cholesterol-related pathways across tumor stages revealed an inconsistent pattern across all tumor types (Supplementary Fig. S4). In lung, uveal, esophagus, testis, and ovarian cancers, no alterations were found across disease stages. In thyroid cancers, all pathways were altered across disease stages. In the remaining tumors, one or more pathways were altered. Nevertheless, despite some pathways showing significant differences according to disease stage in specific tumors, these changes rarely followed a coordinated progression along stages. This indicates that regulation of cholesterol metabolic pathways is not uniform during disease progression and remains highly influenced by the organ of origin.

In addition, cholesterol-related pathways exhibited substantial differences between histological/molecular subtypes in certain tumors (kidney, breast, stomach, and brain) highlighting subtype-specific heterogeneity (Supplementary Fig. S5 and S6).

3.4 Cholesterol impacts patient survival in a pathway-specific manner

Further, pan-cancer association between cholesterol-related pathways and survival was assessed using multivariable Cox proportional hazards models to estimate hazard ratios (HRs), adjusting for stage or grade in the case of glioblastoma (GBM), and low-grade glioma (LGG) (Fig. 3). Higher expression of cholesterol catabolism- and storage-related genes tended to be protective (HR < 1), whereas higher expression of uptake-related genes was associated with increased risk of death (HR > 1). No significant association was observed for biosynthesis- or efflux-related gene expression (Fig. 3). Organ-stratification revealed that catabolism was protective in all tumors showing significant associations, consistent with pan-cancer analysis, including uveal melanoma (UVM), prostate (PRAD), liver (LIHC), kidney chromophobe (KICH), breast (BRCA), and adrenocortical (ACC) tumors (Supplementary Fig. S7b). In contrast to the pan-cancer analysis, stratified analysis for cholesterol storage showed that significant HRs in stomach (STAD) and lung adenocarcinoma (LUAD) were associated with increased risk of death (Supplementary Fig. S7d). Finally, cholesterol uptake HRs across tumor types largely agreed with the pan-cancer analysis, with HRs of ovarian (OV), LGG, GBM, KICH, renal papillary carcinoma (KIRP), and bladder (BLCA) tumors showing higher death risk associated with increased uptake-related gene expression (Supplementary Fig. S7e).

Fig. 3

Pan-cancer prognostic significance of cholesterol pathways expression levels. Multivariable Cox proportional hazards models were used to estimate Pan-Cancer hazard ratios and 95% confidence intervals (CI for each cholesterol metabolism pathways). Models were adjusted for tumor stage or grade (GBM and LGG) and for the remaining cholesterol pathways. Tumor-specific hazard ratios are available in Supplementary Fig. S7

3.5 Cholesterol-related pathways are differentially associated with tumor and immune microenvironment features

To assess associations between cholesterol-related pathway and tumor features, ssGSEA was used to quantify enrichment of key tumor-associated pathways (e.g. cell cycle regulation, stress response, membrane dynamics, metabolism, signaling, differentiation, and inflammation), and Pearson correlations were calculated between their enrichment scores and cholesterol-related pathway scores (Fig. 4a and Supplementary Fig. S8a).

Uptake showed the highest Pearson correlation coefficients, primarily with tumor-associated pathways involved in inflammation, such as Complement System (r = 0.75) and IL-6-JAK-STAT3 Signaling (r = 0.63), and with metabolic pathways, including Peroxissome (r = 0.44), and Fatty Acid Metabolism (r = 0.47), and stress response pathways, such as the Reactive Oxygen Species pathway (r = 0.49) and Apoptosis (r = 0.45) (Fig. 4a). Likewise, one of the highest correlation coefficients related to uptake was Kras signaling Upregulated (r = 0.53), one of the most frequently mutated genes across tumors. Catabolism, storage, and efflux displayed correlation patterns resembling those of uptake, though the correlation coefficients were consistently lower (Fig. 4a). Biosynthesis showed a markedly different correlation pattern, displaying strong positive correlations with metabolism and proliferation pathways, and negative correlations with TGF-b signaling (r = -0.44), VEGFA-VEGFR2 Signaling (r = -0.38), and EGF-EGFR Signaling (r = -0.38) (Fig. 4a). Correlation patterns between tumor-associated and our manually curated cholesterol-related set were largely conserved across tumor types (Supplementary Fig. S9), thus reinforcing cholesterol as a regulator of the identified pathways.

To further evaluate the relevance of the five cholesterol-related pathways in the tumor immune microenvironment, ESTIMATE was used to calculate immune scores, and Consensus TME predicted relative leukocyte proportions subpopulations per sample. Pearson correlations were then calculated between these algorithm scores and cholesterol-related pathway scores (Fig. 4b and Supplementary Fig. S8b). Cholesterol uptake showed a prominent positive correlation with the immune score, predominantly involving T cells: Tregs (r = 0.46), cytotoxic cells (r = 0.41), and CD4+ T cells (r = 0.40), with eosinophils (r = 0.44), and endothelial cells (r = 0.40) (Fig. 4b). Storage, efflux, and catabolism showed positive correlations with most immune cell populations, though weaker than for uptake (Fig. 4b). Regarding biosynthesis, while most correlations with immune cell populations were weak, slightly stronger negative associations were seen with B cells (r = -0.35), neutrophils (r = -0.35), and plasma cells (r = -0.39) (Fig. 4b). Across tumor types, immune cell correlation patterns were largely consistent, except in OV tumors, where correlations with cholesterol biosynthesis were slightly positive, ranging from −0.10 for Fibroblasts to 0.32 for cytotoxic cells (Supplementary Fig. S10).

Lastly, to assess whether cholesterol-related pathways might interact or exert coordinated effects on patient survival, we investigated whether pathways associated with a favorable prognosis could modulate the negative impact of cholesterol uptake on patient prognosis. Pan-cancer samples were initially categorized into uptake-low and uptake-high groups. Each group was further stratified according to the scores of the remaining pathways, yielding corresponding high and low subgroups. Subsequently, survival analyses were performed on these subgroups. As a result, we observed that in the cholesterol uptake-low groups, there is a drastic reduction on patient survival when the catabolism is also low (p-value = 6e-07) (Supplementary Fig. S11b), while the combinations of cholesterol uptake with the other pathways have no effect on patient survival (Supplementary Fig. S11a, c, and d). To investigate the factors underlying this drastic decrease in patient survival, we conducted GSEA on uptake-low samples, comparing groups with low versus high catabolism scores. Enrichment map analysis of the catabolism-low group showed enrichment of pathways mainly related with cell proliferation and DNA replication (Supplementary Fig. S12a). Catabolism-low tumors also showed an increase of fibroblasts and endothelial cells, with immune cells following different trends. Specifically, there was a decrease in neutrophils, plasma cells, and M1-like macrophages, and an increase in cytotoxic cells and regulatory T cells (Supplementary Fig. S12b), suggestive of an immunosuppressive signature. Yet, the increase of cytotoxic T cells led us to further explore the expression of immune checkpoint genes. While most checkpoints (e.g., PDCD1, TIGIT, LAG3, CTLA4) were downregulated in the catabolism-low samples, the gene coding for the immune inhibitory receptor SIGLEC7 was upregulated (Supplementary Fig. S12c).

Fig. 4

Enrichment of tumor-associated pathways and deconvolution of immune cells population. a. Pan-cancer Pearson coefficient correlation between cholesterol metabolism pathways scores and enrichment scores of tumor-associated pathways. Tumor-associated pathways were grouped into seven functional categories; b. Pan-cancer Pearson correlation coefficients between enrichment scores of cell-specific immune signatures, obtained using ESTIMATE (top) and ConsensusTME (bottom), and cholesterol metabolism pathways scores. Cells are sorted according to the strength of their Pearson correlation with uptake, in descending order. Significant correlations (p < 0.05) are indicated by dots; dot size increases with statistical significance. Correlation matrices are available in Supplementary Fig. S8; tumor-specific correlations are available in Supplementary Fig. S9 and S10

3.6 Cholesterol uptake associates with an inflamed microenvironment and with poor prognosis

Given consistent correlations of uptake with both tumor and microenvironment features, this pathway was further explored. For that, samples were grouped by the expression levels of uptake-related genes, and GSEA was performed to identify pathways altered in high versus low contexts. Top-50 enrichment map nodes corresponding to upregulated transcripts in the uptake-high group were exclusively associated with inflammation and immune response (Fig. 5a). While immune-related pathway enrichment persisted across uptake groups, other pathways emerged, particularly synapse-like signaling (uptake-medium versus uptake-low) and organic acid, lipid and xenobiotic metabolism (uptake-high versus uptake-medium) (Supplementary Fig. S13a and S13b, respectively).

Given that a high degree of tumor immune infiltration is commonly associated with a high tumor mutational burden (TMB), we evaluated whether tumors with high uptake also exhibited a high TMB. No significant association between cholesterol uptake and TMB was found (Supplementary Fig. S13c), thus pinpointing cholesterol uptake as an independent regulator of tumor immune infiltration.

Following the pronounced inflammation and immune infiltration in the uptake-high group, we analyzed immune-checkpoint gene expression to explore potential immune modulation pathways. A significant upregulation was observed in the uptake-high group for all the immune-checkpoints analyzed, with exception for Thymocyte selection-associated HMG-box (TOX) (Fig. 5b). Finally, given the concurrent increase in inflammation and immune-checkpoint expression, potentially reflecting activation or exhaustion, we compared overall survival between patients with high and low uptake. This analysis showed reduced survival in the high-uptake group compared with the low-uptake group (Fig. 5c).

Fig. 5

Immune contexture variation and survival outcomes between groups with distinct cholesterol uptake. a. Enrichment map of gene set enrichment analysis (GSEA) of uptake-low versus uptake-high samples. Node size is related to the number of components identified within a gene set. GSEA terms enriched in the uptake-high group are coloured in orange and grouped into nodes with associated terms; b. Expression of the immune supressor genes in both uptake-high and uptake-low groups; p denotes the two-sided p-value obtained from either a Student’s t-test or a Mann–Whitney U test, depending on data distribution, which was assessed using the Shapiro–Wilk test; c. Pan-cancer overall survival curves for uptake-low and high groups. Model was adjusted for the remaining cholesterol pathways and p-value was calculated using Wald test

3.7 Cholesterol uptake positively correlates with KRAS signaling

Given the strong correlation between cholesterol uptake and KRAS signaling, along with the relevance of KRAS in cancer [26, 27], we further investigated this association. First, to assess whether cholesterol uptake associates with other cancer driver genes, we compared mutation frequencies for the ten most commonly mutated cancer genes [28] between the uptake-high and uptake-low groups. KRAS showed the greatest difference in mutation frequency between groups (Supplementary Fig. S14a), revealing a higher mutation rate in tumors with high cholesterol uptake: 55% of uptake-high cases versus less than 23% of uptake-low cases.

To further explore the functional association between cholesterol uptake and KRAS signaling, we compared the KRAS signaling upregulated score between KRAS-mutated and wild-type samples (Fig. 6a) according to the cholesterol uptake group (Fig. 6b). The distribution of KRAS mutations across tumor types included in the cohort is shown in Supplementary Figure S14b. By comparing mutated versus wild-type samples within the same uptake group, mutated samples consistently exhibited higher KRAS signaling upregulated scores. Notably, the comparison of uptake-low KRAS-mutated samples with uptake-high KRAS wild-type revealed significantly higher KRAS signaling upregulated score in the non-mutated uptake-high group. These data suggest that high cholesterol uptake drives greater expression of KRAS pathway-related genes than KRAS mutations do when cholesterol uptake is low. This is further supported by a strong positive correlation between cholesterol uptake and the KRAS signaling upregulated score in both KRAS-mutated and wild-type samples (Supplementary Fig. S15a and S15b), indicating that cholesterol uptake influences KRAS pathway activity independently of KRAS mutational status. This correlation was observed across tumor types except in ACC and LIHC, where it was not significant (Supplementary Fig. S15c).

To further investigate this association, we selected the KRAS downstream effectors from the differentially expressed genes identified in the uptake-low versus uptake-high comparison. In the KRAS wild-type samples, the uptake-high group showed enrichment of PIK3CD and NFKB1 (a PI3K/AKT transcription factor), whereas the uptake-low samples showed enrichment of BRAF (Fig. 6c). Curiously, the uptake-low group had a higher mutation rate of BRAF than the uptake-high group (Supplementary Fig. S14a). In contrast, KRAS-mutated samples showed global upregulation of downstream effectors in uptake-high samples, indicating broad activation rather than a pathway-specific effect (Fig. 6d). These findings show that, unlike KRAS-mutated tumors, in wild-type tumors cholesterol uptake levels dictate preferentially downstream pathway engagement. To evaluate the survival impact of the interaction between uptake and KRAS mutation, pan-cancer multivariable Cox proportional hazards models were fitted. This analysis revealed that in the uptake-high subset, KRAS mutations were associated with significantly reduced patient survival. Conversely, in the uptake-low subset, KRAS mutations did not impact prognosis (Fig. 6e).