Introduction
Colon cancer is one of the most common human malignancies in the world [
1]. Despite the fact that clinical treatment for colon cancer has been improved with the development of immune checkpoint inhibitors (ICIs), the majority of colon cancer patients have limited response to ICI therapies [
2]. Emerging biomarkers such as tumor mutational burden (TMB), inflammatory tumor microenvironment (TME) and microsatellite instability (MSI) have been identified to predict therapeutic benefit in colon cancer [
3]. Unfortunately, drawbacks still remained in current biomarkers [
3]. Therefore, it is important to excavate the predictive biomarkers for immunotherapy response and find a novel strategy for sensitizing ICIs in colon cancer.
Lipid metabolic reprogramming promotes tumor growth, angiogenesis and metastasis [
4]. A prognostic signature of nine lipid metabolism-related genes (LMGs), including CDIPT, MTMR7, PIK3CB, PIK3C2G, ARSE, ARSJ, GLA, GLB1 and UGCG, has been established in diffuse gliomas [
5]. The model including the four LMGs (ABCA1, ACSL1, AGPAT1 and SCD) is proposed as a prognostic marker of colon cancer with stage II [
6], but not effective in all stages of colon cancer [
7]. A very recent study has shown an 8-gene prognostic signature based on LMGs in colon adenocarcinoma [
8]. However, the prognostic prediction model of multiple LMGs in colon cancer has only begun to be appreciated.
Metabolic reprogramming of tumor cells induces metabolic stress in tumor-infiltrating immune cells and stromal cells, and thereby impairs antitumor immune responses [
9]. Targeted reprogramming of lipid metabolism inhibits tumor cell growth, alleviates the immunosuppressive TME and improves response to ICI therapy [
10]. A study has shown that cyclooxygenase (COX) enzyme inhibitor aspirin in combination with anti-programmed cell death protein 1 (PD-1) treatment exhibits a synergistic effect in reducing tumor growth [
11]. Our recent study has also shown that inhibition of COX-2 catalyzed metabolism of arachidonic acid (AA) by melafolone promotes anti-PD-1 therapy in lung cancer through PD-L1 downregulation [
12]. Avasimibe, an inhibitor of cholesterol esterification enzyme, significantly empowers the anti-tumor response of CD8
+ T cells, and its combination with anti-PD-1 antibody has a better efficacy in melanoma [
13]. Some clinical studies have further confirmed that targeted metabolic reprogramming enhances the anti-tumor efficacy of ICIs [
14]. Therefore, it is essential to investigate the new strategy of targeting tumor lipid metabolism to alleviate the immunosuppressive TME and to enhance anti-tumor immunotherapy.
Cytochrome P450 (CYP) 19A1 encodes aromatase, an isoenzyme of estrogen biosynthesis, and is overexpressed in colon cancer tissues [
15]. Aromatase inhibitors, including letrozole, anastrozole and exemestane, inhibit the synthesis of estrogen from androgen by binding to the aromatase, and thereby block tumor cell proliferation [
15]. Aromatase inhibitors are widely used to treat post-menopausal hormone-sensitive breast cancer [
16]. Exemestane exhibits great growth inhibitory potentials in gastric cancer when administered in combination therapy with 5-fluorouracil [
17]. Letrozole plus mTOR inhibitor everolimus results in a better progression free survival in patients with breast cancer, accompanied by decreases in Ki-67 index and tumor-infiltrating regulatory T cells, and an increase in tumor-specific CD8
+ T cells [
18‐
20]. Recently, the combination of exemestane and CTLA-4 monoclonal antibody tremelimumab in breast cancer has entered a phase II clinical trial [
21]. However, the effect of aromatase inhibitors on the immunotherapeutic response of colon cancer is not well understood.
In this study, we utilized The Cancer Genome Atlas (TCGA) cohort as a training set to develop a lipid metabolism-related gene prognostic risk score (LMrisk) in colon cancer, and the significant prognostic values of the model were then validated in three Gene Expression Omnibus (GEO) testing sets. The relationships of LMrisk with 36 immune cell signatures and predictive biomarkers for immunotherapeutic response were evaluated. In particular, the bioinformatic findings were confirmed using human colon cancer tissue samples, in vitro coculture of colon cancer cells with peripheral blood mononuclear cells (PBMCs), and mouse xenograft models of colon cancer. Our study develops a novel lipid metabolism-related risk model to predict immunotherapeutic response, and elucidates the molecular mechanism by which CYP19A1-catalyzed estrogen synthesis mediates immune escape, providing new targets and candidates for sensitizing colon cancer immunotherapy.
Methods
Data and resources
The RNA-sequencing data and clinical information of colon cancer patients were downloaded from the TCGA database (
https://portal.gdc.cancer.gov). The RNA expression profiles contained 494 samples. We obtained 453 colon cancer samples and 41 adjacent normal tissue samples. 435 colon cancer patients with survival data were included in the following study. Gene expression profiles of GSE41258 dataset based on the Affymetrix human genome U133A array platform, GSE38832 and GSE39582 datasets based on the Affymetrix human genome U133 Plus 2.0 array platform and clinical data were downloaded from the GEO database (
https://www.ncbi.nlm.nih.gov/geo/). A list of LMGs was collected from the “metabolism of lipids” in Reactome (
https://reactome.org/download-data/). Genes not included in TCGA or GEO databases were excluded, and 731 genes related to lipid metabolism were obtained.
Differential expression analysis of LMGs and functional annotation
The “edgeR” [
22] package was used to identify differentially expressed LMGs between the tumor and adjacent normal tissue samples. Adjusted
P-value < 0.05 and |log2 (fold change) |> 1 were chosen as the cut-off threshold. The protein–protein interaction network of differentially expressed LMGs was analyzed by STRING database (
https://string-db.org/). Principal component analysis (PCA) was utilized to analyze the expression pattern in the colon cancer and normal tissues. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis were performed with the differentially expressed LMGs by using the “clusterProfiler” R package [
23]. False discovery rate < 0.05 was considered statistically significant.
Construction and verification of the LMrisk
Univariate Cox regression analysis was performed to identify the prognosis-related LMGs. The P-value < 0.05 in univariate Cox regression analysis was considered statistically significant. The “glmnet” R package was used to perform a least absolute shrinkage and selection operator (LASSO)-Cox regression model analysis. The weighted LASSO-Cox coefficients based on individual gene expression levels were used to calculate the lipid metabolism-related risk score (LMrisk) as follows: LMrisk = ∑ (expression of genei × Coefficient of genei). The patients in the TCGA database were stratified into the low- and high-LMrisk groups according to median LMrisk value, and their survival were analyzed using the Kaplan–Meier method. Log-rank test was used to compare the survival curves of two or more groups. The specificity and sensitivity of the LMrisk in predicting 3-, 5- and 10-year survival were determined by receiver operating characteristic (ROC) analysis using the “survivalROC” R package, and the areas under curves (AUC) were calculated. We used a similar approach in the GSE41258, GSE38832 and GSE39582 datasets to verify the applicability of the LMrisk. To study whether the LMrisk is an independent predictor for overall survival of colon cancer patients, univariate and multivariate Cox regression analyses were conducted. The LMrisk, age, gender and TNM stage were used as covariates.
Construction and verification of prognostic nomogram
After testing for collinearity, all independent prognostic parameters were included in the construction of a nomogram to predict 3- and 5-year overall survival of colon cancer patients. Age, TNM stage and LMrisk were used to construct the nomogram using the “rms” and “survival” packages in R. The discrimination power of the predictive model was evaluated by Harrell’s concordance index (C-index). We calculated the C-index with 95% confidence interval using the bootstrap approach with 1000 resamples. Then, calibration curves were drawn to assess the consistency between actual and predicted survival. The nomogram performance was validated by ROC curves at 3 and 5 year using the TNM stage as control.
Analysis of tumor immune signatures and function enrichment for LMrisk
The Estimation of Stromal and Immune cells in Malignant Tumors using Expression data (ESTIMATE) was used to evaluate the immune score, stromal score and ESTIMATE score of each sample [
24]. Based on the gene expression data in the cancer tissues, the xCell and TIMER algorithms were applied to estimate the infiltration of immune cells in each sample [
25]. The R package “maftools” was used to evaluate and sum the mutation data.
Chemicals and reagents
Letrozole (T1590) was purchased from TargetMol (Bellingham, WA, USA). CYP19A1 siRNA (sc-41498) and CYP19A1 antibody (sc-374176) were purchased from Santa Cruz Biotechnology (Santa Cruz, CA, USA). PD-L1 (13684) antibody was purchased from Cell Signaling Technology (Danvers, MA, USA). PE-conjugated anti-CD3 antibody (555340), APC-conjugated anti-CD8 antibody (566852), PE-Cy7-conjugated anti-IFN-γ antibody (557643) were purchased from BD Biosciences (San Jose, CA, USA). FITC-conjugated anti-CD107a antibody (328606) and Zombie NIR (423105) were purchased from BioLegend (San Diego, CA, USA). Opal Polaris 7 Color IHC Detection Kit (NEL861001KT) was purchased from Akoya Bioscience (Menlo Park, CA, USA). Chitosan (CS, deacetylation 98%, Mw = 50 KDa) and sodium tripolyphosphate (TPP) were purchased from Sigma Chemical Co (St. Louis, MO, USA). Hyaluronic acid (HA) was purchased from Meryer Chemical Technology Co Ltd (Shanghai, China).
Human tissue microarray analyses for CYP19A1 and PD-L1 expression
Human colon carcinoma tissue microarray (HCol-Ade180Sur-05) containing 90 patients with colon cancer and adjacent normal tissues were obtained from Shanghai Outdo Biotech Co., Ltd. Each sample dot with a diameter of 1.5 mm and a thickness of 4 μm was prepared according to a standard method. Immunohistochemistry (IHC) of human colon cancer tissue microarrays were performed by using the CYP19A1 (1:100) and PD-L1 (1:300) antibodies. Immunostaining was graded using a two-score system based on intensity score and proportion score as previously described [
26]. The two scores were then multiplied to yield a total immunoreactivity score regarding the protein expression in a sample. IHC score was assessed independently by two pathologists, and a consensus of grading was reached. Patients were divided into high- and low-expression groups according to the median IHC score.
Multiplex immunofluorescence analysis
Tumor tissues from 20 patients with colon cancer were obtained from Zhongnan Hospital, Wuhan University (Hubei, China) between 2020 and 2021. The histological diagnosis for each sample was reconfirmed using microscopic examination of hematoxylin/eosin-stained sections. All samples were collected from patients with informed consent, and all procedures were conducted with the approval of the Ethical Committee of the Medical School of Wuhan University and performed in accordance with relevant regulations and guidelines.
Quantitative multiplex immunofluorescence (mIF) was performed to characterize the immune landscape in human colon cancer tissues using Opal Polaris 7 Color IHC Detection Kits (NEL861001KT, Akoya Bioscience, CA, USA) as previously described [
27]. Briefly, formalin-fixed paraffin-embedded sections were deparaffinized, followed by antigen retrieval with citrate acid buffer (pH 6.0)/Tris–EDTA buffer (pH 9.0) and then blocking with blocking/Ab diluent. Next, slides were incubated with primary antibodies against CYP19A1 (1:200), CD68 (Abcam, ab955, 1:800), CD31 (Abcam, ab24590, 1:300), α-SMA (Abcam, ab7817, 1:300), CD8 (DAKO, IR623, 1:5) and CK (AE1/AE3) (DAKO, IS05330-2, 1:3). Primary antibody was visualized using tyramide signal amplification linked to a specific fluorochrome from the mIF kit for each primary antibody. A stripping procedure based on microwaves was performed for each consecutive antibody staining. The stained slides were scanned using a Vectra® 3 multispectral microscope (Akoya Bioscience). From each slide, Vectra automatically captured the fluorescent spectra from 420 to 720 nm at 20-nm intervals with the same exposure time and then combined the captured images to create a single stack image that retained the particulate spectral signature of all markers. These data were analyzed using InForm 2.6 software.
Cell culture, cell proliferation assays, aromatase activity assays, CYP19A1 siRNA transfection, isolation of PBMCs and their coculture with colon cancer cells, flow cytometry analyses for tumor cell death and CD8
+ T cell function, western blot analyses for CYP19A1 and PD-L1 expression, tumor models and therapeutic efficacy study, and statistical analysis were described in
Supplementary Materials and Methods.
Discussion
Previous studies have shown that the alterations in lipid metabolism including cholesterol metabolism and AA metabolism affect immunotherapeutic response, and are promising biomarkers to predict the efficacy of immunotherapy [
41,
42]. In the present study, two novel observations have been made. First, we used differentially expressed LMGs, including CYP19A1, FABP4, LRP2, SLCO1A2, PPARGC1A and ALOXE3, to construct the LMrisk for predicting prognosis of colon cancer, and found that the LMrisk was associated with the immunosuppressive TME and predictive biomarkers of immunotherapeutic response in colon cancer. To our knowledge, this is the first study that directly demonstrated that a prognostic risk model based on LMGs may predict immunotherapeutic response in colon cancer. Second, we uncovered for the first time that CYP19A1 inhibition downregulated PD-L1, IL-6 and TGF-β expression in colon cancer cells, and thereby enhanced the tumor-killing ability of CD8
+ T cells. These results implicate the LMrisk based on the prognostic model as a predictive biomarker of immunotherapeutic response in colon cancer, highlighting its therapeutic potential for optimizing anti-PD-1 immunotherapy.
ALOXE3, PPARGC1A or FABP4 are prognostic biomarkers in colon cancer [
43‐
45]. However, a single gene is difficult to provide powerful predictive performance for colon cancer patients [
46]. Therefore, it is a trend to use a multiple gene model to predict prognosis of cancer patients [
47]. Indeed, several studies have used multiple LMGs to construct prognostic models in patients with breast cancer, gastric cancer or osteosarcoma [
48‐
50]. Here we integrated CYP19A1, FABP4, LRP2, SLCO1A2, PPARGC1A and ALOXE3 with clinicopathological information of patients to construct a prognostic nomogram in colon cancer. A recent study shows an 8-gene signature based on LMGs including RTN2, FYN, HEYL, FAM69A, FBXL5, HMGN2, LGALS4, STOX1 as a novel marker to predict colon cancer patients’ survival [
8]. We reason that this discrepancy in prognosis-related LMGs could be due to different algorithms (differentially expressed LMGs
vs. all LMGs). Our lipid metabolism-related nomogram has larger C-index than the model of Jiang et al. [
8], suggesting its ability to better predict colon cancer prognosis as compared with gene signature-derived risk score. A previous study shows that nomogram-derived prognosis, together with user-friendly digital interfaces, elevates the accuracy of cancer survival prediction, and thereby allows for seamless incorporation to aid clinical decision making [
47].
High level of MSI is closely related to the improved prognosis of colon cancer patients receiving immunotherapy [
29], and higher TMB results in more neo-antigens, increasing chances for T cell recognition, accompanied by better ICI outcomes [
30]. Here we demonstrated that the LMrisk was positively correlated with TMB and MSI, suggesting that the patients with high LMrisk are more likely to benefit from immunotherapy. Further analysis revealed that TTN mutation was the major reason for the high TMB in the colon cancer patients with high LMrisk. Previous studies show that TTN mutation is related to high immunogenicity and inflammatory tumor immune microenvironment in lung adenocarcinoma, accompanied by favorable objective response and survival with ICI administration [
51,
52]. Along with immunogenicity, inflammatory TME makes colon cancer patients amenable to respond to ICIs [
53]. We also observed more infiltration of macrophages, monocytes, NK cells and CAFs, and higher PD-L1 expression in colon cancer tissues from the patients with high LMrisk. Our extensive functional studies demonstrated that CYP19A1 protein expression was positively correlated with PD-L1 expression and infiltration of macrophages, CAFs and endothelial cells in human colon cancer tissues. Given that strong correlations of the LMrisk with PD-L1 expression, TMB and MSI, we conclude that the LMrisk including CYP19A1 is a promising biomarker for predicting immunotherapeutic response to colon cancer.
Several studies showed that men have a higher incidence of colon cancer than women, suggesting that estrogen may play a protective role in the development of colon cancer [
54]. However, recent epidemiological studies have shown that hormone replacement therapy in postmenopausal women does not inhibit the development of colon cancer [
55,
56]. Moreover, anti-PD-1 or anti-PD-L1 treatment resulted in a trend to greater overall survival and better response rates in individual males with colon cancer as compared with females [
57]. Aromatase encoded by CYP19A1 and GPR30 are highly expressed in colon cancer tissues, and high expression of GPR30 predicts poor prognosis in colon cancer patients [
58,
59]. Here, we demonstrated that CYP19A1 inhibition by letrozole or siRNA reduced production of IL-6 and TGF-β and downregulated PD-L1 expression by inactivating GPR30-Akt signaling, and thereby promoted the proliferation and cytotoxic activity of CD8
+ T cells (Fig. S
6). These data delineate that CYP19A1 inhibition combined with PD-1 antibody represents a promising therapeutic strategy for colon cancer.
According to our analyses, the LMrisk may predict response and outcome of immunotherapy in colon cancer. However, there are still some limitations that should be acknowledged. Firstly, because the mRNA expression data from colon cancer patients with immunotherapy is not available, the predictive ability of LMrisk for immunotherapeutic response is estimated indirectly by urothelial cancer cohorts and biomarkers. Therefore, further well-powered prospective studies are still needed. Secondly, we studied the expression and pathophysiological significance of CYP19A1 in vitro and in vivo, and further studies are needed on other LMGs including FABP4, LRP2, SLCO1A2, PPARGC1A and ALOXE3 in colon cancer.
Conclusions
Collectively, we screened the six LMGs including CYP19A1, FABP4, LRP2, SLCO1A2, PPARGC1A and ALOXE3, and constructed a signature to predict prognosis and immunotherapeutic response in colon cancer, which was extendedly and externally validated. Importantly, CYP19A1 inhibition improves anti-PD-1 immunotherapy for colon cancer and blunts PD-L1-induced anergy of CD8+ T cells. Our findings facilitate prediction for prognosis and immunotherapeutic response in colon cancer, and targeting lipid metabolism reprogramming in the TME may be promising strategies for synergy with anti-PD-1 treatment.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.