Background
Necroptosis is a novel form of regulated necrotic cell death mechanistically mimicking apoptosis and morphologically resembling necrosis [
1,
2]. It is mainly regulated by the key proteins such as RIPK1, RIPK3, and their substrate, mixed-lineage kinase domain-like protein (MLKL) [
3‐
5]. Previous researches have reported the relevance of necroptosis in many human diseases, including inflammatory diseases, neurodegenerative diseases, and cancer etc. [
6‐
8]. In addition, it has been suggested to be involved in cancer initiation, progression, immunity, and chemoresistance, providing novel perspectives and potential targets for cancer therapy, for which several therapeutic agents aiming to treat cancer by inducing or manipulating necroptosis are under investigation [
6,
9].
Colorectal cancer (CRC) is a major lethal malignancy worldwide [
10,
11]. Like other malignancies, tumor microenvironment (TME) plays an indispensable role in CRC tumorigenesis [
12]. Previous reports indicated that myeloid-derived suppressive cell (MDSC), an anti-tumor immune suppressor, accumulates in CRC tissue and promotes cancer metastasis [
13,
14]. In advanced stage CRC, the well-known immune-activated effectors, CD8
+ T cells can be suppressed by IL-17A secretion from Th17 cells [
15]. As the most exciting breakthrough in cancer treatment, immune-checkpoint blockade (ICB) therapy based on CTLA-4 and PD-1, has demonstrated promising efficacy in CRC patients [
16‐
18]. However, only some of those with microsatellite instability high (MSI-H) or mismatch repair deficient (dMMR) status could benefit from ICB therapy [
19]. Therefore, it is necessary and urgent to further investigate the TME characteristics in CRC to identify more effective immunotherapeutic targets.
The involvement of necroptosis has been reported not only in cancer cells but also in other components in the TME [
20,
21]. For example, necroptosis could promote pancreatic tumorigenesis by inducing the expression of CXCL1, a potent chemoattractant for myeloid cells that was highly expressed in a RIP1- and RIP3-dependent manner, which could shape the immune suppressing environment [
22]. Therefore, further exploring the correlation between TME cells infiltration and necroptosis can provide new perspectives for understanding underlying mechanisms and developing cancer therapeutics, such as combination treatment of necroptosis-based therapy and immunotherapy.
By using bulk and single-cell transcriptomic data analysis, we identified two stable necroptosis-related phenotypes in CRC: a phenotype characterized by few TME cells infiltration but with EMT/TGF-β pathways activation, and another recognized as an immune-excluded phenotype [
23]. We further established a scoring system, which could reveal TME characteristics, help accurately determine patients’ survival outcomes, and predict responses to immunotherapy and chemotherapy.
Materials and methods
Preparation of bulk RNA expression datasets
A total of 1003 patients from Gene Expression Omnibus (GEO) database (including GSE33113, GSE39582, GSE14333, and GSE37892) were recruited in this study. We corrected the batch effects of GEO datasets using combat method [
24] and integrated them into a meta-GEO cohort.
Analysis of single-cell RNA data
Single-cell RNA (scRNA) datasets were downloaded from GEO database (including CRC datasets from GSE144735, GSE178318, LUAD datasets from GSE131907). We calculated the score of single-cell using ‘AddModuleScore’ function via signature α and β.
To calculate the risk score of single-cell data, we first averaged gene expression of each patient to represent their bulk gene expression level. Then we calculated their risk score as follows: risk score = Σ (Expi × coefi), according to methods in necroptosis-related gene score (NRG_score).
Thirty-three necroptosis-related genes (NRGs) were retrieved from previous publications [
4,
8]. The details of NRGs are shown in Additional file
11: Table S2.
Consensus molecular clustering by “ConsensusClusterPlus”
We performed consensus clustering with “ConsensusClusterPlus” to identify classifications in CRC patients based on the expression of necroptosis-related genes (NRGs). The final number of clusters was determined by cumulative distribution function (CDF). K = 3 was finally set as the number of clusters. The annotation of clusters of all datasets was shown in Additional file
10: Table S1.
Gene set variation analysis (GSVA) and single-sample gene set enrichment (ssGSEA) analysis
We calculated pathway activities of tumor samples (Fig.
2E and Additional file
3: Fig. S3C) using GSVA R package. The gene-signatures included for analyzing were downloaded from Hallmark gene sets and C2 curated gene sets (MSigDB database v7.4) [
25].
We evaluated immune cell types signature scores using ssGSEA analysis. The immune cell types signature was extracted from the study of Charoentong [
26].
CMS classification for bulk RNA-seq
We utilized CMSclassifier [
27] to classify TCGA-COAD/READ tumor samples. The CMS subtypes of TCGA and GEO databases were shown in Additional file
10: Table S1.
TME infiltration evaluation using ssGSEA, CIBERSORT and ESTIMATE
We adopted the CIBERSORT [
28] deconvolution approach to evaluating the relative abundance of 22 tumor-infiltrating immune cells (TIICs). To confirm the stable TME infiltration patterns of necroptosis-related clusters, we also evaluated immune cell infiltration with cell types from the study of Charoentong [
26] using ssGESA analysis [
29]. In addition, we used ESTIMATE algorithm to calculate tumor purity, immune and stromal scores of each patient.
Somatic mutation analysis
Varscan file format of somatic mutation data were downloaded from
https://portal.gdc.cancer.gov/repository. Copy number variation information was curated from UCSC Xena online. Maftool R package was used to identify mutant genes and calculate TMB level.
Quantitative real-time polymerase chain reaction (RT-qPCR)
We collected 208 pairs of patients’ tissues (including CRC and adjacent non-tumor tissues) from Fudan University Shanghai Cancer Center (FUSCC) in this study. The written informed consent was signed by all patients according to the Institutional Review Boards of FUSCC, and the study was approved by the Ethical Committee of FUSCC.
RNA was extracted from these samples by using TRIzol reagent (Invitrogen, Carlsbad, CA, USA), which was then reversed into complementary DNA (cDNA) with a PrimeScript RT reagent kit (Takara). Then RT-qPCR was performed using SYBR-Green assays (Takara). The data were calculated using the 2
−ΔΔCt value, and normalized with 18 s rRNA. The primer sequences used in our study are shown in Additional file
15: Table S6.
Construction of the prognostic NRG_score
NRG_score was calculated to quantify the expression patterns of NRGs the individual samples. First, the differentially expressed genes (DEGs) were subjected to univariate Cox regression analysis to identify those linked to CRC overall survival. Second, the patients were classified into different necroptosis phenotype-related groups (gene-cluster A, gene-cluster B, and gene-cluster C) for deeper analysis using an unsupervised clustering method based on the expression of prognostic DEGs (Additional file
13: Table S4) and 33 NRGs. Finally, based on necroptosis phenotype-related prognostic genes, the Lasso Cox regression algorithm was used to minimize the risk of over-fitting using the “glmnet” R package [
30]. We analyzed the change trajectory of each independent variable and then used tenfold cross-validation to establish a model. As previously reported [
31], we totally performed 1000 iterations and included 5 gene groups for further screening. A gene model with 13 genes showed the highest frequencies of 726 compared to other four-gene models (Fig.
5A). Thus, this 13-gene model was applied to generate the gene signature for calculating NRG_score, which was calculated as follows:
$$\mathrm{NRG}\_\mathrm{score }=\Sigma (\mathrm{Expi }\times \mathrm{ coefi})$$
Based on the median risk score, a total of 578 patients in the training set were divided into low-risk and high-risk groups in survival analysis. Similarly, the testing and all sets were divided into low- and high-risk groups, each of which was subjected to Kaplan–Meier survival analysis and the generation of receiver operating characteristic (ROC) curves. The NRG_score of TCGA and GEO datasets were shown in Additional file
14: Table S5.
Drug susceptibility analysis
To explore the differences in the therapeutic effects of drugs in CRC patients, we calculated the drug imputed sensitivity score of drugs from Sanger’s Genomics of Drug Sensitivity in Cancer (GDSC) v2 using the “oncoPredict” package [
32].
Kaplan–Meier survival analysis
We plotted the Kaplan–Meier (K-M) survival curve using R package ‘Survminer’ (0.4.6). We stratified samples into high and low gene expression subgroups using surv-cutpoint function.
Statistical analyses
Statistical analysis was performed using R (version 4.0.0) and GraphPad Prism (version 7.04). The Wilcox test, log rank test, Kruskal–Wallis H test, and Pearson’s Chi-square test were performed in this study. Detailed descriptions of statistical tests are specified in the figure legends.
Discussion
Cell death has recently attracted increasing attention for its potential role in triggering anti-tumor immunity [
43]. Like apoptotic cells, emerging researches have showed that necroptotic tumor cells can induce anti-tumor immunity by their interaction with diverse immune cell types [
44,
45]. Although various studies have revealed the regulation of NRGs in TME [
46,
47], a landscape of TME characteristics mediated by NRGs have not been comprehensively understood.
In this study, we introduced necroptosis-related phenotypes of TME in CRC. Based on 33 NRGs and DEGs associated with necroptosis-related phenotypes, we could stratify CRC samples into three molecular phenotypes (NRC1-3). However, we observed that only two classifications kept stable according to their immune infiltration patterns. Therefore, we postulated that there were two stable TME patterns mediated by necroptosis in CRC: a phenotype characterized by few TME cells infiltration but with EMT/TGF-β pathways activation, and another phenotype characterized by remarkable stromal cells infiltration, together with EMT, TGF-β signaling pathway activation, corresponding to the immune-excluded and CMS4-like phenotype. To confirm these two stable phenotypes related to necroptosis, we performed single-cell transcriptomic analyses in CRC datasets and further validated in LUAD datasets. We observed that score of NRC1 represented by score α was increased in tumor metastatic sites, while score β was elevated in TME cells. We thus postulated that EMT phenotype in NRC1 was mainly exhibited on tumor cells, while CMS4-like and EMT phenotype in NRC3 were predominantly caused by its remarkable stromal cell infiltration. What’s more, high α score might be used to predict the risk of CRC metastasis.
Previous reports suggested that immune context of TME could promote EMT. MDSCs, well-known as immature immune cells, are associated with poor prognosis of cancers for suppressing T cells activation [
48]. TGF-β production from MDSCs have been experimentally proved to render a profound impact on tumor metastasis [
49]. Stromal cells such as fibroblasts have been also reported as a major source of TGF-β production [
50,
51]. TGF-β expressed by cancer-associated fibroblast (CAF) (such as myofibroblast) induces recruitment of more fibroblasts, and might thus lead to a pro-tumorigenic and immunotolerant status [
52]. Adaptive immune cells like CD8
+ T cells respond to TGF-β may also cause an immunosuppressive environment. Since NRC3 was infiltrated by stromal cells and MDSCs, patients in NRC3 cannot respond to PD-1/PD-L1 therapy. Fortunately, NRC3 was remarkably infiltrated by activated T cell populations such as CD4
+ and CD8
+ T, which should have been related to anti-tumor immunity. High expression of PD-1/PD-L1 was observed in NRC3, which has been reported to predict response to immune checkpoint inhibitors [
53]. Therefore, intervention targeting on stromal cells and MDSCs, and downregulation of TGF-β may help patients within NRC3 regain an effective response to immunotherapy. Without considering TME, the role of necroptosis in tumor cells has not been comprehensively understood either [
54]. Previous findings showed that RIPK3 was upregulated in late-stage breast tumors, implying a promising role of necroptosis in tumor progression [
54,
55]. In NRC1, we observed upregulation of RIPK3 (Fig.
5A), EMT activity (Additional file
3: Fig. S3A and S3D), and enrichment of advanced stages (15.60%; Fig.
2E), suggesting that RIPK3 may play an indispensable role in CRC progression. Emerging evidences demonstrated that RIPK3 upregulation could potentiate chemotherapeutic effects by inducing necroptosis [
56]. Therefore, RIPK3 may be a key mediator resulting in EMT and chemo-sensitive phenotype of patients within NRC1. Future experimental researches are required to investigate the key regulator RIPK3 in CRC development.
We also constructed a robust and effective prognostic NRG_score and demonstrated its predictive ability in CRC survival by integrated analyses of public databases and a patients’ cohort from FUSCC. Patients with low- and high-risk NRG_score displayed significantly different clinicopathological characteristics, prognosis, immune infiltration and drug susceptibility. We observed that high-risk score group was highly infiltrated by myofibroblast and characterized by TGF-β pathway activation. In contrast, low-risk group was enriched with more cytotoxic T cells. We further explored cytotoxic genes like GZMA and IFNG in public database, confirming the precise predictive ability of low-risk score in response to immunotherapy. Interestingly, the exploration of drug imputed score showed patients in high- and low-risk groups might present different chemotherapeutic efficacy, suggesting that NRG_score could be used for patient selection when considering ADJC and there might be potential molecular targets based on NRGs. Finally, by integrating NRG_score and tumor stage, we established a quantitative nomogram, which further improved the performance and facilitated the use of NRG_score. Overall, the NRG_score we constructed can be an accurate prognostic model for prognosis stratification of CRC patients, and a good predictor for immunotherapy and chemotherapy.
In a nutshell, we comprehensively analyzed the mutations and expression patterns of NRGs in CRC. NRCs and NRG_score were established and their associations with TME were explored. Sensitivity to chemotherapy and response to immunotherapy were probed. These integrated analyses highlighted the main role of necroptosis in TME infiltration of CRC. Moreover, we put forward specific genes related to EMT phenotype on tumor cells, and genes related to stromal cells infiltration in TME, which will provide an interesting insight into the mechanism between necroptosis and TME infiltration. However, there are still some limitations: (1) the study was conducted based on retrospective data, thus, selection bias might be unavoidable; (2) though we validated our findings in validation sets based on public datasets, validation in prospective study will further add credibility to these findings; (3) molecular mechanisms of these observations necessitate exploration in the future.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.