Alterations of the immune microenvironment with age predicts patient prognosis of gastrointestinal tract tumours

Dear Editor: Because aging plays critical roles in immune response and tumour development, alterations of the tumour immune microenvironment with age may be predictive of prognosis. Hence, we analysed single-cell RNA-seq (scRNA-seq) data of gastrointestinal (GI) tract tumours to search for immune cells enriched in elderly patients, and thereby raised an immune index with potential clinical values. We initially constructed a single-cell atlas using scRNA-seq datasets from Gene Expression Omnibus (GEO) for gastric cancer (GC) , colorectal carcinoma (CRC) and oesophageal carcinoma (OSCA), which contained 115 samples and 419 000 cells (Figure S1A,B and Table S1). The lymphoid and myeloid cells were reclustered to subpopulations (Figures S1C–G and S2A,B; Figure 1A,B), and their biased distributions in older (>60 years) and younger (≤60 years) patients were evaluated by the ratio of observed to expected cell numbers (Ro/e) analysis1 (Figure 1C). For each subpopulation showing preference for older patients, the top 100 differentially expressed genes were employed to score patients from TCGA datasets via ssGSEA.2 The following Kaplan–Meier analysis revealed several subpopulations that were inversely correlated to survival (Figure 1D and Table S2). Further analysis of the differentially expressed genes of these subpopulations discovered multiple commonly enriched pathways (Figure S2C–F). Moreover, among these subpopulations, the CD8T_C2 and Proli_T_C1 had high expression of T cell exhaustion markers (Figure S2G), although this may not necessarily stand for a canonical exhaustion state. Taken together, several immune subpopulations are enriched in older GI tract tumour patients and may affect prognosis. We analysed the potential communications between immune subpopulations with CellChat,3, 4 and evaluated strength of intercellular signalling pathway with IMPORTANCE (Figure S3A,B). Among the immune subpopulations with enrichment in older patients and correlation to prognosis, nine subpopulations demonstrated markedly altered signalling in older relative to younger patients (ΔIMPORTANCE > 10) (Figure 2A) and had intensive interactions (Figure S3C). Therefore, the nine subpopulations may constitute an elderly enriched immune meta-cluster (EIM). Deconvolution analysis of spatial transcriptomic data for GC tumours defined distinctive distribution patterns of immune subpopulations, and the pattern with preference for older patients contained several EIM component subpopulations (Figure 2B–D and Figure S3D). The spatial co-enrichment of EIM component subpopulations were observed by multiple methods (Figure S3D,E). Moreover, GI tract tumour patients in TCGA were scored for EIM followed by Kaplan–Meier analysis, which revealed an inverse correlation between EIM and survival (Figure 2E and Figure S3F). Meanwhile, pathway enrichment network analysis of the signature genes of EIM suggested their participation in immune regulation (Figure S3G). We then stained GI tract tumour sections for molecular markers of EIM. A large proportion of CD8+ T cells and CD14+/CD16+ monocytes belonged to EIM, and more immunofluorescent signals of CD8, CD14, and CD16 were detected in tumours from older relative to younger patients (Figure 2F,G). Besides, while lymphoid subpopulations of EIM scored high for T cell exhaustion markers (Figure S2G), elevated PD1 expression was observed in tumours from older patients (Figure S4A,B). Dual staining of these potential EIM markers showed the same trend (Figure S4C–F). Therefore, the EIM in GI tract tumours is enriched in older patients and inversely correlated with prognosis. We applied a machine learning framework composed of four classifiers and tested if EIM could classify elderly patients in TCGA and GEO datasets.5, 6 The datasets were subjected to every classifier for each immune subpopulation, resulting in area under the curve values as outcomes. The classifier with the highest area under the curve values for a specific dataset was employed (Figure 3A). When it came to classifying elderly patients, the EIM was superior to any subpopulations and outperformed the senescence-related gene signatures from CellAge7 (Figure 3B,C and Figure S5A–C). We utilized scRNA-seq data of OSCA and CRC tumours to explore the role of EIM in responses to neoadjuvant anti-PD1-based immunotherapy (Figure S6A,B). After cell type annotation with our scRNA-seq atlas, information including age groups and therapeutic outcomes was assigned to cells (Figure 3D and Figure S6C–E). Ro/e analysis of varied distributions showed that most EIM component subpopulations declined in elderly OSCA patients with complete pathologic response relative to those with incomplete pathologic response or major pathologic response (Figure 3E). Consistently, patients with incomplete pathologic response had increased EIM scores than those with major pathologic response or complete pathologic response (Figure 3F). Similar results were obtained in elderly CRC patients showing response or no response to neoadjuvant immunotherapy (Figure S7A–C). As to elderly CRC patients receiving chemotherapy, those with higher EIM scores had worse survivals (Figure 3G and Figure S7D). In summary, EIM is able to predict aging status and response to treatments in elderly patients with GI tract tumours. We sought to obtain a prognostic index based on EIM by integrating 10 machine learning algorithms to identify concise gene sets8 (Figure 4A). We found that the RSF+Ridge model achieved superior performance and stratified high-/low-risk patients across all cohorts (Figure 4A,B). Notably, our EIM-based model outperformed several classical prognostic signatures in GC and OSCA (Figure 4C). We used the 69 top feature genes from the EIM-based RSF+Ridge machine learning model to generate an EIM index (Table S3), which was applied to stratify patients into high and low groups. Interestingly, whereas younger patients in the TCGA cohort in general were scored low for the EIM index, a small proportion were ascribed to the high EIM index group and demonstrated poorer prognosis (Figure S7E). The same cut point stratified most older patients into the high EIM index group with worse survival (Figure S7E). However, for all patients together, the EIM index negatived predicted prognosis (Figure 4D). In fact, the EIM index relative to chronological age showed a closer correlation with prognosis (Figure 4E). We further evaluated the transcriptional features of tumours by principal component analysis using the top 3000 highly variable genes of each sample (Figure 4F). Surprisingly, the EIM index, rather than the chronological age, classified patients into distinctive groups in principal component analysis (Figure 4F). Furthermore, patients with higher levels of EIM index had higher expression of the known aging markers such as senescence-associated secretory phenotype (SASP), regardless of their chronological ages (Figure 4G). Therefore, the EIM index not only predicts prognosis and but also reflects aging status beyond chronological age. People just begin to utilize single-cell analysis to study the immune features associated with age.9 In this study, through scRNA-seq analysis of immune landscapes in GI tract tumours from older and younger patients, we discovered the EIM with enrichment in the elders and inverse correlation with patient survivals, and consequently raised the EIM index with potential prognostic values. Our findings suggest that alterations of the tumour immune microenvironment in aged patients may be quantified to foresee tumour development, although the causes of such shifts require further studies. F.L., J.C. and J.W. contributed equally to this work. W.Z., H.W., A.Z. and D. L. developed the working hypothesis and scientific concept. W.Z., F.L., J. C. and J.W. designed the experiments, analysed the data and prepared the manuscript. F.L. and J.C. performed the experiments and organized the data. D.L. and X.L. provided surgical specimens. All authors assisted the experiments, helped manuscript preparation and provided scientific input. This study was supported by the Research Funds of Centre for Leading Medicine and Advanced Technologies of IHM No. 2023IHM01042 (to W.Z.) and No. 2023IHM01043 (to H.W.). The authors declare no conflicts of interest. Written informed consent was obtained from all participants, and the related experiments were approved by the Medical Research Ethics Committee of The First Affiliated Hospital of the University of USTC with the permit no. 2023KY454. Ethics approval of use of the public sequencing data from human patients was exempted by the Medical Research Ethics Committee of The First Affiliated Hospital of the University of USTC. All the sequencing data are publicly available. Details are provided in Supporting Information. Codes used in this study are available upon request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.