Spatial transcriptomics to identify intra-tumor heterogeneity in CRC
In this study, we focused on two patients’ spatial transcriptomics data from a previously published CRC research (ST-P1, untreated, ST-P3, NACT with PR) [9]. Firstly, we used unsupervised clustering to cluster similar ST spots and divide into nine distinct clusters (Fig. 1A, B). Secondly, we annotated the clusters according to H&E sections and cell markers, then five morphological regions were identified (Fig. 1A, C, D). The results showed that unsupervised clustering analysis could effectively cluster ST spots with similar features into the same cluster, such as fibroblast and tumor regions (Fig. 1E). In addition, unsupervised clustering analysis could also subdivide tissues sections, which facilitates the discovery of tissue heterogeneity that is invisible to naked eye.
Because each spot of spatial transcriptomics contains more than one cell, its accuracy is lower than that of single-cell sequencing. Therefore, we scored well-defined genesets of 12 cell types [7, 18,19,20,21,22,23,24] by ssGSEA algorithm to recognize cell types that contained in each cluster. The annotation results had demonstrated the effectiveness of using ssGSEA in identifying cell populations of each cluster. As shown in Fig. 1F, fibroblast, enterocyte and smooth muscle were precisely annotated, and the lamina propria is composed of loose connective tissue that contains a variety of cells such as: dendritic cells, NK cells, etc. [25], while the tumor region was significantly enriched with monocytes, NK cells, and epithelial cells, indicating the existence of immune-inflammatory microenvironment in tumor region, which was consistent with previous studies [4, 5]. In all, spatial transcriptome analysis combined with ssGSEA can accurately determine the cell types contained in cell populations and offset the insufficiency of resolution in ST.
CAFs-enriched subgroups are identified in CRC
To analyze the heterogeneity of tumor region, principal component analysis (PCA) was used to cluster tumor region into 5 subclusters (Tumor-subcluster 0 ~ 4) (Fig. 2A). To find the cell types and functions of each cluster, we analyzed cluster-specific differential expression genes (DEGs) and found that subcluster 0 highly expressed fibroblast markers, such as: COL1A2, COL1A1, SPARC and COL3A1, etc. (Fig. 2B). Therefore, we considered subcluster 0 as CAFs-enriched cluster. Li et al. [19] previously identified two types of CAFs at single-cell sequencing of CRC, in which CAF-A expressed genes related to ECM remodeling, while CAF-B expressed cell markers of myo-fibroblasts, such as ACTA2 and TAGLN. Elyada et al. [6] identified two distinct types of CAFs from human pancreatic cancer, named mCAFs and iCAFs, an emerging study [7] also confirmed the presence of two different fibroblasts subtypes in bladder cancer. As seen in Fig. 2C, to confirm the existence of CAFs with specific signatures, we applied immunofluorescence analysis and we detected the expression of PDGFRA (iCAFs marker) and RGS5 (mCAFs marker) in colorectal cancer [7]. Thereafter, we utilized ssGSEA to further investigate the cell subtype of subcluster 0. And the result showed that subcluster 0 enriched with mCAFs (mCAFs-enriched cluster) (Fig. 2D, F).
To determine the functions of mCAFs-enriched cluster, we carried out KEGG pathway enrichment analysis. As shown in Fig. 2E, mCAFs-enriched cluster (Tumor-subcluster 0) was related to ECM-receptor interaction, focal adhesion, and proteoglycans in cancer et al., suggesting that mCAFs has the function of ECM remodeling in TME, which was consistent with previous studies [6, 7, 26]. It was proved again that ssGSEA was robust for spatial transcriptomics data.
In addition, we also conducted in-depth analysis on fibroblast region (Additional file 1: Fig. S1). The result showed that Fibro-subcluster 1 enriched iCAFs, named iCAFs-enriched cluster, and Fibro-subcluster 2 enriched mCAFs (Fig. 2G), because it highly expressed EPCAM, thus we identified it as mCAFs-enriched tumor cluster (Fig. 3D, E). Interestingly, when subclusters annotated by cell markers, we found that marker of CAFs, such as RGS5 and ACTA2 in mCAFs, PDGFRA and CXCL12 in iCAFs, were not specifically expressed in a certain subcluster (Fig. 2F, Additional file 1: Fig. S1D). Therefore, it further indicates that the accuracy of 10× spatial transcriptome data is limited, and spatial transcriptomics combined with multidimensional analysis (such as ssGSEA) can provide more detailed information. In detail, we found that anti-tumor immune cells such as T cells and dendritic cells were significantly enriched in mCAFs-enriched tumor cluster, iCAFs and macrophages were co-enriched in iCAFs-enriched cluster, however the anti-tumor immune cells, especially NK cells, reduced significantly in iCAFs-enriched cluster (Fig. 2G). It has been reported that CAFs can inhibit the immune system by releasing cytokines, chemokines and other compounds, thus leading to tumor metastasis [7, 27], indicating the important role of iCAFs in the immune microenvironment and its value as an anti-cancer drug target.
iCAFs promote oncogenesis by altering tumor microenvironment
We applied ssGSEA enrichment analysis on fibroblast region to reveal the functions of iCAFs in TME, and our results indicated that iCAFs were associated with epithelial mesenchymal transition (EMT), cholesterol homeostasis, bile acid metabolism, and fatty acid metabolism (Fig. 3A). To further explore the transitional relationships of EMT phenotype in distinct clusters (Fig. 3B, C), we examined the expression level of EMT markers in different clusters through pseudotime analysis (Fig. 3F). Strikingly, the results demonstrated an increased expression of EMT signatures in iCAFs-enrich cluster, while lack of expression in tumor site (Fibro-subcluster 2). These results suggest that CAFs, rather than tumor epithelial cells, might be the main culprit in promoting EMT and leading to tumor metastasis [19]. Metabolic reprogramming also plays an important role in tumor proliferation and metastasis [28]. In this study, iCAFs were supposed to be correlated with lipid metabolism through the bioinformatics analysis (Fig. 3A), suggesting that the changes of lipid metabolic activity of iCAFs in TME might be a potential mechanism in promoting tumor tumorigenesis.
Chemotherapy alters the tumor microenvironment
When we integrated two patients’ datasets, we found a patient-specific ST expression pattern (Fig. 4B), so we used CCA [10] to remove the batch effects on ST data. It is noteworthy that there was still obvious heterogeneity between colon1 and colon2 (Fig. 4A), therefore we supposed that chemotherapeutic drugs may be one reason for explaining the differential expression profiling in TME. In detail, we analyzed the cell compositions of two patients, results showed that the putative proportion of iCAFs relatively increased in all clusters (Fig. 4C left, Fig. 4D and Additional file 1: Figs. S2, S3) of colon2 (NACT with PR), while the proportions of some anti-tumor immune cells decreased, such as: NK cells, monocytes, etc. (Fig. 4C left, Fig. 4D and Additional file 1: Figs. S2, S3). In addition, we also examined the metabolism pattern changes caused by chemotherapeutic drugs, and we found that the metabolic activity significantly decreased in colon2 (Fig. 4E). However, when we focused on iCAFs-enriched cluster, we detected that fatty acid metabolic activity did not decrease in colon2 (Fig. 4F), then we supposed this phenomenon may be associated with iCAFs. Perhaps, it is the decrease of these tumor-killing cells and metabolism patterns changes caused by iCAFs that explain the internal mechanisms of drug resistance.
iCAFs is associated with clinical prognosis and immune infiltration
To correlate spatial transcriptomics data with public dataset, we evaluated the clinical significance of iCAFs in TCGA COADREAD cohort. Clinicopathological analysis showed that iCAFs were significantly associated with lymph node invasion (p < 0.05), the higher of iCAFs, the more possibility of lymph node invasion (Fig. 5A–D). In addition, univariate (Fig. 5E, F) and multivariate Cox proportional hazards regression (Fig. 5G) were used to analyze the association of the accumulation of iCAFs with prognosis. Multivariate analysis after adjusting confounding factors showed that the accumulation of iCAFs was an independent prognostic factor for OS (p = 0.035) (Fig. 5G). To explore the differential expression of iCAFs in distinct tissues, we specially detected the expression of iCAFs marker (PDGFRA) in tumor tissues and para-carcinoma tissues (n = 5) using immunohistochemical method, and our results demonstrated that the higher proportion of iCAFs in tumor tissues rather than para-carcinoma tissues (Fig. 5H). These results further indicated that iCAFs were associated with poor prognosis. When we focused on the relationship between iCAFs and immune inflammation, ESTIMATE algorithm was applied to calculate the immune score and stromal score for TCGA cohort, and the results showed that iCAFs was significantly correlated with immune score (r = 0.39, p < 0.05) (Fig. 5I, J), suggesting that iCAFs could interact with immunosuppressive cells [27, 29], thus inhibit the anti-tumor inflammatory response in TME.