Systematic pan-cancer landscape identifies CARM1 as a potential prognostic and immunological biomarker

Background Belonging to the protein arginine methyltransferase (PRMT) family, the enzyme encoded by coactivator associated arginine methyltransferase 1 (CARM1) catalyzes the methylation of protein arginine residues, especially acts on histones and other chromatin related proteins, which is essential in regulating gene expression. Beyond its well-established involvement in the regulation of transcription, recent studies have revealed a novel role of CARM1 in tumorigenesis and development, but there is still a lack of systematic understanding of CARM1 in human cancers. An integrated analysis of CARM1 in pan-cancer may contribute to further explore its prognostic value and potential immunological function in tumor therapy. Results Based on systematic analysis of data in multiple databases, we firstly verified that CARM1 is highly expressed in most tumors compared with corresponding normal tissues, and is bound up with poor prognosis in some tumors. Subsequently, relevance between CARM1 expression level and tumor immune microenvironment is analyzed from the perspectives of tumor mutation burden, microsatellite instability, mismatch repair genes, methyltransferases genes, immune checkpoint genes and immune cells infiltration, indicating a potential relationship between CARM1 expression and tumor microenvironment. A gene enrichment analysis followed shortly, which implied that the role of CARM1 in tumor pathogenesis may be related to transcriptional imbalance and viral carcinogenesis. Conclusions Our first comprehensive bioinformatics analysis provides a broad molecular perspective on the role of CARM1 in various tumors, highlights its value in clinical prognosis and potential association with tumor immune microenvironment, which may furnish an immune based antitumor strategy to provide a reference for more accurate and personalized immunotherapy in the future. Supplementary Information The online version contains supplementary material available at 10.1186/s12863-021-01022-w.

chromatin-associated proteins [1]. With regard to human CARM1 protein, it is composed of an N-terminal pleckstrin homology-like domain (PH-like), a C-terminal transactivase domain, and a central catalytic domain containing the four conserved PRMT motifs (Fig. 1b). The N-and C-terminal domains of CARM1 are vital for substrate recognition and transcription-mediated activation [2], and the motifs in central catalytic domain are essential for binding of the cofactor S-adenosyl methionine (SAM) and the substrate arginine [3]. Long known as a transcriptional coactivator, recent studies have shown that it is also involved in the regulation of metabolism [4][5][6], autophagy [7], RNA regulation [8] and early mammalian development [9]. Recently, accumulating evidence has suggested that CARM1 also has an impact on the occurrence and development of tumors [10][11][12][13][14].
Existing studies on exploring the mechanisms of CARM1 methylation affecting tumor progression have shown that CARM1 is a coactivator of several cancer-related transcription factors and can be involved in promoting tumor cell proliferation and metastasis by methylating cancerrelated transcription factors, including NF-κB, p53, steroid receptors and so on, and its high expression is associated with poor prognosis of tumors [15]. For example, in the most studied breast cancer, CARM1 could methylate the R838 site of lysine demethylase 1 (LSD1) to promote the binding of deubiquitinase USP7, resulting in the ubiquitination and stabilization of LSD1, thereby promoting the invasion and metastasis of breast cancer cells [16]. In addition, CARM1 has been found to be involved in regulating metabolic pathways in tumors. Metabolic reprogramming is a hallmark of cancer. In breast cancer Fig. 1 Basic information about CARM1. a Genomic location of human CARM1. b Protein structure diagram of human CARM1. c According to the reported studies, the carcinogenic pathways CARM1 was involved in across different cancers are shown graphically. The related references are also indicated cells, methylation of the key glycolytic enzyme pyruvate kinase M2 isoform (PKM2) by CARM1 shifts the metabolic balance from oxidative phosphorylation to aerobic glycolysis, producing a large amount of ATP, so as to promote tumor cell proliferation and migration [4]. Nevertheless, CARM1 is up-regulated when glucose starvation, followed by methylation of GAPDH to inhibiting glycolysis, thereby suppressing tumor cell proliferation in liver cancer cells [6], which is to some extent consistent with the results of the correlation analysis between CARM1 expression and liver cancer prognosis described in our work. Current evidence about effects of CARM1 on various cancers has been shown in Fig. 1c and Table 4 [5,6,13,[16][17][18][19][20][21][22][23][24][25].
However, researches on CARM1 in cancers are only started in recent years, and limited to several kinds of tumors. There is still no systematic pan-cancer evidence about the relationship between CARM1 and multiple tumor types based on big clinical data. Our work, for the first time, used multiple databases containing The Cancer Genome Atlas (TCGA) project, cBioPortal, Human Protein Altas (HPA) and so on to conduct a comprehensive pan-cancer analysis of CARM1. A group of factors, such as gene expression, survival status, genetic alteration, tumor mutation burden (TMB), microsatellite instability (MSI), methyltransferases genes, immune infiltration, and relevant cellular pathway, are included to investigate the potential associations between CARM1 and the pathogenesis and clinical prognosis of different cancers, providing a basis for further understanding the role of CARM1 in tumor immunotherapy.

Expression profile of CARM1 across normal tissues and cancer samples
In this work, we aimed to investigate the role of human CARM1 in tumorigenesis and development. As mentioned above, CARM1 protein is usually composed of a N-terminal pH like domain (cl17171), a C-terminal transactivase domain, and a central catalytic domain (cl17173), and its structure is conserved among most species (Fig. S1a, see Additional file 1, e.g., H. sapiens, M. mulatta, R. norvegicus, etc.). The evolutionary relationship of CARM1 protein among different species is also shown in the phylogenetic tree data (Fig. S1b).
Firstly, the physiologic CARM1 gene expression levels across normal tissues were observed combining HPA, GTEx and Function annotation of the mammalian genome 5 (FANTOM5) datasets. As shown in Fig. 2a, CARM1 expression is the highest in skeletal muscle with high RNA tissue specificity, while other detected tissues express relatively low level of CARM1, especially the blood cell lineage. When analyzing the expression of CARM1 in different blood cells, a low RNA blood cell specificity could be observed (Fig. S2b, see Additional file 2). The CARM1 expression levels in various cancer cell lines were also analyzed. The result shows that all cancers expressed CARM1, with the highest expression level in ovarian cancer, followed by endometrial cancer and colorectal cancer (Fig. S2a).
In addition, HPA, TCGA and CPTAC datasets were used to evaluate CARM1 expression at protein level. We obtained the immunohistochemistry (IHC) results from HPA and compared them with the CARM1 gene expression data provided by TCGA. As shown in Fig. 3a, the analysis results from the two databases are basically consistent. The staining results of BRCA, LUAD, LUSC present strong or medium CARM1 staining, while the corresponding normal tissues show low or moderate staining. On the contrary, normal kidney tissues have low or moderate staining, while KICH samples have no CARM1 staining. Furthermore, the results of the CPTAC dataset indicate higher expression of CARM1 protein in the primary tissues of KIRC and colon cancer than in normal tissues (Fig. 3b), and increase from grade I to grade II in KIRC patients (Fig. S2c). It is noteworthy that although there are no significant correlations of protein expression between primary tissues of breast cancer, ovarian cancer, lung adenocarcinoma, UCEC and related normal tissues, the expression of protein in normal and other subtypes of breast cancer is significantly higher than that of luminal subtype. In addition, the CARM1 protein expression The expression of CARM1 in tumors and normal tissues from TCGA project were compared by TIMER2. *P < 0.05, **P < 0.01, ***P < 0.001. c For the types of ACC, DLBC, LGG, OV, SARC, TGCT, THYM and UCS in TCGA project, the corresponding normal tissues in GTEx database were used as controls. * P < 0.05 Fig. 3 Analysis of CARM1 total protein expression data. a Comparison of CARM1 gene expression data from TCGA (left) with IHC results of HPA (right). The CARM1 RNA expression is up-regulated in BRCA, LUAD, LUSC and down-regulated in KICH, which is consistent with the results of IHC. b Data from CPTAC dataset indicate KIRC and colon cancer samples express higher level of CARM1 total protein than normal tissues. ***P < 0.001 level of 21-40 years old group in ovarian cancer patients is up-regulated compared with other age groups, which may be a potential feature of this group.
Using the "Pathological Stage Plot" module of GEPIA2, we also found that the expression of CARM1 is related to the pathological stages of the following carcinomas, comprising ACC (adrenocortical carcinoma), ESCA, KICH and UCS (uterine carcinosarcoma) (Fig. 4a, all P < 0.05), but no significant difference is observed in other tumors (Fig. S3, see Additional file 3).

Prognostic value of CARM1 in pan-cancers
To explore the correlation between CARM1 expression and prognosis of patients with different tumors, TCGA and GEO were used and cancer cases were divided into high-expression and low-expression groups according to the expression levels of CARM1. As shown in  (Fig. 4c).
Furthermore, Kaplan-Meier Plotter tool was also used to identify the prognostic value of CARM1 in the five types of tumors shown in Fig. S4 (see Additional file 8), which manifest a correlation between high expression CARM1 and poor OS, PPS, FP prognosis for gastric cancer and lung cancer. As for ovarian cancer, low CARM1 expression is related to poor PFS, while the relationship between CARM1 expression and OS, PPS prognosis are not detected. Additionally, CARM1 is a high-risk gene in breast cancer (OS, P = 0.019; DMFS, P = 0.00033; PPS, P = 0.0038), while it is a low-risk We also performed a subgroup survival analysis using selected clinical factors and observed different conclusions. Significantly, highly expressed CARM1 is linked to poor prognosis for estrogen receptor (ER) positive subgroup of breast cancer cases. As for patients in grade II or lymph node negative status, CARM1 overexpression may be a poor prognostic factor (Table 1). For gastric cancer patients with lymph node metastasis, highly expressed CARM1 is associated with poor OS, FP and PPS prognosis (Table S1, see Additional file 4). Notably, consistent with the overall analysis results of liver cancer cases aforementioned, CARM1 overexpression is a beneficial prognostic factor in most subgroup analyses, especially in patients with hepatitis virus infection or in low grade, and it may turn into a deleterious prognostic factor when the disease developed into high grade (Table S2, see   Additional file 5). More details about prognosis of these five tumors can be found in Table 1, Table S1-S4 (see Additional files 4, 5, 6, 7).

Genetic alteration analysis of CARM1 across cancers
Gene alteration features of different cancers in TCGA were further investigated using the cBioPortal tool. Among all cancers, ovarian cancers present the highest alteration frequency of CARM1 (> 8%) with "amplification" as the primary type (Fig. 5a). It is worth noting that all cases of uterine carcinosarcoma (~ 7%), adrenocortical carcinoma (~ 3.5%) and mesothelioma (> 2%) with gene variation have "amplification" mutation type, while all cases of DLBCL (> 4%) have copy number deletion of CARM1. Figure 5b further shows the types, loci and number of cases of CARM1 gene variations. As the main type of gene change, there are 65 missenses in CARM1, among which 419 sites in the methyltransferase domain Table 1 Subgroup analysis on the correlation of CARM1 expression and prognosis of breast cancer cases ER Estrogen receptor, PR Progesterone receptor, HER2 human epidermal growth factor receptor 2, NA not available data, HR hazard ratio The P value marked bold indicates that the prognosis of the low expression group is better than that of the high expression group, while the P value marked bold and italic indicates that the prognosis of the high expression group is better than that of the low expression group. NS, P > 0.05; * P < 0.05; ** P < 0.01; *** P < 0.001 had the maximum R (arginine) mutations, translation from R to W (tryptophan), L (leucine) and Q (glutamine), respectively. Moreover, the potential association between CARM1 genetic alterations and clinical prognosis was analyzed in various cancers cases (Table 2). Compared with the cases with CARM1 change, Cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC) cases without altered CARM1 show better prognosis in PFS (P = 0.0239) and DFS (P = 2.892e-5), but not OS (P = 0.883) and DSS (P = 0.870). These results have been shown in Fig. 5c. In addition, the prognosis of the group without genetic changes is significantly better than that of the relative group in COAD. Serving as emerging prognostic and immunotherapeutic response biomarkers for a variety of tumors, the quantification of TMB and MSI have entered the exploratory stage of clinical transformation [26,27]. Herein, we analyzed the correlation between CARM1 mRNA expression and TMB, MSI. As shown in Fig. 6a, there is a positive correlation between CARM1 expression and TMB for LUAD (P = 0.00031), pancreatic adenocarcinoma (PAAD, P = 0.049), SARC STAD (P = 0.017) and UVM (P = 0.014), but is negatively correlated with that of SKCM (P = 0.0038), HNSC (P = 0.0022), READ (P = 1.3e-07) and DLBC (P = 0.0044). It should be noted that both TMB and MSI of LUAD, SARC and STAD are positively correlated with CARM1 expression, which deserves further study.

CARM1 expression is related to DNA repair genes and methyltransferase expression in various tumor samples
Correlation between mutation indexes TMB, MSI and CARM expression prompted us to further explore the potential relationship between CARM1 expression and tumorigenesis mechanism. MMRs, the intracellular mismatch repair mechanisms, the loss of function of its key genes will lead to unrepairable DNA replication errors, and then result in somatic mutations [28]. Here, utilizing TCGA expression profile data, we evaluated the relationship between CARM1 expression and mutation of five major MMRs genes including MLH1, MSH2, MSH6, PMS2, EPCAM. Except COAD, AML (acute myeloid leukemia), READ, UCS and UVM, the expression of CARM1 is positively correlated with MMRs genes mutation in almost all types of tumors from TCGA, and the results of MLH1, MSH2, MSH6 and PMS2 are more significant (Fig. 7a).
As a form of DNA chemical modification involving the transfer of methyl group onto the C5 position of the  6 Correlation between CARM1 mRNA expression and TMB, MSI. Based on TCGA dataset, the potential correlation between CARM1 expression and TMB (a), MSI (b) is explored. TMB was calculated according to total mutation rate per million base pairs in each cancer, and MSI was counted by total incidence of deletion or insertion in repeating sequences per million base pairs. The partial correlation values are marked. Spearman correlation test, P < 0.05 is considered significant. Red font represents positive correlation and blue font represents negative correlation Fig. 7 Correlation analysis of CARM1 expression with MMR genes mutation and methyltransferases genes expression. a Correlation between CARM1 mRNA expression and five major MMR genes mutation. The lower triangle in each block represents the coefficient calculated by Pearson's correlation test, and the upper triangle represents log 10 transformed P-value. * P < 0.05, ** P < 0.01, *** P < 0.001. b Co-expression analysis of CARM1 and methyltransferases. The outer circle represents different tumor types, the second circle represents the four methyltransferases (DNMT1: red, DNMT2: blue, Dnmt3a: green, DNMT3b: Purple), the third circle represents co-expression correlation coefficient, and the fourth circle represents P value. P < 0.05 is considered significant cytosine to form 5-methylcytosine under the action of DNA methyltransferase, DNA methylation can change genetic performance without changing DNA sequence, which can change chromatin structure, DNA conformation, DNA stability and the interaction between DNA and protein, so as to regulate gene expression [29]. Here, analysis of the correlation between CARM1 and four methyltransferases (DNMT1, DNMT2, DNMT3a, DNMT3b) expression was conducted for each tumor to explore whether CARM1 expression is related to epigenetics. The result shows that CARM1 and methyltransferases are significantly co-expressed in almost all tumors (Fig. 7b), which is an interesting phenomenon worth further exploring.

Correlation between TME and CARM1 expression Correlation between tumor-infiltrating immune cells and CARM1 expression
Tumor microenvironment (TME) is an environment conducive to tumor cell survival established by tumor cells that evade early immune surveillance by remodeling local immune cells and stromal cells, and a full understanding of TME provides us with valuable clues to develop more effective therapeutic strategies. Currently, there is still a lack of research on the association of CARM1 methylation and immune cell infiltration. In an impressive study, CARM1 inactivation was found to activate innate immunity in melanoma resistant cell lines with high CARM1 expression, making them more sensitive to T cell immunity and immune checkpoint blockade [30]. Moreover, it is eye-catching that CARM1-KO T cells exert more effective antitumor effects than wild-type, indicating that CARM1 inhibition enables immunotherapy of resistant tumors by dual effects on tumor cells and T cells, which has great clinical translational value. And the analysis results presented below in this study also manifest there is a large correlation between immune cells infiltration and CARM1 expression in the TME of some tumors, pointing out that investigating the role of CARM1 in the field of tumor immunity in-depth is a direction worth exploring. A growing number of researches show that tumor-infiltrating immune cells serve as a vital part in TME, affecting the occurrence and development of tumors significantly [31,32]. It is of great significance to further explore the pan cancerous relationship between these immune cells and CARM1 expression. As a significant part of TME, cancer-associated fibroblasts in tumor stroma have been reported to be involved in regulating the function of a variety of tumor infiltrating immune cells [33]. Therefore, several algorithms were applied to investigate the relationship between CARM1 expression and cancer-associated fibroblasts infiltration. As shown in Fig. 8a, in most TCGA tumors, the expression of CARM1 is positively correlated with the infiltration of cancer-related fibroblasts, especially in ACC, KIRC, MESO, THYM and UVM. The representative scatterplots produced using one algorithm are also presented in the Figure. Because tumor infiltrating lymphocytes are independent predictors of sentinel lymph node status and survival in cancer [32], analogous analysis on potential correlation between CD8 + T cells infiltration and CARM1 expression is also conducted. The results show that CD8 + T cells infiltration is negatively correlated with CARM1 expression in ESCA, HNSC, LUSC, PAAD, SKCM and THYM cases (Fig. 8b). Similar trend could be found between CARM1 expression and other immune cells infiltration in ESCA, LUSC, SKCM and THYM in Table 3, which presents more comprehensive details. On the contrary, a positive correlation trend could be observed in BLCA, BRCA, KIRC, KIRP, LGG, LIHC, LUAD, PAAD, PCPG, PRAD and THCA.

Correlation between TME scores and CARM1 expression
In order to conduct more in-depth research on the pancancer relationship between TME and CARM1 expression, ESTIMATE algorithm was used to analyze the relationship between stromal and immune scores and gene expression level among 33 tumors from TCGA. The results of the top three tumors with the highest correlation coefficient have been shown in Fig. 9, which reveal that CARM1 expression is significantly negatively correlated with immune scores in LUSC, SARC, testicular germ cell tumors (TGCT), indicating that the content of immune cells decreases while the level of CARM1 expression escalates (Fig. 9a). Similar results were also observed in the stromal score of LUSC and SARC, while the opposite correlation was shown in KIRC (Fig. 9b).
Genetic instability of tumor cells often leads to a large number of mutations, and the expression of nonsynonymous mutations could produce tumor specific antigens called tumor neoantigens [34]. Because they are not expressed in normal tissues, neoantigens have high immunogenicity and can activate T cells to trigger immune response, which have become a potential new target of tumor immunotherapy. Here, we counted the

Correlation between immune checkpoints and CARM1 expression
Tumor cells induce immunosuppression through various ways to achieve immune escape, and tumor immunotherapy is a treatment method to control and eliminate tumors by restarting and maintaining tumor immune cycle and restoring normal anti-tumor immune response, in which immune checkpoint inhibitor is an important aspect [35]. Herein, we also conducted a correlation analysis between CARM1 and checkpoint genes expression and found that CARM1 expression is highly correlated with CD276 in various cancer types (Fig. S5, see Additional file 9). Additionally, CARM1 expression has a certain correlation with the expression of multiple immune checkpoints in KICH, KIRC, KIRP, LGG, LIHC and THCA. In contrast, the expression of CARM1 is negatively correlated with most immune checkpoint molecules in SKCM and TGCT.

CARM1-associated genes enrichment analysis
In order to further study the biological significance of CARM1 gene in tumorigenesis, we screened CARM1 binding protein and expression related genes, and carried  out a series of pathway enrichment analysis. STRING tool was applied to obtain the top 50 CARM1 binding proteins, which have been shown in the form of interaction network in Fig. 10a. Then, we used the GEPIA2 to acquire the first 100 genes related to CARM1 expression, and the top 6 genes with the highest correlation are shown in the form of scatter diagram in Fig. 10b.
Corresponding heatmap data also indicate that CARM1 is positively correlated with the above 6 genes in most cancer types (Fig. 10c). Combined with the above two databases, we conducted KEGG enrichment analysis. The results show that the effect of CARM1 on tumor pathogenesis may be related to transcriptional misregulation and viral carcinogenesis (Fig. 10d).

Discussion
CARM1 has been well known as a transcriptional coactivator, it has also been found to be crucial in regulation of metabolism, autophagy, RNA regulation and early mammalian development. In addition, increasing evidences indicate that CARM1 exerts an impact on the occurrence and development of tumors. After literature search, we found there is still a lack of research reports on pan-cancer analysis of CARM1. Therefore, based on the data from TCGA, CPTAC and other databases, we comprehensively detected the potential significance of CARM1 expression in various cancers from the perspectives of gene expression, gene alteration, immune microenvironment and related signaling pathways. Overexpression of CARM1 in most tumors was first verified, which is also associated with the pathological stage of some tumors, such as ACC, ESCA, KICH and UCS. Survival analysis results from GEPIA2 indicate that high expression of CARM1 is a significant adverse prognostic factor in ACC, BLCA, LGG, MESO, SKCM and other tumors.
Overall, according to the analysis conducted by Kaplan Meier plotter, among the five tumors provided by the website, the low expression CARM1 group has a better clinical prognosis. However, subgroup analysis shows that the prognosis value of the CARM1 expression level vary in some subgroups. For example, the effect of CARM1 expression on prognosis in breast cancer is related to the state of HER2. In the HER2 − group, the prognosis of CARM1 low expression group is better Fig. 10 CARM1-associated genes enrichment analysis. a Based on STRING tool, the top 50 CARM1-binding proteins and the interaction network were obtained. b Using GEPIA2 method, the top 100 CARM1 related genes in TCGA project were gained, and the expression correlation between CARM1 and the top 6 target genes was analyzed. c Corresponding heatmap data was obtained utilizing TIMER2.0 online tool and further identifies CARM1 is positively correlated with the above 6 genes in most cancer types. d Combined with the binding and related proteins, KEGG enrichment analysis was conducted, showing clearly that the role of CARM1 in tumor pathogenesis may be related to transcriptional misregulation and viral carcinogenesis than that of high expression group, while the prognosis of HER2 + subgroup is the opposite, which has been confirmed in previous studies [36]. In patients with hepatocellular carcinoma, high expression of CARM1 has a better prognosis in patients with hepatitis virus infection, while low expression of CARM1 is associated with favorable clinical prognosis of OS, FP and PPS, especially in high stage, specific TNM grading and pathological classification. Additionally, the analysis data obtained from cBioPortal tool shows that the mutation of CARM1 gene is related to the poor prognosis of colon cancer, which has been proved to be highly expressed CARM1 protein by CPTAC analysis tool. Therefore, on the premise of fully considering subgroup factors, CARM1 expression level is expected to be a good prognostic index. This study also analyzed a series of immune related factors of CARM1. Based on the results of immune cells infiltration, immune and matrix score as well as co-expression analysis of MMRs, methyltransferases genes and immune checkpoint genes, we found that CARM1 potentially affects the tumor immune microenvironment in most tumors, especially ACC, LUAD, LUSC, STAD, HNSC, THYM, etc. It is a direction worth exploring to clarify how CARM1 affects tumor immunity.
Our study conducts a comprehensive analysis of CARM1 in pan-cancer, which could provide clues for detecting its prognostic value and potential immunological function in tumor therapy. Information on various indicators suggesting the potential significance of CARM1 in different tumors has been summarized in Table 4, where the overall impact and conclusions about CARM1 on a certain tumor can be quickly found. Nevertheless, there are still some limitations in the present study. Although the correlation analysis between the gene expression of CARM1 and immune related factors implies the relevancy between them, it is not enough to capture detailed interaction. The concrete mechanisms of CARM1 affecting tumor immune microenvironment still needs further experimental verification.

Conclusions
In the light of big data analysis based on multiple databases, we revealed that the expression level and mutation degree of CARM1 are significantly related with clinical prognosis of patients with various tumors, indicating that CARM1 is expected to become an effective prognostic index. In the process of exploring mechanisms of CARM1 involved in tumor progression, we correlated probable causes from the perspective of several immune related elements and signaling pathways, which would be conducive to provide clues to support further molecular mechanism exploration, and may furnish an immune based antitumor strategy to provide a reference for more accurate and personalized immunotherapy in the future.

Acquisition of gene information and protein structure
The detailed genomic location information of CARM1 gene was obtained by querying UCSC website (http:// genome. ucsc. edu/). Then, the protein structure diagrams of CARM1 containing conserved regions in different species were gained and analyzed by using the Homologene module in NCBI website (https:// www. ncbi. nlm. nih. gov/ homol ogene/). Additionally, the phylogenetic tree of CARM1 in different species was also acquired by using the COBALT online tool of NCBI (https:// www. ncbi. nlm. nih. gov/ tools/ cobalt/).

Gene expression analysis at mRNA level
The expression data in different tissues, blood cells and tumor cell lines under physiological conditions were obtained by inputting "CARM1" in HPA database (https:// www. prote inatl as. org/ human prote ome/ patho logy). Using the "Gene_DE" module of TIMER2 (tumor immune estimation resource, version 2) website (http:// timer. cistr ome. org/), the expression difference of CARM1 between tumors and corresponding normal tissues in TCGA project was observed. As for tumors without normal control, "Expression analysis-Box Plots" module of GEPIA (Gene Expression Profiling Interactive Analysis, version 2) was applied to acquire expression data from GTEx (Genotype-Tissue expression) database, under the settings of "P-value cutoff = 0.01, log2FC (fold change) cutoff =1" and "Match TCGA normal and GTEx data" (http:// gepia2. cancer-pku. cn/) [37]. We also utilize the "pathological staging map" module of GEPIA2 to obtain the violin plot of CARM1 expression in different pathological stages (stage I, II, III and IV) of all TCGA tumors.

Gene expression analysis at protein level
To evaluate the CARM1 expression difference from the protein expression level, immunohistochemistry images of tumor tissues and corresponding normal tissues were downloaded from HPA and analyzed. Further, we obtained CPTAC (clinical proteome tumor analysis alliance) dataset from the UALCAN portal (http:// ualcan. path. uab. edu/ analy sis-prot. html) and conducted protein expression analysis of various tumors.

Survival prognosis analysis
Utilizing the "Survival Map" and "Survival Analysis" module of GEPIA 2, the OS (overall survival) and DFS  [13], [16], [17], [18]  Favorable in some subgroups [6], [22] Pro-proliferative; Antiproliferative  NS not significance, P > 0.05, NA not available, PC positive correlation, NC negative correlation Superscript # indicates that more detailed subgroup analysis data on this index are provided in the supplementary material. In most tumors, CARM1 expression is correlated with MMRs and methyltransferases gene expression, which can be seen in Fig. 7. Considering that there are too many content about the correlation between CARM1 expression and immune cell infiltration, detailed information can be found in Table 3 and will not be summarized here. NS, P > 0.05; * P < 0.05; ** P < 0.01; *** P < 0.001 (disease-free survival) significance map data and survival map of CARM1 in all TCGA tumors can be obtained respectively. Cutoff-high (50%) and cutoff-low (50%) values were used as the expression thresholds to split the high-expression and low-expression cohorts. The association between CARM1 expression and survival in pan-cancer was also verified by Kaplan-Meier Plotter (https:// kmplot. com/ analy sis/), which pools the different GEO datasets for a series of analyses of OS, DMFS (distant metastasis-free survival), RFS (relapse-free survival), PPS (post-progression survival), FP (first progression), DSS (disease-specific survival), and PFS (progress-free survival). The five types of tumor cases were split by setting "autoselect best cutoff " and the hazard ratio (HR), log-rank P-value and 95% confidence intervals were computed, as well as the Kaplan-Meier survival plots were generated. We also set up other grouping factors to obtain subgroup analysis data of CARM1 mRNA expression and prognosis.

Gene alteration and survival analysis
Analysis of gene alteration of CARM1 in pan-cancer was conducted by querying the Cbioportal tool (http:// www. cbiop ortal. org/) [38]. The "TCGA Pan Cancer Atlas Studies" module was selected to get the genetic alteration characteristics of CARM1. The "comparison" module was also used to obtain survival prognosis data of cancer cases from TCGA (with or without CARM1 gene alteration), and Kaplan Meier plots with log-rank P-value were generated.

Tumor immune microenvironment analysis
To explore the correlation between gene expression and immunotherapeutic response biomarkers TMB and MSI, sangerbox tool was used with the query of "CARM1" (http:// sange rbox. com/ Tool). TMB was calculated as the total mutation incidences per million base pair, and MSI was counted by the number of insertion or deletion events that occurred in repeating sequences of genes. Spearman method was used and the P-value as well as partial correlation value were obtained. We also conducted a co-expression analysis between CARM1 and mismatch repair genes (MMRs), methyltransferases as well as acknowledged immune checkpoints genes respectively. The images were modified using the software Adobe Illustrator.
To predict the presence of infiltration stromal or immune cells in pan-cancer tissues, R-package "estimate" and "limma" were used to calculate the scores of immune and stromal cells. As a database derived-web tool for immune cell infiltration calculation, TIMER provides the infiltration scores of many common types of immune cells [39]. Herein, we downloaded the infiltration data from it and used to test the correlation with CARM1 expression. The TIMER, EPIC, MCPCOUNTER, XCELL, CIBERSORT, CIBERSORT-ABS and QUANTISEQ algorithms were applied for immune cells infiltration calculation.

CARM1-associated genes enrichment analysis
On the STRING website (https:// string-db. org/), we firstly input the gene name and the organism "Homo sapiens", then set the parameters as Network type [full network], meaning of network edges [evidence], active interaction sources [experiments], minimum required interaction score [medium confidence (0.400)] and max number of interactors to show [("no more than 50 interactors" in 1st shell] and finally obtained the top 50 binding proteins network of CARM1. According to the data sets of all TCGA tumors and normal tissues, the "Similar Gene Detection" module of GEPIA2 was used to obtain the top 100 CARM1 related genes. The "Correlation Analysis" module of GEPIA2 was also used to perform a pairwise gene Pearson correlation analysis between CARM1 and the selected genes. The log2 TPM was applied for the dot plot and the P-value as well as the correlation coefficient (R) were indicated. After that, the "Gene_Corr" module of TIMER2 was applied to generate a heatmap containing the partial correlation Fcorand P-value in the purity-adjusted Spearman's rank correlation.
Combining the two sets of data, KEGG (Kyoto Encyclopedia of genes and genomes) pathway analysis was carried out by uploading the gene lists to DAVID (https:// david. ncifc rf. gov/), Database for annotation, visualization, and integrated discovery). The analysis results are visualized with "ggplot2" R software package.