NMI expression data in different types of normal tissues were obtained from the genotype-tissue expression (GTEx, http://commonfund.nih.gov/GTEx/) database. The expression data of NMI in tumor and normal tissues, and the clinical information of patients were obtained from The Cancer Genome Atlas (TCGA, http://cancergenome.nih.gov) and GTEx databases. NMI expression data in tumor cell lines were obtained from the Human Protein Atlas (HPA, https://www.proteinatlas.org/) database. The above data were analyzed by R software (version 3.6.3) and visualized by the ggplot2 R package. A p value <0.05 was set as the critical value and was considered statistically significant by the Wilcoxon test.

The correlation between the expression of NMI and the 6 immune types of tumors (C1 (wound healing); C2 (IFN-gamma dominant); C3 (inflammatory); C4 (lymphocyte depleted); C5 (immunologically quiet); C6 (TGF-β dominant) and their unique molecular types were obtained from the TISIDB (http://cis.hku.hk/TISIDB/) database.

The STRING database was used to analyze the top 50 proteins that strongly interact with NMI, and the parameters were set as follows: meaning of network edges: evidence, active interaction sources: Textmining, Experiments, Databases, minimum required interaction score: medium confidence (0.400). Subsequently, we performed Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses of the above proteins using the clusterprofiler package and visualized them using the ggplot2 package.

Receiver operating characteristic (ROC) curves were used to characterize the correlation between NMI expression and tumor diagnosis. Analysis was performed using the proc package and visualized using the ggplot2 package. An area under the curve (AUC) > 0.7 is regarded as having diagnostic accuracy, and an AUC >0.9 is regarded as having high diagnostic accuracy.

The value of NMI in the prognosis of tumor patients is characterized by its correlation with overall survival (OS), disease specific survival (DSS) and progression free interval (PFI). Correlation analysis of OS, DSS, PFI and clinical characteristics was performed in lower grade glioma (LGG) tumor types. Analysis was performed using the survival package and visualized using the survminer package. The median level of NMI expression was set as the cutoff value and statistical significance was tested by Cox regression analyses.

The correlation between NMI expression and the infiltration of 28 immune cell subtypes was analyzed at the pan-cancer level by searching the TISIDB database. Subsequently, the correlation between the abundance of T cells, B cells, DCs, macrophages, CD8⁺ T cells, NK cells, cytotoxic cells and Treg cells and NMI expression in LGG was analyzed by the R package (GSVA). The Wilcoxon rank sum test was used for significance analysis.

The correlation between NMI expression and 45 immunostimulatory molecules and 24 immunosuppressive molecules was analyzed by the TISIDB database.

We analyzed the relationship between the expression of NMI and the nomenclature of various types of antitumor drugs through the RNAactDrug database (http://bio-bigdata.hrbmu.edu.cn/RNAactDrug), and showed the top 20 drugs with the strongest correlation. Pearson and Spearman correlation methods were used to analyze the correlation.

Using the GTEx database, we analyzed the expression of NMI in normal tissues. We found that NMI was expressed in most tissues and organs, and it was the most highly expressed in the spleen and liver (Figure 1A). Subsequently, by analyzing the HPA database, we found that NMI was also expressed in a variety of tumor cell lines, especially in bone marrow-derived tumor cells (Figure 1B). Compared with adjacent normal tissues, NMI was significantly upregulated in 15 tumor types, including bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), GBM, head and neck squamous cell carcinoma (HNSC), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), stomach adenocarcinoma (STAD) and uterine corpus endometrial carcinoma (UCEC), while it was downregulated in kidney chromophobe (KICH), pheochromocytoma and paraganglioma (PCPG) and prostate adenocarcinoma (PRAD) (Figure 1C). Compared with the normal organization of GTEx database, NMI was significantly upregulated in 24 tumor types, including BLCA, BRCA, CESC, CHOL, COAD, ESCA, GBM, HNSC, KIRC, KIRP, acute myeloid leukemia (LAML), LGG, LIHC, LUAD, LUSC, ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), rectum adenocarcinoma (READ), skin cutaneous melanoma (SKCM), STAD, testicular germ cell tumors (TGCT), thyroid carcinoma (THCA), UCEC and uterine carcinosarcoma (UCS). However it was downregulated in KICH, PCPG and PRAD (Figure 1D).

In the past, the classification of tumors was mainly based on pathological classification. With the rapid development of genomics and immunotherapy, molecular typing and immunotyping are urgently needed to guide the treatment of patients. Therefore, we first analyzed the correlation between NMI expression and tumor molecular type at the pan-cancer level. We found that the expression of NMI was mainly related to 11 types of tumors, including BRCA (Figure 2A), COAD (Figure 2B), HNSC (Figure 2C), KIRP (Figure 2D), LGG (Figure 2E), LUSC (Figure 2F), OV (Figure 2G), PCPG (Figure 2H), PRAD (Figure 2I), STAD (Figure 2J) and UCEC (Figure 2K).

Next, we focused on analyzing the relationship between NMI expression and tumor immunotyping. We found that the expression of NMI was correlated with the immunotyping of 24 types of cancer. Among the 20 cancer types, high expression of NMI was mainly associated with C2 type, including BLCA, BRCA, CESC, CHOL, COAD, ESCA, HNSC, KIRC, KIRP, LUAD, LUSC, MESO, OV, PCPG, PRAD, SARC, SKCM, STAD, UCEC and UCS (Figure 3A). This suggests that the effect of NMI on tumorigenesis and progression may be mainly mediated by the expression of IFN-γ, which is also consistent with previous studies (Zhu et al., 1999). Interestingly, we observed that the highest expression of NMI was associated with C4 type (lymphocyte depleted) in PAAD and READ (Figure 3B), while the highest expression of NMI was associated with C6 type (TGF-β dominant) in LIHC and LGG (Figure 3C). This may be related to the small sample size of this subtype, but it also indicates that NMI may play different roles in different types of tumors.

NMI proteins have attracted the attention of researchers because of their extensive interactions with transcription factors. Therefore, to further explore the protein interaction network of NMI, we analyzed the STRING database and obtained 50 proteins that closely interacted with NMI (Figure 4A). Subsequently, we performed KEGG and GO analysis of these proteins and visualized them (Figure 4B). GO enrichment analyses showed that NMI was mainly associated with “Kinase regulator activity”, “protein kinase regulator activity”, “transferase complex”, “transferring phosphorus-containing groups”, “protein kinase complex” and “serine/threonine protein kinase complex” (Figure 4C). KEGG enrichment analysis showed that NMI was mainly associated with “JAK-STAT signaling pathway”, “Human T-cell leukemia virus 1 infection”, “Measles” and “Hepatitis” (Figure 4D). Currently, many studies have established that NMI interacts with all signal transducers and activators of transcription (STATs) except STAT2 and enhances STAT-mediated transcription in response to the cytokines IL-2 and IFN-γ (Wang et al., 2017).

To further clarify whether NMI can be used as a biomarker for tumor diagnosis, ROC curves were used to explore the value of NMI in tumor diagnosis. We found that NMI accurately (AUC>0.7) predicted 8 types (Supplementary Figures S1A–H) and had higher accuracy (AUC>0.9) in 11 cancer types including CESC (AUC = 0.961) (Figure 5A), CHOL (AUC = 0.981) (Figure 5B), GBM (AUC = 0.979) (Figure 5C), KICH (AUC = 0.974) (Figure 5D), LAML (AUC = 0.965) (Figure 5E), LGG (AUC = 0.905) (Figure 5F), OV (AUC = 0.955) (Figure 5G), PAAD (AUC = 0.973) (Figure 5H), READ (AUC = 0.900) (Figure 5I), STAD (AUC = 0.935) (Figure 5J) and TGCT (AUC = 0.908) (Figure 5K).

To determine whether NMI has predictive value for tumor prognosis, we analyzed the relationship between NMI expression and the OS, DSS and PFI of tumor patients. We found that NMI was highly correlated with three prognostic indicators in LGG, LUAD and SKCM. For LGG, high expression of NMI was associated with worse OS (hazard ratio (HR) = 3.30, p < 0.001), DSS (HR = 3.46, p < 0.001) and PFI (HR = 2.25, p < 0.001) (Figure 6A). For LUAD, high expression of NMI was associated with worse OS (HR = 1.50, p = 0.006), DSS (HR = 1.58, p = 0.015) and PFI (HR = 1.60, p < 0.001) (Figure 6B). However, in SKCM, high expression of NMI was associated with better OS (HR = 0.51, p < 0.001), DSS (HR = 0.46, p < 0.001) and PFI (HR = 0.71, p = 0.003) (Figure 6C). In addition, the high expression of NMI was associated with poor PFI in ACC (HR = 3.52, p < 0.001) (Supplementary Figure S2A) and GBM (HR = 1.44, p = 0.041) (Supplementary Figure S2B) and better PFI in COAD (HR = 0.68, p = 0.029) (Supplementary Figure S2C) and READ (HR = 0.50, p = 0.04) (Supplementary Figure S2D). The high expression of NMI was also associated with poor OS (HR = 1.57, p = 0.032) and DSS (HR = 1.64, p = 0.038) of PAAD (Supplementary Figure S2E), poor OS (HR = 1.47, p = 0.029) of LIHC (Supplementary Figure S2F) and poor OS (HR = 0.40, p = 0.005) of osteosarcoma (Supplementary Figure S2G).

Through the above analysis, we found that the expression of NMI was significantly correlated with the diagnosis and prognosis of LGG patients. Therefore, we further analyzed the correlation between NMI expression and the clinical characteristics of LGG patients. We found that the expression of NMI was closely related to tumor histological type, 1p/19q codeletion, WHO grade, primary therapy outcome and isocitrate dehydrogenase (IDH) status (Figure 7A) but not to age, race, sex or laterality (Supplementary Figure S3A). In addition, we further explored the impact of NMI expression in different clinical subtypes on OS in LGG patients. The results showed that higher NMI expression was associated with worse OS in 1p/19q codeletion = non–codel (HR = 2.55, p < 0.001) (Figure 7B), age >40 years (HR = 3.52, p < 0.001) (Figure 7C), WHO grade = G2 (HR = 2.36, p = 0.021) (Figure 7D), laterality = left (HR = 3.89, p < 0.001) (Figure 7E), sex = male (HR = 3.35, p < 0.001) (Figure 7F), IDH status = mutation (HR = 1.83, p = 0.01) (Figure 7G), primary therapy outcome = progressive disease (PD) (HR = 2.2, p = 0.001) (Figure 7H), histological type = astrocytoma (HR = 2.64, p = 0.001) (Figure 7I), and race = white (HR = 3.47, p < 0.001) (Figure 7J).

Immunotherapy is the fastest developing type of cancer treatment. Finding suitable biomarkers before and after treatment has become an urgent issue. A previous analysis suggested that NMI is closely related to the diagnosis and prognosis of tumor patients. We found that the expression of NMI was highly correlated with the abundance of various tumor infiltrating immune cells, especially in LGG, KICH, and UCS tumor types (Figure 8A). Next, we focused on analyzing the relationship between NMI expression and immune cell infiltration in LGG. The results showed that the high expression of NMI was accompanied by high infiltration of T cells, B cells and macrophages, but not DCs (Figure 8B). Interestingly, the expression of NMI was also associated with major innate and adaptive toxic cells (CD8⁺ T cells and NK cells) (Figure 8C). The high expression of NMI was accompanied by an overall increase in toxic cells, and the abundance of Treg cells decreased significantly.

Immune checkpoint blockade (ICB) therapy is a new model of immunotherapy, that has profoundly changed the prognosis of many types of tumors. Therefore, we further explored the relationship between the expression of NMI and the expression of immune checkpoint molecules. We found that the expression of NMI was related to the expression of 45 immunostimulatory molecules, especially the immunostimulatory molecules CD28, CD40, ICOS and the CD28 ligand CD80/86 (Figure 9A). In addition, we also found that the expression of NMI was related to 24 immunosuppressive molecules, especially PDCD-1, LAG-3, CTLA-4, TIGIT and IDO1 (Figure 9B). This suggests that the abnormal expression of NMI may be a biomarker for ICB treatment.

We explored the relationship between NMI mRNA expression and various types of antitumor drugs by searching the RNAactDrug database. We found that the top 20 drugs most closely related to NMI expression were Pcliainib, bosutinib, AICA Ribonucleotide, WHI-P97, idelalisib, BIX02189, TPCA-1, afatinib, SNX-2112, AT-7519, FR-180204, methotrexate, enzastaurin, OSU-03012, AZD7762, gemcitabine, AS605240, ZSTK474, CP466722 and VNLG/124 (Table 1).