Butyrophilin-like 9 expression is associated with outcome in lung adenocarcinoma

Expression profiles of BTNs in LUAD were determined using the GEPIA database [19]. Default settings were used with |Log2FC| > 1 and p-value Cutoff < 0.01 used as the cutoff criteria to determine differentially expressed genes. Log2(TPM + 1) was transformed for gene expression profile; jitter size was set at 0.4 for the plot. Notably, TCGA and GTEx datasets were included in the analysis. mRNA expression profile of BTNL9 in different tumor types was determined using the Oncomine database [17] using default settings with a P-value of 0.01, a fold change of 1.5, and a top 10% gene ranking.

Median gene expression was used as the cutoff point for survival analysis. Survival analysis of BTNL9 was performed using three online databases, including KM plotter [14], UALCAN [15], and OncoLnc databases [16]. TissGDB is a tissue-specific gene annotation database in cancer [18]. Forest plot were generated using Cox proportional hazard ratio (HR), and overall survival (OS) and relapse-free survival (RFS) of 28 cancer types were performed using 95% CI.

TIMER is a comprehensive resource for systematically analyzing immune infiltration in diverse cancer types based on the TCGA dataset [13, 20]. The Gene_DE module from TIMER was used to calculate BTNL9 mRNA expression in cross-carcinoma (*: P-value < 0.05; **: P-value < 0.01; ***: P-value < 0.001). The expression profile of BTNL9 in LUAD and its correlation with six immune infiltration cells, including B cells, CD4+ T cells, CD8+ T cells, macrophages, neutrophils, and DCs, were analyzed using Gene and Survival module in TIMER. Gene_Corr module was used to determine the correlation between BTNL9 expression and B and DC cells [21]. The immune score and stromal score of each TCGA tumor sample were estimated using Sangerbox (http://​sangerbox.​com/​Index).

miRMap [22], TargetScan [23], and miRWalk [24] were used to predict miRNAs that can bind to BTNL9. Predicted miRNAs obtained from the three databases were further verified using the starBase database [25].

Computational Analysis of Resistance (CARE) is a software that uses compound screening data to identify genome-scale biomarkers for targeted therapeutic strategies. Pearson correlation analysis between the gene expression profile of the cancer sample and the CARE scoring vector was used to group the patient as a responder or a non-responder [26].

We acquired TCGA LUAD RNA-seq data from the University of California, Santa Cruz (UCSC) Xena Browser (https://​xenabrowser.​net/​). After screening, samples with missing clinical data and 0 days overall survival time were excluded, and a total of 501 samples were included. Next, we randomly divided the TCGA-LUAD cohort (n = 501) in a 7 to 3 ratio into a training (n = 352) and testing dataset (n = 149). We performed the R package “rms” to construct a nomogram based on the TNM stage and expression profile of BTNL9 using the training dataset. To evaluate the usefulness of the nomogram, the R package “ROCsurvival” was used to construct ROC for the prediction of the 1-, 3- and 5- year OS. The R package “ggDCA” was executed to create a decision analysis curve to evaluate the clinical utility of the nomogram. Finally, R package “rms” was applied to perform a calibration curve to evaluate the precision for predicting 1-, 3- and 5-year OS prediction of the LUAD cohort.

The relationship between BTNL9 expression and single cancer cell biological behavior of LUAD was determined using Pearson correlation analysis and Spearman’s correlation analysis of the correlation between BTNL9 and tumor mutation burden (TMB). In all the studies, P < 0.05 was considered statistically significant.

Gene expression profile of BTNs, including BTN1A1, BTN2A1, BTN2A2, BTN2A3P, BTN3A1, BTN3A2, BTN3A3, BTNL2, BTNL3, BTNL8, BTNL9, BTNL10, and SKINTL was evaluated in normal and tumor lung tissues (Fig. 2A). Analysis showed that expression levels of BTNL8 and BTNL9 were significantly lower in tumor tissues compared with that of normal tissues (Fig. 2A). Furthermore, the expression level of BTNL9 was significantly negatively correlated with the clinical stage, lymph node metastasis stage, and p53 mutation. Concurrently, the expression level of BTNL8 was significantly negatively correlated with the clinical stage and N stage (Fig. 2B). However, survival analysis showed that BTNL8 was not significantly correlated with OS, whereas BTNL9 was significantly correlated with OS in LUAD (Fig. 2C). Validation of the prognosis value of BTNL9 in LUAD cohorts using PrognoScan [27] showed that BTNL9 expression was significantly correlated with RFS and OS in the GSE31210 dataset (n = 204). In addition, the expression level of BTNL9 was significantly associated with OS in GSE3141 cohort (n = 111) (Supplementary Table 1). Analysis using OncoLnc, UALCAN, and KM plotter showed that high expression of BTNL9 is significantly associated with better OS in LUAD (Fig. 2D). Although the two survival curves crossover occurred after 150 months (Fig. 2C), it is well beyond 5-years (60 months), and survival curves in the verification databases didn’t show crossover. Thus, we considered the results in this study reliable and stable. These findings imply that BTNL9 is a critical immune checkpoint of BTNs in LUAD.

Furthermore, the clinical distribution of BTNL9 was explored. TCGA-LUAD dataset was divided into the high and low groups based on the median gene expression level of BTNL9. The clinical characteristics, including age, gender, race, TNM staging, ECOG score, EGFR mutations, KRAS mutations, and radiotherapy information, were compared between the two groups. Analysis showed no significant difference in these clinical characteristics between the two groups (data not shown).

To further understand BTNL9 expression in pan-cancer, analysis of the dataset was performed using the Oncomine database. The findings showed that BTNL9 expression level was significantly lower in breast cancer, one colon cancer cohort, lung cancer, kidney cancer, and crabtree uterus cancer than normal tissues. However, BTNL9 expression was significantly higher in the brain and CNS cancer, colorectal cancer, esophageal cancer, leukemia, and lymphoma, than in normal tissue (Fig. 3A and Supplementary Table 2). Further, the BTNL9 expression profile was explored using TCGA RNA sequencing data (TIMER). The BTNL9 expression level was significantly downregulated in Bladder Urothelial Carcinoma (BLCA), Breast invasive carcinoma (BRCA), Cholangiocarcinoma (CHOL), Esophageal carcinoma (ESCA), Glioblastoma multiforme (GBM), Head and Neck squamous cell carcinoma (HNSC), Kidney renal papillary cell carcinoma (KIRP), LUAD, and LUSC compared with normal tissues. In contrast, BTNL9 was significantly increased in Colon adenocarcinoma (COAD), Kidney Chromophobe (KICH), and Kidney renal clear cell carcinoma (KIRC) compared with normal tissues (Fig. 3B). Analysis using the TissGDB database gave a correlation coefficient of BTNL9 expression with LUAD’s RFS HR of 0.87 [95% CI (0.79, 0.96)], and that for the correlation between BTNL9 expression and OS HR was 0.87 [95% CI (0.8,0.96)] (Fig. 3C, D).

A previous study reports that the expression level of BTNL9 in LUAD is significantly lower than that in normal tissues. DNA methylation is a biological process through which methyl groups are added to DNA molecules by Methyltransferases (DNMTs). DNA methylation of a gene promoter region functions by inhibiting gene transcription. Correlation analysis between BTNL9 expression and DNA methylation marker DNMTs (including DNMT1, DNMT2, DNMT3A, DNMT3B) was conducted using the GEPIA tool. Analysis showed that expression of BTNL9 in normal lung tissue was positively correlated with DNMTs (r = 0.35, P = 0.0059); however, there was no correlation with DNMTs in LUAD (r = − 0.019, P = 0.67) (Fig. 4A, B). These findings show that DNA methylation may be involved in the pathogenesis of LUAD. To further explore the upstream regulation mechanisms of the BTNL9 expression, miRNAs that bind to BTNL9 were predicted by using miRMap [22], TargetScan [23], and miRWalk [24] databases. A total of 248 miRNAs common predicted miRNAs from the three databases were obtained (Fig. 4C) and used starBase [25] to validate the predicted binding miRNAs. Analysis showed that, hsa-miR-30b-3p, hsa-miR-4709-3p and hsa-miR-6514-3p were significantly positively correlated with BTNL9 expression (r = 0.312, P = 5.25E-13, r = 0.277, P = 1.74E-10, and r = 0.103, P = 0.02, respectively, Fig. 4D). In addition, the three miRNAs were highly expressed and significantly correlated with higher OS of LUAD patients (HR = 0.66, P = 0.0058, HR = 0.63, P = 0.0023, and HR = 0.73, P = 0.036, respectively) (Fig. 4D). Although the two survival curves of hsa-miR-4709-3p and hsa-miR-6514-3p crossover occurred after 100 months, it is well beyond 5-years (60 months). Thus, we considered the results in this study reliable. Furthermore, 18 lncRNAs were predicted to bind to BTNL9 using LncMap [28] database (Supplementary Table 3). These findings were verified using LncACTdb2.0 [29] database. LncRNA AP001462.6 was predicted to bind to BTNL9, and the high expression level of AP001462.6 was significantly correlated with a high OS of LUAD patients (P = 0.049) (Fig. 4E).

Moreover, proteins implicated in binding BTNL9 were analyzed using the STRING [30] database. Analysis showed a total of 7 proteins that bind to BTNL9 (Fig. 4F). The top 2 binding proteins, including HTRA4 and TM4SF19, were predicted using the cytoHubba module of Cytoscape [31] (Fig. 4G). HTRA4 gene encodes a member of the HtrA protease family. HTRA4 plays a role as a secreted oligomeric chaperone protease to degrade misfolded secretory proteins [19]. We hypothesized that low expression of BTNL9 in LUAD might be related to degradation through ubiquitination. Analysis using Ubibrowser [32] database showed that E3 (MARCH8) ligases could bind the substrate BTNL9 (Supplementary Table 4). In addition, BTNL9 has a potential E3 recognizing domain and two potential E3 identifying motifs (Fig. 4H, I).

Gene Set Enrichment Analysis (GSEA) analysis for KEGG and HALLMARK was performed using the Sangerbox tool to explore the two groups’ biological pathways. The findings showed that the top 3 significantly enriched KEGG pathways in the high BTNL9 expression group were vascular smooth muscle contraction, phosphatidylinositol signaling system, and abc transporters (Fig. 5A). On the other hand, the top 4 significantly enriched KEGGs pathways in the low BTNL9 expression group were pathways implicated in Parkinson’s disease, oxidative phosphorylation, DNA replication, and proteasome pathways (Fig. 5B). GSEA for the HALLMARK pathway showed that the top 3 pathways associated with high BTNL9 expression were bile acid metabolism, heme metabolism, and Wnt/beta-catenin signaling pathways. Further, the top 4 pathways associated with low BTNL9 expression were E2F targets, glycolysis, myc targets v1, and mTORC1 signaling (Fig. 5C, D). These findings imply that BTNL9 is involved in LUAD metabolic reprogramming.

Metabolic reprogramming is a hallmark of cancer, and intrinsic and extrinsic factors contribute to various metabolic phenotypes in tumors. As cancer develops from pre-tumor lesions to local, clinically obvious malignant tumors to metastatic cancer, metabolism changes the phenotype and dependence [33]. Single-cell RNA (scRNA) analysis of LUAD using CancerSEA [34] database (Fig. 5E) showed that BTNL9 expression is significantly negatively correlated with tumor malignant features including invasion (r = − 0.53, P < 0.0001), metastasis (r = − 0.35, P = 0.011), EMT (r = − 0.47, P = 0.0006), proliferation (r = − 0.37, P = 0.0086), Hypoxia (r = − 0.36, P = 0.011), and DNA damage (r = − 0.34, P = 0.017) (Fig. 5F). This finding implies that low expression of BTNL9 is significantly associated with the malignant features of LUAD.

Spearman’s correlation analysis of the correlation between expression of BTNL9 and tumor mutation burden (TMB) in the TCGA LUAD cohort showed that BTNL9 is significantly negatively correlated with TMB (P = 1.4E-9) (Fig. 6A). Analysis of somatic mutation pattern of BTNL9 in LUAD using the SangerBox tool showed that the mutation frequency of BTNL9 in LUAD was 1.41% (Fig. 6B). Genetic mutations are implicated in the tumor microenvironment (TME) [35]; therefore, the relationship between the expression of BTNL9 and the immune score was determined using the ESTIMATE algorithm in the SangerBox tool. Analysis showed that BTNL9 was significantly positively correlated with ImmuneScore (r = 0.129, P = 0.003) and ESTIMATEScore (r = 0.106, P = 0.016). However, the expression of BTNL9 was not significantly correlated with StromalScore (Fig. 6C-E).

Analysis of the correlation between gene expression of BTNL9 and infiltrating level of immune cells using TIMER [13] database showed that BTNL9 was negatively correlated with tumor purity (r = − 0.124, P = 5.6E-03). On the other hand, gene expression of BTNL9 was significantly positively correlated with B cells (r = 0.24, P = 8.88E-8), CD4⁺T (r = 0.283, P = 2.24E-10) and macrophages (r = 0.209, P = 3.35E-6) (Fig. 6F). Moreover, survival analysis showed that high expression of BTNL9 in B cells (P = 0.000) and DC cells (P = 0.048) was correlated with better OS for LUAD (Fig. 6G).

A detailed analysis of TME infiltrated DC and B cells using the TIMER database showed that DC and its subtypes cDCs1 and cDCs2 [36] are associated with BTNL9 expression before and after purity adjustment. GEPIA database analysis showed that normal lung tissue was not correlated with DC and its subtypes cDCs1 and cDCs2. However, DC and its subtypes were significantly positively correlated with LUAD (Table 1), implying that DCs regulated by BTNL9 may participate in LUAD immune response. B cells are heterogeneous and include two subtypes: naïve B cells and plasma B cells [37]. TIMER analysis showed that total B cells and naïve B cells were significantly correlated with BTNL9 expression before and after purity adjustment. However, plasma B cells were not associated with BTNL9 expression before and after purity adjustment. GEPIA analysis showed that total B cells and naïve B cells were not correlated with BTNL9 expression in normal lung tissues; however, they were significantly positively correlated with BTNL9 expression in LUAD tissues. Plasma B cells showed no correlation with BTNL9 in both normal tissues and LUAD tissues (Table 1), indicating that BTNL9 may play a role in promoting naïve B cell antitumor immune response.

BTNL9 expression was significantly positively correlated with CARE scores for several compounds retrieved from Cancer Cell Line Encyclopedia (CCLE), Genomics of Drug Sensitivity in Cancer (GDSC, previously named CGP), and The Cancer Therapeutics Response Portal (CTRP) cohorts, mainly including antiangiogenic tyrosine kinase inhibitors Axitinib, Nilotinib, Sorafenib, Pazopanib, Masitinib, and Ki8751 (Fig. 7, and Table 2). These findings show that immune checkpoint inhibitors based on BTNL9 plus antiangiogenic tyrosine kinase inhibitors could be developed as a potential chemotherapy-free combination treatment strategy for LUAD.

We used the R package “survival V3.2–10” to construct a Cox model, including known important clinical variables for OS, such as TNM stage, primary therapy outcome, and BTNL9 expression. We also used multivariate analysis to explore whether BTNL9 expression was an independent OS factor for TCGA-LUAD patients. The results demonstrated that higher BTNL9 expression significantly (p = 0.049) and independently increased OS (HR = 0.67, 95% CI 0.45–0.99) (Table 3).

A nomogram predicting the 1-, 3- and 5- year OS for TCGA-LUAD was constructed based on BTNL9 expression and TNM stage (Fig. 8A). We built the ROC for the training dataset and the testing dataset and calculated the area under the ROC (AUC) to validate the accuracy of the nomogram. The AUCs for 1-, 3- and 5-year OS were 0.642, 0.645, and 0.607 in the training set (Fig. 8B); 0.727, 0.545, and 0.631 in the testing set (Fig. 8C). These results suggested that the nomogram showed a consistent accuracy in the training and testing dataset. We then conducted decision curve analysis (DCA) to evaluate the clinical usefulness, and the result showed that the nomogram provided an additional benefit compared to the “treat-all” and “treat-none” strategies in both the training and testing dataset (Fig. 8D-E). Finally, to compare the consistency of the model predictions with actual clinical outcomes, calibration curves for 1-year, 3-year, and 5-year OS were created for the training and testing dataset (supplementary Fig. 1A-F). The calibration curves showed consistent agreement between the predicted and observed values for 1-, 3- and 5-year OS.