2.1 Clinical specimens and data processing

The studies involving human participants and the experiments including any relevant details were admitted by the Ethics Committee of the Fourth Hospital of Hebei Medical University (No. 2019MEC100). All methods were confirmed to the Code of Ethics of the Declaration of Helsinki. The written informed consent to participate was obtained from participants. In total, 29 pairs of LUAD and matched adjacent normal lung tissues obtained from the Fourth Hospital of Hebei Medical University (Shijiazhuang, China) during December 2021 to July 2022 were frozen in liquid nitrogen until further analysis. The clinicopathological information was listed in Table S1.

First, publicly available RNA sequencing data (FPKM values) of LUAD were obtained from TCGA (https://portal.gdc.cancer.gov/projects/TCGA-LUAD); level 3 gene expression profile data of 594 LUAD cases (59 normal samples and 535 LUAD samples) and corresponding clinical data were included. Second, we merged the data through Perl (mRNA_merge.pl). To normalize the FPKM values, the gene expression data were transformed to log2 (value + 1). To further validate the results, the related data were extracted from the GEO database (https://www.ncbi.nlm.nih.gov/geo/): GSE10072 (33 LUAD and 33 paired adjacent nontumor samples), GSE19188 (45 LUAD and 65 healthy lung samples), GSE33532 (36 LUAD and paired adjacent nontumor samples), GSE50081 (128 LUAD samples), and GSE37745 (106 LUAD samples).

2.2 Immune and stromal score generation

The immune and stromal scores of LUAD were acquired from ESTIMATE (https://bioinformatics.mdanderson.org/estimate/), and the samples were grouped into low- and high-score groups referring to the stromal and immune median scores.

2.3 Survival analysis

The relationship between the stromal and immune scores and overall survival (OS) rate was revealed by Kaplan–Meier survival analysis using the “survival” package in R software.

2.4 Score comparison among different clinical statuses

Wilcoxon rank sum or Kruskal–Wallis rank sum tests were used depending on the count of clinical status obtained from TCGA for the comparison.

2.5 TMERG identification

The LUAD samples were grouped into high score or low score according to the median of immune score and stromal score, respectively. The package “limma” in R software was utilized to conduct the difference analysis of the between gene expression, and TMERGs were excavated by the comparison between the high-score and the low-score samples.¹⁵ A false discovery rate (FDR) < 0.05 and |log fold change (logFC)| > 2 were defined as the selection criteria. The “VennDiagram” package was utilized to find common DEGs.

2.6 Functional analysis

Gene Ontology (GO; http://www.geneontology.org/) functional and Kyoto Encyclopedia of Genes and Genomes (KEGG; http://www.kegg.jp/kegg/pathway.html) pathway analyses were performed to understand the functions of the common up- and downregulated DEGs in LUAD through the “clusterProfiler” R package.^16-19

2.7 Protein–protein interaction (PPI) network and TME-related prognostic hub gene identification

To obtain the interactions between the common up- and downregulated DEGs, a PPI network was constructed through STRING (https://string-db.org/)²⁰ and rebuilt through Cytoscape software (version 3.7.2). The minimum required interaction score was set to 0.950. The Molecular Complex Detection plugin of Cytoscape was applied to screen out the significant modules from the PPI network (degree cutoff = 2, node score cutoff = 0.2, k-core = 2, and maximum depth = 100). Furthermore, genes with node degrees ≥10 were defined as hub genes. The “survival” package in R software was used to conduct a univariate Cox regression, and the top 30 TME-related prognostic DEGs were determined by the P value from small to large. The hazard ratios (HRs) and 95% confidence intervals (CIs) were generated by a Cox analysis. Venn diagrams were generated to find the intersection between PPI hub genes and TME-related prognostic DEGs.

2.8 TME-related prognostic hub gene analysis

The differential gene expression between the LUAD and control samples was evaluated by a Wilcoxon test. Furthermore, the difference in paired gene expression data between the LUAD and normal samples was analyzed by a Wilcoxon test. The association between gene expression and the clinical status of LUAD was analyzed by a Tukey honestly significant difference (HSD) test, Dunn's test, or Wilcoxon rank sum test according to the clinical status. The association between prognostic hub gene expression and OS was calculated through the packages “survminer” and “survival” in R software based on TCGA and GEO data. The survival rate of LUAD patients with different clinical features were generated using the “survminer” and “survival” packages in R software. Univariate and multivariate Cox analyses were used to determine the potential prognostic factors.

2.9 Diagnostic value of HLA-DMA

The time ROC analysis was used to compare the predictive accuracy of HLA-DMA. Multivariate and univariate cox were performed to develop the nomogram. A nomogram was constructed based on the results of multivariate Cox proportional hazards analysis to predict the 1-, 3-, and 5-year overall recurrence through “rms” R package.

2.10 Survival-associated HLA-DMA-related gene screening and functional analysis

A Spearman's correlation analysis was used to determine the HLA-DMA-related genes by R software. Then, the top 25 positively and 25 negatively HLA-DMA-related genes were ordered by the Spearman's correlation coefficient, and a heatmap of the top 50 HLA-DMA-related genes was generated by the package “ggplot2” in R. The “VennDiagram” package was utilized to screen the intersection genes of the top 500 HLA-DMA-related genes ordered by the Spearman's correlation coefficient and top 500 survival-associated genes ordered by HR. Next, the intersection genes were subjected to GO and KEGG pathway analyses through the “clusterProfiler” R package. A PPI network of the intersection genes was constructed through STRING and rebuilt by the “igraph” R package. An intersection gene co-expression matrix was produced by the “ggplot2” R package. The relationships between HLA-DMA and the intersecting genes were calculated through a Spearman's correlation test.

2.11 HLA-DMA-related TIICs

Single-sample GSEA (ssGSEA) was performed by using the GSVA package to obtain the fraction of TIICs. Then, all samples were divided into high- and low-expression groups according to the median expression of HLA-DMA. A Mann–Whitney U test was conducted to compare the difference in TIIC infiltration between the high- and low-HLA-DMA expression groups. The relationship between HLA-DMA expression and TIIC infiltration was calculated by a Spearman's correlation analysis.

2.12 Significantly mutated genes and immunotherapy response based on HLA-DMA expression

The TMB was calculated by tmb function of “maftools” package in R. SIGLEC15, TIGIT, CD274, HAVCR2, PDCD1, CTLA4, LAG3, and PDCD1LG2 were selected to be immune checkpoint-related transcripts, and the expression values of the eight genes were got from the TCGA-LUAD data. The statistical difference of the eight genes in HLA-DMA high/low and normal groups was compared through the Kruskal–Wallis test. Potential ICB response was forecast by Tumor Immune Dysfunction and Exclusion (TIDE) algorithm. The data of mutations were downloaded and visualized using the “maftools” package. Genes with higher mutational frequency detected of LUAD patient in histogram were shown.

2.13 Cell culture

The A549 and H1975 human LUAD cell lines and BEAS-2B human normal lung fibroblast cell lines used in this study were purchased from the Cell Bank of Type Culture Collection of the Chinese Academy of Science (Shanghai Institute of Cell Biology, China) and authenticated by STR typing. The A549 were cultured in Ham's F-12K medium with 10% fetal bovine serum (FBS, Gibco, Thermo Fisher Scientific, Waltham, MA). H1975 were cultured in RPMI 1640 medium (Gibco, Life Technologies, California, USA) with 10% fetal bovine serum. The BEAS-2B were cultured in were cultured in RPMI 160 medium (Invitrogen, Shanghai, China) supplemented with 10% FBS. The cells were placed in a CO₂ incubator (SANYO Electric Co., Ltd., Japan) with constant 90% humidity and 5% CO₂.

2.14 Cell transfection

The full-length HLA-DMA cDNA sequence was synthesized and cloned into pcDNA3.1 plasmid vector (NewHelix, China). The pcDNA3.1-HLA-DMA vector or pcDNA3.1 plasmid vector was transfected into A549 and H1975 cell lines with Lipo8000™ transfection Reagent (Beyotime, China), and the cells that stably expressed HLA-DMA were identified by a Western blot analysis.

2.15 Cell proliferation assay

Cell proliferation was examined with a Cell Counting Kit-8 (CCK-8, Beyotime, China) assay according to the manufacturer's protocols. A549 and H1975 cells were transfected with the corresponding vector, seeded into 96-well plates and cultured for 24, 48, and 72 h. Next, CCK-8 reagent (approximately 10 μL) was added to each well, and the cells were incubated for an additional 1 or 2 h. This experiment was performed in triplicate.

2.16 Cell cycle analysis

Cells were collected and fixed in 70% ethanol at 4°C overnight. Then, the cells were treated with RNase A (50 μg/mL, Fuyuanbio, China) and stained with DAPI Staining Solution (DAPI, 10 μg/mL, Beyotime, China) for 30 min at 37°C after washing with PBS. The distribution of the cell cycle phases was calculated by a FACSCalibur flow cytometer (Becton, Dickinson and Company, CA). The phase ratio (%) was regarded as the percentage of cells in the G1/S/G2 phase. Each experiment was conducted in triplicate.

2.17 Real-time quantitative reverse transcription PCR (RT-qPCR)

The total RNA was extracted from 15 paired NSCLC tissues using TRIzol reagent (Invitrogen, Carlsbad, California, USA). The total RNA was then transcribed to cDNA using Servicebio RT First Strand cDNA Synthesis Kit (G3330; Wuhan, China) according to the manufacturer's instructions. The reverse transcription procedure was: 5 min at 25 and 42°C for 30 min and then 5 s at 85°C. Real-time quantitative reverse transcription PCR (RT-qPCR) was performed on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad, China). The RT-qPCR system contained 7.5 μL 2 × SYBR Green qPCR Master Mix, 0.75 μL forward primer, 0.75 μL reverse primer, 2 μL cDNA product, and 4.0 μL water nuclease-free. The reaction protocol was 95°C for 30 s, followed by 40 cycles at 95°C for 15 s and 60°C for 30 s. The primers were synthesized by Servicebio (Wuhan, China). The 2^−ΔCq method was used to demonstrate the expression levels of HLA-DMA. All experiments were conducted in triplicate. The primer sequences used for RT-qPCR were GAPDH: 5′-GGAAGCTTGTCATCAATGGAAATC-3′,5′-TGATGACCCTTTTGGCTCCC-3′; HLA-DMA: 5′-TGTCCAGAGGGTTTCCTATCG-3′,5′-CTGCCAGTTCACTGTCAGCAT-3′.

2.18 Western blotting

Five LUAD samples with paired adjacent normal lung tissue samples and LUAD cells were lysed with a protease inhibitor, phosphatase inhibitor, and RIPA cleavage buffer (Thermo Fisher Scientific, Shanghai, China). The protein lysates were separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) (Servicebio, Wuhan, China) and imprinted on polyvinylidene fluoride (PVDF) membranes (Millipore, Billerica, MA, USA) for analysis. The samples were incubated with anti-HLA-DMA (1:1000 dilution, #PA5-22365; Thermo Fisher, Shanghai, China) and anti-GAPDH (1:2000 dilution, GB15002; Servicebio, Wuhan, China) antibodies overnight at 4°C. Horseradish peroxidase (HRP)-conjugated secondary antibodies (1:5000 dilution; GB25301, Servicebio, Wuhan, China) were added for 2 h at room temperature. Protein bands were visualized using ECL Plus Western Blotting Substrate under chemiluminescence system (Tanon 5200). The analysis of Western blot result was conducted by Image J software.

2.19 Immunofluorescence

Nine paraffin-embedded LUAD and paracarcinoma tissues were used for immunofluorescence (IF). Each tissue was incubated overnight with a primary antibody against HLA-DMA (1:100 dilution, #PA5-22365; Thermo Fisher, Shanghai, China) at 4°C. After washing with PBS, coverslips were treated with Cy3 conjugated Goat Anti-rabbit IgG (1:300 dilution, GB21303, Servicebio, Wuhan, China) and DAPI staining. The images were photographed under a Nikon Eclipse C1 Ortho-Fluorescent Microscopy (Nikon Instruments Inc., Japan). The results were analyzed and scored by three pathologists who were blinded to the classification of the samples. ImageJ software was used to analyze the intensity of staining using a semiquantitative integration method.

2.20 Statistical analysis

All data were exhibited as mean ± standard deviation (SD). The results of HLA-DMA in LUAD patients and paracarcinoma tissues in Western blot analysis were detected by one-way ANOVA to test difference between LUAD and LUAD groups or different cell lines. The difference of HLA-DMA expression level in IF between NSCLC and paired normal tissue was determined by paired-samples Test. The difference of HLA-DMA mRNA expression level between NSCLC and paired normal tissue was determined by Wilcoxon signed rank test. The correlation between HLA-DMA and S phase-related genes was conducted by Spearman correlation analysis. Statistical significance was considered when P < 0.05.

3.1 Stromal and immune scores of the study population and correlation with clinical features

Workflow of the study is presented in Figure S1. The stromal and immune scores were ranged from −1959.31 to 2098.77 and −1355.85 to 2905.3. The immune scores of the LUAD patients were positively correlated with the OS rate (log-rank P = 0.01 in Figure S2A). However, the correlation between the stromal scores and OS rate showed no significant differences at the level of P < 0.05 (Figure S2B).

The immune score was negatively correlated with the T and M classification (P = 0.005 in Figure S2C and P = 0.035 in Figure S2E), but there were no significant differences between the immune score and N classification (P = 0.433 in Figure S2D). Higher immune scores were markedly correlated with earlier stages (P = 0.034 in Figure S2F). The correlation between the stromal scores and T, N or stages classification indicated no significant differences at the level of P < 0.05 in Figure S2G,H,J, respectively. The stromal score only showed a negative correlation with the M classification (P = 0.014 in Figure S2I). Overall, the fraction of TME components were connected with the metastasis and tumor size of LUAD.

3.2 TME-related DEG screening and functional enrichment analysis

According to the median of stromal score, the samples were divided into high/low-score group, 2192 upregulated genes and 2680 downregulated genes were found (Figure S3A). Similarly, 1589 upregulated genes and 161 downregulated genes of immune score were obtained (Figure S3B). The Venn diagrams indicate that 620 DEGs were commonly upregulated, and 35 DEGs were commonly downregulated (Figure S3C,D).

The KEGG functional analysis indicated that these DEGs were involved in cytokine–cytokine receptor interactions (Figure S3E). The GO functional analysis verified that these DEGs were enriched in the immune system process and immune response, indicating that the DEGs were closely associated with the LUAD immune response (Figure S3F–H).

3.3 HLA-DMA as a TME-related prognostic hub gene and gene expression validation

The PPI network consisted of 472 nodes and 416 edges. Then, the network was constructed by Cytoscape, and the top 3 significant modules are displayed in Figure S4A. Among the 472 nodes, 21 central node genes were regarded as hub genes based on the criterion of a filtering degree ≥10 (Figure S4B). The top 30 factors ranked by the P value were showed in Figure S4C. Finally, an intersection analysis between the 21 central node genes and the top 30 factors was performed, and the common gene HLA-DMA was ultimately selected as the prognostic hub gene (Figure S4D).

HLA-DMA in the normal samples was obviously higher than that in the LUAD group (log-rank P = 2.6e-08; Figure S5A). In addition, paired HLA-DMA expression in LUAD was lower than that in the normal controls (P = 5.8e-07; Figure S5B). Furthermore, HLA-DMA expression was significantly related to the T stage (P = 1.3e-04; Figure S5C), initial treatment outcome (P = 0.01, Figure S5D), OS event (P = 2e-03, Figure S5E), age (P = 4e-04, Figure S5F) and smoking (P = 4.3e-06; Figure S5G). Data in GSE10072, GSE19188, and GSE33532 proved that HLA-DMA level in LUAD was higher in normal control (P < 0.05; Figure S5H–J). The comparison between LUAD and control demonstrated that HLA-DMA expression was lower in LUAD (P < 0.05; Figure S5K–M).

The HLA-DMA expression in tumors was obviously lower than that in matched adjacent normal lung tissues in Western blot results (P < 0.001; Figure S6A). The qRT-PCR indicated lower HLA-DMA expression in LUAD (P = 6.1e-05; Figure S6B). In addition, the IF also exhibited that the HLA-DMA was downregulated in LUAD (P < 0.001; S6C,D). Besides, the expression of HLA-DMA was obviously higher in BEAS-2B cells compared with that in A549 and H1975 cells (P < 0.001; Figure S6E). The above findings verified that the HLA-DMA was decreased in LUAD.

3.4 Higher HLA-DMA expression indicates longer OS in LUAD

As shown in Figure S7A, higher HLA-DMA expression was related to a longer OS in the LUAD patients (HR = 0.61, P = 0.001). Then, the subgroup analysis shown in Figure S7B–E indicated that a higher expression of HLA-DMA was significantly related to a better prognosis in patients with the following clinical statuses: age over 65 years (P = 0.019), T2 (P < 0.001), Stage I (P = 0.007) and smoking (P = 0.005). And patients in higher HLA-DMA group had a longer OS (P < 0.05, Figure S7F–G) in the validation cohort. The univariate Cox analysis showed the high HLA-DMA level was obviously correlated with a longer OS (HR = 0.599, 95% CI = 0.446–0.804) (Table 1). These outcomes suggest that higher HLA-DMA patients had a longer OS.

3.5 Diagnostic value of HLA-DMA

The area under the 1-year ROC curve (AUC) was 0.657, as shown in Figure S8A. Cox regression demonstrated that HLA-DMA was an independent risk factor (Figure S8B,C). A nomogram containing HLA-DMA expression and clinical features was constructed to predict the 1-, 3-, and 5-year survival probability (Figure S8D). The calibration analysis showed that the curves at 1-, 3- and 5-year were close to the ideal line and verified the accuracy of the nomogram (Figure S8E).

3.6 Functional analysis of survival-associated HLA-DMA-related genes

A heatmap was constructed to show the top 50 genes positively (red) and negatively (blue) related to HLA-DMA (Figure 1A). A Venn diagram screened 32 survival-associated HLA-DMA-related genes (Figure 1B), and functional analyses of these 32 genes revealed that cell cycle signaling and spindle and chromosome segregation were enriched (Figure 1C). In addition, further analysis of the cell cycle-related functions revealed that 16 HLA-DMA-related genes were enriched in positive regulation of cell cycle process (Figure 1D). Figure 1E further demonstrated the eight HLA-DMA-related genes were involved in G1/S phase and G2/M phase transition, and Figure 1F showed that six HLA-DMA-related genes were enriched in the cell cycle checkpoint. The PPI network showed a stronger enrichment network that was closely related to cell cycle (Figure 1G). The gene coexpression correlation analysis demonstrated that the proteins had a significant positive correlation with each other (Figure 1H). The relationship between HLA-DMA and cell cycle-related genes was calculated by a Spearman's correlation test. The results showed that 12 genes among the cell cycle-related genes were negatively related to HLA-DMA expression (r < −0.4, P < 0.001; Figure 2A–L), indicating that HLA-DMA might participate in the regulation of the cell cycle.

3.7 HLA-DMA may inhibit NSCLC cell proliferation by inducing S phase arrest

The expression of HLA-DMA was increased after the transfection with the pcDNA3.1-HLA-DMA compared with empty pcDNA3.1 vector or control in H1975 and A549 cells (P < 0.001; Figure 3A). The increased expression of HLA-DMA led to an obvious reduction in the viability of H1975 and A549 cells at 48 and 72 h time point according to the CCK-8 assay (P < 0.05; Figure 3B,C).

Flow cytometry was used to examine the changes in cell cycle induced in H1975 and A549 cells. The percentage of cells in S phase decreased from 47.5 to 33 in empty vector transfected A549 cells and pcDNA3.1-HLA-DMA vector transfected A549 cells. In the meanwhile, the percentage of cells in S phase decreased from 41.9 to 32.7 in empty vector transfected H1975 cells and pcDNA3.1-HLA-DMA vector transfected H1975 cells. The upregulation of HLA-DMA markedly decreasing the ratio of cells in the S phase cells (P < 0.001; Figure 3D), resulting in cell cycle arrest compared with the control. The specific functions of S phase-related markers were provided in Table S2. Furthermore, the correlation between HLA-DMA and S phase-related markers was shown in Figure 3E, higher HLA-DMA expression accompanied lower level of CCNC/CCND1/RBL1/CCNE2/CDK2/E2F1(P < 0.05; Figure 3E). These results indicated that HLA-DMA may inhibit cell proliferation by arresting cell cycle at S phase.

3.8 HLA-DMA-related TIICs and ICB response prediction

A Wilcoxon rank sum test was performed to compare the immune infiltration difference between the two groups (high vs. low), and the results are shown in Figure 4A. Compared with the HLA-DMA high group, the infiltration levels of aDCs, B cells, CD8 T cells, cytotoxic cells, DCs, eosinophils, iDCs, macrophages, mast cells, neutrophils, NK cells, pDCs, T cells, T helper cells, TFH cells, Th1 cells, Th17 cells, and Tregs were obviously lower in the low group (P < 0.01). In contrast, Th2 cell and Tgd infiltration were higher in the low group (P < 0.05). The Spearman's correlation analysis revealed a correlation between HLA-DMA expression and the immune infiltration level (Figure 4B–P), and Th2 cells were negatively associated with the HLA-DMA expression levels (r < −0.3, P < 0.001). In contrast, 14 types of TIICs were positively correlated with the HLA-DMA expression levels, including aDCs, B cells, CD8 T cells, cytotoxic cells, DCs, iDCs, macrophages, mast cells, neutrophils, pDCs, T cells, TFH cells, Th1 cells, and Tregs (r > 0.3, P < 0.001).

The high HLA-DMA expression patients were accompanied with higher TMB (Figure 5A; P < 0.001). In Figure 5B, high HLA-DMA group with lower TIDE score, indicating better ICB response and longer survival time. The HLA-DMA high group accompanied with higher immune checkpoint-relevant genes expression (Figure 5C, P < 0.001), which suggested that high HLA-DMA group may have better immunotherapy response.

3.9 Genomic alterations in high- and low- HLA-DMA patients

The top 15 mutated genes in the high- and low-HLA-DMA expression (Figure 6A) patients were compared. The mutation frequencies of genes, including TTN(45%) and TP53(47%), were higher in the patients with low HLA-DMA expression. Missense mutations were the top common type of mutation, single nucleotide polymorphism (SNP) was the fore, and the primary type of mutation was T > G (Figure 6B). In Figure 6C, the amount of mutations per sample was presented, and each kind of mutation was represented by a color in the box diagram.