Leveraging mitochondrial dynamics-related risk signatures to predict the prognosis and tumor microenvironment of lung adenocarcinoma Authors: Li Xu, Danting Zheng, Jisong Zhang, Huihui Hu, Liangliang Dong, Enguo Chen Published: 2026 ABSTRACT AbstractBackgroundLung adenocarcinoma (LUAD) is a highly heterogeneous disease, bringing daunting challenges in prognosis prediction. Alterations in mitochondrial dynamics (MD) are crucial in tumor generation and progression. Therefore, this study is the first to build a prognostic model based on mitochondrial dynamics-related genes (MDRGs) to predict microenvironment and potential drugs for LUAD.MethodsLUAD transcriptomic data were sourced from the Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO), respectively. First, differentially expressed genes (DEGs) were screened between normal and LUAD samples in the TCGA cohort. The intersection of DEGs and MDRGs yielded differentially expressed MDRGs (DE-MDRGs). Then, a prognostic model was constructed through univariate Cox regression, LASSO Cox regression, and multivariate Cox regression analysis. A nomogram was graphed using clinical factors and the MD-related risk score (MDRS). The predictive performance of the model and the nomogram was evaluated using ROC curves. Finally, analyses on tumor microenvironment (TME), mutation, and predicted in vitro drug response were undertaken.ResultsA prognostic model based on 8 MDRGs (SLC52A3, HMGA2, CPS1, GLS2, CYP27A1, CFTR, STAR, and DRP2) was established, demonstrating relatively accurate predictive ability. The low-MDRS group had higher levels of immune cell infiltration, such as aDCs, B cells, neutrophils, and tumor-infiltrating lymphocytes. We also discovered that the tumor mutation burden in the high-MDRS group was considerably higher than that in the low-MDRS group. Additionally, the low-MDRS group was more sensitive to AZD8055, ZM447439, ERK-6604, SB505124, Tozasertib, and BMS-754807, while the high-MDRS group was more sensitive to BI-2536 and Venetoclax.ConclusionThis work set up a prognostic model for LUAD based on 8 MDRGs, pinpointing promising biomarkers and targets for LUAD treatment.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-026-05038-5.Keywords: Lung adenocarcinoma, Mitochondrial dynamics, Prognosis, Risk signature, Tumor microenvironment Introduction Globally and in China, lung cancer (LC) is the most frequently diagnosed and deadliest cancer among all malignancies, with lung adenocarcinoma (LUAD) causing roughly 50% of LC cases as the most prevalent pathological type [1]. The management of LUAD necessitates stage-adapted therapeutic strategies: early-staged disease is optimally treated with surgical resection, whereas advanced or metastatic LUAD requires multimodal systemic therapies including surgical resection combined with radiotherapy, chemotherapy, immunotherapy, and targeted therapy [2–4]. These treatment modalities may ameliorate disease-related symptoms and extend overall survival (OS) in affected patients. However, LUAD patients frequently develop therapeutic resistance, characterized by chemoresistance and evasion of apoptotic pathways, resulting in disease recurrence and adverse prognosis [5, 6]. Contemporary data indicate the 5-year OS rate in LUAD cohorts persists at < 20% [7, 8]. Consequently, the development of predictive models and novel biomarkers is imperative for prognosis prediction and therapeutic efficacy assessment. Mitochondrial dynamics (MD) refers to the perpetual processes of fission and fusion within the mitochondrial network. This dynamic equilibrium is essential for maintaining cellular homeostasis, particularly in energy metabolism and other critical physiological processes, including cellular motility, differentiation, cell cycle progression, senescence, and apoptosis [9, 10]. Cancer is a major mitochondrial metabolic disease, with mitochondrial dysfunction manifested as changes in structure, function, and dynamics, which are believed to be related to tumorigenesis [11]. In LUAD, the dysregulation of MD is closely linked to tumor initiation and progression. Alterations in MD modulate cellular energy metabolism, programmed cell death, and tumor microenvironment (TME), consequently impacting prognosis and therapeutic responsiveness in LUAD [12]. Additionally, the protein optic atrophy 1 (OPA1), linked with mitochondrial fusion, is proven to be a key player in tumor angiogenesis [13] and to maintain mitochondrial homeostasis in LUAD while potentially facilitating tumor immune escape from CD8+ T cell-mediated cytotoxicity via multiple mechanisms [14]. While MD is critically involved in LUAD tumorigenesis and development, the prognostic value of mitochondrial dynamics-related genes (MDRGs) in LUAD remains to be systematically investigated. This study established a novel 8-gene prognostic model related to MD in the TCGA-LUAD cohort and verified it in GSE72094 and GSE26939. Furthermore, the TME infiltration, tumor mutations, and predicted in vitro drug response were analyzed. In summary, our work identified and validated the MDRG characteristics in LUAD and screened out potential prognostic biomarkers. Materials and methods Data acquisition The LUAD dataset, containing 59 normal and 541 tumor samples, was obtained from the Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/) database. Table S1 lists detailed clinical characteristics of the TCGA-LUAD cohort. The GSE72094 dataset (including 398 LUAD samples) and the GSE26939 dataset (115 LUAD samples) were retrieved from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/). Table S2 and Table S3 list the relevant clinical information of GSE72094 and GSE26939, respectively. Furthermore, we conducted a comparative analysis of shared clinical information from the TCGA and GEO datasets (Table S4 and Table S5). Using the “tableone” R package [15], we constructed baseline characteristic tables where categorical variables were presented as frequency percentages and continuous variables as mean values ± standard deviations. To evaluate differences between groups, we performed chi-square tests on categorical variables and t-tests on continuous variables. Subsequently, we utilized the “gtsummary” R package [16] to summarize the variables and calculate corresponding P-values. MDRGs, including genes related to mitochondrial fusion and mitochondrial fission, were sourced from the GeneCard website (https://www.genecards.org/). Genes with a relevance score > 10 were retained, resulting in a total of 1123 genes (Table S6). Difference analysis and building and validation of the prognostic model The R package “edgeR” [17] was applied in the differential analysis on normal and LUAD samples in the TCGA cohort (FDR < 0.05, |log2(FC)|> 1), yielding differentially expressed genes (DEGs). First, standardized factors were calculated using the trimmed mean of M-values (TMM) method to correct for compositional bias among samples. Next, the estimateTagwiseDisp function was employed to assess gene expression quantification dispersion, with the empirical Bayesian method applied to shrink the dispersion estimates. Finally, Benjamini-Hochberg (BH) correction was implemented on raw p-values to perform multiple testing control, ultimately yielding the corrected false discovery rate (FDR). The intersection of DEGs and MDRGs was taken to generate differentially expressed MDRGs (DE-MDRGs). Univariate Cox regression analysis was undertaken on DE-MDRGs to screen for genes with statistically significant (P < 0.05) survival rates. Subsequently, LASSO regression analysis was finished by utilizing the “glmnet” R package [18] to further screen candidate genes. In LASSO regression analysis, tenfold cross-validation was employed to determine the optimal regularization parameter λ. The minimum value of λ.min can identify models with optimal predictive performance, while the maximum value of λ.1se facilitates the construction of more concise and interpretable models by retaining only the most significant features. Given that the primary objective of this study is to maximize prediction accuracy, we ultimately selected λ.min as the regularization parameter. In the implementation of tenfold cross-validation, the training dataset was randomly divided into 10 subsets. During each cross-validation round, 9 subsets were used as training sets to fit the LASSO model, while the remaining subset served as a validation set to evaluate model performance. For each λ value, we calculated the average cross-validation error (expressed as the mean of likelihood bias) on the validation set. By comparing the average cross-validation errors across different λ values, we selected the optimal λ value that minimized the error (λ.min). Finally, based on multivariate Cox regression analysis, the final prognostic model was built. Combined with the expression values and regression coefficients of the feature genes, the MD-related risk score (MDRS) for each patient was calculated using the formula: Patients in the training set, GSE26939 and GSE72094 data sets, fell into high-MDRS and low-MDRS groups, with the median MDRS as the cutoff value. Kaplan–Meier (K-M) curves were plotted for survival analysis of the two MDRS groups (high and low). This study employed the positive K-M method to evaluate follow-up completeness. The TCGA dataset showed a 63.4% dropout rate with a median follow-up duration of 1.83 years, while the GSE72094 dataset had a 71.6% dropout rate and a median follow-up duration of 2.26 years. The GSE26939 dataset demonstrated a 42.6% dropout rate with a median follow-up duration of 2.75 years. The accuracy of the model was assessed by ROC analysis implemented in the “timeROC” R package [19]. In this process, we first utilized the K-M estimator to estimate the censoring distribution. Subsequently, when calculating the ROC curves for 1-year, 3-year, and 5-year intervals, the “timeROC” R package marginal method was utilized. This approach involved weighting the data based on the K-M estimated censoring distribution, thus ensuring that the obtained ROC curves accurately reflected the model’s predictive performance at different time points. Nomogram construction and verification To assist in predicting the 1-year, 3-year, and 5-year OS rates of LUAD patients, we integrated clinical features (age, TMN, gender, and tumor stage) and MDRS by utilizing the “rms” R package [20] to construct a prognostic nomogram. 1/3/5-year calibration curves were graphed to evaluate the prognostic performance of the nomogram. TME analysis and immunotherapy response prediction The function of the TME in cancer was investigated by assessing immune cell infiltration levels using CIBERSORT. Through ssGSEA, we assessed immune cell infiltration levels and immune function and compared scores between high and low-MDRS groups. In both CIBERSORT and ssGSEA analyses, a series of data processing operations was performed using mRNA FPKM data. For CIBERSORT analysis, the avereps function was employed to calculate the average expression values of duplicated genes for consolidation. Subsequently, genes with expression means ≤ 0 across all samples were excluded to remove non-expressed genes, thereby enhancing both the efficiency and accuracy of the analysis. The RNA-seq count data were transformed using the voom function in the “limma” R package to generate an expression matrix suitable for linear model analysis. In the ssGSEA analysis, the avereps function was employed to handle duplicate genes and remove genes with zero expression mean values, thereby eliminating interference from meaningless expression values. Finally, the results were standardized by setting the ssgseaParam parameter (…, normalize = TRUE). Additionally, immunophenoscore (IPS, https://tcia.at/home) and umor Immune Dysfunction and Exclusion (TIDE, http://tide.dfci.harvard.edu/) scores were collected to predict the immunotherapy response of patients in the two groups with different MDRSs. Mutation analysis Single-nucleotide variation (SNV) data for LUAD were obtained from the TCGA database. With the use of the “maftools” R package [21], the map of top-ranked mutation genes in the high-MDRS group and low-MDRS group was graphed, which also displayed mutation types and frequencies in different groups. Moreover, the mutation rates of model genes in the two groups were analyzed, and the tumor mutation burden (TMB) between the two groups was compared. Specifically, we calculated the total number of nonsynonymous mutations (including missense, nonsense, and frameshift mutations) in the sample. The total number of mutations was divided by the size of the exonic region covered by sequencing (in Mb) to calculate the number of mutations per Mb, namely TMB. The total_per Mb values were then log-transformed to make the data distribution more normal, thereby facilitating statistical analysis. Drug prediction Based on the data from the Cancer Drug Sensitivity Genomics 2 (GDSC2) database (https://www.cancerrxgene.org/), the “oncoPredict” R package [22] was used to predict in vitro drug response (shown by IC50 value) of samples from the TCGA dataset, and to analyze the differences in chemotherapy response among high- and low-MDRS groups. Results Differential analysis and the building and validation of the prognostic model Differential expression analysis was undertaken on normal samples and LUAD samples in the TCGA cohort, with 5576 DEGs determined. The intersection of DEGs and MDRGs yielded 201 DE-MDRGs (Fig. 1A, Table S7). Subsequently, 63 prognosis-associated genes were determined through univariate Cox regression analysis (p < 0.05) (Table S8). Further LASSO regression analysis was undertaken to identify 16 candidate genes (Fig. 1B, C). Finally, a prognostic model of 8 MDRGs was established based on multivariate Cox regression analysis (Fig. 1D). Differential expression analysis and construction of the prognostic model. A Venn diagram of the intersection of DEGs and MDRGs. B LASSO coefficient spectrum. C Cross-validation in LASSO regression to optimize parameter selection. D Forest plot of feature genes obtained from multivariate Cox regression analysis The classification of patients in the training set, GSE26939 and GSE72094 data sets, was conducted according to MDRS values calculated by the formula, with the median serving as the cutoff value between low- and high-MDRS groups. Through K-M analysis, we discovered that the high-MDRS group had a lower survival rate than the low-MDRS group in the TCGA cohort, GSE26939, and GSE72094 datasets (Fig. 2A–C), with the difference being statistically significant (p < 0.01). Specifically, the risk ratio (HR) in the TCGA training set was 2.08 with a 95% confidence interval (CI) of 1.54–2.81, while HRs in the validation sets of GSE26939 and GSE72094 were 2.16 (95% CI 1.31–3.55) and 2.27 (95% CI 1.54–3.35), respectively. Subsequently, using ROC analysis, the 1/3/5-year AUC values in the TCGA cohort were 0.77, 0.70, and 0.66, respectively (Fig. 2D). In the GSE26939 validation set, the AUC values were 0.69, 0.66, and 0.70, respectively (Fig. 2E). In the GSE72094 validation set, the AUC values were 0.69, 0.69, and 0.84 (Fig. 2F). The data indicated that the prognostic model we constructed performed well in predicting. In addition, a prognostic nomogram was plotted based on MDRS and various clinical characteristics to predict 1, 3, and 5 years of OS in LUAD patients, providing a convenient and intuitive tool for clinical treatment decision-making (Fig. 2G). The calibration curve revealed a high consistency between the survival rate predicted by the nomogram and the actual survival rate (Fig. 2H). Validation of the prognostic model and construction and validation of the nomogram. A–C K-M survival curves for high- and low-MDRS groups using the TCGA (A), GSE26939 (B), and GSE72094 (C) datasets. D-F ROC curve analysis based on the TCGA (D), GSE26939 (E), and GSE72094 (F) cohorts. G The 1/3/5-year prognostic nomogram was constructed by combining risk scores and clinical factors (Age, Gender, Stage, and TMN). H Calibration curve of the nomogram Analysis of tumor immune The CIBERSORT analysis results demonstrated the different abundance of immune cell infiltration between the low-MDRS group and the high-MDRS group. The low-MDRS group had considerably higher levels of resting memory CD4+ T cells and memory B cells compared to the high-MDRS group, while M0 Macrophages were more infiltrated in the high-MDRS group (p < 0.05) (Fig. 3A). The results of ssGSEA presented two parts: immune cells and immune functions. Specifically, the low-MDRS group had a higher infiltration abundance of pDCs, aDCs, B cells, iDCs, helper T cells, neutrophils, mast cells, and TILs (p < 0.05) (Fig. 3B). In the low-MDRS group, the levels of Check-point, HLA, and T cell co-stimulation were also higher, and all were statistically significant (p < 0.05) (Fig. 3C). Additionally, we discovered that the low-MDRS group had higher IPS and lower TIDE scores (p < 0.01) (Fig. 3D, E). Patients in the low-MDRS group had a higher possibility of benefiting from immunotherapy. The correlation analysis further demonstrated that the MDRS was significantly negatively correlated with positive- and negative-immune phenotypes of CTLA-4, and was significantly positively correlated with the TIDE score (p < 0.05) (Figure S1A–B). This suggests that patients with high-MDRS exhibit more pronounced immune dysfunction and immune escape characteristics, and their potential response to CTLA-4 blockade and other immunotherapies is generally lower. Immune-related analysis. A Comparison of tumor immune infiltration between the high-MDRS group and the low-MDRS group using CIBERSORT. B, C Comparison of immune cell infiltration (B) and immune function scores (C) between different MDRS groups using ssGSEA. D IPS score. E TIDE score. ns indicates no significance; * indicates P < 0.05; * * indicates P < 0.01; * * * indicates P < 0.001; * * * indicates P < 0.0001 Analysis of mutation landscape The analysis of gene mutation profiles for high- and low-MDRS groups showed that the most common mutation types in both groups were missense mutations. However, the mutation frequencies of TP53, TTN, MUC16, and CSMD3 in the high-MDRS group were significantly higher than those in the low-MDRS group (Figs. 4A, B; S2A), with mutation rates as 54% vs 43% (TP53), 47% vs 41% (TTN), 44% vs 35% (MUC16), and 43% vs 34% (CSMD3), respectively. Further comparison of mutation types revealed that TP53 exhibited missense mutations, nonsense mutations, splice site mutations, frameshift deletion mutations, and Multi Hit mutations in both high- and low-MDRS groups. Compared with the low-MDRS group, the high-MDRS group had more abundant missense mutations and splice site mutations, and specifically exhibited frameshift insertion mutations, while the low-MDRS group specifically exhibited in-frame deletion mutations. TTN was mainly characterized by missense mutations and Multi Hit mutations in both groups, with a higher proportion of these mutations and the addition of splice site mutations in the high-MDRS group. MUC16 detected missense mutations, Multi Hit mutations, and splice site mutations in both groups. The high-MDRS group had more diverse mutation types, with newly added frameshift deletion and frameshift insertion mutations, while the low-MDRS group exhibited specific nonsense mutations. CSMD3 exhibited missense mutations, Multi Hit mutations, nonsense mutations, and splice site mutations in both groups, with a higher proportion of missense and Multi Hit mutations in the high-MDRS group (Figure S2A). Further analysis of differential mutation genes revealed that compared with the low-MDRS group, the mutation frequency of more genes was significantly higher in the high-MDRS group (p < 0.05) (Figure S2B-C). Correlation analysis indicated that TMB was significantly positively correlated with the MDRS (p < 0.05) (Figure S2D). Further comparison of the mutation rates of model genes between the two groups revealed that the model gene with the highest mutation rate in both the high-MDRS group and the low-MDRS group was CPS1 (13% vs 8%) (Fig. 4C). The TMB was significantly higher in the high-MDRS group than in the low-MDRS group (p < 0.001) (Fig. 4D). Based on the above results, it is suggested that tumors with high MDRS have higher genomic instability and a unique set of driver gene mutations. These characteristics may jointly drive their more aggressive biological phenotype. Additionally, it should be noted that although the high-MDRS group has a higher TMB, the immunosuppressive microenvironment characteristics (such as the enrichment of M0 macrophages and a higher TIDE score) may counteract the potential immunotherapy benefits brought by the high mutation load. Mutation landscape analysis. A, B Mutation profiles of the high-MDRS group (A) and the low-MDRS group (B). Insets highlight the loci of key recurrent mutations (e.g., TP53, CSMD3). The mutation classification statistics chart: the leftmost chart in the first row; the mutation type statistics chart: the middle chart in the first row; the detailed statistics chart of single nucleotide polymorphism mutations: the rightmost chart in the first row; the statistics chart of the number and proportion of mutations for all samples: the leftmost chart in the second row; the statistics chart of the number of different mutation types; the middle chart in the second row; the statistics chart of the top 10 mutated gene mutation results: the rightmost chart in the second row. C Comparison of mutation rates of model genes between the high-MDRS group and the low-MDRS group. D Differences in TMB between the two MDRS groups Chemotherapy drug prediction The relationship between MDRS and predicted in vitro chemotherapy response was further elucidated based on the GDSC2 database. The results showed that there were significant differences in drug treatment response between the high-MDRS group and the low-MDRS group. We discovered that the low-MDRS group was more sensitive to AZD8055, BMS-754807, ERK-6604, SB505124, Tozasertib, and ZM447439 than the high-MDRS group, while the high-MDRS group was more sensitive to Venetoclax and BI-2536 (p < 0.001) (Fig. 5). Analysis of predicted in vitro drug response between patients in the low-MDRS group and the high-MDRS group Discussion Over the past several decades, substantial scientific advancements have been achieved in LUAD. The advent of multidisciplinary treatment strategies and innovative therapeutic drugs has transformed the therapeutic landscape, offering improved prognoses for patients [23]. LUAD patients, however, still experience dismal survival time and a high mortality rate [24]. MD dysregulation is tightly linked to the development of various cancers [25, 26], which brings new research directions for cancer treatment. Therefore, this study sets up an MD-related prognostic model using bioinformatics methods, which can accurately predict the prognosis and immune cell infiltration levels of LUAD patients. The work provides promising biomarkers for the prognosis prediction and treatment of LUAD. The prognostic model we constructed consisted of 8 MDRGs (SLC52A3, HMGA2, CPS1, GLS2, CYP27A1, CFTR, STAR, and DRP2), most of which are linked with the occurrence, development, or prognosis of LUAD. The HMGA2 gene is one of the members of the chromatin remodeling family and an oncogene in several malignant tumors [27, 28]. HMGA2 is upregulated in LUAD patients and has a bearing on reduced OS [29]. Moreover, HMGA2 can also regulate circular RNA ASPH to boost tumor growth in LUAD [30]. In the urea cycle, CPS1 is the first rate-limiting mitochondrial enzyme, and its expression level is greatly elevated in LUAD tissues, impacting dismal prognosis [31, 32]. In KRAS-mutant/LKB1-deficient LUAD, CPS1 also drives pyrimidine synthesis and supports tumor growth [33]. CYP27A1 belongs to the cytochrome P450 oxidase family, and its expression is reduced in LUAD, which is linked with adverse prognosis of patients [34]. CFTR acts as a tumor suppressor to repress LUAD cells from growth, invasion, and migration and reinforce apoptosis, serving as a key survival predictor for smokers and non-smokers [35, 36]. Furthermore, GATA6 and CFTR are down-regulated in LUAD. GATA6 activates CFTR by modulating the Arachidonic Acid metabolic pathway, hindering the progression of LUAD. The GATA6/CFTR axis may thus be a therapeutic target for LUAD patients [37]. Therefore, these genes may be promising biomarkers for LUAD and offer valuable adjuncts for LUAD early prediction and treatment. More importantly, the multi-gene risk model constructed in this study suggests that there may be extensive synergistic effects among these genes. They can collectively form a core network that influences the prognosis of LUAD. For instance, HMGA2, as a transcriptional regulator, can widely promote tumor proliferation and invasion [28–30]. CPS1 in LUAD drives pyrimidine synthesis and supports metabolic reprogramming [38], and its high expression is closely associated with poor prognosis. Although the direct transcriptional regulation between HMGA2 and CPS1 has not been clearly confirmed, there may be certain functional synergy between the two. The proliferative tumor program driven by HMGA2 may rely on the pyrimidine metabolism supported by CPS1 to provide raw materials for rapid proliferation. Therefore, their co-occurrence in the model may help identify the aggressive LUAD subtypes that rely on “metabolic-transcriptional synergy”. Additionally, in cholesterol metabolism, STAR is responsible for transporting cholesterol to mitochondria, while CYP27A1 catalyzes its conversion to 27-hydroxycholesterol, which has been found to have specific signaling functions [39], and their functions are closely coupled. The deletion of CFTR may exacerbate mitochondrial dysfunction by disrupting cellular lipid homeostasis, reducing fatty acid β-oxidation, and inhibiting mitochondrial respiratory chain function [40]. Moreover, CFTR mutations may also disrupt intracellular cholesterol homeostasis, interfering with the transport mediated by STAR and the catalytic process of CYP27A1, ultimately affecting steroid hormone balance and tumor progression [41]. In summary, our model indicates that genes such as SLC52A3, GLS2, DRP2, etc. may play essential roles in the occurrence and development of LUAD through interactions with the above network. However, whether HMGA2 directly regulates the transcription of CPS1, as well as how CFTR precisely affects mitochondrial function, still requires further experimental verification. In the future, in vitro co-deletion or co-overexpression experiments can be conducted on HMGA2 and CPS1 (or CFTR) in the LUAD cell line to assess potential synthetic lethality or synergistic effects. Subsequently, metabolomics analysis can be performed to verify their impact on pyrimidine synthesis and mitochondrial function. The model we constructed was proven to be accurate in predicting the prognosis of LUAD patients, so we probed into the underlying mechanisms. Traditional cancer treatment targets specific proteins in specific disease pathways, but there is increasing evidence that multi-angle treatment strategies are needed [42–44]. Immune cell infiltration assumes a central role in the TME, dramatically affecting patient prognosis [45]. We compared the abundance of immune cell infiltration between the high-MDRS group and the low-MDRS group. The tumors with high MDRS had lower immune cell infiltration, particularly T cells and B cells, but had highly infiltrated M0 macrophages. This suggests that high-MDRS tumors have the so-called cold tumor characteristics, with repressed anti-tumor activity [46]. Low levels of immune cell infiltration can make tumor cells more likely to evade immune surveillance, thereby boosting tumor progression, which may partially explain the considerably reduced survival rate of LUAD patients in the high-MDRS group. Moreover, IPS, TIDE, and TMB are crucial biomarkers for tumor immunotherapy [47–49]. This study shows that tumors with high MDRS have a higher TMB, but are accompanied by a lower IPS and a higher TIDE score. Additionally, the high infiltration of immunosuppressive M0 macrophages and the low infiltration of T cells all indicate that they are in an immunosuppressive state. This immunosuppressive microenvironment may weaken the predictive value of TMB for the response to immunotherapy. Patients with high MDRS may need to adopt combined treatment strategies targeting immunosuppression. Since the human body is a complex regulatory system, these results require further research to be thoroughly analyzed. Drug-assisted therapy can improve patient prognosis. In the treatment of non-small cell lung cancer (NSCLC), combination chemotherapy is usually the first-line treatment, typically including a platinum-based drug (such as cisplatin or carboplatin) combined with other drugs with different mechanisms [49]. In addition, for T790M mutant NSCLC, osimertinib, as a first-line treatment, greatly elevates OS and enhances patient outcomes [50, 51]. Nevertheless, the emergence of drug resistance represents the principal obstacle in LUAD management, frequently resulting in disease recurrence and progression [52]. Consequently, precision assessment of individual predicted in vitro drug response profiles is critical for optimizing therapeutic prognosis and survival in LUAD patients. Our study evaluated the relationship between different MDRS groups and the predicted in vitro drug response. Patients in the low-MDRS group were more suitable for the treatment of AZD8055, ZM447439, ERK-6604, SB505124, Tozasertib, and BMS-754807, while the high-MDRS group had a more sensitive reaction to BI-2536 and Venetoclax. These findings set up a basis for personalized treatment of LUAD patients and are of great importance for improving drug treatment efficiency. For example, Tozasertib (VX-680) is a specific molecular inhibitor targeting Aurora kinases that interfere with the cell cycle and curb tumor growth in multiple cancer types [53, 54]. In addition, BMS-754807, a small molecule inhibitor of insulin-like growth factor-1R/IR, is cytotoxic to NSCLC cells and potentiates the effect of platinum-based chemotherapy in the human A549 cell line [55, 56]. However, whether the therapeutic effects of these drugs in LUAD are influenced by intra-tumoral MD, as well as their safety and efficacy, still needs to be explored in further research. Interestingly, the predicted sensitivity of the high-MDRS group to Venetoclax (a BCL-2 inhibitor) is particularly intriguing, as it aligns with the model’s indication of mitochondrial dysfunction and altered apoptotic signaling in this group. Conversely, the low-MDRS group’s sensitivity to AZD8055 (an mTOR inhibitor) may reflect their relatively immune-active tumor microenvironment, which could be more responsive to therapies targeting proliferative and metabolic pathways. We herein utilized bioinformatics methods to screen genes in LUAD and selected 8 DMRGs to create a prognostic model that is useful in predicting the prognosis and immune infiltration abundance of LUAD patients. Therefore, the work not only guides personalized treatment of LUAD but also offers an instrumental foundation for research on the TME, suggesting potential therapeutic targets. However, there are still some limitations worth further discussion. Primarily, there is a lack of additional publicly available datasets to validate our findings. Therefore, it is necessary to design comprehensive, prospective multicenter studies to validate the prognostic performance of MDRGs. Secondly, in view of the lack of treatment-related information in the current data set, future research will focus on the collection and integration of treatment response data to further optimize the model. Additionally, our work preliminarily reveals the relationship between MDRGs and the TME and the prognosis of LUAD, but the underlying mechanisms behind it need further investigation. Supplementary Information Supplementary Material 1. Figure S1: Correlation analysis of MDRS and indicators related to immunotherapy response. A Correlation analysis between MDRS and indicators of immunotherapy response prediction score (IPS). B Correlation analysis between MDRS and TIDE score of tumor immune dysfunction and rejection. Supplementary Material 2. Figure S2: Differential analysis at the genetic mutation level between high- and low- MDRS groups. A Comparison of the distribution of top 10 mutated genes in common mutated genes between high- and low-MDRS groups. B Forest plot showing the genes with significant differences in mutation frequency between high- and low-MDRS groups and their odds ratio (OR) and confidence intervals. C The mutated genes with significant differences statistically between high- and low-MDRS groups, with the mutation gene, mutation type, mutation frequency and significance level reflected in the figure. D The correlation between tumor mutational burden (TMB) and model risk score (MDRS). Supplementary Material 3. Table S1: Information table of clinical samples in the TCGA-LUAD dataset Supplementary Material 4. Table S2: Information table of clinical samples in the GSE72094 dataset Supplementary Material 5. Table S3: Information table of clinical samples in the GSE26939 dataset Supplementary Material 6. Table S4: Comparative analysis of clinical information in the TCGA and GSE72094 data sets. Supplementary Material 7. Table S5: Comparative analysis of clinical information in the TCGA and GSE26939 data sets. Supplementary Material 8. Table S6: 1123 MDRGs. Supplementary Material 9. Table S7: Intersection of DEGs and MDRGs, resulting in 201 DE-MDRGs. Supplementary Material 10. Table S8: 63 prognostic genes screened by univariate Cox regression analysis Supplementary Material 11. Table S9: The code information used in this study Author contributions Li Xu and Danting Zheng conceived and designed the study. Jisong Zhang and Huihui Hu collected the data. Liangliang Dong and Enguo Chen analyzed and interpreted the data. Li Xu contributed to the manuscript writing and editing. All authors read and approved the final manuscript. Funding Not applicable. Data availability The datasets analyzed in this study are publicly available. The TCGA-LUAD dataset was obtained from the Cancer Genome Atlas database through the Genome Data Sharing Center platform (https://portal.gdc.cancer.gov/). The GSE72094 and GSE26939 datasets were obtained from the Gene Expression Omnibus Database (website: https://www.ncbi.nlm.nih.gov/geo/). All relevant clinical information used in this study is listed in Tables S1-S3. The code information used in this study is shown in Table S9. Declarations Ethics approval and consent to participate Ethical approval and consent were not required as this study was based on publicly available data. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Li Xu and Danting Zheng have contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1. Figure S1: Correlation analysis of MDRS and indicators related to immunotherapy response. A Correlation analysis between MDRS and indicators of immunotherapy response prediction score (IPS). B Correlation analysis between MDRS and TIDE score of tumor immune dysfunction and rejection. Supplementary Material 2. Figure S2: Differential analysis at the genetic mutation level between high- and low- MDRS groups. A Comparison of the distribution of top 10 mutated genes in common mutated genes between high- and low-MDRS groups. B Forest plot showing the genes with significant differences in mutation frequency between high- and low-MDRS groups and their odds ratio (OR) and confidence intervals. C The mutated genes with significant differences statistically between high- and low-MDRS groups, with the mutation gene, mutation type, mutation frequency and significance level reflected in the figure. D The correlation between tumor mutational burden (TMB) and model risk score (MDRS). Supplementary Material 3. Table S1: Information table of clinical samples in the TCGA-LUAD dataset Supplementary Material 4. Table S2: Information table of clinical samples in the GSE72094 dataset Supplementary Material 5. Table S3: Information table of clinical samples in the GSE26939 dataset Supplementary Material 6. Table S4: Comparative analysis of clinical information in the TCGA and GSE72094 data sets. Supplementary Material 7. Table S5: Comparative analysis of clinical information in the TCGA and GSE26939 data sets. Supplementary Material 8. Table S6: 1123 MDRGs. Supplementary Material 9. Table S7: Intersection of DEGs and MDRGs, resulting in 201 DE-MDRGs. Supplementary Material 10. Table S8: 63 prognostic genes screened by univariate Cox regression analysis Supplementary Material 11. Table S9: The code information used in this study Data Availability Statement The datasets analyzed in this study are publicly available. The TCGA-LUAD dataset was obtained from the Cancer Genome Atlas database through the Genome Data Sharing Center platform (https://portal.gdc.cancer.gov/). The GSE72094 and GSE26939 datasets were obtained from the Gene Expression Omnibus Database (website: https://www.ncbi.nlm.nih.gov/geo/). All relevant clinical information used in this study is listed in Tables S1-S3. The code information used in this study is shown in Table S9.