Abstract

   Understanding the tumor microenvironment (TME) and the role of long
   noncoding RNAs (lncRNAs) in gastric adenocarcinoma (GA) is crucial, as
   these elements not only influence tumor progression but also provide
   opportunities for more precise prognostic assessments and tailored
   therapeutic interventions. This study identified mitochondrial
   autophagy-related lncRNAs, constructed a robust prognostic risk model,
   and explored the relationship between immune microenvironment
   characteristics and therapeutic responses. The model’s performance was
   evaluated using ROC curves, Kaplan–Meier survival analysis, and
   nomograms. Our results demonstrate that the model outperforms
   traditional clinical factors, such as age and stage, in predicting
   patient outcomes. Immune cell analysis revealed distinct correlations
   with risk scores, and several immune checkpoint genes exhibited
   differential expression between risk groups. Drug sensitivity analysis
   suggested that low-risk patients could benefit more from ICIs,
   Oxaliplatin, Irinotecan, Afatinib, and Dabrafenib, while high-risk
   patients showed higher sensitivity to IGF1R3801, JQI, WZ4003 and
   NU7441. The identified lncRNA-based risk model provides a reliable
   prognostic tool for GA patients and highlights distinct immune
   microenvironment profiles that may influence treatment responses. These
   findings contribute to developing personalized therapeutic strategies
   targeting lncRNAs and the TME in GA.

Supplementary Information

   The online version contains supplementary material available at
   10.1007/s12672-025-02042-z.

   Keywords: Gastric adenocarcinoma, Mitophagy, lncRNAs, Prognostic model

Introduction

   Gastric adenocarcinoma (GA), the most common histological subtype of
   gastric cancer, accounts for approximately 95% of all gastric cancer
   cases [[30]1]. Despite advancements in treatment, which typically
   involves a combination of surgery and chemotherapy, patient outcomes
   remain poor due to persistent challenges such as drug resistance,
   metastasis, and recurrence [[31]2]. A key factor contributing to this
   poor prognosis is the high heterogeneity of GA. This heterogeneity
   manifests in diverse signaling pathway activation, variable immune
   profiles, and differing disease susceptibilities among patients. Such
   complexity not only complicates treatment but also highlights the need
   for more precise prognostic tools and individualized therapeutic
   strategies [[32]3]. Given these challenges, the identification of novel
   prognostic biomarkers for GA is critical. Biomarkers capable of
   accurately stratifying patients based on their unique disease
   characteristics could facilitate more tailored prognostic assessments
   and personalized treatment approaches. This precision-driven strategy
   holds significant potential to improve patient outcomes and address the
   pressing need for more effective GA management.

   Macroautophagy (commonly referred to as autophagy) is a catabolic
   process essential for maintaining cellular homeostasis by degrading and
   recycling intracellular components, including entire organelles
   [[33]4]. A specialized form of autophagy, mitophagy, selectively
   eliminates damaged or dysfunctional mitochondria through the
   autophagosome-lysosome system [[34]5, [35]6]. This targeted
   mitochondrial turnover has emerged as a critical mechanism in cancer,
   as tumor cells rely on functional mitochondria to sustain their growth
   and adapt to various microenvironmental stresses [[36]7–[37]9]. Recent
   evidence underscores the pivotal role of mitophagy in cancer
   progression by facilitating mitochondrial renewal, thereby enabling
   tumor cells to survive chemotherapeutic and targeted therapies that
   induce mitochondrial damage [[38]10]. Despite these advances, the
   precise roles and regulatory mechanisms of mitophagy in mediating
   resistance to anticancer therapies remain poorly understood. Unraveling
   these mechanisms holds the potential to provide valuable insights into
   overcoming therapeutic resistance in cancer treatment. In recent years,
   mitophagy has attracted growing interest in cancer research,
   particularly in GA, where it serves as a mechanism for selectively
   removing damaged mitochondria [[39]11]. Initial evidence linking
   autophagy to cancer persistence came from observations that certain
   tumor tissues exhibit elevated levels of LC3 puncta and lipidated LC3
   (LC3-II), indicators of autophagosome accumulation [[40]12]. However,
   such static, tissue-based measurements only reflect the abundance of
   autophagosomes and cannot distinguish between increased autophagy
   initiation and defects in autophagosome turnover.

   Long noncoding RNAs (lncRNAs) are a diverse class of RNA transcripts
   exceeding 200 nucleotides in length that do not code for proteins
   [[41]13–[42]15]. Unlike other types of RNA, lncRNAs have recently been
   identified as crucial regulators of gene expression, influencing a
   range of physiological and pathological processes [[43]16–[44]18].
   Emerging evidence suggests that lncRNAs play significant roles in
   cancer biology by modulating pathways involved in cancer initiation and
   progression, acting either as oncogenes that promote tumor growth or as
   tumor suppressors that inhibit it [[45]19, [46]20]. This dual role
   highlights the precision and complexity of lncRNAs in cancer
   development, making them potential targets for new therapeutic
   strategies.

   Here, we show that mitophagy-related lncRNAs may play a crucial role in
   the occurrence and progression of GA and have potential prognostic
   predictive value. Thus, this study aims to construct a prognostic model
   for GA based on mitophagy-related lncRNAs to provide a reference for
   personalized treatment of GA patients. The innovation of this study
   lies in constructing a prognostic model based on mitophagy-related
   lncRNAs, exploring the role of lncRNAs in GA, particularly their
   potential impact on the tumor microenvironment, immune function, and
   drug response. Our study not only provides new insights into biomarker
   research for GA but also offers scientific evidence for developing
   personalized treatment strategies.

Results

Correlation and differential expression analyses

   In this study, we obtained gene expression data and corresponding
   clinical information for 375 GA samples and 21 normal samples from The
   Cancer Genome Atlas (TCGA) database. To identify long non-coding RNAs
   (lncRNAs) associated with mitochondrial autophagy, we first retrieved
   40 mitochondrial autophagy-related genes from the Pathway Unification
   database (Supplementary Data 1). These genes formed the foundation for
   subsequent correlation analyses to identify relevant lncRNAs. Using
   Pearson correlation analysis, we identified 7,466 lncRNAs significantly
   correlated with mitochondrial autophagy-related genes (Supplementary
   Data 2). Next, we performed differential expression analysis between
   tumor and normal samples, revealing 5,097 lncRNAs that were
   differentially expressed in GA compared to normal tissues
   (Supplementary Data 3). To visualize these findings, a heatmap
   (Fig. [47]1A) presents the expression profiles of the most
   significantly differentially expressed lncRNAs across all samples,
   while a volcano plot (Fig. [48]1B) highlights the significantly
   upregulated and downregulated lncRNAs in tumor tissues. These analyses
   provide valuable insights into the potential role of these lncRNAs as
   prognostic markers for GA.

Fig. 1.

   [49]Fig. 1
   [50]Open in a new tab

   Differential expression of lncRNAs between GA samples and normal
   samples. A A heatmap shows the expression patterns of the top
   differentially expressed lncRNAs across GA samples and normal samples.
   Red indicates high expression, and blue indicates low expression. B A
   volcano plot shows the differentially expressed lncRNAs between tumor
   and normal samples. Red and green dots indicate significantly
   upregulated and downregulated lncRNAs, respectively, while black dots
   represent non-significant lncRNAs

MKLs score was an independent prognostic factor

   Univariate Cox regression analysis of the differentially expressed
   lncRNAs identified eight lncRNAs significantly associated with the
   overall survival of gastric adenocarcinoma (GA) patients (Fig. [51]2A).
   The results of various machine learning algorithms used in the analysis
   are summarized in Fig. [52]2B. This analysis identified 8 mitophagy key
   lncRNAs (MKLs) as core components of the model: HAGLR, LINC01210,
   LINC01094, LINC00592, LINC00705, LEF1-AS1, C3orf36, and AC005498.3. The
   survival analysis revealed significant differences in survival times
   between high-risk and low-risk patient groups, as shown in Fig. [53]2C
   (TCGA) and Fig. [54]2D (GEO). These MKLs demonstrated strong
   associations with GA patient survival outcomes, underscoring their
   potential utility as prognostic biomarkers. To further investigate the
   relationship between clinical characteristics (age, gender, grade, and
   stage) and patient prognosis, we performed univariate Cox regression
   analysis. Age (p = 0.022, HR = 1.026, 95% CI [1.003–1.039]), stage
   (p < 0.001, HR = 1.479, 95% CI [1.193–1.833]), and risk score
   (p < 0.001, HR = 5.762, 95% CI [3.530–9.406]) were significantly
   associated with overall survival (Fig. [55]3A). Further multivariate
   Cox regression analysis confirmed that age, stage, and risk score were
   independent prognostic factors for GA patient survival (Fig. [56]3B).
   These findings highlight their robust predictive value, emphasizing the
   potential of the identified MKLs and clinical characteristics as tools
   for improving GA prognostic assessments.

Fig. 2.

   [57]Fig. 2
   [58]Open in a new tab

   Performance of the Prognostic Model and Machine Learning Algorithms. A
   Hazard ratios of the eight selected lncRNAs (LINC00592, LEF1-AS1,
   LINC00964, LINC01210, Corf36, LINC00705, AC005943.8, and HAGLR) used in
   the prognostic model are displayed. The red squares represent the
   hazard ratios (HR), with confidence intervals shown as horizontal
   lines. Significant associations with survival outcomes are indicated by
   the p-values. B Comparison of various machine learning algorithm
   combinations in constructing the prognostic model. The C-index values
   for TCGA and GEO cohorts are listed for each algorithm combination. The
   highest-performing algorithms, such as RSF + GBM and CoxBoost + GBM,
   achieved the best predictive performance. C Kaplan–Meier survival
   curves for high- and low-risk groups in the TCGA cohort. Patients in
   the high-risk group exhibited significantly worse overall survival
   compared to the low-risk group (p < 0.001). D Kaplan–Meier survival
   curves for high- and low-risk groups in the GEO cohort. Similar to the
   TCGA cohort, high-risk patients had significantly worse overall
   survival compared to low-risk patients (p = 0.008)

Fig. 3.

   [59]Fig. 3
   [60]Open in a new tab

   Prognostic Performance of the Risk Model in TCGA and GEO Cohorts. A
   Univariate Cox regression analysis of clinical features (age, gender,
   grade, stage) and the risk score in TCGA patients. The hazard ratios
   (HR) and confidence intervals are displayed, with the risk score
   showing a strong association with survival outcomes (p < 0.001). B
   Multivariate Cox regression analysis of clinical features and the risk
   score in TCGA patients. The risk score remains an independent
   prognostic factor with a significant hazard ratio (p < 0.001). C
   Comparison of the concordance index (C-index) for the risk score and
   clinical characteristics (age, gender, grade, stage) over time. The
   risk score consistently outperforms clinical characteristics,
   demonstrating its superior prognostic value. D Time-dependent ROC
   curves for the risk score in the TCGA cohort. The AUCs for predicting
   1-year, 3-year, and 5-year survival are 0.685, 0.736, and 0.832,
   respectively, indicating high predictive accuracy. E ROC curve for the
   risk score predicting 2-year survival in the GEO cohort. The AUC is
   0.623, reflecting moderate predictive performance. F ROC curve for the
   risk score predicting 6-year survival in the GEO cohort. The AUC is
   0.603, showing slightly lower performance for long-term prediction. G
   Nomogram based on risk score and clinical variables predicts survival
   probabilities for 1-, 3-, and 5-year overall survival. H Calibration
   curves for the nomogram demonstrate agreement between predicted and
   observed survival outcomes across 1-, 3-, and 5-year time points, with
   a C-index of 0.680 (95% CI 0.632–0.729)

Construction and evaluation of MKLs prognostic model

   The prognostic model composed of 8 lncRNAs exhibited a C-index that was
   significantly higher than that of clinical-pathological features
   (Fig. [61]3C). In TCGA patients, the AUCs for 1-year, 3-year, and
   5-year survival times were 0.685, 0.736, and 0.832, respectively
   (Fig. [62]3D). Validation using the [63]GSE62254 dataset showed AUCs of
   0.623 and 0.603 for 2-year and 6-year survival, respectively
   (Fig. [64]3E and F). However, due to the specific survival time
   distribution of patients in this dataset, the AUCs for 1-year, 3-year,
   and 5-year survival were slightly lower, all close to 0.6. The
   constructed nomogram (Fig. [65]3G) suggested that the predictions for
   1-year, 3-year, and 5-year survival times were consistent with the
   observed survival outcomes (Fig. [66]3H).

High-risk patients with significant B cell and immune-related pathways

   Differential expression analysis identified significant differences in
   gene expression between the high-risk and low-risk groups
   (Fig. [67]4A). Gene Ontology (GO) enrichment analysis revealed that the
   differentially expressed genes (DEGs) were significantly enriched in
   various biological processes, cellular components, and molecular
   functions (Fig. [68]4B). KEGG pathway analysis highlighted pathways
   such as B-cell receptor signaling, hematopoietic cell lineage, and
   calcium signaling as significantly enriched in these DEGs
   (Fig. [69]4C). Gene set enrichment analysis (GSEA) was performed to
   explore the biological pathways associated with the high- and low-risk
   groups. In the high-risk group, pathways related to cell adhesion
   molecules, chemokine signaling, and complement cascades were
   significantly enriched (Fig. [70]4D). Conversely, the low-risk group
   exhibited enrichment in pathways associated with the cell cycle and
   immune response (Fig. [71]4E), suggesting distinct biological processes
   underlying the different risk groups.

Fig. 4.

   [72]Fig. 4
   [73]Open in a new tab

   Differential expression, functional enrichment, and GSEA between high-
   and low-risk groups. A Heatmap showing the DEGs between the high-risk
   and low-risk groups. Red indicates higher expression, while blue
   indicates lower expression. B Circular barplot showing GO enrichment
   analysis of DEGs, including Biological Processes, Cellular Components,
   and Molecular Functions. C Bar plot summarizing the KEGG pathway
   enrichment analysis results. The most significantly enriched pathways
   include cytoskeleton regulation, ECM-receptor interaction, focal
   adhesion, and pathways related to cancer. The color gradient represents
   the q-value, with darker bars indicating higher significance. D GSEA
   shows pathways significantly enriched in the high-risk group, including
   cell adhesion molecules and chemokine signaling pathways. E GSEA
   reveals pathways significantly enriched in the low-risk group, such as
   cell cycle and immune-related pathways

Somatic mutation between high and low-risk GA patients

   Somatic mutation profiles were comprehensively analyzed for both
   high-risk and low-risk GA patient groups to uncover mutation patterns
   associated with mitophagy-related lncRNAs (Fig. [74]5). In the
   high-risk group (Fig. [75]5A), mutations were observed in 178 of 194
   patients (91.75%). The most frequently mutated genes included TTN
   (51%), TP53 (41%), and MUC16 (28%), with a significant prevalence of
   missense mutations and frame-shift deletions. Other notable mutations
   were detected in LRP1B (26%), ARID1A (25%), and CSMD3 (24%). The total
   mutation burden (TMB) varied substantially among individuals within the
   high-risk cohort. Similarly, in the low-risk group (Fig. [76]5B),
   somatic mutations were identified in 151 of 167 patients (90.42%). The
   key genes exhibiting mutations were consistent with those in the
   high-risk group, including TTN (49%), TP53 (44%), and MUC16 (32%).
   Additional genes, such as LRP1B (27%), ARID1A (25%), and CSMD3 (25%),
   also showed significant mutation frequencies.

Fig. 5.

   [77]Fig. 5
   [78]Open in a new tab

   Somatic mutation analysis in GA patients. A Oncoplot showing the
   distribution and types of somatic mutations in high-risk group
   patients. B Oncoplot displaying the mutation landscape in low-risk
   group patients

High MKLs score was associated with low TMB and good immune infiltration

   Tumor mutation burden (TMB) was found to be negatively correlated with
   the risk score (R = − 0.28, p = 2e-07), indicating that patients with
   higher risk scores tend to have lower TMB (Fig. [79]6A). A significant
   difference in TMB between the high-risk and low-risk groups was
   observed (p = 3.8e-07, Fig. [80]6B). Kaplan–Meier survival analysis
   showed that patients with high TMB had significantly better survival
   outcomes compared to those with low TMB (p = 0.020, Fig. [81]6C).
   Moreover, when combining TMB and risk score, the survival analysis
   revealed significant differences between the four subgroups (high
   TMB + high risk, high TMB + low risk, low TMB + high risk, low
   TMB + low risk, p < 0.001, Fig. [82]6D), underscoring the prognostic
   value of integrating these factors. Analysis of the tumor
   microenvironment revealed significant differences between the high-risk
   and low-risk groups. The ESTIMATE score, which reflects the overall
   level of stromal and immune components, was significantly higher in the
   high-risk group (p = 0.0012, Fig. [83]6E). Both the immune score
   (p = 0.4, Fig. [84]6F) and stromal score (p < 1.7e-07, Fig. [85]6G)
   were also significantly elevated in the high-risk group, suggesting
   that the tumor microenvironment may contribute to the prognostic
   differences between the two groups.

Fig. 6.

   [86]Fig. 6
   [87]Open in a new tab

   TMB analysis, survival analysis, and tumor microenvironment scores. A
   Scatter plot showing the correlation between TMB and risk score. B
   Boxplot comparing TMB between the high-risk and low-risk groups. C
   Kaplan–Meier survival curve showing the difference in overall survival
   between patients with high TMB and low TMB. D Kaplan–Meier survival
   analysis combining TMB and risk score. E Boxplot comparing ESTIMATE
   scores between high-risk and low-risk groups. F Boxplot comparing
   immune scores between high-risk and low-risk groups. G Boxplot
   comparing stromal scores between high-risk and low-risk groups

Immune microenvironment and drug sensitivity

   To investigate the relationship between mitophagy-related lncRNAs and
   the immune microenvironment, we analyzed immune cell infiltration and
   immune checkpoint expression in high- and low-risk groups (Fig. [88]7).
   Immune Cell Infiltration (Fig. [89]7A): Correlation analysis revealed
   distinct immune cell infiltration patterns associated with the risk
   groups. Infiltration of T cell CD4 + memory, T cell CD4 + Th1, and
   cancer-associated fibroblasts (analyzed via XCELL) was positively
   correlated with risk, while infiltration of Macrophage M1 showed a
   negative correlation. These findings suggest a differential immune cell
   composition between the risk groups, potentially reflecting variations
   in the tumor microenvironment (TME). Immune Checkpoint Expression
   (Fig. [90]7B): Differential expression analysis of immune
   checkpoint-related genes demonstrated significantly higher expression
   of checkpoint molecules such as CD44, TNFSF18, CD276, and ICOSLG in the
   high-risk group compared to the low-risk group (p < 0.05). These
   results highlight the potential immune evasion mechanisms that may
   contribute to tumor progression in high-risk patients. TCGA Subtype
   Analysis (Fig. [91]7C): Patients were further stratified based on TCGA
   immune subtypes (C1–C4). A significant association was observed between
   risk groups and immune subtypes (p = 0.025). The majority of low-risk
   patients belonged to the C2 subtype (immune active, 59%), whereas
   high-risk patients were more evenly distributed across C1
   (immune-depleted, 33%) and C2 (49%). Notably, a small fraction of
   patients in the high-risk group fell into the C3 (inflammatory, 15%)
   and C4 (lymphocyte-depleted, 3%) subtypes, further underscoring the
   heterogeneity of the TME in these groups.

Fig. 7.

   [92]Fig. 7
   [93]Open in a new tab

   Correlation analysis of immune cells and differential expression of
   immune checkpoint genes. A Correlation between immune cell infiltration
   and risk scores. Various immune cells, assessed using different
   computational tools (e.g., xCell, TIMER, QUANTISEQ, MCPcounter, EPIC,
   and CIBERSORT), show distinct correlations with risk scores. Positive
   correlations (e.g., stromal cells, macrophages) and negative
   correlations (e.g., certain T cells) highlight the differing immune
   microenvironments between high- and low-risk groups. B Differential
   expression of immune checkpoint-related genes between high- and
   low-risk groups. Genes such as CD44, TNFSF18, PDCD1LG2, CD40LG, and
   others exhibit significant differences in expression levels, suggesting
   potential implications for immunotherapy responsiveness. Statistical
   significance is denoted by asterisks (*p < 0.05, **p < 0.01,
   ***p < 0.001). C Distribution of TCGA patients across immune subtypes
   (C1–C4) according to risk groups. The table highlights the proportion
   of patients in each immune subtype and their corresponding risk groups
   (low or high)

   Tumor Immune Dysfunction and Exclusion (TIDE) Score (Fig. [94]8A): The
   TIDE analysis indicated a significant difference in immune escape
   potential between the high-risk and low-risk groups. High-risk patients
   exhibited higher TIDE scores compared to low-risk patients (p < 0.001),
   suggesting that the high-risk group may be more prone to immune evasion
   mechanisms, which could contribute to their poorer prognosis. Drug
   Sensitivity Analysis (F[95]ig. [96]8B–I): To explore the potential
   therapeutic implications of the risk model, we evaluated the
   sensitivity of both risk groups to various chemotherapeutic agents and
   targeted therapies. Drug sensitivity analysis suggested that low-risk
   patients could benefit more from ICIs, Oxaliplatin, Irinotecan,
   Afatinib, and Dabrafenib, while high-risk patients showed higher
   sensitivity to IGF1R3801, JQI, WZ4003 and NU7441.

Fig. 8.

   [97]Fig. 8
   [98]Open in a new tab

   Drug sensitivity analysis. A TIDE score comparison between low-risk and
   high-risk groups. B–I Boxplots showing the sensitivity of different
   risk groups to chemotherapy and targeted drugs

Discussion

   Recent advancements have emphasized the critical role of
   autophagy-related gene networks in cancer prognosis and therapy.
   Prognostic models based on autophagy-related gene pairs have been
   successfully applied to stratify patients with lung adenocarcinoma
   based on survival outcomes and immune microenvironment characteristics
   [[99]21]. These prognostic models of autophagy-related genes based on
   cancer data sets have diagnostic value for the prognosis of LUAD
   patients, and the combination of clinical characteristics improves the
   accuracy of the models [[100]22, [101]23]. Similarly, programmed cell
   death-associated gene signatures have shown promise in predicting
   responses to immune checkpoint inhibitors by characterizing the tumor
   microenvironment [[102]24]. Our study highlights the critical roles of
   MKLs in GA and their potential to improve prognostic precision and
   guide personalized treatment strategies. In cancer, lncRNAs interact
   with protein complexes to regulate transcription and stabilize mRNA
   [[103]25, [104]26]. They are involved in a variety of biological
   processes, including the maintenance of stem cell characteristics and
   the promotion of tumor progression [[105]27, [106]28]. By elucidating
   the relationship between MKLs and mitophagy, we provide a mechanistic
   basis for their association with GA progression. Furthermore, the
   integration of MKL-based risk scores with immune microenvironment
   profiling highlights how these lncRNAs influence key biological
   pathways, including immune evasion mechanisms. The findings emphasize
   the potential for targeting MKLs in therapeutic interventions, such as
   improving immune checkpoint blockade efficacy or reversing
   chemoresistance, thereby paving the way for personalized treatment
   strategies.

   Our research identified 8 key lncRNAs that show significant
   differential expression in GA, effectively distinguishing high-risk
   from low-risk patients and closely associated with patient survival.
   Some of these lncRNAs have already been highlighted in previous
   studies, suggesting their potential role in the onset and progression
   of various cancers. In screening mitochondrial autophagy-related
   lncRNAs, we found that lncRNAs such as HAGLR and LINC00705 have
   significant expression in GA and are associated with prognosis. As a
   classical oncogenic lncRNA, HAGLR has been extensively studied in
   various tumors [[107]29, [108]30]. These previous studies are
   consistent with our findings and further support the prognostic model
   for GA based on mitophagy-related lncRNAs. Our lncRNA-based prognostic
   model provides significant clinical advantages over traditional
   prognostic factors, such as TNM staging, by incorporating molecular and
   immune-related characteristics into risk stratification. While TNM
   staging primarily evaluates tumor size, lymph node involvement, and
   metastasis, it does not account for the molecular heterogeneity of
   gastric adenocarcinoma. In contrast, our model stratifies patients
   based on the expression of mitochondrial autophagy-related lncRNAs,
   which reflect underlying biological processes associated with tumor
   progression and immune microenvironment modulation. For example, our
   model effectively differentiates patients with similar TNM stages but
   varying survival outcomes, as demonstrated by the significantly higher
   C-index and AUC values. Moreover, the integration of risk scores with
   clinical features improves individualized treatment planning, enabling
   better identification of patients who may benefit from specific
   therapies, such as immunotherapy or targeted agents.

   The success of cancer immunotherapy depends on a thorough understanding
   of the TME and the mechanisms of immune evasion, which involve a
   complex network of interactions among the tumor, stroma, and
   infiltrating immune cells [[109]31]. Zhou et al. previously used a
   consensus clustering algorithm to classify tumor samples and evaluate
   the similarities and differences in mitophagy-related gene expression
   and immune profiles between tumor samples [[110]32]. Our immune cell
   function analysis revealed significant differences between high- and
   low-risk groups, highlighting distinct immune microenvironment
   profiles. For example, high-risk patients exhibited increased
   infiltration of T cell CD4 + memory, T cell CD4 + Th1, and
   cancer-associated fibroblasts, while low-risk patients showed elevated
   levels of Macrophage M1, which are crucial for anti-tumor immunity.
   Gene Set Enrichment Analysis (GSEA) further identified immune-related
   pathways, such as chemokine signaling and complement cascades, as
   enriched in the high-risk group, suggesting a potential role in immune
   evasion mechanisms. Conversely, pathways associated with immune
   activation, such as T-cell receptor signaling, were enriched in the
   low-risk group. These findings suggest that the differences in immune
   microenvironment profiles may influence prognosis and treatment
   responses. Importantly, the enrichment of immune checkpoint genes such
   as PDCD1LG2 and TIGIT in the high-risk group provides a foundation for
   exploring targeted immunotherapies, such as immune checkpoint
   inhibitors, in these patients, consistent with the findings of Masugi
   et al., who reported an association between high PDCD1LG2 expression
   and poor prognosis in colorectal cancer [[111]33]. Additionally, genes
   such as TIGIT and CD244, which are also differentially expressed, have
   been linked to immune evasion and tumor progression [[112]34, [113]35].
   By linking GSEA findings to specific biological processes and
   therapeutic opportunities, our study offers valuable insights into the
   mechanisms underpinning immune modulation in gastric adenocarcinoma.

   Mitochondrial autophagy, which responds to various extracellular
   signals, plays a role in numerous biological processes, including
   mitochondrial depolarization, hypoxia, and development [[114]9,
   [115]36, [116]37]. Previous studies have indicated that autophagy
   contributes to chemotherapy resistance in small cell lung cancer
   (SCLC), with significantly elevated autophagy levels observed in
   chemoresistant SCLC cell lines [[117]38, [118]39]. However, limited
   research has specifically explored the relationship between
   mitochondrial autophagy and GA resistance. Risk scores can be valuable
   tools for screening therapeutic drugs in patients with GA, as they help
   stratify patients according to likely treatment responses. For example,
   a recent study found that high-risk GA patients exhibited greater
   sensitivity to chemotherapy agents, such as rapamycin, suggesting that
   these patients might benefit more from this treatment option [[119]40].
   In terms of drug sensitivity analysis, our study indicated that
   patients in the low-risk group are more sensitive to ICIs, Oxaliplatin,
   Irinotecan, Afatinib, and Dabrafenib, while those in the high-risk
   group show greater response to IGF1R3801, JQI, WZ4003 and NU7441.

Limitation

   Despite the significant findings in this study, some limitations
   remain. First, although we validated the model’s efficacy using the
   TCGA database, external independent cohort validation is lacking. To
   further validate the model's applicability, future studies could
   attempt multi-center research using different databases or clinical
   samples. Additionally, this study mainly relied on gene expression data
   without considering the impact of epigenetic modifications or protein
   levels. Future research could incorporate proteomics or epigenetic data
   to explore the role of mitochondrial autophagy-related lncRNAs in GA.

Conclusions

   In summary, the mitochondrial autophagy-related lncRNA-based prognostic
   model demonstrated effective performance in survival prediction, immune
   microenvironment analysis, and drug sensitivity prediction. Our
   research provides new insights into individualized treatment for GA and
   foundational data to support future research on the mechanisms of
   mitochondrial autophagy in cancer.

Methods

Acquisition of gene expression data and clinical information

   Using Perl (version 5.32.1.1), gene expression data and clinical
   information of GA patients were obtained from the TCGA database
   ([120]https://cancergenome.nih.gov/). Mitochondrial autophagy-related
   genes were downloaded from the Pathway Unification database
   ([121]https://pathcards.genecards.org/).

Screening of lncRNA and construction of prognostic model

   Pearson correlation analysis was conducted using R software (version
   4.2.1) to identify lncRNAs associated with mitochondrial autophagy
   genes (parameters set at P < 0.001, |Correlation Coefficient|> 0.2.
   Differential expression analysis of mitochondrial autophagy-related
   lncRNAs between tumor and normal tissues was performed (cut-off
   criteria set as |log2fold change (logFC)|> 0.5, P-value < 0.05;
   FDR < 0.05). Univariate Cox regression analysis was then applied to the
   differentially expressed lncRNAs, with a filtering condition of
   P < 0.01. Subsequently, various machine learning algorithms were
   applied, including RSF, GBM, CoxBoost, LASSO, survivalSVM, StepCox,
   Enet, Ridge, SuperPC, and plsRcox, to construct the optimal predictive
   model and the risk score was calculated by the linear Predictor
   function of the best model after screening. The predictive performance
   of the model was then validated using the [122]GSE62254 dataset.

Model performance evaluation and nomogram construction

   Using univariate and multivariate Cox analyses, the risk score
   calculated by the model was compared with clinicopathological
   characteristics for prognosis. The survivalROC package was used to plot
   ROC curves and calculate the area under the curve (AUC) for 1-, 3-, and
   5-year outcomes, assessing the accuracy of the prognostic model.
   Survival time differences between low- and high-risk groups were
   assessed, and Kaplan–Meier curves were plotted. A nomogram was
   constructed by integrating patients’ gender, age, grade, T, N, M
   stages, Stage staging, and risk scores. The concordance between the
   nomogram’s prediction and actual survival, as well as the C-index
   value, were evaluated to assess accuracy. A heatmap was used to display
   the expression levels of lncRNAs in the model across high- and low-risk
   groups.

GO, KEGG, and GSEA analyses

   Using R software, differential genes between high- and low-risk groups
   were identified (parameters set as |log2fold change (logFC)|> 1,
   P-value < 0.05; FDR < 0.05). Heatmaps were generated for the
   differential genes, and GO and KEGG analyses were conducted to obtain
   information on functions and pathways. Subsequently, using
   c2.cp.kegg.Hs.symbols.gmt, c5.go.Hs.symbols.gmt, and risk score data,
   GSEA enrichment analysis was performed in R to examine pathway biases
   in high- and low-risk groups.

Relationship between TMB, microenvironment, and risk score

   After calculating the TMB data of GA patients from TCGA using R, the
   differences in TMB results between high- and low-risk groups were
   examined. A boxplot was used to display TMB differences between the
   high- and low-risk groups. By combining survival data, survival
   differences between high and low TMB expressions in high- and low-risk
   groups were obtained. Finally, ESTIMATEScore, ImmuneScore, and
   StromalScore of the GA microenvironment were compared between high- and
   low-risk groups.

Immune microenvironment and drug efficacy differential analysis

   To clarify the relationship between the immune microenvironment and
   risk scores, Wilcoxon signed rank test and Spearman correlation
   analysis were conducted in R. Differences in immune cell distribution
   among GA patients were obtained using XCELL, TIMER, QUANTISEQ,
   MCPcounter, EPIC, CIBERSORT, and CIBERSORT-ABS methods (set criteria:
   P < 0.05). Differences in the expression levels of immune checkpoint
   inhibition-related genes between low- and high-risk groups were
   compared. Subsequently, based on the immune subtypes classified by
   TCGA: C1 (Wound Healing), C2 (IFN-γ Dominant), C3 (Inflammatory), C4
   (Lymphocyte Deprived), C5 (Immunologically Quiet), and C6 (TGF-β
   Dominant), the differences in these immune subtypes were evaluated
   between high- and low-risk patient groups. Immunotherapy-related
   metrics for GA samples were obtained using the Tumor Immune Dysfunction
   and Exclusion (TIDE) database ([123]http://tide.dfci.harvard.edu). R
   was then utilized to analyze the differences in TIDE scores between the
   high- and low-risk groups, thereby assessing the efficacy of antitumor
   immunotherapy. Finally, the oncoPredict package was used to score drug
   sensitivity to assess the clinical treatment response differences of GA
   patients in lncRNA prognostic model groups.

Supplementary Information

   [124]12672_2025_2042_MOESM1_ESM.csv^ (297B, csv)

   Supplementary Material 1. Data 1 Mitochondrial autophagy-related genes
   from the Pathway Unification database
   [125]12672_2025_2042_MOESM2_ESM.csv^ (18.2MB, csv)

   Supplementary Material 2. Data 2 LncRNAs significantly correlated with
   mitochondrial autophagy-related genes
   [126]12672_2025_2042_MOESM3_ESM.csv^ (477KB, csv)

   Supplementary Material 3. Data 3 Differential expression of lncRNA
   between tumor and normal tissues

Acknowledgements