Integrative multi-omics reveals a prognostic and therapeutic landscape of synthetic lethality-associated signatures in lung adenocarcinoma

Zerong Li; Wenmei Qiao; Bin Fan; Fang Qiu; Wei Su

doi:10.1016/j.isci.2026.117098

. 2026 Aug 3;29(8):117098. doi: 10.1016/j.isci.2026.117098

Integrative multi-omics reveals a prognostic and therapeutic landscape of synthetic lethality-associated signatures in lung adenocarcinoma

Zerong Li ^2,⁵, Wenmei Qiao ^3,⁵, Bin Fan ², Fang Qiu ², Wei Su ^1,^4,^6,^∗

PMCID: PMC13469817 PMID: 42597961

Summary

Lung adenocarcinoma (LUAD) exhibits considerable heterogeneity and therapeutic resistance. Here, we integrated single-cell RNA sequencing, spatial transcriptomics, and machine learning to characterize synthetic lethality (SL)-associated transcriptional activity in LUAD. We quantified SL activity across malignant epithelial cells and stratified them into high-, dominant-, and low-SL groups. High-SL cells were enriched in advanced-stage and metastatic samples and showed reduced differentiation potential. Through multiple machine learning algorithms, we identified 13 high-SL signature genes, with IFRD2 (interferon-related developmental regulator 2) among the top contributors. Experimental validation confirmed IFRD2 upregulation in high-malignancy cell lines, and IFRD2 knockdown modulated sensitivity to the PARP inhibitor niraparib. The predicted compound prostaglandin A1 exhibited selective cytotoxicity against LUAD cells and induced DNA damage, consistent with an SL-related mechanism. Our findings establish a framework for identifying SL-associated vulnerabilities and provide a resource of candidate genes and compounds for further mechanistic investigation in LUAD precision therapy.

Keywords: lung adenocarcinoma, synthetic lethality, single-cell RNA sequencing, machine learning, drug sensitivity

Graphical abstract

Highlights

•
Integrative multi-omics reveals SL-associated landscape in LUAD
•
High-SL cells mark advanced, metastatic tumors with reduced differentiation
•
IFRD2 drives HSL signature and modulates niraparib sensitivity
•
Prostaglandin A1 shows selective cytotoxicity via DNA damage mechanism

Health sciences; medicine; medical specialty; health informatics; internal medicine; oncology

Introduction

Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancer cases and is characterized by a slow rate of growth and metastasis compared with small cell lung cancer (SCLC).¹ Among NSCLC subtypes, lung adenocarcinoma (LUAD) is the most prevalent, representing nearly 50% of cases.² Current clinical management of LUAD is largely stage dependent: early-stage (I–II) patients are primarily treated with surgical resection; stage III patients often require a combination of chemoradiotherapy and immunotherapy; and stage IV patients with actionable driver mutations typically benefit from targeted therapies against EGFR, ALK, or ROS1.³^,⁴ Despite these advances, the emergence of resistance to targeted agents, such as EGFR tyrosine kinase inhibitors (TKIs), and the significant toxicity associated with conventional chemotherapy underscore the limitations of current treatment paradigms.⁵^,⁶ These challenges highlight an urgent need to identify novel therapeutic approaches and molecular targets to improve outcomes in LUAD.

Synthetic lethality (SL) refers to a phenomenon in which the loss (or inhibition) of either of two genes individually is tolerated by cells, but their simultaneous loss (or inhibition) leads to cell death.⁷ A classic example is the combination of BRCA1/2 mutations with PARP inhibitors (e.g., olaparib).⁷ BRCA1/2 mutations result in homologous recombination repair deficiency, while PARP inhibitors block single-strand DNA repair via the base excision repair (BER) pathway.⁸ This dual deficiency prevents tumor cells from repairing DNA damage, ultimately causing apoptosis, whereas normal cells with intact BRCA1/2 remain unaffected by PARP inhibition.⁸ The major molecular mechanisms underlying SL include DNA damage repair dependencies, compensatory signaling pathway interactions, cell cycle regulatory defects, and metabolic vulnerabilities.⁹^,¹⁰

In recent years, SL has made remarkable progress in the field of cancer precision therapy. By targeting tumor-specific genetic defects, SL strategies spare normal cells and significantly reduce the broad-spectrum toxicity associated with traditional chemotherapy.¹¹ Moreover, SL approaches simultaneously disrupt alternative survival pathways, offering a promising solution to overcome therapeutic resistance in cancer.¹² The integration of high-throughput CRISPR screening has accelerated the discovery of new SL gene pairs, and the incorporation of artificial intelligence with multi-omics data enables the precise prediction of drug sensitivity, highlighting the promising future of SL-based therapeutic strategies.¹³ While our previous multi-omics framework that characterized copy number variation (CNV)-driven malignant subpopulations and established a pathomics-based prognostic model in LUAD,¹⁴ there remains a lack of comprehensive analysis to systematically characterize synthetic lethal interactions in LUAD.

Therefore, we performed an integrative analysis of large-scale single-cell RNA sequencing (scRNA-seq) datasets and machine learning algorithms to systematically characterize SL activity in LUAD. By identifying key regulatory pathways and high-SL (HSL) signature genes, we aimed to uncover potential therapeutic targets and provide novel insights into the prognostic strategies for LUAD. In accordance with the TITAN Guidelines 2025, we have ensured transparency in our AI-driven analyses by explicitly reporting all AI tools, model parameters, and data processing steps and by maintaining rigorous human oversight over all computational outputs. All code and detailed parameter settings are documented to enable reproducibility, and we have adhered to the ethical principles outlined in the TITAN framework, including bias mitigation and data de-identification.¹⁵

Results

Single-cell transcriptomic profiling of LUAD

To dissect the cellular heterogeneity within LUAD and its microenvironment, we performed scRNA-seq on samples collected from multiple anatomical sites, including primary tumors, normal lung tissues, lymph nodes, and pleural effusions. After stringent quality control (Figure S1) and normalization, a total of 178,739 high-quality cells were selected for subsequent analysis. Unsupervised dimensionality reduction using uniform manifold approximation and projection (UMAP) revealed a total of 28 transcriptionally distinct clusters (Figures 1A and 1B). Cell type annotation identified multiple major cell populations within the LUAD microenvironment based on canonical marker genes. T cells were characterized by high expressions of CD3D and CD3E, B cells by CD79A and MS4A1, and natural killer (NK) cells by NKG7 and KLRD1. Myeloid populations included macrophages (Mac; CD68 and MARCO), monocytes (Mon; LYZ and FCN1), and plasmacytoid dendritic cells (PDC; LILRA4 and CLEC4C). Plasma cells were defined by the expressions of IGHG1 and JCHAIN. Epithelial cells (Epi) were marked by EPCAM and KRT7, fibroblasts (Fib) by ACTA2 and PDGFRA, and endothelial cells (Endo) by PECAM1 and VWF. Additionally, a small subset of cycling cells expressing TOP2A and MKI67 were detected (Figures 1C, 1D, and S2).

The cellular composition varied markedly across different tissue types (Figure 1E). In normal lymph nodes (nLNs), T cells dominated, with substantial proportions of B and NK cells. Normal lung tissues (nLung) showed increased proportions of myeloid and epithelial cells, accompanied by a reduction in T cells. In tumor lung tissues (tLung), epithelial cells were highly enriched. Pleural effusion (PE) samples remained T cell dominant, with NK and B cells as secondary populations. Metastatic lymph nodes (mLNs) exhibited an increased proportion of epithelial cells. In tumor biopsy samples obtained from advanced-stage LUAD patients via bronchoscopy (tL/B), epithelial cells were highly abundant, while T cell infiltration was markedly reduced, suggesting a strongly immunosuppressive microenvironment.

SL activity is elevated in LUAD and enriched in malignant populations

To explore the role of SL in LUAD, we first analyzed bulk RNA-seq data. SL activity was significantly higher in tumor samples than in normal lung tissues (Wilcoxon test, p = 5.8e−13; Figure 2A). Furthermore, SL activity was elevated in advanced-stage (III–IV) LUAD relative to early-stage (I–II) disease (Wilcoxon test, p = 0.0096; Figure 2B), with a progressive increase observed from stage I to stage IV (Kruskal-Wallis test, p = 0.0059; Figure 2C). Survival analysis across multiple cohorts (TCGA-LUAD, GSE50081, GSE31210, and GSE3141) demonstrated that patients with high SL activity had significantly poor overall survival (Figure 2D).

SL activity analysis in LUAD

(A) Comparison of SL activity between tumor and normal lung tissues in bulk RNA-seq data.

(B and C) SL activity distribution across different clinical stages.

(D) Kaplan-Meier survival analysis stratified by SL activity.

(E) CNV-based identification of malignant and normal epithelial cells.

(F) Dot plot of SL activity across major cell types.

(G) UMAP visualization of SL activity in single-cell populations.

(H) Density and distribution of SL activity among cell types.

(I) SL activity comparison across different tissue types.

We next evaluated SL activity at the scRNA-seq level. Given that LUAD originates from epithelial cells, we isolated epithelial subsets and performed CNV analysis to distinguish malignant epithelial cells from normal epithelial cells (Figure 2E). Using multiple scoring algorithms, we comprehensively assessed SL activity across cell types. The results revealed that the SL activity was predominantly enriched in malignant epithelial cells (Figure 2F), which was further confirmed by UMAP visualization (Figure 2G). Density and distribution analyses showed malignant cells as the primary population with elevated SL activity, while macrophages, monocytes, and endothelial cells exhibited moderate levels (Figure 2H). Notably, comparison across tissue types indicated significantly higher SL activity in tL/B samples than in tLung samples (Kruskal-Wallis, p < 2.2e−16; Figure 2I), suggesting that SL activity in malignant cells progressively increases during disease progression, consistent with observations from bulk RNA-seq analysis.

Heterogeneity of SL activity in LUAD

To investigate the heterogeneity of SL activity within LUAD malignant cells, we first assessed its spatial distribution across tumor sections using spatial transcriptomics. From adenocarcinoma in situ (AIS) to minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC), SL activity exhibited a progressive gradient within tumor core regions (Figure 3A). This pattern was consistent with UMAP visualization derived from scRNA-seq data (Figure 3B). Based on SL scores, malignant cells were stratified into three groups: low-SL (LSL) cells (scoring < 2.615, 25%), dominant-SL (DTSL) cells, and HSL cells (scoring > 3.379, 75%) (Figure 3C). The distribution of these groups mirrored the observed activity gradient (Figure 3D).

CytoTRACE analysis further revealed that cells with higher SL activity displayed elevated CytoTRACE scores (Figures 3E–F), and a significant positive correlation was observed between SL activity and CytoTRACE scores (R = 0.47, p < 0.001; Figure 3G), suggesting that increased SL activity may characterize cells in a less differentiated, more aggressive state. Similarly, scPagwas TRS analysis showed that HSL cells had significantly higher TRS scores (p < 0.001; Figures 3H and 3I), with a strong positive correlation between SL activity and TRS scores (R = 0.69, p = 0.001; Figure 3J), highlighting transcriptional plasticity associated with high SL activity.

Notably, Ro/e analysis indicated that the proportion of HSL cells increased progressively along anatomical sites representing disease progression, following the pattern of tLung → PE → mLN → tL/B. This spatial enrichment of HSL cells in advanced lesion sites suggests a potential link between high SL activity and enhanced malignant potential during tumor dissemination (Figure 3K).

Distinct cell-cell communication, functional states, and drug sensitivity profiles of HSL cells

To further delineate the biological differences among SL activity groups, we first investigated cell-cell communication patterns using CellChat analysis. The interaction network revealed that the HSL cells exhibited a marked increase in both the number and strength of interactions with various cell types compared with LSL cells (Figure 4A). Quantitative assessments highlighted a substantial gain in interaction pairs involving HSL malignant cells, particularly with macrophages and monocytes (Figures 4B and 4C), suggesting enhanced crosstalk between HSL cells and the tumor microenvironment (TME).

Cell-cell communication, functional pathways, and drug sensitivity profiles of SL activity groups in LUAD

(A) CellChat networks show the number and strength of interactions across cell types in HSL and LSL groups.

(B) Bar plots quantifying interaction pairs between malignant cells and other cell types.

(C) Scatterplot of outgoing versus incoming interaction strengths for each cell type.

(D) Heatmap of oncogenic pathway enrichment among SL groups.

(E) GSVA results highlighting differential pathway activities between HSL and LSL cells.

(F) Bubble plot showing metabolic pathway enrichment across SL groups.

(G) Volcano plot of predicted drug sensitivities, identifying compounds with increased or decreased sensitivity in HSL cells.

(H) UMAP visualization of drug response scores in malignant cells.

Pathway enrichment analysis indicated that HSL cells were significantly associated with the activation of oncogenic signaling pathways, including Wnt and MAPK, alongside inflammatory and androgen response pathways (Figure 4D). Moreover, GSVA and metabolic pathway enrichment analyses uncovered pronounced metabolic reprogramming in HSL cells. Pathways such as glycolysis, glutathione metabolism, fatty acid degradation, and glycosaminoglycan biosynthesis were notably upregulated, potentially supporting tumor proliferation and immune evasion (Figures 4E and 4F).

To evaluate therapeutic vulnerabilities, drug sensitivity prediction identified several compounds with differential responses in HSL cells. Notably, HSL cells demonstrated reduced sensitivity to compounds such as KU-C103875 (sig-3099) and KU-C104131 (sig-4255) but displayed heightened sensitivity to “baccatin-III” (sig-5747) and “prostaglandin-A1” (sig-6785), with the strongest sensitivity observed for prostaglandin-A1 (PGA1) (Figure 4G). In the updated UMAP visualization, PGA1 response scores exhibited an obvious overlap with HSL cell distributions (Figure 4H).

Uncovering LUAD driver gene modules through hdWGCNA

To identify potential driver genes in LUAD, we performed hdWGCNA. A soft-threshold power of 8 was selected to achieve scale-free topology and optimal network connectivity (Figure 5A). Based on this threshold, hdWGCNA identified multiple gene modules, represented by distinct colors in the dendrogram (Figure 5B).

Identification of LUAD driver gene modules using hdWGCNA

(A) Scale-free topology and network connectivity plots for soft-threshold selection.

(B) Dendrogram showing gene modules with assigned colors.

(C) Violin plots of kME values for hub genes.

(D) UMAP plots displaying the expression patterns of module genes in malignant cells.

(E) Gene network diagrams of hub genes in the blue, brown, and turquoise modules.

Among these, four modules (turquoise, blue, brown, and yellow) were prioritized due to their strong co-expression patterns and potential relevance to LUAD pathology. Module membership (kME) analysis highlighted key hub genes within each module (Figure 5C). UMAP visualization further revealed distinct expression patterns of these module genes across malignant cell populations (Figure 5D).

Module-trait correlation analysis identified the blue, brown, and cyan modules (the latter being a distinct module separate from the turquoise module identified earlier) as critical networks associated with HSL cells (Figure 5E; Table S1). The top 100 hub genes from each module were extracted for functional annotation. The blue module included CLDN4, CDH1, MUC1, and SDC1, which are involved in cell-cell adhesion, tight junction integrity, and epithelial-mesenchymal transition (EMT), indicating their potential roles in driving tumor invasion and metastasis. The brown module contained RPL3, RPS17, and EEF1B2, encoding ribosomal proteins and translation elongation factors associated with increased protein synthesis and rapid tumor proliferation. The turquoise module comprised NDUFS5 and UQCC2, which are linked to mitochondrial function, oxidative phosphorylation, and cellular energy metabolism, implicating their roles in metabolic reprogramming within HSL cells.

Signature gene selection for HSL cells in LUAD

To screen the signature genes of HSL cells, we first compared the gene expression profiles between the HSL and LSL cells. Differential expression analysis identified 193 upregulated genes in HSL cells (Figure 6A; Table S2); of these genes, 75 overlapped with HSL hub driver genes and were considered potential signature genes (Figure 6B; Table S3).

Screening of HSL signature genes using multiple machine learning analysis

(A) Differential expression analysis between HSL and LSL cells.

(B) Venn diagram showing the overlap of upregulated genes and HSL hub driver genes.

(C) Correlation analysis of candidate genes with the HSL phenotype.

(D) ABESS feature selection results.

(E) Random forest feature selection results.

(F) GBM feature selection results.

(G) Decision tree (DT) feature selection results.

(H) LASSO regression feature selection results.

(I) Intersection of signature genes identified by all machine learning methods.

Correlation analysis ranked candidate genes based on their association with the HSL phenotype, yielding 71 positively correlated genes for further evaluation (Figure 6C; Table S4). We then applied multiple machine learning algorithms, including adaptive best subset selection (ABESS; Figure 6D), random forest (Figure 6E), gradient boosting machine (GBM; Figure 6F), decision tree (DT; Figure 6G), and least absolute shrinkage and selection operator (LASSO) regression (Figure 6H). The signature genes identified using each method are provided in Tables S5, S6, S7, S8, and S9.

Finally, intersection analysis of all machine learning outputs revealed 13 shared signature genes (Figure 6I), namely FAM195A, RBCK1, SNF8 (sucrose non-fermenting 8 homolog), CYC1 (cytochrome c1), STUB1 (STIP1 homology and U-box containing protein 1), SLC25A39 (solute carrier family 25 member 39), AHCY (S-adenosylhomocysteine hydrolase), IFRD2 (interferon-related developmental regulator 2), PITX1 (paired like homeodomain 1), SRM (spermidine synthase), NME4 (NDP kinase D), UBE2S (ubiquitin-conjugating enzyme E2 S), and STOML2 (stomatin like 2).

Expression validation and prognostic analysis of HSL signature genes

To demonstrate the representativeness of the signature genes identified through machine learning, we validated their expression patterns across multiple data levels. At the single-cell level, these genes exhibited the highest expression in late-stage samples (tL/B) (Figure 7A). They were predominantly expressed in malignant epithelial cells, with markedly lower expression in the immune and stromal cell populations (Figure 7B). Within malignant cell subsets, these signature genes were highly enriched in HSL cells, whereas their expression was minimal in LSL cells (Figure 7C).

In The Cancer Genome Atlas (TCGA)-LUAD cohort, paired expression analysis revealed that 10 signature genes (AHCY, CYC1, IFRD2, NME4, PITX1, SLC25A39, SNF8, SRM, STOML2, and UBE2S) were significantly upregulated in tumor tissues compared to adjacent normal tissues (Figure 7D). Further, LUAD patients were stratified into high- and low-expression groups for each gene based on optimal cutoff values. Kaplan-Meier survival analysis showed that high expression of 9 signature genes (AHCY, CYC1, NME4, PITX1, RBCK1, SLC25A39, SNF8, SRM, and UBE2S) was associated with significantly poorer overall survival (Figure 7E).

Machine learning benchmark for HSL cell prediction model

To evaluate the predictive performance of individual signature genes for HSL cell classification, receiver operating characteristic (ROC) analysis was performed. The area under the curve (AUC) values ranged from 0.673 (NME4) to 0.746 (SRM), suggesting that all signature genes possessed moderate predictive ability (Figure 8A).

To enhance prediction accuracy and establish a robust HSL cell recognition model, we benchmarked multiple machine learning algorithms, including logistic regression, support vector machine (SVM), random forest, and gradient boosting. Among these, the random forest model (classif.ranger) achieved the highest mean AUC during cross-validation, demonstrating superior performance (Figure 8B). ROC analysis further validated the robustness of the random forest model compared with the other classifiers (Figure 8C). Precision-recall (PR) and ROC curves for both the development and test sets indicated strong discriminative power of the final model (Figure 8D). Confusion matrix analysis revealed well-balanced sensitivity and specificity across the training and test datasets (Figure 8E). Decision curve analysis showed that the proposed model provided greater net clinical benefit than classifying all patients as positive or negative (Figure 8F). Model interpretability analysis using SHAP (SHapley Additive exPlanations) values identified IFRD2, CYC1, SRM, RBCK1, and STUB1 as major contributors, with positive correlations between feature values and predicted probabilities (Figure 8H). Feature importance analysis further confirmed IFRD2, CYC1, and SRM as key predictors of HSL cell identity, underscoring their potential as therapeutic targets for future LUAD treatment strategies (Figure 8G).

Validation of HSL signature genes by RT-qPCR and western blotting

To experimentally validate the expression of the 13 computationally identified signature genes, we performed quantitative reverse-transcription PCR (RT-qPCR) analysis in three cell lines: Beas-2b (a non-tumorigenic lung epithelial cell line, representing a low-malignancy/LSL-like phenotype) and A549 and H1299 (LUAD cell lines, representing high-malignancy/HSL-like phenotypes). The results revealed distinct expression patterns across the gene set (Figure 9A–M). RT-qPCR validation revealed that, IFRD2 exhibited the most pronounced and consistent upregulation, with its mRNA expression being significantly higher in both A549 and H1299 cells than in Beas-2B cells (p < 0.001 for both comparisons; Figure 9H). This finding strongly aligns with its top ranking in the prior SHAP feature importance analysis. Another gene, SRM, also showed an elevated expression trend in the adenocarcinoma cell lines (Figure 9I). In contrast, the expression of CYC1, another top-ranked gene by SHAP analysis, did not show significant upregulation in this in vitro validation (Figure 9D). The marked upregulation of IFRD2 at the mRNA level was further confirmed at the protein level by western blot analysis. Protein expression of IFRD2 was substantially stronger in A549 and H1299 cell lysates than in Beas-2B, whereas the internal control GAPDH remained constant (Figure 9N). This concordance between transcriptional and translational evidence solidifies IFRD2 as a robustly validated molecular feature associated with the high-SL phenotype in LUAD.

Validation of signature gene expression by RT-qPCR and western blotting

(A–M) RT-qPCR analysis of the mRNA expressions of (A) FAM195A (MCRIP2, MAPK regulated corepressor interacting protein 2), (B) RBCK1 (RANBP2-type and C3HC4-type zinc finger containing 1), (C) SNF8 (SNF8 subunit of ESCRT-II), (D) AHCY (adenosylhomocysteinase), (E) SLC25A39 (solute carrier family 25 member 39), (F) STUB1 (STIP1 homology and U-box containing protein 1), (G) CYC1 (cytochrome c1), (H) IFRD2 (interferon-related developmental regulator 2), (I) SRM (spermidine synthase), (J) NME4 (NME/NM23 nucleoside diphosphate kinase 4), (K) STOML2 (stomatin like 2), (L) UBE2S (ubiquitin-conjugating enzyme E2 S), and (M) PITX1 (paired like homeodomain 1).

(N) Expression of IFRD2, detected by western blotting (GAPDH, internal control).

∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

IFRD2 deficiency alone is dispensable for cell proliferation and apoptosis

To investigate the role of IFRD2 in cell proliferation and apoptosis, we transfected cells with small interfering RNAs (siRNAs) targeting IFRD2. Quantitative real-time PCR (real-time qPCR) results showed that si-IFRD2-2 exhibited the highest knockdown efficiency and was, therefore, selected for all subsequent experiments (Figure 10A). Successful knockdown of IFRD2 was further confirmed at the protein level by western blotting (Figure 10B).

CCK8 assays revealed that IFRD2 knockdown alone had no significant effect on the viability of the normal lung epithelial cell line Beas-2B or the LUAD cell lines A549 and H1299 (Figure 10C). Furthermore, western blot analysis demonstrated that IFRD2 knockdown alone did not induce appreciable changes in the expression levels of the apoptosis-related proteins Bcl2 and Bax (Figures 10D–10F).

Knockdown of IFRD2 attenuates sensitivity to synthetic lethal agents in LUAD cells

To further elucidate the role of IFRD2 in HSL cells, we selected niraparib, a classic synthetic lethal agent. CCK8 assays showed that the IC₅₀ values of niraparib were 58.63 nM in Beas-2B cells, 40.53 nM in A549 cells, and 27.21 nM in H1299 cells, indicating that cells with higher IFRD2 expression were more sensitive to the synthetic lethal drug (Figures 11A–11C). Therefore, in subsequent experiments, we aimed to knock down IFRD2 in malignant LUAD cell lines with high IFRD2 expression and then assess changes in tumor cell proliferation, migration, and apoptosis. EdU (5-ethynyl-2′-deoxyuridine) staining results revealed that, compared with the DMSO control group, niraparib treatment significantly reduced the proliferation of A549 (Figure 11D) and H1299 cells (Figure 10F), whereas IFRD2 knockdown partially restored cell proliferation. Furthermore, wound healing assays showed that niraparib markedly inhibited cell migration, and IFRD2 knockdown rescued the migratory capacity of LUAD cells (Figures 11E and 11F). Collectively, these results demonstrated that the LUAD cell lines with HSL sensitivity are vulnerable to niraparib and that IFRD2 knockdown attenuates this sensitivity, suggesting that IFRD2 is likely a key factor maintaining HSL sensitivity in LUAD cells.

Knockdown of IFRD2 attenuates sensitivity to synthetic lethal agents in lung adenocarcinoma cells

(A–C) The IC₅₀ values of niraparib in Beas-2B (A), A549 (B), and H1299 (C) cells.

(D and E) The proliferation capacity of A549 (D) and H1299 (E) cells, as determined by EdU experiment.

(F and G) Wound healing assay for evaluation of the migration of A549 (F) and H1299 (G) cells.

(H) Expression of the apoptosis-related proteins Bcl2 and Bax in A549 cells, as detected by western blotting; n = 3.

(I) Expression of the apoptosis-related proteins Bcl2 and Bax in H1299 cells, as detected by western blotting; n = 3.

∗p < 0.05; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

Prostaglandin-A1 may be an effective therapy for LUAD

PGA1 is a metabolite of cyclooxygenase 2 and has been reported to modulate cardiovascular diseases, neurodegenerative diseases, and exert anti-tumor effects.¹⁶^,¹⁷^,¹⁸ To validate the accuracy of our drug sensitivity prediction, Beas-2B, A549, and H1299 cells were treated with various concentrations of PGA1. CCK8 assay results showed that the IC₅₀ values of PGA1 were 53.5 μM in Beas-2B cells, 13.8 μM in A549 cells, and 11.8 μM in H1299 cells. Compared with the classic synthetic lethal agent niraparib, the IC₅₀ values of PGA1 in these three cell lines were considerably higher, which indicates its low translational potential as a therapeutic agent for LUAD (Figures 12A–12C). However, the IC₅₀ value of PGA1 in normal lung epithelial cells was 3.87- to 4.53-fold higher than in the LUAD cell lines that indicated its stronger cell targeting specificity. We next preliminarily explored the potential mechanism of action of PGA1. Because SL primarily induces cell death through DNA damage, we examined the expression changes of γH2AX, a marker of DNA double-strand breaks, by western blotting (Figure 12D). The results showed that the protein level of γH2AX was significantly elevated following PGA1 treatment, suggesting that PGA1 may also induce cell death via SL.

Prostaglandin-A1 may be an effective therapy for lung adenocarcinoma

(A–C) The IC₅₀ values of prostaglandin A1 on Beas-2B (A), A549 (B), and H1299 (C) cells.

(D) Expression level of γH2AX, examined by western blotting; n = 3. ∗∗∗p < 0.001.

Discussion

In recent years, although therapeutic strategies for LUAD have expanded considerably, significant challenges remain. Most patients with EGFR mutations develop resistance to EGFR-TKIs within 12–18 months of treatment, often due to secondary mutations such as T790M and C797S.¹⁹^,²⁰ Similarly, patients harboring ALK and ROS1 rearrangements frequently experience comparable resistance.²¹^,²² Moreover, only approximately 20%–30% of patients achieve durable responses to PD-1/PD-L1 inhibitors, with the immunosuppressive TME of LUAD—often characterized as “immune cold”—further limiting immunotherapy efficacy.⁵ The inherent genetic and epigenetic heterogeneity of LUAD also undermines the effectiveness of traditional single-target approaches. SL offers a novel precision medicine strategy by targeting tumor-specific dependency pathways while sparing normal cells.²³ Beyond the established application of PARP inhibitors, promising preclinical evidence has demonstrated the potential of combining KRAS-targeted therapy with CDK4/6 inhibitors such as palbociclib.²⁴ With the clinical success of PARP inhibitors and ongoing investigations of new SL targets, SL-based approaches hold great potential to become an integral component of individualized LUAD treatment. However, further studies are needed to elucidate the underlying mechanisms of SL in LUAD.

In this study, we interpreted the scRNA-seq data from 58 LUAD samples, and a total of 178,739 high-quality cells were selected. Then, we analyzed the association between SL activity and survival probability through bulk sequencing, finding that higher SL activity is associated with poorer OS and higher pathological stage. Further, by single-cell analysis, we found that HSL cells account for a higher proportion in samples from patients in the terminal stage. Notably, cells with higher SL activity displayed elevated CytoTRACE scores and higher TRS scores, indicating a less differentiated and more aggressive state. Malignant tumors are often characterized by genomic instability, such as the loss of DNA repair genes, amplification of oncogenes, and deletion of tumor suppressor genes (e.g., TP53, PTEN, RB1, SMARCA4, and ARID1A).²⁵^,²⁶^,²⁷ These alterations render tumor cells highly dependent on specific survival pathways, making them particularly sensitive to corresponding SL targets. This suggests that highly malignant tumors may have greater potential for SL-based therapies, providing a critical entry point for precision oncology.

Further, we investigated the signature gene expression between normal tissues and LUAD tumor tissues and found that the expression levels of CYC1, SRM, SNF8, AHCY, SLC25A39, UBE2S, NME4, and PITX1 were much higher in tumor tissues, and the survival probability was significantly lower in the high-expression subgroups, indicating worse survival time. CYC1, SLC25A39, and NME4 are key enzymes involved in the regulation of metabolism and mitochondrial function.²⁸^,²⁹ CYC1, a component of mitochondrial respiratory chain complex III, participates in electron transport and ATP synthesis.³⁰ Its overexpression supports the high energy demands of tumor cells, promoting cell proliferation and metastasis³¹^,³²; however, specific studies in LUAD are still lacking. SLC25A39 is a mitochondrial membrane transporter responsible for shuttling reduced glutathione (GSH) from the cytosol into the mitochondria.³³^,³⁴ In colorectal cancer (CRC), high SLC25A39 expression promotes cell proliferation and migration while suppressing apoptosis, and its inhibition has been shown to slow tumor growth in experimental models.³⁵ However, the role of SLC25A39 in LUAD remains largely unexplored. NME4, localized within mitochondria, is involved in nucleoside diphosphate kinase activity and mitochondrial dynamics.³⁶ Its downregulation has been associated with EMT, increased invasiveness, and poor prognosis.³⁷^,³⁸ In LUAD, previous studies showed that LUAD patients with high NME4 expression had poorer prognosis, along with reduced CD8⁺ T cells and more infiltration of CD3⁺ T cells and CD20⁺ B cells, which could be identified as a novel oncogene associated with immune exclusion and may serve as a new target for LUAD intervention and immunotherapy.³⁹^,⁴⁰ UBE2S is a key enzyme in the ubiquitin-proteasome pathway and is responsible for mediating protein ubiquitination and degradation.⁴¹ It is overexpressed in various cancers, where it promotes cell cycle progression, proliferation, and metastasis.⁴²^,⁴³^,⁴⁴ In LUAD tissues, UBE2S is upregulated and closely associated with poor prognosis, highlighting its potential as both a prognostic biomarker and a therapeutic target.⁴⁵^,⁴⁶ AHCY regulates the cellular methylation cycle by converting S-adenosylhomocysteine (SAH) into homocysteine, thereby influencing epigenetic regulation and cancer cell stemness.⁴⁷ In breast cancer, its high expression has been associated with poor prognosis⁴⁸; however, its role in LUAD remains unexplored. SNF8, a component of the ESCRT-II complex, is involved in endosomal protein sorting and membrane protein degradation,⁴⁹ whose specific functional roles in tumor development remain unclear. SRM catalyzes the synthesis of spermidine, and its overexpression can promote tumor cell proliferation as well as regulate inflammation and the cell cycle.⁵⁰^,⁵¹ PITX1, a homeobox transcription factor, regulates TERT and RAS signaling pathways, and its overexpression has been shown to induce apoptosis in multiple cancer types.⁵²^,⁵³ Notably, through SHAP feature importance analysis, we also found gene IFRD2, which was highly expressed in LUAD tumor tissues but showed non-significant survival differences between high and low subgroups. IFRD2 is a nuclear-localized translational repressor that binds to the P/E site of the ribosome, blocking mRNA entry into the exit channel and thereby inhibiting protein translation.⁵⁴^,⁵⁵ IFRD2 exhibits low specificity across various cancer cell lines and does not display high expression in any particular tumor type.⁵⁶ Although copy number deletions of IFRD2 are observed in many lung cancer cell lines, loss-of-function mutations are rare, suggesting that its function may be preserved to support tumor progression.⁵⁷ Critically, our experimental validation confirmed that IFRD2 expression was significantly upregulated at both the mRNA and protein levels in high-malignancy LUAD cell models (A549 and H1299) compared with a low-malignancy control (Beas-2b). This functional evidence solidifies IFRD2 not merely as a computational marker but as a bona fide molecular feature of the HSL phenotype. In LUAD, the role of IFRD2 remains largely unexplored, and future studies may uncover its potential as a novel regulator of tumor stress responses and translational homeostasis. In addition, we experimentally examined the functional role of IFRD2, the top-ranked HSL signature gene. IFRD2 knockdown alone did not significantly affect cell proliferation or apoptosis in either normal cells or LUAD cells, a finding that aligns with the core principle of SL-single gene loss is tolerated. However, IFRD2 knockdown substantially attenuated the anti-proliferative and anti-migratory effects of niraparib, directly demonstrating that IFRD2 contributes to maintaining HSL sensitivity. Furthermore, the differential drug sensitivity profile predicted for HSL cells, particularly the heightened sensitivity to compounds like PGA1, was experimentally tested. Consistent with the computational prediction, PGA1 exhibited selective cytotoxicity: its IC₅₀ values in Beas-2B (normal lung epithelial cells) were 3.9- to 4.5-fold higher than those in LUAD cell lines (A549 and H1299), indicating a favorable therapeutic window. Although the absolute IC₅₀ values of PGA1 (11.8–53.5 μM) were considerably higher than those of the classic synthetic lethal agent niraparib (27.2–58.6 nM), this reduced potency may be offset by its superior tumor selectivity. Mechanistically, PGA1 treatment led to a marked elevation of γH2AX, a well-established marker of DNA double-strand breaks, suggesting that PGA1, like PARP inhibitors, may induce cell death through DNA damage-driven SL. Collectively, these experimental data validate the bioinformatic predictions, establish IFRD2 as a functionally relevant HSL-associated gene, and identify PGA1 as a promising lead compound for HSL-targeted therapy. Future work should focus on delineating the precise molecular mechanism by which IFRD2 modulates synthetic lethal responses and evaluating the in vivo efficacy of PGA1 in preclinical LUAD models, such as patient-derived xenografts or organoids.

Importantly, the experimental confirmation of IFRD2 overexpression in HSL-like cells provides immediate and strong rationale for future mechanistic investigations. Subsequent studies should focus on elucidating whether IFRD2 plays a causal role in maintaining the high-SL state, for instance, by leveraging genetic knockdown or knockout models to assess its impact on tumor cell viability, DNA damage response, and drug sensitivity. Furthermore, the differential drug sensitivity profile predicted for HSL cells, particularly the heightened sensitivity to compounds like PGA1, warrants experimental testing in cell lines and patient-derived organoids stratified by IFRD2 expression or SL activity.

Limitations of the study

There are several limitations in our study. First, while we have provided initial in vitro validation for the elevated expression of IFRD2 in high-malignancy cell models, the functional and therapeutic roles of this gene, as well as the other signature genes, require comprehensive mechanistic investigation using genetic perturbation models (e.g., knockdown/knockout) and in vivo studies. Second, although we analyzed potential key genes and associated molecular pathways, other broader determinants influencing LUAD progression may not have been captured, warranting further studies to elucidate intratumoral heterogeneity and prevention strategies. Third, when validating the 13 signature genes using TCGA paired data and Kaplan-Meier analysis, not all genes showed positive results, possibly due to the limited sample size of the TCGA cohort. Finally, effective preclinical validation models, technological optimizations, and rigorous clinical trials are needed to translate these potential targets into clinical applications.

Resource availability

Lead contact

Request for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Wei Su (suwei4582@cqu.edu.cn).

Materials availability

Models generated in this study are available upon request from the lead contact.

Data and code availability

•
This paper analyzes existing, publicly available data, accessible at TCGA (https://portal.gdc.cancer.gov/), genome-wide association study (GWAS, https://opengwas.io/datasets/), and Gene expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/).
•
This paper does not report original code.
•
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was financially supported by the National Natural Science Foundation of China (82500094), the Hospital Pharmacy Research Fund (Fosunpharma Fund) of Guangdong Pharmaceutical Association (2025A02018), the Hospital Pharmacy Research Fund (Yangzijiang Fund) of Guangdong Pharmaceutical Association (2025A01015), and the Shenzhen High-level Hospital Construction Fund.

Author contributions

Conceptualization, Z.L., W.Q., and B.F.; methodology, W.Q., B.F., and F.Q.; investigation, Z.L., W.Q., B.F., F.Q., and W.S.; writing – original draft, Z.L., W.Q., and W.S.; writing – review & editing, W.S., B.F., and W.Q.; funding acquisition, W.S. and Z.L.; resources, W.S. and F.Q.; supervision, W.S. and Z.L.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used DeepSeek to refine the English language and ensure consistent formatting. After using this tool, the authors reviewed and edited the content as necessary and take full responsibility for the final publication.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies

IFRD2 antibody	Immunoway	Cat#YN0225
Bcl2 antibody	Immunoway	Cat#YM8319
Bax antibody	Immunoway	Cat#YM8175
Histone H2A.X (Phospho Ser139)	Immunoway	Cat#YM8195
GAPDH Polyclonal antibody	Proteintech	Cat#10494-1-AP; RRID: AB_2263076
Beta Actin Polyclonal antibody	Proteintech	Cat# 20536-1-AP
Goat Anti-Rabbit IgG-HRP	Immunoway	Cat#RS0002

Chemicals, peptides, and recombinant proteins

Niraparib	MCE	Cat#HY-10619
Olaparib	MCE	Cat#HY-10162
Prostaglandin (PG) A1	Yeasen Biotechnology	Cat#60806ES03

Critical commercial assays

Lipofectamine 2000 Transfection Kit	Life Technologies	Cat#L3000015
TRIzoL Reagent	Invitrogen	Cat#15596018CN
High-Capacity cDNA Reverse Transcription Kit	Vazyme	Cat#R323-01
BeyoClickTM EdU Cell Proliferation Kit	Beyotime	Cat#C0071S
PowerUp SYBR Green Master Mix	Vazyme	Cat#Q711-02
Cell Counting Kit-8	Biosharp	Cat#CCK8-BS350B
RIPA buffer	Beyotime	Cat#P0013B
Protease Inhibitor Cocktail	Solarbi	Cat#329–98-6
Phosphatase Inhibitors	Selleck	Cat#R323-01
BCA assay	Beyotime	Cat#P0010
PVDF membrane	Millipore	Cat#IPVH00010
BeyoClickTM EdU Cell Proliferation Kit	Beyotime	Cat#C0071S
Cell lysis buffer for Western	Beyotime	Cat#P0013
Triton X-100	SolarbioLIFE SCIENCES	Cat#IT9100
Goat serum	Biosharp	Cat#BL1097A
DAPI	Bioshap	Cat#Bl105A; CAS:28718-90-3

Deposited data

Gene expression data	TCGA	https://portal.gdc.cancer.gov/
The overall survival data	TCGA-LUAD	https://portal.gdc.cancer.gov/
PPIs data	UniProt; STRING	https://www.uniprot.org/uniprotkb; https://cn.string-db.org/
MR outcome dataset	GWAS	https://gwas.mrcieu.ac.uk/datasets/ukb-b-14521/

Experimental models: Cell lines

Human: A549	Laboratory preservation	N/A
Human: NCI-H1299	Laboratory preservation	N/A
Human: BEAS-2B	Zhong Qiao Xin Zhou Biotechnology	N/A

Oligonucleotides

Primers for Q-PCR	This paper	Table S13
siRNA for IFRD2	This paper	Table S13

Software and algorithms

R (v4.3.3)	Posit	https://www.r-project.org/
RStudio (v2023.12.1)	Posit	https://www.rstudio.com/
Graphpad Prism 10.0	GraphPad Software Co., Ltd.	https://www.graphpad.com/
Uniform manifold approximation and projection (UMAP)	OmniAnalyzer Pro	N/A
Seurat	OmniAnalyzer Pro	N/A
Inferred copy number variation (inferCNV)	OmniAnalyzer Pro	N/A
Differentially expressed gene (DEG) analysis	OmniAnalyzer Pro	N/A

Open in a new tab

Experimental model and study participant details

Cell culture

The Beas-2B, A549, and H1299 cell lines were obtained from Laboratory preservation and Zhong Qiao Xin Zhou Biotechnology (Shanghai Zhong Qiao Xin Zhou Biotechnology Co., Ltd.). The cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium or Dulbecco’s modified Eagle medium (DMEM) (Gibco, Grand Island, NY, USA), supplemented with 10% fetal bovine serum (FBS) (HyClone, South Logan, UT, USA) and 1% penicillin-streptomycin (Gibco, Grand Island, NY, USA) in a humidified atmosphere at 37°C and 5% CO₂.

Method details

Acquisition and processing of multi-omics data

scRNA-seq data were obtained from the GEO database (GSE131907 ⁵⁸). Spatial transcriptomics (stRNA-seq) data were sourced from the study by Jianfei Zhu et al.⁵⁹ Bulk RNA-seq data were retrieved from The Cancer Genome Atlas (TCGA-LUAD cohort) and three GEO datasets (GSE50081,⁶⁰ GSE31210,⁶¹ GSE3141 ⁶²). Additionally, summary statistics for scPagwas analysis were downloaded from the GWAS Catalog (https://gwas.mrcieu.ac.uk/datasets/ukb-b-14521/), with full dataset details provided in Table S10.

All datasets underwent strict quality control and preprocessing to ensure reliability for downstream analyses. For scRNA-seq data, quality control was performed to filter out low-quality cells and potential doublets. Cells were retained if they expressed between 200 and 7,000 genes, contained <20% mitochondrial gene content, <3% hemoglobin gene content, and had a total UMI count below 100,000. A total of 178,739 high-quality cells remained after filtering. Data normalization, dimensionality reduction via PCA, and clustering were performed using the Seurat package,⁶³ with batch effects corrected by Harmony.⁶⁴

For spatial transcriptomics data, spots with extremely low UMI counts or high mitochondrial gene expression were removed. Preprocessing was also conducted in Seurat, employing SCTransform normalization and unsupervised clustering to define spatial regions. H&E histology and marker gene expression were used to annotate cell types, and spatial expression patterns were visualized using “SpatialDimPlot” and “SpatialFeaturePlot.⁵⁹”

For bulk RNA-seq data, only LUAD samples with complete clinical annotations were included.

Synthetic lethality activity scoring

We first curated a set of 146 SL-related genes from the SynLethDB database (https://synlethdb.sist.shanghaitech.edu.cn/v2/#/disease), as listed in Table S11. To quantify SL activity at the single-cell level, four complementary scoring algorithms were applied: AUCell,⁶⁵ UCell,⁶⁶ singscore,⁶⁷ and AddModuleScore.⁶⁸ Each algorithm provides a unique approach for evaluating gene set enrichment within individual cells. The resulting scores were normalized and aggregated to compute a comprehensive SL score for each cell. It is important to emphasize that the composite SL score quantifies the transcriptional activity of SL-related pathways rather than functional synthetic lethal interactions. This expression-based proxy is justified by the principle that compensatory pathway engagement, a prerequisite for synthetic lethality, is detectable at the transcriptomic level and has been shown to predict therapeutic vulnerability in multiple cancer contexts. Given the epithelial origin of LUAD, CNV inference was performed on epithelial cell subsets during scoring using the R package “inferCNV”⁶⁹ to differentiate malignant from normal epithelial cells.

Subsequently, malignant epithelial cells were stratified into three subgroups based on quartile thresholds of their composite SL scores: the lowest 25th percentile defined as low SL score (LSL) cells, the 25th–75th percentiles as dominant SL score (DTSL) cells, and the highest 25th percentile as high SL score (HSL) cells.

CytoTRACE evaluation and scPagwas analysis

CytoTRACE was used to evaluate differentiation states of malignant cells based on the principle that transcriptional diversity decreases as cells mature.⁷⁰ This analysis provided a quantitative assessment of cellular plasticity and helped characterize subpopulations with high developmental potential. To further investigate the genetic drivers of LUAD heterogeneity, we applied scPagwas (R package “scPagwas”), which integrates scRNA-seq expression data with GWAS summary statistics.⁷¹ By correlating gene expression trends with GWAS-derived SNPs, scPagwas identifies functional variants that may contribute to tumor evolution and differentiation dynamics.

Cell–cell interaction analysis

Cell-cell communication analysis was performed using the “CellChat” R package to investigate ligand-receptor interactions between HSL cells and other cell populations in the scRNA-seq dataset.⁷² Communication networks were constructed to map interaction patterns, and the netVisual_circle function was used to visualize overall signaling strength and directionality. Key ligand-receptor pairs mediating interactions were identified and visualized using the netVisual_bubble function, highlighting signaling pathways enriched in HSL-associated interactions.

HdWGCNA analysis

We applied hdWGCNA to identify potential drivers of tumor progression within HSL cells.⁷³ Malignant epithelial cells classified as HSL were extracted from the scRNA-seq dataset for network construction. A weighted gene co-expression network was generated by calculating pairwise gene correlations and detecting co-expression modules. Module–trait association analysis was then performed to identify modules strongly linked to the HSL phenotype. Hub driver genes were defined based on intra-module connectivity, highlighting critical regulators that may contribute to LUAD malignancy and SL-associated tumor heterogeneity.

Machine learning-based identification of HSL cell signature genes

To identify robust signature genes associated with HSL cells, we applied multiple machine learning algorithms, including Least Absolute Shrinkage and Selection Operator (LASSO), Adaptive Best Subset Selection (ABESS), Random Forest (RF), Gradient Boosting Machine (GBM), and Decision Tree (DT). HSL and LSL cells were extracted from the scRNA-seq dataset and randomly split into training (70%) and validation (30%) sets for model construction and evaluation.

LASSO regression was conducted using the “glmnet” R package, with optimal lambda determined by 10-fold cross-validation to minimize overfitting. ABESS was applied under default parameters to select sparse gene sets with high predictive value. Random forest, GBM, and DT models were implemented via the “mlr3” R package. Gene importance was assessed using model-specific metrics such as mean decrease in accuracy and Gini index. Genes prioritized by these algorithms were ranked and visualized to highlight their relevance to HSL cell classification.

Machine learning benchmark and SHAP analysis

Eight machine learning algorithms were benchmarked using the “mlr3” R package, including k-nearest neighbor (KNN), linear discriminant analysis (LDA), naive Bayes (NB), random forest (Ranger), logistic regression, recursive partitioning and regression trees (RPART), support vector machine (SVM), and extreme gradient boosting (XGBoost). Hyperparameters were tuned via 5-fold internal cross-validation, and model performance was evaluated using 10-fold external cross-validation.⁷⁴ Confusion matrices were generated to assess classification accuracy in both training and validation sets, and decision curve analysis was performed to compare the net clinical benefit of different models. The model achieving the highest average area under the curve (AUC) was selected as the optimal framework for HSL signature prediction.

To interpret model predictions, SHAP (SHapley Additive exPlanations) analysis was applied to quantify gene contributions.⁷⁵ Mean absolute SHAP values (mean_phi) were calculated across all samples to rank feature importance. Genes with the highest SHAP values were prioritized for downstream biological interpretation.

Drug sensitivity analysis

Drug sensitivity prediction was performed using the R package “Beyondcell”,⁷⁶ which estimates drug response at the single-cell level by integrating gene expression profiles with pharmacogenomic datasets. HSL and LSL cells were used as input to predict drug sensitivity scores derived from the Genomics of Drug Sensitivity in Cancer (GDSC) database.

The beyondcellScore function was applied to calculate residual mean and switch point values, enabling the identification of drugs with distinct sensitivity patterns across cell populations. Drugs were classified as high-sensitivity, low-sensitivity, or differential-sensitivity based on their distribution in HSL cells. Top candidate compounds were visualized using scatterplots and UMAP density maps to highlight their association with malignant cell clusters.

Experimental validation of signature gene expression

To experimentally validate the expression of the 13 signature genes identified through computational analyses, we selected Beas-2b cells as a low-malignancy lung epithelial cell model (LSL group), as well as A549 and H1299 cells as high-malignancy lung adenocarcinoma cell models (HSL group).

Total RNA was isolated using Trizol reagent (Cat. No. 15596026, Invitrogen) according to the manufacturer’s instructions. Genomic DNA was eliminated from the RNA samples using DNase I, followed by reverse transcription to cDNA using the High-Capacity cDNA Reverse Transcription Kit (Vazyme, R323-01). Relative mRNA expression levels were quantified using the QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific, United States) with PowerUp SYBR Green Master Mix (Vazyme, TRIzoL Reagent). Custom-designed PCR primers (Shanghai Sangon Biotech Co., Ltd.) were used for qPCR, and their sequences are provided in Table S12.

Protein lysates were prepared using ice-cold RIPA buffer (Beyotime, P0013B) supplemented with protease inhibitor cocktail (Solarbio, 329–98-6) and phosphatase inhibitors (Selleck, B15001), with protein concentration determined by BCA assay (Beyotime, P0010). Equal protein aliquots (30 μg/lane) were resolved on 10% SDS-PAGE gels and electrotransferred to PVDF membranes (0.45 μm, IPVH00010, Millipore). Membranes were blocked, followed by overnight incubation with primary antibodies (4°C) and 1 h incubation with HRP-conjugated secondary antibodies, with detection using ECL substrate (BL520A, Biosharp) on a Bio-Rad ChemiDoc. Protein expression levels were normalized to GAPDH (internal control) using ImageJ for densitometric analysis. Primary antibodies were obtained from Bioworld Technology. Antibody dilutions were optimized as: goat anti-rabbit IgG-HRP (1:2000), and anti-IFRD2 (Immunoway, YN0225, 1:1000) in blocking buffer.

RNA transfection

Beas-2B cells were cultured in DMEM containing 10% FBS and 1% penicillin-streptomycin solution. A549 and H1299 cells were cultured in RPMI-1640 containing 10% FBS and 1% penicillin-streptomycin solution. During the transfection process, the complete medium was replaced with DMEM or RPMI-1640. Next, the plasmid or siRNA was mixed with Lipofectamine 2000 (Life Technologies, USA) and Opti-MEM (Invitrogen, USA), added to plates, and co-incubated for 6 h.

CCK-8 assay

Beas-2B, A549 and H1299 cells were seeded into 96-well plates at a density of 5×10³ cells per well and cultured overnight in complete medium. After the treatments, the culture medium was replaced with 100 μL of fresh complete medium containing 10 μL of CCK-8 solution per well. The plates were incubated at 37 °C for 1–4 h, and the absorbance at 450 nm was measured using a microplate reader. The cell viability was calculated relative to the control group.

EdU staining

Cells were incubated with 10 μM EdU (5-ethynyl-2′-deoxyuridine) for 2 h at 37°C. The cells were then fixed with 4% paraformaldehyde, permeabilized with 0.5% Triton X-100, and stained with Click-iT reaction cocktail according to the manufacturer’s instructions. Cell nuclei were counterstained with DAPI. The percentage of EdU-positive cells was visualized and quantified using a fluorescence microscope.

Wound healing assay

The cell monolayer was then scratched using a sterile 200 μL pipette tip to create a uniform wound. After washing twice with PBS to remove detached cells, the culture medium was replaced with serum-free medium (DMEM or RPMI-1640) to eliminate the effect of cell proliferation. Wound closure was photographed at 0, 12 and 24 h under an inverted microscope at the same positions.

Quantification and statistical analysis

All statistical calculations and data plotting were conducted using R software (4.3.3) and GraphPad Prism 10.0. For comparisons between two groups, the Wilcoxon rank-sum test was applied, while the Kruskal-Wallis test was used for comparisons involving more than two groups. Survival differences were assessed using Kaplan-Meier analysis with log rank tests. Correlation analyses were conducted using Spearman’s rank correlation coefficient unless otherwise specified. P-values were adjusted for multiple testing using the Benjamini-Hochberg method where appropriate. A two-sided P-value <0.05 was considered statistically significant.

Footnotes

Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.117098.

Supplemental information

Document S1. Figures S1 and S2

mmc1.pdf^{(669.8KB, pdf)}

Data S1. Uncropped western blots corresponding to Figures 9–12

mmc2.pdf^{(200.6KB, pdf)}

Table S1. Top 100 genes in each HSL cell phenotype-related module with module assignment and kME values

mmc3.xlsx^{(37.8KB, xlsx)}

Table S2. Upregulated genes in HSL cells with differential expression statistics (p value, log₂FC, percentage expressed, and adjusted p value)

mmc4.xlsx^{(40.9KB, xlsx)}

Table S3. 75 potential signature genes identified by integrative analysis across multiple selection methods

mmc5.xlsx^{(27.2KB, xlsx)}

Table S4. 71 signature genes positively correlated with HSL cell abundance, along with correlation coefficients and p values

mmc6.xlsx^{(28.2KB, xlsx)}

Table S5. Signature genes selected using the ABESS model

mmc7.xlsx^{(26.6KB, xlsx)}

Table S6. Signature genes selected using the random forest model, with their importance scores

mmc8.xlsx^{(27.7KB, xlsx)}

Table S7. Signature genes selected using the gradient boosting machine model, with their importance scores

mmc9.xlsx^{(28.2KB, xlsx)}

Table S8. Signature genes selected using the decision tree model, with their overall importance values

mmc10.xlsx^{(27.8KB, xlsx)}

Table S9. Signature genes selected using LASSO regression, with their regression coefficients

mmc11.xlsx^{(27.5KB, xlsx)}

Table S10. Detailed information of all enrolled datasets, including disease type, sample size, survival data, data type, and accession URLs

mmc12.xlsx^{(28.1KB, xlsx)}

Table S11. 146 synthetic lethality-related genes retrieved from the SynLethDB database

mmc13.xlsx^{(27.8KB, xlsx)}

Table S12. Primer sequences used for qPCR validation experiments

mmc14.xlsx^{(27.7KB, xlsx)}

Table S13. siRNA duplex sequences (sense and antisense) targeting IFRD2

mmc15.xlsx^{(27.3KB, xlsx)}

References

1.Wang M., Herbst R.S., Boshoff C. Toward personalized treatment approaches for non-small-cell lung cancer. Nat. Med. 2021;27:1345–1356. doi: 10.1038/s41591-021-01450-2. [DOI] [PubMed] [Google Scholar]
2.Travis W.D., Garg K., Franklin W.A., Wistuba I.I., Sabloff B., Noguchi M., Kakinuma R., Zakowski M., Ginsberg M., Padera R., et al. Evolving concepts in the pathology and computed tomography imaging of lung adenocarcinoma and bronchioloalveolar carcinoma. J. Clin. Oncol. 2005;23:3279–3287. doi: 10.1200/JCO.2005.15.776. [DOI] [PubMed] [Google Scholar]
3.Shergalis A.G., Hu S., Bankhead A., Neamati N. Role of the ERO1-PDI interaction in oxidative protein folding and disease. Pharmacol. Ther. 2020;210 doi: 10.1016/j.pharmthera.2020.107525. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Guo H., Zhang J., Qin C., Yan H., Liu T., Hu H., Tang S., Tang S., Zhou H. Biomarker-Targeted Therapies in Non-Small Cell Lung Cancer: Current Status and Perspectives. Cells. 2022;11:3200. doi: 10.3390/cells11203200. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Spella M., Stathopoulos G.T. Immune Resistance in Lung Adenocarcinoma. Cancers (Basel) 2021;13 doi: 10.3390/cancers13030384. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Zeng J., Ding X., Ding J., Wang X. Histological transformation into SCLC: An important resistance mechanism of NSCLC upon immunotherapy. Front. Immunol. 2023;14 doi: 10.3389/fimmu.2023.1275957. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Lord C.J., Ashworth A. PARP inhibitors: Synthetic lethality in the clinic. Science. 2017;355:1152–1158. doi: 10.1126/science.aam7344. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Murai J., Pommier Y. BRCAness, Homologous Recombination Deficiencies, and Synthetic Lethality. Cancer Res. 2023;83:1173–1174. doi: 10.1158/0008-5472.CAN-23-0628. [DOI] [PubMed] [Google Scholar]
9.Ryan C.J., Devakumar L.P.S., Pettitt S.J., Lord C.J. Complex synthetic lethality in cancer. Nat. Genet. 2023;55:2039–2048. doi: 10.1038/s41588-023-01557-x. [DOI] [PubMed] [Google Scholar]
10.Setton J., Zinda M., Riaz N., Durocher D., Zimmermann M., Koehler M., Reis-Filho J.S., Powell S.N. Synthetic Lethality in Cancer Therapeutics: The Next Generation. Cancer Discov. 2021;11:1626–1635. doi: 10.1158/2159-8290.CD-20-1503. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Huang A., Garraway L.A., Ashworth A., Weber B. Synthetic lethality as an engine for cancer drug target discovery. Nat. Rev. Drug Discov. 2020;19:23–38. doi: 10.1038/s41573-019-0046-z. [DOI] [PubMed] [Google Scholar]
12.Parameswaran S., Kundapur D., Vizeacoumar F.S., Freywald A., Uppalapati M., Vizeacoumar F.J. A Road Map to Personalizing Targeted Cancer Therapies Using Synthetic Lethality. Trends Cancer. 2019;5:11–29. doi: 10.1016/j.trecan.2018.11.001. [DOI] [PubMed] [Google Scholar]
13.Ye B.-J., Li D.-F., Li X.-Y., Hao J.L., Liu D.J., Yu H., Zhang C.D. Methylation synthetic lethality: Exploiting selective drug targets for cancer therapy. Cancer Lett. 2024;597 doi: 10.1016/j.canlet.2024.217010. [DOI] [PubMed] [Google Scholar]
14.Li Z., Qiao W., Yu S., Fan B., Yang M., Wu M., Qiu F., Wang J. Integrating computational pathology and multi-transcriptomics to characterize lung adenocarcinoma heterogeneity and prognostic modeling. Int. J. Surg. 2025;111:5162–5181. doi: 10.1097/JS9.0000000000002639. [DOI] [PubMed] [Google Scholar]
15.Agha R.A., Mathew G., Rashid R., Kerwan A. Transparency In The reporting of Artificial INtelligence – the TITAN guideline. Premier J. Sci. 2025;10 [Google Scholar]
16.Xu G.B., Yang L.Q., Guan P.P., Wang Z.Y., Wang P. Prostaglandin A1 Inhibits the Cognitive Decline of APP/PS1 Transgenic Mice via PPARγ/ABCA1-dependent Cholesterol Efflux Mechanisms. Neurotherapeutics. 2019;16:505–522. doi: 10.1007/s13311-018-00704-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Yang G., Chen L. An Update of Microsomal Prostaglandin E Synthase-1 and PGE2 Receptors in Cardiovascular Health and Diseases. Oxid. Med. Cell. Longev. 2016;2016:5249086–5249089. doi: 10.1155/2016/5249086. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Tanikawa M., Yamada K., Tominaga K., Morisaki H., Kaneko Y., Ikeda K., Suzuki M., Kiho T., Tomokiyo K., Furuta K., et al. Potent prostaglandin A1 analogs that suppress tumor cell growth through induction of p21 and reduction of cyclin E. J. Biol. Chem. 1998;273:18522–18527. doi: 10.1074/jbc.273.29.18522. [DOI] [PubMed] [Google Scholar]
19.Coudray N., Ocampo P.S., Sakellaropoulos T., Narula N., Snuderl M., Fenyö D., Moreira A.L., Razavian N., Tsirigos A. Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning. Nat. Med. 2018;24:1559–1567. doi: 10.1038/s41591-018-0177-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Song H., Liu D., Wang L., Liu K., Chen C., Wang L., Ren Y., Ju B., Zhong F., Jiang X., et al. Methyltransferase like 7B is a potential therapeutic target for reversing EGFR-TKIs resistance in lung adenocarcinoma. Mol. Cancer. 2022;21:43. doi: 10.1186/s12943-022-01519-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Qin Z., Yue M., Tang S., Wu F., Sun H., Li Y., Zhang Y., Izumi H., Huang H., Wang W., et al. EML4-ALK fusions drive lung adeno-to-squamous transition through JAK-STAT activation. J. Exp. Med. 2024;221 doi: 10.1084/jem.20232028. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Speel E.-J.M., Dafni U., Thunnissen E., Hendrik Rüschoff J., O'Brien C., Kowalski J., Kerr K.M., Bubendorf L., Sansano I., Joseph L., et al. ROS1 fusions in resected stage I-III adenocarcinoma: Results from the European Thoracic Oncology Platform Lungscape project. Lung Cancer. 2024;194 doi: 10.1016/j.lungcan.2024.107860. [DOI] [PubMed] [Google Scholar]
23.Lee J.S., Nair N.U., Dinstag G., Chapman L., Chung Y., Wang K., Sinha S., Cha H., Kim D., Schperberg A.V., et al. Synthetic lethality-mediated precision oncology via the tumor transcriptome. Cell. 2021;184:2487–2502.e13. doi: 10.1016/j.cell.2021.03.030. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Wang G., Cao R., Wang Y., Qian G., Dan H.C., Jiang W., Ju L., Wu M., Xiao Y., Wang X. Simvastatin induces cell cycle arrest and inhibits proliferation of bladder cancer cells via PPARγ signalling pathway. Sci. Rep. 2016;6 doi: 10.1038/srep35783. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.O'Connor M.J. Targeting the DNA Damage Response in Cancer. Mol. Cell. 2015;60:547–560. doi: 10.1016/j.molcel.2015.10.040. [DOI] [PubMed] [Google Scholar]
26.Dagogo-Jack I., Shaw A.T. Tumour heterogeneity and resistance to cancer therapies. Nat. Rev. Clin. Oncol. 2018;15:81–94. doi: 10.1038/nrclinonc.2017.166. [DOI] [PubMed] [Google Scholar]
27.Shen Z. Genomic instability and cancer: an introduction. J. Mol. Cell Biol. 2011;3:1–3. doi: 10.1093/jmcb/mjq057. [DOI] [PubMed] [Google Scholar]
28.Gaignard P., Menezes M., Schiff M., Bayot A., Rak M., Ogier de Baulny H., Su C.H., Gilleron M., Lombes A., Abida H., et al. Mutations in CYC1, encoding cytochrome c1 subunit of respiratory chain complex III, cause insulin-responsive hyperglycemia. Am. J. Hum. Genet. 2013;93:384–389. doi: 10.1016/j.ajhg.2013.06.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Schlattner U., Tokarska-Schlattner M., Epand R.M., Boissan M., Lacombe M.-L., Kagan V.E. NME4/nucleoside diphosphate kinase D in cardiolipin signaling and mitophagy. Lab. Invest. 2018;98:228–232. doi: 10.1038/labinvest.2017.113. [DOI] [PubMed] [Google Scholar]
30.Chiang Y.L., King T.E. Cytochrome c1 complexes. J. Biol. Chem. 1979;254:1845–1853. [PubMed] [Google Scholar]
31.Chishiki M., Takagi K., Sato A., Miki Y., Yamamoto Y., Ebata A., Shibahara Y., Watanabe M., Ishida T., Sasano H., Suzuki T. Cytochrome c1 in ductal carcinoma in situ of breast associated with proliferation and comedo necrosis. Cancer Sci. 2017;108:1510–1519. doi: 10.1111/cas.13251. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Sato A., Takagi K., Miki Y., Yoshimura A., Hara M., Ishida T., Sasano H., Suzuki T. Cytochrome c1 as a favorable prognostic marker in estrogen receptor-positive breast carcinoma. Histol. Histopathol. 2019;34:1365–1375. doi: 10.14670/HH-18-130. [DOI] [PubMed] [Google Scholar]
33.Shi X., DeCiucis M., Grabinska K.A., Kanyo J., Liu A., Lam T.T., Shen H. Dual regulation of SLC25A39 by AFG3L2 and iron controls mitochondrial glutathione homeostasis. Mol. Cell. 2024;84:802–810.e6. doi: 10.1016/j.molcel.2023.12.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Chen X., Gan B. SLC25A39 links mitochondrial GSH sensing with iron metabolism. Mol. Cell. 2024;84:616–618. doi: 10.1016/j.molcel.2023.12.037. [DOI] [PubMed] [Google Scholar]
35.Zhang W., Ou Z., Tang T., Yang T., Li Y., Wu H., Li L., Liu M., Niu L., Zhu J. Up-regulated SLC25A39 promotes cell growth and metastasis via regulating ROS production in colorectal cancer. J. Cancer. 2024;15:5841–5854. doi: 10.7150/jca.98844. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Lacombe M.-L., Tokarska-Schlattner M., Boissan M., Schlattner U. The mitochondrial nucleoside diphosphate kinase (NDPK-D/NME4), a moonlighting protein for cell homeostasis. Lab. Invest. 2018;98:582–588. doi: 10.1038/s41374-017-0004-5. [DOI] [PubMed] [Google Scholar]
37.Lacombe M.-L., Lamarche F., De Wever O., Padilla-Benavides T., Carlson A., Khan I., Huna A., Vacher S., Calmel C., Desbourdes C., et al. The mitochondrially-localized nucleoside diphosphate kinase D (NME4) is a novel metastasis suppressor. BMC Biol. 2021;19:228. doi: 10.1186/s12915-021-01155-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Ferrucci V., Lomada S., Wieland T., Zollo M. PRUNE1 and NME/NDPK family proteins influence energy metabolism and signaling in cancer metastases. Cancer Metastasis Rev. 2024;43:755–775. doi: 10.1007/s10555-023-10165-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Zhang L., Cui Y., Zhou G., Zhang Z., Zhang P. Leveraging mitochondrial-programmed cell death dynamics to enhance prognostic accuracy and immunotherapy efficacy in lung adenocarcinoma. J. Immunother. Cancer. 2024;12 doi: 10.1136/jitc-2024-010008. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Wang W., Dong M., Cui J., Xu F., Yan C., Ma C., Yi L., Tang W., Dong J., Wei Y. NME4 may enhance non-small cell lung cancer progression by overcoming cell cycle arrest and promoting cellular proliferation. Mol. Med. Rep. 2019;20:1629–1636. doi: 10.3892/mmr.2019.10413. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Zhang R., Li C., Zhang S., Kong L., Liu Z., Guo Y., Sun Y., Zhang C., Yong Y., Lv J., et al. UBE2S promotes glycolysis in hepatocellular carcinoma by enhancing E3 enzyme-independent polyubiquitination of VHL. Clin. Mol. Hepatol. 2024;30:771–792. doi: 10.3350/cmh.2024.0236. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Guo Y., Chen X., Zhang X., Hu X. UBE2S and UBE2C confer a poor prognosis to breast cancer via downregulation of Numb. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.992233. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Hu W., Li M., Chen Y., Gu X. UBE2S promotes the progression and Olaparib resistance of ovarian cancer through Wnt/β-catenin signaling pathway. J. Ovarian Res. 2021;14:121. doi: 10.1186/s13048-021-00877-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Xiao K., Peng S., Lu J., Zhou T., Hong X., Chen S., Liu G., Li H., Huang J., Chen X., Lin T. UBE2S interacting with TRIM21 mediates the K11-linked ubiquitination of LPP to promote the lymphatic metastasis of bladder cancer. Cell Death Dis. 2023;14:408. doi: 10.1038/s41419-023-05938-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Sun D., Duan X., Li N., Qiao O., Hou Y., Ma Z., Liu S., Gong Y., Liu Z. Construction of ubiquitination-related risk model for predicting prognosis in lung adenocarcinoma. Sci. Rep. 2025;15 doi: 10.1038/s41598-025-92177-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Du Z., Zhang T., Lin Y., Dong G., Li A., Wang Z., Zhang Y., Giamas G., Stebbing J., Zhu L., Peng L. A prognostic model of drug tolerant persister-related genes in lung adenocarcinoma based on single cell and bulk RNA sequencing data. Heliyon. 2023;9 doi: 10.1016/j.heliyon.2023.e20708. [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Vizán P., Di Croce L., Aranda S. Functional and Pathological Roles of AHCY. Front. Cell Dev. Biol. 2021;9 doi: 10.3389/fcell.2021.654344. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Bondarev N., Ivanenko K., Khabusheva E., Lebedev T., Manukhov I., Prassolov V. MGL S3 Chimeric Enzyme Drives Apoptotic Death of EGFR-Dependent Cancer Cells through ERK Downregulation. Int. J. Mol. Sci. 2022;23 doi: 10.3390/ijms232112807. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Gmeiner W.H., Reinhold W.C., Pommier Y. Genome-wide mRNA and microRNA profiling of the NCI 60 cell-line screen and comparison of FdUMP[10] with fluorouracil, floxuridine, and topoisomerase 1 poisons. Mol. Cancer Ther. 2010;9:3105–3114. doi: 10.1158/1535-7163.MCT-10-0674. [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Yu S., Zhao Y., Liu Q., Wang J., Fu J., Li R., Yuan Y., Yan X., Su J. Spermidine synthase promotes liver cancer progression in a paracrine manner by altering the macrophage immunometabolic state. Bioorg. Chem. 2025;155 doi: 10.1016/j.bioorg.2025.108135. [DOI] [PubMed] [Google Scholar]
51.Miao H., Ou J., Peng Y., Zhang X., Chen Y., Hao L., Xie G., Wang Z., Pang X., Ruan Z., et al. Macrophage ABHD5 promotes colorectal cancer growth by suppressing spermidine production by SRM. Nat. Commun. 2016;7 doi: 10.1038/ncomms11716. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Zhao J., Xu Y. PITX1 plays essential functions in cancer. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1253238. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Huang Z., Liu D., Zhang Y., Lu W., Hu L., Zhang J., Xie L., Chen S. PITX1 as a grading, prognostic and tumor-infiltrating immune cells marker for chondrosarcoma: a public database-based immunoassay and tissue sample analysis. Front. Oncol. 2025;15 doi: 10.3389/fonc.2025.1477649. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Barton A.R., Sherman M.A., Mukamel R.E., Loh P.-R. Whole-exome imputation within UK Biobank powers rare coding variant association and fine-mapping analyses. Nat. Genet. 2021;53:1260–1269. doi: 10.1038/s41588-021-00892-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Cheluvappa R. Identification of New Potential Therapies for Colitis Amelioration Using an Appendicitis-Appendectomy Model. Inflamm. Bowel Dis. 2019;25:436–444. doi: 10.1093/ibd/izy332. [DOI] [PubMed] [Google Scholar]
56.Latif F., Duh F.M., Bader S., Sekido Y., Li H., Geil L., Zbar B., Minna J.D., Lerman M.I. The human homolog of the rodent immediate early response genes, PC4 and TIS7, resides in the lung cancer tumor suppressor gene region on chromosome 3p21. Hum. Genet. 1997;99:334–341. doi: 10.1007/s004390050368. [DOI] [PubMed] [Google Scholar]
57.Gong Z., Zhang X., Cui J., Chen W., Huang X., Yang Q., Li T., Zhang W. IFRD2, a target of miR-2400, regulates myogenic differentiation of bovine skeletal muscle satellite cells via decreased phosphorylation of ERK1/2 proteins. J. Muscle Res. Cell Motil. 2024;45:253–262. doi: 10.1007/s10974-024-09677-5. [DOI] [PubMed] [Google Scholar]
58.Kim N., Kim H.K., Lee K., Hong Y., Cho J.H., Choi J.W., Lee J.I., Suh Y.L., Ku B.M., Eum H.H., et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. Nat. Commun. 2020;11:2285. doi: 10.1038/s41467-020-16164-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Zhu J., Fan Y., Xiong Y., Wang W., Chen J., Xia Y., Lei J., Gong L., Sun S., Jiang T. Delineating the dynamic evolution from preneoplasia to invasive lung adenocarcinoma by integrating single-cell RNA sequencing and spatial transcriptomics. Exp. Mol. Med. 2022;54:2060–2076. doi: 10.1038/s12276-022-00896-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Der S.D., Sykes J., Pintilie M., Zhu C.Q., Strumpf D., Liu N., Jurisica I., Shepherd F.A., Tsao M.S. Validation of a histology-independent prognostic gene signature for early-stage, non-small-cell lung cancer including stage IA patients. J. Thorac. Oncol. 2014;9:59–64. doi: 10.1097/JTO.0000000000000042. [DOI] [PubMed] [Google Scholar]
61.Okayama H., Kohno T., Ishii Y., Shimada Y., Shiraishi K., Iwakawa R., Furuta K., Tsuta K., Shibata T., Yamamoto S., et al. Identification of genes upregulated in ALK-positive and EGFR/KRAS/ALK-negative lung adenocarcinomas. Cancer Res. 2012;72:100–111. doi: 10.1158/0008-5472.CAN-11-1403. [DOI] [PubMed] [Google Scholar]
62.Bild A.H., Yao G., Chang J.T., Wang Q., Potti A., Chasse D., Joshi M.B., Harpole D., Lancaster J.M., Berchuck A., et al. Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature. 2006;439:353–357. doi: 10.1038/nature04296. [DOI] [PubMed] [Google Scholar]
63.Butler A., Hoffman P., Smibert P., Papalexi E., Satija R. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 2018;36:411–420. doi: 10.1038/nbt.4096. [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Korsunsky I., Millard N., Fan J., Slowikowski K., Zhang F., Wei K., Baglaenko Y., Brenner M., Loh P.R., Raychaudhuri S. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods. 2019;16:1289–1296. doi: 10.1038/s41592-019-0619-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Long E., Yin J., Shin J.H., Li Y., Li B., Kane A., Patel H., Sun X., Wang C., Luong T., et al. Context-aware single-cell multiomics approach identifies cell-type-specific lung cancer susceptibility genes. Nat. Commun. 2024;15:7995. doi: 10.1038/s41467-024-52356-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Li C., Hao R., Li C., Liu L., Ding Z. Integration of single-cell and bulk RNA sequencing data using machine learning identifies oxidative stress-related genes LUM and PCOLCE2 as potential biomarkers for heart failure. Int. J. Biol. Macromol. 2025;300 doi: 10.1016/j.ijbiomac.2025.140793. [DOI] [PubMed] [Google Scholar]
67.Fan C., Chen F., Chen Y., Huang L., Wang M., Liu Y., Wang Y., Guo H., Zheng N., Liu Y., et al. irGSEA: the integration of single-cell rank-based gene set enrichment analysis. Brief. Bioinform. 2024;25 doi: 10.1093/bib/bbae243. [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Liu J., Shi Y., Zhang Y. Multi-omics identification of an immunogenic cell death-related signature for clear cell renal cell carcinoma in the context of 3P medicine and based on a 101-combination machine learning computational framework. EPMA J. 2023;14:275–305. doi: 10.1007/s13167-023-00327-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Luo L., Yang P., Mastoraki S., Rao X., Wang Y., Kettner N.M., Raghavendra A.S., Tripathy D., Damodaran S., Hunt K.K., et al. Single-cell RNA sequencing identifies molecular biomarkers predicting late progression to CDK4/6 inhibition in patients with HR+/HER2- metastatic breast cancer. Mol. Cancer. 2025;24:48. doi: 10.1186/s12943-025-02226-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Gulati G.S., D'Silva J.P., Liu Y., Wang L., Newman A.M. Profiling cell identity and tissue architecture with single-cell and spatial transcriptomics. Nat. Rev. Mol. Cell Biol. 2025;26:11–31. doi: 10.1038/s41580-024-00768-2. [DOI] [PubMed] [Google Scholar]
71.Ma Y., Deng C., Zhou Y., Zhang Y., Qiu F., Jiang D., Zheng G., Li J., Shuai J., Zhang Y., et al. Polygenic regression uncovers trait-relevant cellular contexts through pathway activation transformation of single-cell RNA sequencing data. Cell Genom. 2023;3 doi: 10.1016/j.xgen.2023.100383. [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Jin S., Plikus M.V., Nie Q. CellChat for systematic analysis of cell-cell communication from single-cell transcriptomics. Nat. Protoc. 2025;20:180–219. doi: 10.1038/s41596-024-01045-4. [DOI] [PubMed] [Google Scholar]
73.Dai L., Fan G., Xie T., Li L., Tang L., Chen H., Shi Y., Han X. Single-cell and spatial transcriptomics reveal a high glycolysis B cell and tumor-associated macrophages cluster correlated with poor prognosis and exhausted immune microenvironment in diffuse large B-cell lymphoma. Biomark. Res. 2024;12:58. doi: 10.1186/s40364-024-00605-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Huang R.H., Hong Y.K., Du H., Ke W.Q., Lin B.B., Li Y.L. A machine learning framework develops a DNA replication stress model for predicting clinical outcomes and therapeutic vulnerability in primary prostate cancer. J. Transl. Med. 2023;21:20. doi: 10.1186/s12967-023-03872-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Qi X., Wang S., Fang C., Jia J., Lin L., Yuan T. Machine learning and SHAP value interpretation for predicting comorbidity of cardiovascular disease and cancer with dietary antioxidants. Redox Biol. 2025;79 doi: 10.1016/j.redox.2024.103470. [DOI] [PMC free article] [PubMed] [Google Scholar]
76.Fustero-Torre C., Jiménez-Santos M.J., García-Martín S., Carretero-Puche C., García-Jimeno L., Ivanchuk V., Di Domenico T., Gómez-López G., Al-Shahrour F. Beyondcell: targeting cancer therapeutic heterogeneity in single-cell RNA-seq data. Genome Med. 2021;13:187. doi: 10.1186/s13073-021-01001-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Document S1. Figures S1 and S2

mmc1.pdf^{(669.8KB, pdf)}

Data S1. Uncropped western blots corresponding to Figures 9–12

mmc2.pdf^{(200.6KB, pdf)}

Table S1. Top 100 genes in each HSL cell phenotype-related module with module assignment and kME values

mmc3.xlsx^{(37.8KB, xlsx)}

Table S2. Upregulated genes in HSL cells with differential expression statistics (p value, log₂FC, percentage expressed, and adjusted p value)

mmc4.xlsx^{(40.9KB, xlsx)}

Table S3. 75 potential signature genes identified by integrative analysis across multiple selection methods

mmc5.xlsx^{(27.2KB, xlsx)}

Table S4. 71 signature genes positively correlated with HSL cell abundance, along with correlation coefficients and p values

mmc6.xlsx^{(28.2KB, xlsx)}

Table S5. Signature genes selected using the ABESS model

mmc7.xlsx^{(26.6KB, xlsx)}

Table S6. Signature genes selected using the random forest model, with their importance scores

mmc8.xlsx^{(27.7KB, xlsx)}

Table S7. Signature genes selected using the gradient boosting machine model, with their importance scores

mmc9.xlsx^{(28.2KB, xlsx)}

Table S8. Signature genes selected using the decision tree model, with their overall importance values

mmc10.xlsx^{(27.8KB, xlsx)}

Table S9. Signature genes selected using LASSO regression, with their regression coefficients

mmc11.xlsx^{(27.5KB, xlsx)}

Table S10. Detailed information of all enrolled datasets, including disease type, sample size, survival data, data type, and accession URLs

mmc12.xlsx^{(28.1KB, xlsx)}

Table S11. 146 synthetic lethality-related genes retrieved from the SynLethDB database

mmc13.xlsx^{(27.8KB, xlsx)}

Table S12. Primer sequences used for qPCR validation experiments

mmc14.xlsx^{(27.7KB, xlsx)}

Table S13. siRNA duplex sequences (sense and antisense) targeting IFRD2

mmc15.xlsx^{(27.3KB, xlsx)}

Data Availability Statement

•
This paper analyzes existing, publicly available data, accessible at TCGA (https://portal.gdc.cancer.gov/), genome-wide association study (GWAS, https://opengwas.io/datasets/), and Gene expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/).
•
This paper does not report original code.
•
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

[bib1] 1.Wang M., Herbst R.S., Boshoff C. Toward personalized treatment approaches for non-small-cell lung cancer. Nat. Med. 2021;27:1345–1356. doi: 10.1038/s41591-021-01450-2. [DOI] [PubMed] [Google Scholar]

[bib2] 2.Travis W.D., Garg K., Franklin W.A., Wistuba I.I., Sabloff B., Noguchi M., Kakinuma R., Zakowski M., Ginsberg M., Padera R., et al. Evolving concepts in the pathology and computed tomography imaging of lung adenocarcinoma and bronchioloalveolar carcinoma. J. Clin. Oncol. 2005;23:3279–3287. doi: 10.1200/JCO.2005.15.776. [DOI] [PubMed] [Google Scholar]

[bib3] 3.Shergalis A.G., Hu S., Bankhead A., Neamati N. Role of the ERO1-PDI interaction in oxidative protein folding and disease. Pharmacol. Ther. 2020;210 doi: 10.1016/j.pharmthera.2020.107525. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib4] 4.Guo H., Zhang J., Qin C., Yan H., Liu T., Hu H., Tang S., Tang S., Zhou H. Biomarker-Targeted Therapies in Non-Small Cell Lung Cancer: Current Status and Perspectives. Cells. 2022;11:3200. doi: 10.3390/cells11203200. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib5] 5.Spella M., Stathopoulos G.T. Immune Resistance in Lung Adenocarcinoma. Cancers (Basel) 2021;13 doi: 10.3390/cancers13030384. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib6] 6.Zeng J., Ding X., Ding J., Wang X. Histological transformation into SCLC: An important resistance mechanism of NSCLC upon immunotherapy. Front. Immunol. 2023;14 doi: 10.3389/fimmu.2023.1275957. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib7] 7.Lord C.J., Ashworth A. PARP inhibitors: Synthetic lethality in the clinic. Science. 2017;355:1152–1158. doi: 10.1126/science.aam7344. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] 8.Murai J., Pommier Y. BRCAness, Homologous Recombination Deficiencies, and Synthetic Lethality. Cancer Res. 2023;83:1173–1174. doi: 10.1158/0008-5472.CAN-23-0628. [DOI] [PubMed] [Google Scholar]

[bib9] 9.Ryan C.J., Devakumar L.P.S., Pettitt S.J., Lord C.J. Complex synthetic lethality in cancer. Nat. Genet. 2023;55:2039–2048. doi: 10.1038/s41588-023-01557-x. [DOI] [PubMed] [Google Scholar]

[bib10] 10.Setton J., Zinda M., Riaz N., Durocher D., Zimmermann M., Koehler M., Reis-Filho J.S., Powell S.N. Synthetic Lethality in Cancer Therapeutics: The Next Generation. Cancer Discov. 2021;11:1626–1635. doi: 10.1158/2159-8290.CD-20-1503. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] 11.Huang A., Garraway L.A., Ashworth A., Weber B. Synthetic lethality as an engine for cancer drug target discovery. Nat. Rev. Drug Discov. 2020;19:23–38. doi: 10.1038/s41573-019-0046-z. [DOI] [PubMed] [Google Scholar]

[bib12] 12.Parameswaran S., Kundapur D., Vizeacoumar F.S., Freywald A., Uppalapati M., Vizeacoumar F.J. A Road Map to Personalizing Targeted Cancer Therapies Using Synthetic Lethality. Trends Cancer. 2019;5:11–29. doi: 10.1016/j.trecan.2018.11.001. [DOI] [PubMed] [Google Scholar]

[bib13] 13.Ye B.-J., Li D.-F., Li X.-Y., Hao J.L., Liu D.J., Yu H., Zhang C.D. Methylation synthetic lethality: Exploiting selective drug targets for cancer therapy. Cancer Lett. 2024;597 doi: 10.1016/j.canlet.2024.217010. [DOI] [PubMed] [Google Scholar]

[bib14] 14.Li Z., Qiao W., Yu S., Fan B., Yang M., Wu M., Qiu F., Wang J. Integrating computational pathology and multi-transcriptomics to characterize lung adenocarcinoma heterogeneity and prognostic modeling. Int. J. Surg. 2025;111:5162–5181. doi: 10.1097/JS9.0000000000002639. [DOI] [PubMed] [Google Scholar]

[bib15] 15.Agha R.A., Mathew G., Rashid R., Kerwan A. Transparency In The reporting of Artificial INtelligence – the TITAN guideline. Premier J. Sci. 2025;10 [Google Scholar]

[bib16] 16.Xu G.B., Yang L.Q., Guan P.P., Wang Z.Y., Wang P. Prostaglandin A1 Inhibits the Cognitive Decline of APP/PS1 Transgenic Mice via PPARγ/ABCA1-dependent Cholesterol Efflux Mechanisms. Neurotherapeutics. 2019;16:505–522. doi: 10.1007/s13311-018-00704-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib17] 17.Yang G., Chen L. An Update of Microsomal Prostaglandin E Synthase-1 and PGE2 Receptors in Cardiovascular Health and Diseases. Oxid. Med. Cell. Longev. 2016;2016:5249086–5249089. doi: 10.1155/2016/5249086. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib18] 18.Tanikawa M., Yamada K., Tominaga K., Morisaki H., Kaneko Y., Ikeda K., Suzuki M., Kiho T., Tomokiyo K., Furuta K., et al. Potent prostaglandin A1 analogs that suppress tumor cell growth through induction of p21 and reduction of cyclin E. J. Biol. Chem. 1998;273:18522–18527. doi: 10.1074/jbc.273.29.18522. [DOI] [PubMed] [Google Scholar]

[bib19] 19.Coudray N., Ocampo P.S., Sakellaropoulos T., Narula N., Snuderl M., Fenyö D., Moreira A.L., Razavian N., Tsirigos A. Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning. Nat. Med. 2018;24:1559–1567. doi: 10.1038/s41591-018-0177-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib20] 20.Song H., Liu D., Wang L., Liu K., Chen C., Wang L., Ren Y., Ju B., Zhong F., Jiang X., et al. Methyltransferase like 7B is a potential therapeutic target for reversing EGFR-TKIs resistance in lung adenocarcinoma. Mol. Cancer. 2022;21:43. doi: 10.1186/s12943-022-01519-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib21] 21.Qin Z., Yue M., Tang S., Wu F., Sun H., Li Y., Zhang Y., Izumi H., Huang H., Wang W., et al. EML4-ALK fusions drive lung adeno-to-squamous transition through JAK-STAT activation. J. Exp. Med. 2024;221 doi: 10.1084/jem.20232028. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib22] 22.Speel E.-J.M., Dafni U., Thunnissen E., Hendrik Rüschoff J., O'Brien C., Kowalski J., Kerr K.M., Bubendorf L., Sansano I., Joseph L., et al. ROS1 fusions in resected stage I-III adenocarcinoma: Results from the European Thoracic Oncology Platform Lungscape project. Lung Cancer. 2024;194 doi: 10.1016/j.lungcan.2024.107860. [DOI] [PubMed] [Google Scholar]

[bib23] 23.Lee J.S., Nair N.U., Dinstag G., Chapman L., Chung Y., Wang K., Sinha S., Cha H., Kim D., Schperberg A.V., et al. Synthetic lethality-mediated precision oncology via the tumor transcriptome. Cell. 2021;184:2487–2502.e13. doi: 10.1016/j.cell.2021.03.030. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib24] 24.Wang G., Cao R., Wang Y., Qian G., Dan H.C., Jiang W., Ju L., Wu M., Xiao Y., Wang X. Simvastatin induces cell cycle arrest and inhibits proliferation of bladder cancer cells via PPARγ signalling pathway. Sci. Rep. 2016;6 doi: 10.1038/srep35783. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib25] 25.O'Connor M.J. Targeting the DNA Damage Response in Cancer. Mol. Cell. 2015;60:547–560. doi: 10.1016/j.molcel.2015.10.040. [DOI] [PubMed] [Google Scholar]

[bib26] 26.Dagogo-Jack I., Shaw A.T. Tumour heterogeneity and resistance to cancer therapies. Nat. Rev. Clin. Oncol. 2018;15:81–94. doi: 10.1038/nrclinonc.2017.166. [DOI] [PubMed] [Google Scholar]

[bib27] 27.Shen Z. Genomic instability and cancer: an introduction. J. Mol. Cell Biol. 2011;3:1–3. doi: 10.1093/jmcb/mjq057. [DOI] [PubMed] [Google Scholar]

[bib28] 28.Gaignard P., Menezes M., Schiff M., Bayot A., Rak M., Ogier de Baulny H., Su C.H., Gilleron M., Lombes A., Abida H., et al. Mutations in CYC1, encoding cytochrome c1 subunit of respiratory chain complex III, cause insulin-responsive hyperglycemia. Am. J. Hum. Genet. 2013;93:384–389. doi: 10.1016/j.ajhg.2013.06.015. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib29] 29.Schlattner U., Tokarska-Schlattner M., Epand R.M., Boissan M., Lacombe M.-L., Kagan V.E. NME4/nucleoside diphosphate kinase D in cardiolipin signaling and mitophagy. Lab. Invest. 2018;98:228–232. doi: 10.1038/labinvest.2017.113. [DOI] [PubMed] [Google Scholar]

[bib30] 30.Chiang Y.L., King T.E. Cytochrome c1 complexes. J. Biol. Chem. 1979;254:1845–1853. [PubMed] [Google Scholar]

[bib31] 31.Chishiki M., Takagi K., Sato A., Miki Y., Yamamoto Y., Ebata A., Shibahara Y., Watanabe M., Ishida T., Sasano H., Suzuki T. Cytochrome c1 in ductal carcinoma in situ of breast associated with proliferation and comedo necrosis. Cancer Sci. 2017;108:1510–1519. doi: 10.1111/cas.13251. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib32] 32.Sato A., Takagi K., Miki Y., Yoshimura A., Hara M., Ishida T., Sasano H., Suzuki T. Cytochrome c1 as a favorable prognostic marker in estrogen receptor-positive breast carcinoma. Histol. Histopathol. 2019;34:1365–1375. doi: 10.14670/HH-18-130. [DOI] [PubMed] [Google Scholar]

[bib33] 33.Shi X., DeCiucis M., Grabinska K.A., Kanyo J., Liu A., Lam T.T., Shen H. Dual regulation of SLC25A39 by AFG3L2 and iron controls mitochondrial glutathione homeostasis. Mol. Cell. 2024;84:802–810.e6. doi: 10.1016/j.molcel.2023.12.008. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib34] 34.Chen X., Gan B. SLC25A39 links mitochondrial GSH sensing with iron metabolism. Mol. Cell. 2024;84:616–618. doi: 10.1016/j.molcel.2023.12.037. [DOI] [PubMed] [Google Scholar]

[bib35] 35.Zhang W., Ou Z., Tang T., Yang T., Li Y., Wu H., Li L., Liu M., Niu L., Zhu J. Up-regulated SLC25A39 promotes cell growth and metastasis via regulating ROS production in colorectal cancer. J. Cancer. 2024;15:5841–5854. doi: 10.7150/jca.98844. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib36] 36.Lacombe M.-L., Tokarska-Schlattner M., Boissan M., Schlattner U. The mitochondrial nucleoside diphosphate kinase (NDPK-D/NME4), a moonlighting protein for cell homeostasis. Lab. Invest. 2018;98:582–588. doi: 10.1038/s41374-017-0004-5. [DOI] [PubMed] [Google Scholar]

[bib37] 37.Lacombe M.-L., Lamarche F., De Wever O., Padilla-Benavides T., Carlson A., Khan I., Huna A., Vacher S., Calmel C., Desbourdes C., et al. The mitochondrially-localized nucleoside diphosphate kinase D (NME4) is a novel metastasis suppressor. BMC Biol. 2021;19:228. doi: 10.1186/s12915-021-01155-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib38] 38.Ferrucci V., Lomada S., Wieland T., Zollo M. PRUNE1 and NME/NDPK family proteins influence energy metabolism and signaling in cancer metastases. Cancer Metastasis Rev. 2024;43:755–775. doi: 10.1007/s10555-023-10165-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib39] 39.Zhang L., Cui Y., Zhou G., Zhang Z., Zhang P. Leveraging mitochondrial-programmed cell death dynamics to enhance prognostic accuracy and immunotherapy efficacy in lung adenocarcinoma. J. Immunother. Cancer. 2024;12 doi: 10.1136/jitc-2024-010008. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib40] 40.Wang W., Dong M., Cui J., Xu F., Yan C., Ma C., Yi L., Tang W., Dong J., Wei Y. NME4 may enhance non-small cell lung cancer progression by overcoming cell cycle arrest and promoting cellular proliferation. Mol. Med. Rep. 2019;20:1629–1636. doi: 10.3892/mmr.2019.10413. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib41] 41.Zhang R., Li C., Zhang S., Kong L., Liu Z., Guo Y., Sun Y., Zhang C., Yong Y., Lv J., et al. UBE2S promotes glycolysis in hepatocellular carcinoma by enhancing E3 enzyme-independent polyubiquitination of VHL. Clin. Mol. Hepatol. 2024;30:771–792. doi: 10.3350/cmh.2024.0236. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib42] 42.Guo Y., Chen X., Zhang X., Hu X. UBE2S and UBE2C confer a poor prognosis to breast cancer via downregulation of Numb. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.992233. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib43] 43.Hu W., Li M., Chen Y., Gu X. UBE2S promotes the progression and Olaparib resistance of ovarian cancer through Wnt/β-catenin signaling pathway. J. Ovarian Res. 2021;14:121. doi: 10.1186/s13048-021-00877-y. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib44] 44.Xiao K., Peng S., Lu J., Zhou T., Hong X., Chen S., Liu G., Li H., Huang J., Chen X., Lin T. UBE2S interacting with TRIM21 mediates the K11-linked ubiquitination of LPP to promote the lymphatic metastasis of bladder cancer. Cell Death Dis. 2023;14:408. doi: 10.1038/s41419-023-05938-2. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib45] 45.Sun D., Duan X., Li N., Qiao O., Hou Y., Ma Z., Liu S., Gong Y., Liu Z. Construction of ubiquitination-related risk model for predicting prognosis in lung adenocarcinoma. Sci. Rep. 2025;15 doi: 10.1038/s41598-025-92177-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib46] 46.Du Z., Zhang T., Lin Y., Dong G., Li A., Wang Z., Zhang Y., Giamas G., Stebbing J., Zhu L., Peng L. A prognostic model of drug tolerant persister-related genes in lung adenocarcinoma based on single cell and bulk RNA sequencing data. Heliyon. 2023;9 doi: 10.1016/j.heliyon.2023.e20708. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib47] 47.Vizán P., Di Croce L., Aranda S. Functional and Pathological Roles of AHCY. Front. Cell Dev. Biol. 2021;9 doi: 10.3389/fcell.2021.654344. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib48] 48.Bondarev N., Ivanenko K., Khabusheva E., Lebedev T., Manukhov I., Prassolov V. MGL S3 Chimeric Enzyme Drives Apoptotic Death of EGFR-Dependent Cancer Cells through ERK Downregulation. Int. J. Mol. Sci. 2022;23 doi: 10.3390/ijms232112807. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib49] 49.Gmeiner W.H., Reinhold W.C., Pommier Y. Genome-wide mRNA and microRNA profiling of the NCI 60 cell-line screen and comparison of FdUMP[10] with fluorouracil, floxuridine, and topoisomerase 1 poisons. Mol. Cancer Ther. 2010;9:3105–3114. doi: 10.1158/1535-7163.MCT-10-0674. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib50] 50.Yu S., Zhao Y., Liu Q., Wang J., Fu J., Li R., Yuan Y., Yan X., Su J. Spermidine synthase promotes liver cancer progression in a paracrine manner by altering the macrophage immunometabolic state. Bioorg. Chem. 2025;155 doi: 10.1016/j.bioorg.2025.108135. [DOI] [PubMed] [Google Scholar]

[bib51] 51.Miao H., Ou J., Peng Y., Zhang X., Chen Y., Hao L., Xie G., Wang Z., Pang X., Ruan Z., et al. Macrophage ABHD5 promotes colorectal cancer growth by suppressing spermidine production by SRM. Nat. Commun. 2016;7 doi: 10.1038/ncomms11716. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib52] 52.Zhao J., Xu Y. PITX1 plays essential functions in cancer. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1253238. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib53] 53.Huang Z., Liu D., Zhang Y., Lu W., Hu L., Zhang J., Xie L., Chen S. PITX1 as a grading, prognostic and tumor-infiltrating immune cells marker for chondrosarcoma: a public database-based immunoassay and tissue sample analysis. Front. Oncol. 2025;15 doi: 10.3389/fonc.2025.1477649. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib54] 54.Barton A.R., Sherman M.A., Mukamel R.E., Loh P.-R. Whole-exome imputation within UK Biobank powers rare coding variant association and fine-mapping analyses. Nat. Genet. 2021;53:1260–1269. doi: 10.1038/s41588-021-00892-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib55] 55.Cheluvappa R. Identification of New Potential Therapies for Colitis Amelioration Using an Appendicitis-Appendectomy Model. Inflamm. Bowel Dis. 2019;25:436–444. doi: 10.1093/ibd/izy332. [DOI] [PubMed] [Google Scholar]

[bib56] 56.Latif F., Duh F.M., Bader S., Sekido Y., Li H., Geil L., Zbar B., Minna J.D., Lerman M.I. The human homolog of the rodent immediate early response genes, PC4 and TIS7, resides in the lung cancer tumor suppressor gene region on chromosome 3p21. Hum. Genet. 1997;99:334–341. doi: 10.1007/s004390050368. [DOI] [PubMed] [Google Scholar]

[bib57] 57.Gong Z., Zhang X., Cui J., Chen W., Huang X., Yang Q., Li T., Zhang W. IFRD2, a target of miR-2400, regulates myogenic differentiation of bovine skeletal muscle satellite cells via decreased phosphorylation of ERK1/2 proteins. J. Muscle Res. Cell Motil. 2024;45:253–262. doi: 10.1007/s10974-024-09677-5. [DOI] [PubMed] [Google Scholar]

[bib58] 58.Kim N., Kim H.K., Lee K., Hong Y., Cho J.H., Choi J.W., Lee J.I., Suh Y.L., Ku B.M., Eum H.H., et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. Nat. Commun. 2020;11:2285. doi: 10.1038/s41467-020-16164-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib59] 59.Zhu J., Fan Y., Xiong Y., Wang W., Chen J., Xia Y., Lei J., Gong L., Sun S., Jiang T. Delineating the dynamic evolution from preneoplasia to invasive lung adenocarcinoma by integrating single-cell RNA sequencing and spatial transcriptomics. Exp. Mol. Med. 2022;54:2060–2076. doi: 10.1038/s12276-022-00896-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib60] 60.Der S.D., Sykes J., Pintilie M., Zhu C.Q., Strumpf D., Liu N., Jurisica I., Shepherd F.A., Tsao M.S. Validation of a histology-independent prognostic gene signature for early-stage, non-small-cell lung cancer including stage IA patients. J. Thorac. Oncol. 2014;9:59–64. doi: 10.1097/JTO.0000000000000042. [DOI] [PubMed] [Google Scholar]

[bib61] 61.Okayama H., Kohno T., Ishii Y., Shimada Y., Shiraishi K., Iwakawa R., Furuta K., Tsuta K., Shibata T., Yamamoto S., et al. Identification of genes upregulated in ALK-positive and EGFR/KRAS/ALK-negative lung adenocarcinomas. Cancer Res. 2012;72:100–111. doi: 10.1158/0008-5472.CAN-11-1403. [DOI] [PubMed] [Google Scholar]

[bib62] 62.Bild A.H., Yao G., Chang J.T., Wang Q., Potti A., Chasse D., Joshi M.B., Harpole D., Lancaster J.M., Berchuck A., et al. Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature. 2006;439:353–357. doi: 10.1038/nature04296. [DOI] [PubMed] [Google Scholar]

[bib63] 63.Butler A., Hoffman P., Smibert P., Papalexi E., Satija R. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 2018;36:411–420. doi: 10.1038/nbt.4096. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib64] 64.Korsunsky I., Millard N., Fan J., Slowikowski K., Zhang F., Wei K., Baglaenko Y., Brenner M., Loh P.R., Raychaudhuri S. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods. 2019;16:1289–1296. doi: 10.1038/s41592-019-0619-0. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib65] 65.Long E., Yin J., Shin J.H., Li Y., Li B., Kane A., Patel H., Sun X., Wang C., Luong T., et al. Context-aware single-cell multiomics approach identifies cell-type-specific lung cancer susceptibility genes. Nat. Commun. 2024;15:7995. doi: 10.1038/s41467-024-52356-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib66] 66.Li C., Hao R., Li C., Liu L., Ding Z. Integration of single-cell and bulk RNA sequencing data using machine learning identifies oxidative stress-related genes LUM and PCOLCE2 as potential biomarkers for heart failure. Int. J. Biol. Macromol. 2025;300 doi: 10.1016/j.ijbiomac.2025.140793. [DOI] [PubMed] [Google Scholar]

[bib67] 67.Fan C., Chen F., Chen Y., Huang L., Wang M., Liu Y., Wang Y., Guo H., Zheng N., Liu Y., et al. irGSEA: the integration of single-cell rank-based gene set enrichment analysis. Brief. Bioinform. 2024;25 doi: 10.1093/bib/bbae243. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib68] 68.Liu J., Shi Y., Zhang Y. Multi-omics identification of an immunogenic cell death-related signature for clear cell renal cell carcinoma in the context of 3P medicine and based on a 101-combination machine learning computational framework. EPMA J. 2023;14:275–305. doi: 10.1007/s13167-023-00327-3. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib69] 69.Luo L., Yang P., Mastoraki S., Rao X., Wang Y., Kettner N.M., Raghavendra A.S., Tripathy D., Damodaran S., Hunt K.K., et al. Single-cell RNA sequencing identifies molecular biomarkers predicting late progression to CDK4/6 inhibition in patients with HR+/HER2- metastatic breast cancer. Mol. Cancer. 2025;24:48. doi: 10.1186/s12943-025-02226-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib70] 70.Gulati G.S., D'Silva J.P., Liu Y., Wang L., Newman A.M. Profiling cell identity and tissue architecture with single-cell and spatial transcriptomics. Nat. Rev. Mol. Cell Biol. 2025;26:11–31. doi: 10.1038/s41580-024-00768-2. [DOI] [PubMed] [Google Scholar]

[bib71] 71.Ma Y., Deng C., Zhou Y., Zhang Y., Qiu F., Jiang D., Zheng G., Li J., Shuai J., Zhang Y., et al. Polygenic regression uncovers trait-relevant cellular contexts through pathway activation transformation of single-cell RNA sequencing data. Cell Genom. 2023;3 doi: 10.1016/j.xgen.2023.100383. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib72] 72.Jin S., Plikus M.V., Nie Q. CellChat for systematic analysis of cell-cell communication from single-cell transcriptomics. Nat. Protoc. 2025;20:180–219. doi: 10.1038/s41596-024-01045-4. [DOI] [PubMed] [Google Scholar]

[bib73] 73.Dai L., Fan G., Xie T., Li L., Tang L., Chen H., Shi Y., Han X. Single-cell and spatial transcriptomics reveal a high glycolysis B cell and tumor-associated macrophages cluster correlated with poor prognosis and exhausted immune microenvironment in diffuse large B-cell lymphoma. Biomark. Res. 2024;12:58. doi: 10.1186/s40364-024-00605-w. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib74] 74.Huang R.H., Hong Y.K., Du H., Ke W.Q., Lin B.B., Li Y.L. A machine learning framework develops a DNA replication stress model for predicting clinical outcomes and therapeutic vulnerability in primary prostate cancer. J. Transl. Med. 2023;21:20. doi: 10.1186/s12967-023-03872-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib75] 75.Qi X., Wang S., Fang C., Jia J., Lin L., Yuan T. Machine learning and SHAP value interpretation for predicting comorbidity of cardiovascular disease and cancer with dietary antioxidants. Redox Biol. 2025;79 doi: 10.1016/j.redox.2024.103470. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib76] 76.Fustero-Torre C., Jiménez-Santos M.J., García-Martín S., Carretero-Puche C., García-Jimeno L., Ivanchuk V., Di Domenico T., Gómez-López G., Al-Shahrour F. Beyondcell: targeting cancer therapeutic heterogeneity in single-cell RNA-seq data. Genome Med. 2021;13:187. doi: 10.1186/s13073-021-01001-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Integrative multi-omics reveals a prognostic and therapeutic landscape of synthetic lethality-associated signatures in lung adenocarcinoma

Zerong Li

Wenmei Qiao

Bin Fan

Fang Qiu

Wei Su

Summary

Graphical abstract

Highlights

Introduction

Results

Single-cell transcriptomic profiling of LUAD

Figure 1.

SL activity is elevated in LUAD and enriched in malignant populations

Figure 2.

Heterogeneity of SL activity in LUAD

Figure 3.

Distinct cell-cell communication, functional states, and drug sensitivity profiles of HSL cells

Figure 4.

Uncovering LUAD driver gene modules through hdWGCNA

Figure 5.

Signature gene selection for HSL cells in LUAD

Figure 6.

Expression validation and prognostic analysis of HSL signature genes

Figure 7.

Machine learning benchmark for HSL cell prediction model

Figure 8.

Validation of HSL signature genes by RT-qPCR and western blotting

Figure 9.

IFRD2 deficiency alone is dispensable for cell proliferation and apoptosis

Figure 10.

Knockdown of IFRD2 attenuates sensitivity to synthetic lethal agents in LUAD cells

Figure 11.

Prostaglandin-A1 may be an effective therapy for LUAD

Figure 12.

Discussion

Limitations of the study

Resource availability

Lead contact

Materials availability

Data and code availability

Acknowledgments

Author contributions

Declaration of interests

Declaration of generative AI and AI-assisted technologies in the writing process

STAR★Methods

Key resources table

Experimental model and study participant details

Cell culture

Method details

Acquisition and processing of multi-omics data

Synthetic lethality activity scoring

CytoTRACE evaluation and scPagwas analysis

Cell–cell interaction analysis

HdWGCNA analysis

Machine learning-based identification of HSL cell signature genes

Machine learning benchmark and SHAP analysis

Drug sensitivity analysis

Experimental validation of signature gene expression

RNA transfection

CCK-8 assay

EdU staining

Wound healing assay

Quantification and statistical analysis

Footnotes

Supplemental information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases