Skip to main content
NCBI home page
Animal Models and Experimental Medicine logoLink to Animal Models and Experimental Medicine
. 2026 Jul 26:10.1002/ame2.70258. Online ahead of print. doi: 10.1002/ame2.70258

Systematic pathway‐level analysis defines conserved transcriptional divergence between primary lung tumors and cell line models

Pritam Bera 1, Rajesh Raju 1, Debodipta Das 1,✉
PMCID: PMC13402649  PMID: 42503847

Abstract

Background

Cancer cell lines are widely used in preclinical oncology, yet their fidelity in modeling primary tumors remains uncertain.

Methods

We performed a systematic pathway‐level transcriptomic comparison of lung adenocarcinoma (LUAD) primary tumors from The Cancer Genome Atlas (TCGA; n = 513) and LUAD cell lines from the Cancer Cell Line Encyclopedia (CCLE; n = 45). Single‐sample gene set enrichment analysis (ssGSEA) combined with L1‐penalized logistic regression was applied to identify a discriminatory pathway signature.

Results

Five Hallmark pathways (E2F targets, interferon‐gamma response, epithelial‐mesenchymal transition, coagulation, and G2M checkpoint) distinguished primary lung tumors from cell line models. Cell lines exhibited significant enrichment of proliferation‐associated pathways, whereas primary tumors retained immune‐ and microenvironment‐related programs. Tumor–normal comparisons revealed significant pathway alterations in TCGA samples that were not fully recapitulated in CCLE models. Gene‐level analyses confirmed depletion of stromal and vascular markers in cell lines. CCLE samples also demonstrated significantly higher tumor mutational burden and increased frequencies of TP53 and STK11 mutations. These pathway‐level distinctions were conserved in lung squamous cell carcinoma, indicating generalizability across major NSCLC subtypes.

Conclusion

These findings define a pathway‐based framework for evaluating transcriptional fidelity between cell lines and primary tumors, highlighting fundamental transcriptional divergence with implications for rational model selection in lung cancer research.

Keywords: cancer cell lines, LASSO feature selection, lung adenocarcinoma, pathway‐level transcriptomics, single‐sample GSEA (ssGSEA), tumor microenvironment divergence

Lung cancer cell lines diverge substantially from primary tumors at the transcriptional level. Using single‐sample gene set enrichment analysis and L1‐penalized feature selection across TCGA‐LUAD and CCLE‐LUAD, we identified five Hallmark pathways (E2F targets, G2M checkpoint, IFNγ response, coagulation, and EMT) that discriminated primary tumors from cell lines, with cell lines enriched for proliferative programs and tumors retaining immune, stromal, extracellular matrix, and vascular signatures. CCLE samples showed higher mutational burden with elevated TP53 and STK11 mutation frequencies. Similar pathway distinctions were seen in lung squamous cell carcinoma, supporting generalizability across NSCLC subtypes.

graphic file with name AME2-9999-0-g006.jpg

1. INTRODUCTION

Cancer cell lines are foundational tools in biomedical research, enabling mechanistic investigation, target validation, and preclinical therapeutic screening across diverse experimental platforms. 1 , 2 , 3 Despite their widespread use and substantial contributions to cancer biology, a persistent challenge remains: efficacy observed in cell line models may not reliably predict clinical outcomes. 1 , 4 , 5 Oncology drugs exhibit the highest attrition rate among all therapeutic areas, with only 6.7% achieving U.S. Food and Drug Administration (FDA) approval from phase I clinical trials. 6 These observations have prompted ongoing evaluation of how well cancer cell lines represent the biological complexity of human tumors.

The advent of large‐scale molecular profiling initiatives, most notably The Cancer Genome Atlas (TCGA) 7 and the Cancer Cell Line Encyclopedia (CCLE), 8 , 9 has allowed systematic comparisons between primary tumors and cancer cell lines. Several comparative analyses have documented compositional differences, noting that cell lines lack stromal, immune, and vascular components present in primary tumors and exhibit altered pathway‐level activity patterns. 10 , 11 , 12 These studies have advanced our understanding of tumor‐cell line relationships across multiple cancer types. Here, we sought to complement this knowledge to identify a minimal discriminatory signature that consistently distinguishes primary tumors from cell line models in lung adenocarcinoma.

Lung adenocarcinoma (LUAD), the most common histological subtype of lung cancer and the leading cause of cancer‐related mortality worldwide, 13 , 14 represents a particularly important context for understanding cell line fidelity. Comprehensive molecular characterization by TCGA identified key driver genes and pathway alterations in LUAD. 15 Despite advances in molecularly targeted therapies and immunotherapy, the prognosis for advanced lung cancer remains distressingly poor, with a low five‐year survival rate, 16 underscoring the continued need for effective therapeutic strategies.

Given the extensive reliance on cell lines in preclinical oncology, a systematic comparison of pathway‐level features between primary tumors and in vitro models is essential to guide their appropriate use. In this study, we performed pathway‐based transcriptomic analyses to identify a discriminatory set of pathways that distinguish primary lung adenocarcinoma samples from The Cancer Genome Atlas (TCGA‐LUAD) and lung adenocarcinoma cell lines from the Cancer Cell Line Encyclopedia (CCLE‐LUAD). By integrating single‐sample gene set enrichment analysis (ssGSEA) with L1‐penalized feature selection (LASSO), we identified five Hallmark gene sets that discriminated between primary tumor tissues and cell lines. These findings provide a pathway‐centric framework for understanding tumor–cell line divergence and may facilitate the rational selection of experimental models tailored to specific biological questions in lung cancer research.

2. METHODS

2.1. Data acquisition and preprocessing

Gene expression (RNA‐seq) data and accompanying annotations for lung adenocarcinoma (TCGA‐LUAD) were retrieved from the TOIL recompute dataset available via the TCGA Pan‐Cancer Atlas hub on the UCSC Xena platform (https://xena.ucsc.edu/). Corresponding gene expression and clinical annotation for LUAD cell lines (CCLE‐LUAD) were downloaded from the Dependency Map (DepMap) portal (Public 24Q4 release; https://depmap.org/portal/). Expression values were provided in transcripts per million (TPM) format and log2‐transformed using dataset‐specific pseudo‐counts (0.001 for TCGA‐LUAD and 1 for CCLE‐LUAD). For cross‐dataset comparability, only protein‐coding genes common to both datasets were retained, and expression values were uniformly converted to log2 (TPM + 1) prior to downstream analyses. The TCGA‐LUAD cohort comprised 513 primary tumor samples and 58 adjacent normal lung tissues, whereas the CCLE‐LUAD dataset included 45 lung adenocarcinoma cell lines and 6 normal lung‐derived cell lines.

Additionally, gene expression data for lung squamous cell carcinoma (TCGA‐LUSC) were obtained from the same TCGA Pan‐Cancer Atlas resource, and corresponding CCLE‐LUSC cell lines were retrieved from the DepMap portal (Public 24Q4 release). The TCGA‐LUSC dataset comprised 498 primary tumor samples and 50 adjacent normal tissues, and the CCLE‐LUSC dataset comprised 16 tumor cell lines and 6 normal lung cell lines. Expression data for LUSC samples were processed using the same normalization and transformation procedures described above for LUAD.

2.2. Pathway activity quantification by single‐sample gene set enrichment analysis (ssGSEA)

Pathway activity in TCGA‐LUAD and CCLE‐LUAD samples was assessed using single‐sample gene set enrichment analysis (ssGSEA) as implemented in the ssGSEA2 17 , 18 package (v 1.0.0) in R (v 4.5.1). The Hallmark gene set collection from the Molecular Signatures Database (MSigDB; n = 50) 19 served as the primary reference panel. In addition, an ontology‐based analysis was performed using canonical pathway gene sets from the KEGG database 20 (KEGG LEGACY subset) available in MSigDB. The ssGSEA2 package was run with default settings, specifying a minimum gene set overlap of five genes.

2.3. LASSO‐based pathway feature selection

To identify pathways that most effectively distinguished TCGA primary tumors from CCLE cell lines, we applied L1‐penalized logistic regression (Least Absolute Shrinkage and Selection Operator; LASSO) to the ssGSEA normalized enrichment score (NES) matrix derived from the Hallmark gene sets. The binary outcome variable indicated sample origin (TCGA tumor vs. CCLE cell line). An analogous analysis was performed independently using KEGG pathway sets.

To minimize bias arising from unequal sample sizes, a class‐balanced design was adopted. Specifically, TCGA‐LUAD primary tumors with matched adjacent normal lung tissues (n = 58) and CCLE‐LUAD tumor cell lines (n = 45) were included in the modeling. Samples were randomly divided into training (80%) and independent test (20%) sets using stratified sampling to preserve class proportions. Within the training set, a 10‐fold stratified cross‐validation (CV) strategy was implemented. For each outer fold, L1‐penalized logistic regression (LogisticRegressionCV, scikit‐learn v 1.6.1), 21 conducted an internal 10‐fold CV to select the optimal regularization parameter (C) by maximizing the receiver operating characteristic area under the curve (ROC‐AUC). Pathways with non‐zero coefficients were considered selected features, and only those identified in at least five of the ten outer folds were retained to ensure selection stability.

To assess the predictive performance of the LASSO‐derived pathway set (Hallmark), three supervised machine learning classifiers, Random Forest, 22 Support Vector Machine (SVM), 23 and Extreme Gradient Boosting (XGBoost), 24 were trained on the 80% training data and evaluated on the held‐out 20% test set. Model performance was quantified using accuracy, precision, recall, and F 1‐score. Detailed descriptions of hyperparameter optimization, optimal model parameters, and additional model evaluation analyses are provided in the Data S1. Owing to the limited size of the independent test partition, these classification results are presented as exploratory rather than definitive benchmarking.

2.4. Gene‐level differential expression analysis

Genes comprising the LASSO‐selected Hallmark pathways were extracted from the MSigDB Hallmark gene set collection. The union of these genes was used for downstream gene‐level analysis. Differential expression of log2 (TPM + 1) values was compared between TCGA‐LUAD primary tumors with available matched adjacent normal lung tissues (n = 58) and CCLE‐LUAD tumor cell lines (n = 45) using Welch's t‐test. Resulting p‐values were corrected for multiple testing using the Benjamini–Hochberg (BH) false discovery rate (FDR) procedure. 25 Genes with an FDR <0.05 and an absolute log2 fold change (|log2FC|) >1 were considered significantly differentially expressed.

2.5. Tumor mutational burden analysis and mutation frequency comparison

Annotated somatic mutation data for TCGA pan‐cancer 7 samples were retrieved as Mutation Annotation Format (MAF) files from the Genomic Data Commons (GDC) Pan‐Cancer Atlas portal. Corresponding MAF files for CCLE cell lines were obtained from the DepMap portal (DepMap Public 24Q4 release). Tumor Mutational Burden (TMB) was calculated for TCGA‐LUAD tumors and CCLE‐LUAD cell lines using the tmb() function in the maftools R package (v 2.24.0), which estimates the number of non‐synonymous somatic mutations per megabase of coding region. Differences in TMB distributions between the two groups were evaluated using a two‐sided Wilcoxon rank‐sum test.

For gene‐level analyses, seven recurrent driver genes mutated in at least 10% of TCGA‐LUAD tumors 15 (i.e., TP53, KRAS, EGFR, KEAP1, STK11, NF1, and BRAF) were selected for targeted comparison. Differences in mutation frequencies between TCGA‐LUAD and CCLE‐LUAD samples were assessed using two‐sided Fisher's exact tests, with p‐values adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method.

3. RESULTS

3.1. Pathway‐level signatures distinguish primary lung adenocarcinoma from cancer cell lines

To determine whether transcriptomic profiles of CCLE‐LUAD cell lines recapitulate those of primary TCGA‐LUAD tumors at the pathway level, we performed single‐sample gene set enrichment analysis (ssGSEA) on 50 MSigDB Hallmark gene sets across 58 TCGA tumor samples and 45 CCLE cell lines. Principal component analysis (PCA) of the resulting normalized enrichment scores (NES) demonstrated distinct separation between TCGA and CCLE samples (Figure 1A), suggesting systematic differences in pathway‐level transcriptional programs between primary tumors and in vitro models.

FIGURE 1.

FIGURE 1

Open in a new tab

Pathway‐level transcriptomic profiles discriminate primary lung adenocarcinoma tumors from cancer cell lines. (A) Principal component analysis (PCA) of ssGSEA‐derived normalized enrichment scores (NES) for 50 MSigDB Hallmark gene sets across TCGA‐LUAD paired primary tumors (blue, n = 58) and CCLE‐LUAD cell lines (red, n = 45), demonstrating segregation between the two cohorts. (B) Mean LASSO coefficients (across selected cross‐validation folds) highlighting five pathways consistently retained in ≥5 folds as discriminative features. (C) Confusion matrix for Random Forest, Support Vector Machine (SVM), and XGBoost classifiers trained on the five LASSO‐selected gene sets and evaluated on an independent held‐out test set (n = 21).

To identify a pathway signature driving this separation, we applied L1‐penalized logistic regression with nested cross‐validation to the ssGSEA‐derived NES of the 50 Hallmark gene sets. Five pathways were reproducibly selected across cross‐validation folds: E2F targets, interferon‐gamma (IFNγ) response, epithelial‐mesenchymal transition (EMT), coagulation, and G2M checkpoint (Figure 1B). Using these five gene sets as features for supervised classification, the held‐out test set (n = 21) demonstrated consistent separation between TCGA and CCLE samples and comparable predictive performance across Random Forest, Support Vector Machine, and XGBoost classifiers (Figure 1C and Figure S1). Together, these findings indicate that this pathway signature captures fundamental transcriptional differences between primary LUAD tumors and corresponding cell line models.

3.2. LASSO‐selected pathways exhibit distinct activity patterns between primary tumors and cancer cell lines

To further characterize the five discriminative pathways, we compared their ssGSEA‐derived normalized enrichment scores (NES) between TCGA‐LUAD tumors (n = 58; included in the LASSO analysis) and CCLE‐LUAD tumor cell lines (n = 45). All five pathways differed significantly between the two groups (Figure 2A). G2M checkpoint and E2F targets were significantly enriched in CCLE samples (FDR = 5.55 × 10−17 for both), consistent with the enhanced proliferative capacity of cultured cell lines. In contrast, coagulation (FDR = 5.55 × 10−17), IFNγ response (FDR = 7.78 × 10−14), and EMT (FDR = 2.40 × 10−10) exhibited higher activity in TCGA tumors, reflecting immune‐related and microenvironment‐associated processes that are only partially represented in vitro. Comparable patterns were observed when the analysis was expanded to include all TCGA‐LUAD tumors (n = 513) and CCLE samples (Figure S2).

FIGURE 2.

FIGURE 2

Open in a new tab

Differential pathway activity between primary tumors and cell lines reveals divergent tumor–normal regulatory patterns. (A) Comparison of ssGSEA‐derived normalized enrichment scores (NES) for the five LASSO‐selected Hallmark gene sets between TCGA‐LUAD paired primary tumors (red, n = 58) and CCLE‐LUAD cell lines (purple, n = 45). (B) Pathway activity across TCGA normal lung tissues (blue, n = 58), TCGA unpaired LUAD tumors (red, n = 513), CCLE normal lung cell lines (green, n = 6), and CCLE‐LUAD cell lines (purple, n = 45). Statistical significance in both panels was assessed using the two‐sided Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction.

We next assessed differences between tumor and adjacent normal tissue samples within the TCGA‐LUAD cohort. Both paired analyses (58 tumor–normal pairs; Figure S3) and unpaired comparisons (513 tumors vs. 58 normal tissues) demonstrated significant upregulation of G2M checkpoint (FDR = 8.10 × 10−27) and E2F targets (FDR = 1.83 × 10−30) in tumors (Figure 2B). Conversely, coagulation and IFNγ response were significantly reduced in tumors relative to normal tissues (FDR = 3.15 × 10−8 and 1.87 × 10−4, respectively). EMT did not show a significant difference in the unpaired analysis (Figure 2B) but was significantly upregulated in the paired tumor–normal comparison (FDR = 7.85 × 10−3) (Figure S3). The discordant EMT significance between paired and unpaired analyses likely reflects the well‐documented heterogeneity of EMT programs across individual primary tumors. 26

In contrast, no significant differences across these five pathways were detected between CCLE tumor cell lines and matched normal lung cell lines. Together, these findings highlight discordance in pathway activity between primary tumors and cell line models, suggesting that certain tumor‐associated biological processes are not fully recapitulated in vitro.

3.3. Gene‐level analysis reveals divergent expression of pathway constituents between primary tumors and cell lines

To delineate the molecular basis of pathway‐level divergence, we conducted differential expression analysis of 808 unique genes comprising the five LASSO‐selected Hallmark pathways. Comparison between TCGA‐LUAD tumors (n = 58) and CCLE‐LUAD cell lines (n = 45) identified 459 significantly differentially expressed genes (|log2FC| >1, FDR <0.05) (Table S1). Of these, 262 genes exhibited higher expression in TCGA tumors, whereas 197 showed lower expression in tumors and higher expression in CCLE cell lines (Figure 3A).

FIGURE 3.

FIGURE 3

Open in a new tab

Gene‐level differential expression highlights stromal and extracellular matrix enrichment in primary tumors. (A) Volcano plot depicting differential expression of 808 genes constituting the five LASSO‐selected Hallmark pathways, comparing TCGA‐LUAD tumors (n = 58) with CCLE‐LUAD cell lines (n = 45). Red dots denote genes significantly upregulated in TCGA (log2FC >1, FDR <0.05), blue dots indicate genes significantly downregulated in TCGA (log2FC <−1, FDR <0.05), and gray dots represent non‐significant genes. (B) Hierarchical clustering heatmap of the top 10 differentially expressed genes ranked by statistical significance. Rows correspond to genes, and columns represent individual samples.

The genes most significantly enriched in the TCGA cohort included extracellular matrix and stromal components such as LUM (FDR = 4.52 × 10−39), BGN (FDR = 6.79 × 10−58), COL1A2 (FDR = 1.06 × 10−47), DCN (FDR = 8.16 × 10−56), and ACTA2 (FDR = 9.86 × 10−40), as well as the vascular markers PECAM1 (FDR = 1.53 × 10−46) and VWF (FDR = 1.65 × 10−40). Hierarchical clustering of the top 10 significantly differentially expressed genes distinctly separated TCGA and CCLE samples (Figure 3B), with TCGA tumors showing uniformly elevated expression of stromal and microenvironment‐associated signatures that were depleted in cell lines. Conversely, genes enriched in CCLE predominantly reflected cell‐intrinsic proliferative programs, consistent with the increased pathway activity of G2M checkpoint and E2F targets observed at the pathway level.

3.4. KEGG pathway analysis reveals consistent pathway‐level distinctions and tumor‐normal discordance

Application of the LASSO feature selection framework to KEGG pathway annotations identified seven discriminative pathways: cell cycle, complement and coagulation cascades, neuroactive ligand receptor interaction, ABC transporters, purine metabolism, progesterone‐mediated oocyte maturation, and systemic lupus erythematosus (Figure 4A).

FIGURE 4.

FIGURE 4

Open in a new tab

KEGG pathway‐based ssGSEA identifies discriminatory signatures and divergent tumor–normal regulatory patterns. (A) Mean LASSO coefficients for seven KEGG pathways consistently selected in ≥5 cross‐validation folds. (B) Comparison of KEGG pathway activity (ssGSEA‐derived normalized enrichment scores, NES) between TCGA‐LUAD paired tumors (red, n = 58) and CCLE‐LUAD cell lines (purple, n = 45). (C) Tumor–normal comparison of three selected KEGG pathways in TCGA‐LUAD, contrasting tumors (red, n = 58) with adjacent normal tissues (blue, n = 58). (D) Comparative analysis of the same three KEGG pathways across TCGA normal tissues (blue, n = 58), TCGA tumors (red, n = 513), CCLE normal lung cell lines (green, n = 6), and CCLE‐LUAD cell lines (purple, n = 45). Statistical significance in panels (B–D) was assessed using the two‐sided Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction for multiple testing. Reported FDR values reflect significance across all comparisons shown in panels (B–D).

Comparison of TCGA‐LUAD tumors (n = 58) with CCLE‐LUAD cell lines (n = 45) demonstrated significant divergence (FDR <0.05) across all seven pathways (Figure 4B). KEGG cell cycle activity was elevated in CCLE (FDR = 6.23 × 10−17), consistent with the enrichment of G2M checkpoint and E2F targets observed in the Hallmark analysis (Figure 2A). In contrast, KEGG complement and coagulation cascades pathway was significantly suppressed in CCLE compared to TCGA (FDR = 2.52 × 10−16). The concordant enrichment patterns across KEGG and Hallmark gene sets, particularly for cell cycle‐ and coagulation‐related programs, underscore consistent pathway‐level differences between primary tumors and cell lines.

Tumor–normal comparisons recapitulated the regulatory trends observed in the Hallmark pathway analysis (Figure 4C,D). In TCGA, the KEGG cell cycle was significantly upregulated in tumors relative to adjacent normal tissue (FDR = 2.35 × 10−10 for paired comparisons; 1.11 × 10−21 for unpaired comparisons). Conversely, the KEGG complement and coagulation cascades pathway was significantly downregulated in tumors (FDR = 5.22 × 10−9 for paired; 6.84 × 10−23 for unpaired analyses). ABC transporters, functionally associated with epithelial–mesenchymal transition (EMT) and drug transport mechanisms, 27 exhibited modest yet significant tumor‐associated elevation (FDR = 5.07 × 10−3 in paired and 0.03 in unpaired analyses). In contrast, CCLE tumor–normal comparisons showed no significant differences for any of these three pathways (FDR >0.1 for all) (Figure 4D), highlighting the limited capacity of cell line models to recapitulate tumor–normal transcriptional contrasts observed in primary tissues.

3.5. Cell line models exhibit higher mutational burden and divergent driver gene mutation frequencies compared to primary tumors

To assess differences in somatic mutational landscapes between LUAD primary tumors and cell line models, we compared tumor mutational burden (TMB) across TCGA and CCLE samples. CCLE cell lines displayed a significantly higher TMB compared to TCGA primary tumors (p = 3.05 × 10−7) (Figure 5A), consistent with increased accumulation of somatic alterations during in vitro propagation.

FIGURE 5.

FIGURE 5

Open in a new tab

Elevated tumor mutational burden and distinct driver gene mutation frequencies in LUAD cell lines compared with primary tumors. (A) Comparison of tumor mutational burden between TCGA‐LUAD tumors (red, n = 513) and CCLE‐LUAD cell lines (purple, n = 45). Statistical significance was determined by the Wilcoxon rank‐sum test (***p < 0.001). (B) Radar plot illustrating mutation frequencies of seven recurrent LUAD driver genes (TP53, STK11, NF1, KRAS, KEAP1, EGFR, and BRAF) in TCGA (red line) and CCLE (purple line). Adjusted p‐values (p.adj), calculated using Fisher's exact test with Benjamini–Hochberg correction, are shown for each gene.

We next evaluated mutation frequencies of seven recurrent LUAD driver genes (TP53, KRAS, EGFR, KEAP1, STK11, NF1, and BRAF) previously reported in TCGA genomic profiling studies. Fisher's exact test identified significant differences in mutation prevalence for TP53 and STK11 between the TCGA and CCLE cohorts (Figure 5B). Both genes showed higher mutation frequencies in CCLE cell lines (BH‐adjusted p‐values = 0.002 and 0.013, respectively). In contrast, KRAS, EGFR, KEAP1, NF1, and BRAF showed no statistically significant differences in mutation frequency between the two datasets.

3.6. Conserved pathway‐level differences between primary tumors and cell lines across lung cancer histologies

To evaluate whether the pathway‐level divergence observed in LUAD is histology‐specific or represents a broader characteristic of lung cancers, we extended the analysis to lung squamous cell carcinoma (LUSC). We analyzed ssGSEA‐derived normalized enrichment scores (NES) for the five LASSO‐selected Hallmark gene sets across TCGA‐LUSC tumors (n = 498), TCGA‐LUSC adjacent normal tissues (n = 50), CCLE‐LUSC cell lines (n = 16), and CCLE normal lung cell lines (n = 6) (Figure 6). The enrichment patterns closely resembled those observed in LUAD (Figure 2B), demonstrating that the pathway‐level distinctions between primary tumors and cell line models are conserved across major lung cancer histologies.

FIGURE 6.

FIGURE 6

Open in a new tab

Conserved pathway‐level differences between primary tumors and cell lines in lung squamous cell carcinoma (LUSC). Comparison of ssGSEA‐derived normalized enrichment scores (NES) for the five LASSO‐selected Hallmark pathways across TCGA‐LUSC normal tissues (blue, n = 50), TCGA‐LUSC tumors (red, n = 498), CCLE normal lung cell lines (green, n = 6), and CCLE‐LUSC cell lines (purple, n = 16). Statistical significance was evaluated using the Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction. FDR values indicate the significance of tumor–normal and cross‐cohort differences.

Within TCGA‐LUSC, G2M checkpoint and E2F targets showed strong tumor‐associated enrichment (FDR = 1.23 × 10−30 for both), reflecting proliferative signaling in primary tumors. Conversely, coagulation and IFNγ response were significantly downregulated in tumors compared to adjacent normal tissues (FDR = 4.75 × 10−19 and 8.77 × 10−16, respectively). EMT did not demonstrate a significant tumor–normal difference (FDR = 0.12). In contrast, tumor–normal comparisons within CCLE‐LUSC revealed no statistically significant pathway alterations (FDR >0.05), although G2M checkpoint and E2F targets maintained consistently elevated baseline activity across tumor samples.

4. DISCUSSION

Cancer cell lines are indispensable for mechanistic investigation and preclinical drug discovery; however, their ability to faithfully model primary tumor biology remains a subject of ongoing debate. Through systematic, pathway‐level transcriptomic analyses of TCGA and CCLE lung adenocarcinoma cohorts, we identified fundamental differences between primary tumors and in vitro models, observations that extend to lung squamous cell carcinoma as well. Applying an L1‐penalized feature selection framework, we defined a five‐pathway signature that distinguishes primary tumors from cell lines and captures divergent pathway activity across proliferative, immune, and stromal programs.

Our results are consistent with prior pan‐cancer analyses, which show that cell lines are enriched for proliferation‐related signatures, whereas primary tumors retain prominent microenvironmental components. 11 , 12 We extend these observations by performing sample‐level pathway quantification to enable direct tumor–normal comparisons and by defining a discriminative pathway signature specific to lung cancers, with emphasis on LUAD. In TCGA, primary tumors exhibited significant pathway alterations compared to adjacent normal tissue, particularly in the Hallmark coagulation and IFNγ response pathways, as well as in the KEGG complement and coagulation cascades. In contrast, CCLE‐derived tumor cell lines showed no analogous differences when compared with matched normal lung cell lines. This discordance cannot be attributed solely to the absence of stromal and immune compartments; rather, it likely reflects broader divergence driven by loss of tumor microenvironmental context and genetic drift during prolonged in vitro culture, resulting in distinct selective pressures in vivo versus in vitro. 28

The relative depletion of EMT signatures and enrichment of ABC transporters in LUAD cell lines compared with TCGA tumors likely reflect loss of tumor microenvironmental regulation, selection for highly proliferative epithelial clones during cell line establishment, adaptation to artificial culture stress, drug resistance‐associated selective pressures, and the absence of stromal transcript contributions. 29 , 30 , 31

The tumor microenvironment is a highly structured and dynamic ecosystem in which cancer‐associated fibroblasts, immune cells, and endothelial cells collectively regulate tumor initiation, progression, and antitumor immunity. 32 , 33 In LUAD, immune cell infiltration within the tumor microenvironment has prognostic significance, and higher levels of T‐ and B‐cell‐associated immune signatures are linked to improved overall survival. 34 , 35 , 36

More broadly, reduced stromal and immune pathway activity in cell lines underscores the limitations of conventional two‐dimensional culture systems, which lack the complex cellular ecosystem of primary tumors. 37 In vivo, the tumor microenvironment, comprising cancer‐associated fibroblasts, immune cells, endothelial cells, and extracellular matrix, critically shapes tumor progression, therapeutic resistance, angiogenesis, and immune evasion. 33 , 38 , 39 The absence of these contextual cues in vitro likely diminishes the translational relevance of cell line‐based drug screening by failing to capture microenvironment‐dependent determinants of therapeutic response.

The significant enrichment of Hallmark E2F targets and the G2M checkpoint, as well as KEGG cell cycle pathways, in cell lines compared to primary tumors requires cautious interpretation. Although these signatures reflect the high proliferative capacity of cultured cells, they may not fully recapitulate in vivo tumor biology, as cancer cell lines are systematically enriched for cell cycle‐associated transcriptional programs, whereas primary tumors retain immune and microenvironmental expression signals largely absent in vitro. 10 , 12 , 40 Consistent with their clinical relevance, E2F family members define high‐risk subgroups in lung cancer, 41 and elevated CDK1 activity, which is essential for mitotic entry at the G2/M transition, is associated with poor prognosis. 42 Moreover, prior studies have linked higher E2F and G2M pathway scores to tumor aggressiveness across multiple cancer types. 43 , 44 Nevertheless, our findings show that these proliferation‐associated pathways are generally more active in cell lines than in primary tumors, likely reflecting in vitro adaptation, where the absence of stromal and immune interactions, cytokine signaling, and intact tissue architecture removes key regulatory constraints. 45 The elevated mutation burden observed in CCLE samples, together with enrichment of TP53 and STK11 alterations, may further contribute to dysregulated cell cycle control in vitro.

Beyond transcriptional divergence, resistance to molecularly targeted therapies in NSCLC is also driven by stromal and immune components of the tumor microenvironment, including cancer‐associated fibroblasts and tumor‐associated macrophages. 46 , 47 Conventional two‐dimensional culture systems fail to capture key features of the tumor microenvironment, such as multicellular interactions and tissue organization, thereby limiting their ability to accurately model therapy response. 48

Differences in the prevalence of genomic and driver gene alterations between TCGA primary tumors and CCLE cell lines carry important implications for experimental model selection. 49 Notably, TP53, the most frequently mutated gene in human cancers, exhibits a significantly higher mutation frequency in CCLE samples. 50 Because TP53 mutations substantially reshape cellular responses to DNA damage, metabolic stress, and anticancer therapies, 51 their overrepresentation in cell lines may bias preclinical studies toward p53‐deficient biology, thereby incompletely capturing the heterogeneity of primary disease. Similarly, the overrepresentation of STK11 mutations in cell lines is noteworthy, given that STK11 loss has been associated with the development of immunosuppressive tumor microenvironments and diminished responsiveness to immune checkpoint blockade, biological features that are difficult to recapitulate in conventional two‐dimensional culture systems lacking immune components. 52 , 53

The preservation of pathway‐level distinctions across both lung adenocarcinoma and lung squamous cell carcinoma indicates that transcriptional divergence between primary tumors and cell lines is a generalizable feature of major non‐small cell lung cancer (NSCLC) subtypes rather than a histology‐specific phenomenon.

Several important limitations merit discussion. Our analyses were based on bulk RNA sequencing, which reflects averaged transcriptional signals across heterogeneous cell populations; single‐cell approaches may better delineate cellular subpopulations selectively retained or lost during cell line establishment. Although we identified pathway‐ and gene‐level differences, we did not experimentally assess how these transcriptional divergences influence drug response or functional phenotypes. Furthermore, the held‐out test set was limited in size, and classification performance may be further evaluated in larger independent cohorts.

Despite these limitations, our findings have practical implications for cancer research. The five‐pathway ssGSEA signature may serve as a reference profile against which the transcriptional fidelity of candidate lung adenocarcinoma cell lines can be assessed prior to model selection. Conventional cell lines may be appropriate for studying cell‐autonomous proliferation or DNA damage responses, whereas investigations of immune evasion, stromal crosstalk, angiogenesis, or coagulation‐related processes may require more physiologically relevant systems.

Overall, our systematic pathway‐level analysis demonstrates that lung cancer cell lines and primary tumors occupy distinct transcriptional states marked by divergent pathway regulation. These differences reflect both the absence of microenvironmental context in vitro and intrinsic shifts in tumor cell biology. Careful recognition of these distinctions and context‐specific model selection will strengthen the translational relevance of preclinical lung cancer research.

AUTHOR CONTRIBUTIONS

Pritam Bera: Conceptualization; data curation; formal analysis; investigation; methodology; writing – original draft; writing – review and editing. Rajesh Raju: Funding acquisition; project administration; writing – review and editing. Debodipta Das: Conceptualization; data curation; investigation; methodology; project administration; supervision; writing – original draft; writing – review and editing.

FUNDING INFORMATION

This research did not receive any specific grant from funding agencies in the public, commercial, or not‐for‐profit sectors.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests.

ETHICS STATEMENT

This study used publicly available, de‐identified datasets obtained from The Cancer Genome Atlas (TCGA) and the Cancer Cell Line Encyclopedia (CCLE). No animals or human participants were directly involved, and ethical approval was therefore not required for this work. All analyses were conducted in compliance with the data usage policies and ethical guidelines of the respective public repositories.

Supporting information

Figure S1: Classifier evaluation metrics. (A) Calibration curves for Random Forest, SVM, and XGBoost on the test set. (B) Brier scores assessing probabilistic calibration of each classifier. (C) Learning curves showing training and 5‐fold cross‐validation accuracy across incremental training set sizes.

AME2-9999-0-s005.pdf (155.3KB, pdf)

Figure S2: Pathway activity differences between TCGA‐LUAD tumors and CCLE‐LUAD tumors. Comparison of pathway activity (ssGSEA‐derived NES) for the five LASSO‐selected pathways between TCGA tumors (red, n = 513) and CCLE tumors (purple, n = 45). Statistical significance was determined using the Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction.

AME2-9999-0-s004.pdf (6.3MB, pdf)

Figure S3: Differential pathway activity between TCGA‐LUAD tumor–normal paired samples. Comparison of ssGSEA‐derived NES for the five LASSO‐selected Hallmark pathways between TCGA‐LUAD paired tumors (red, n = 58) and their matched adjacent normal tissues (blue, n = 58). Statistical significance was evaluated using the Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction.

AME2-9999-0-s002.pdf (1.5MB, pdf)

Data S1.

AME2-9999-0-s003.pdf (126.1KB, pdf)

Table S1: Gene‐level differential expression of LASSO‐selected Hallmark pathway genes in TCGA‐LUAD tumors versus CCLE‐LUAD cell lines. Differential expression was assessed between 58 TCGA‐LUAD tumors and 45 CCLE‐LUAD cell lines. Rows highlighted in red represent the 459 genes meeting the significance threshold (FDR <0.05 and |log2FC| >1).

AME2-9999-0-s001.xlsx (84.5KB, xlsx)

ACKNOWLEDGMENTS

The authors gratefully acknowledge the investigators and participants of The Cancer Genome Atlas (TCGA) and the Cancer Cell Line Encyclopedia (CCLE) projects for generating and making these valuable datasets publicly available. The authors also acknowledge Yenepoya (Deemed to be University) for providing the infrastructure and institutional support necessary to conduct this research.

DATA AVAILABILITY STATEMENT

All data used in this study were obtained from publicly available repositories. Detailed descriptions of the datasets and their sources are provided in the manuscript.

REFERENCES

  • 1. Gillet J‐P, Calcagno AM, Varma S, et al. Redefining the relevance of established cancer cell lines to the study of mechanisms of clinical anti‐cancer drug resistance. Proc Natl Acad Sci. 2011;108:18708‐18713. doi: 10.1073/pnas.1111840108 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Li J, Zhao W, Akbani R, et al. Characterization of human cancer cell lines by reverse‐phase protein arrays. Cancer Cell. 2017;31:225‐239. doi: 10.1016/j.ccell.2017.01.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Sharma SV, Haber DA, Settleman J. Cell line‐based platforms to evaluate the therapeutic efficacy of candidate anticancer agents. Nat Rev Cancer. 2010;10:241‐253. doi: 10.1038/nrc2820 [DOI] [PubMed] [Google Scholar]
  • 4. Gillet J‐P, Varma S, Gottesman MM. The clinical relevance of cancer cell lines. JNCI J Natl Cancer Inst. 2013;105:452‐458. doi: 10.1093/jnci/djt007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Weinstein JN. Cell lines battle cancer. Nature. 2012;483:544‐545. doi: 10.1038/483544a [DOI] [PubMed] [Google Scholar]
  • 6. Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014;32:40‐51. doi: 10.1038/nbt.2786 [DOI] [PubMed] [Google Scholar]
  • 7. The Cancer Genome Atlas Research Network , Weinstein JN, Collisson EA, et al. The cancer genome atlas pan‐cancer analysis project. Nat Genet. 2013;45:1113‐1120. doi: 10.1038/ng.2764 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Barretina J, Caponigro G, Stransky N, et al. The cancer cell line encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012;483:603‐607. doi: 10.1038/nature11003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9. Ghandi M, Huang FW, Jané‐Valbuena J, et al. Next‐generation characterization of the cancer cell line encyclopedia. Nature. 2019;569:503‐508. doi: 10.1038/s41586-019-1186-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Ertel A, Verghese A, Byers SW, Ochs M, Tozeren A. Pathway‐specific differences between tumor cell lines and normal and tumor tissue cells. Mol Cancer. 2006;5:55. doi: 10.1186/1476-4598-5-55 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Sanders LM, Chandra R, Zebarjadi N, et al. Machine learning multi‐omics analysis reveals cancer driver dysregulation in pan‐cancer cell lines compared to primary tumors. Commun Biol. 2022;5:1367. doi: 10.1038/s42003-022-04075-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Yu K, Chen B, Aran D, et al. Comprehensive transcriptomic analysis of cell lines as models of primary tumors across 22 tumor types. Nat Commun. 2019;10:3574. doi: 10.1038/s41467-019-11415-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74:229‐263. doi: 10.3322/caac.21834 [DOI] [PubMed] [Google Scholar]
  • 14. Travis WD, Brambilla E, Noguchi M, et al. International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society international multidisciplinary classification of lung adenocarcinoma. J Thorac Oncol. 2011;6:244‐285. doi: 10.1097/JTO.0b013e318206a221 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. The Cancer Genome Atlas Research Network . Comprehensive molecular profiling of lung adenocarcinoma. Nature. 2014;511:543‐550. doi: 10.1038/nature13385 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Li Y, Yan B, He S. Advances and challenges in the treatment of lung cancer. Biomed Pharmacother. 2023;169:115891. doi: 10.1016/j.biopha.2023.115891 [DOI] [PubMed] [Google Scholar]
  • 17. Krug K, Mertins P, Zhang B, et al. A curated resource for phosphosite‐specific signature analysis. Mol Cell Proteomics. 2019;18:576‐593. doi: 10.1074/mcp.TIR118.000943 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Subramanian A, Tamayo P, Mootha VK, et al. Gene set enrichment analysis: a knowledge‐based approach for interpreting genome‐wide expression profiles. Proc Natl Acad Sci. 2005;102:15545‐15550. doi: 10.1073/pnas.0506580102 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Liberzon A, Birger C, Thorvaldsdóttir H, Ghandi M, Mesirov JP, Tamayo P. The molecular signatures database Hallmark gene set collection. Cell Syst. 2015;1:417‐425. doi: 10.1016/j.cels.2015.12.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Kanehisa M. KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 2000;28:27‐30. doi: 10.1093/nar/28.1.27 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit‐Learn: Machine Learning in Python. 2012. doi: 10.48550/ARXIV.1201.0490 [DOI]
  • 22. Breiman L. Random forests. Mach Learn. 2001;45:5‐32. doi: 10.1023/A:1010933404324 [DOI] [Google Scholar]
  • 23. Cortes C, Vapnik V. Support‐vector networks. Mach Learn. 1995;20:273‐297. doi: 10.1023/A:1022627411411 [DOI] [Google Scholar]
  • 24. Chen T, Guestrin C. XGBoost: a scalable tree boosting system. In proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794. ACM; 2016. doi: 10.1145/2939672.2939785 [DOI] [Google Scholar]
  • 25. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B Stat Methodol. 1995;57:289‐300. doi: 10.1111/j.2517-6161.1995.tb02031.x [DOI] [Google Scholar]
  • 26. Malagoli Tagliazucchi G, Wiecek AJ, Withnell E, Secrier M. Genomic and microenvironmental heterogeneity shaping epithelial‐to‐mesenchymal trajectories in cancer. Nat Commun. 2023;14:789. doi: 10.1038/s41467-023-36439-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Jiang Z‐S, Sun Y‐Z, Wang S‐M, Ruan J‐S. Epithelial‐mesenchymal transition: potential regulator of ABC transporters in tumor progression. J Cancer. 2017;8:2319‐2327. doi: 10.7150/jca.19079 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Trastulla L, Noorbakhsh J, Vazquez F, McFarland J, Iorio F. Computational estimation of quality and clinical relevance of cancer cell lines. Mol Syst Biol. 2022;18:e11017. doi: 10.15252/msb.202211017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Fan J, To KKW, Chen Z‐S, Fu L. ABC transporters affects tumor immune microenvironment to regulate cancer immunotherapy and multidrug resistance. Drug Resist Updat. 2023;66:100905. doi: 10.1016/j.drup.2022.100905 [DOI] [PubMed] [Google Scholar]
  • 30. Fontana R, Mestre‐Farrera A, Yang J. Update on epithelial‐mesenchymal plasticity in cancer progression. Annu Rev Pathol Mech Dis. 2024;19:133‐156. doi: 10.1146/annurev-pathmechdis-051222-122423 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Nieto MA, Huang RY‐J, Jackson RA, Thiery JP. EMT: 2016. Cell. 2016;166:21‐45. doi: 10.1016/j.cell.2016.06.028 [DOI] [PubMed] [Google Scholar]
  • 32. Fang Y, Kong Y, Rong G, Luo Q, Liao W, Zeng D. Systematic investigation of tumor microenvironment and antitumor immunity with IOBR. Med Research. 2025;1:136‐140. doi: 10.1002/mdr2.70001 [DOI] [Google Scholar]
  • 33. de Visser KE, Joyce JA. The evolving tumor microenvironment: from cancer initiation to metastatic outgrowth. Cancer Cell. 2023;41:374‐403. doi: 10.1016/j.ccell.2023.02.016 [DOI] [PubMed] [Google Scholar]
  • 34. Iglesia MD, Parker JS, Hoadley KA, Serody JS, Perou CM, Vincent BG. Genomic analysis of immune cell infiltrates across 11 tumor types. J Natl Cancer Inst. 2016;108:djw144. doi: 10.1093/jnci/djw144 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Gao P, Zou K, Xiao L, et al. High expression of PTGES3 is an independent predictive poor prognostic biomarker and correlates with immune infiltrates in lung adenocarcinoma. Int Immunopharmacol. 2022;110:108954. doi: 10.1016/j.intimp.2022.108954 [DOI] [PubMed] [Google Scholar]
  • 36. Wang W, Ren S, Wang Z, Zhang C, Huang J. Increased expression of TTC21A in lung adenocarcinoma infers favorable prognosis and high immune infiltrating level. Int Immunopharmacol. 2020;78:106077. doi: 10.1016/j.intimp.2019.106077 [DOI] [PubMed] [Google Scholar]
  • 37. Jubelin C, Muñoz‐Garcia J, Griscom L, et al. Three‐dimensional in vitro culture models in oncology research. Cell Biosci. 2022;12:155. doi: 10.1186/s13578-022-00887-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. Chakravarthy A, Khan L, Bensler NP, Bose P, de Carvalho DD. TGF‐β‐associated extracellular matrix genes link cancer‐associated fibroblasts to immune evasion and immunotherapy failure. Nat Commun. 2018;9:4692. doi: 10.1038/s41467-018-06654-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Winkler J, Abisoye‐Ogunniyan A, Metcalf KJ, Werb Z. Concepts of extracellular matrix remodelling in tumour progression and metastasis. Nat Commun. 2020;11:5120. doi: 10.1038/s41467-020-18794-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Warren A, Chen Y, Jones A, et al. Global computational alignment of tumor and cell line transcriptional profiles. Nat Commun. 2021;12:22. doi: 10.1038/s41467-020-20294-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Sun C‐C, Zhou Q, Hu W, et al. Transcriptional E2F1/2/5/8 as potential targets and transcriptional E2F3/6/7 as new biomarkers for the prognosis of human lung carcinoma. Aging. 2018;10:973‐987. doi: 10.18632/aging.101441 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Wang J, Zhang Y, Xu S, et al. Prognostic significance of G2/M arrest signaling pathway proteins in advanced non‐small cell lung cancer patients. Oncol Lett. 2015;9:1266‐1272. doi: 10.3892/ol.2015.2842 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Chida K, Oshi M, Roy AM, et al. E2F target score is associated with cell proliferation and survival of patients with hepatocellular carcinoma. Surgery. 2023;174:307‐314. doi: 10.1016/j.surg.2023.04.030 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Oshi M, Newman S, Tokumaru Y, et al. High G2M pathway score pancreatic cancer is associated with worse survival, particularly after margin‐positive (R1 or R2) resection. Cancer. 2020;12:2871. doi: 10.3390/cancers12102871 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Mirabelli P, Coppola L, Salvatore M. Cancer cell lines are useful model systems for medical research. Cancer. 2019;11:1098. doi: 10.3390/cancers11081098 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Huang Z, Xiao Z, Yu L, Liu J, Yang Y, Ouyang W. Tumor‐associated macrophages in non‐small‐cell lung cancer: from treatment resistance mechanisms to therapeutic targets. Crit Rev Oncol Hematol. 2024;196:104284. doi: 10.1016/j.critrevonc.2024.104284 [DOI] [PubMed] [Google Scholar]
  • 47. Zhang C, Zhou W, Xu H, et al. Cancer‐associated fibroblasts promote EGFR‐TKI resistance via the CTHRC1/glycolysis/H3K18la positive feedback loop. Oncogene. 2025;44:1400‐1414. doi: 10.1038/s41388-025-03318-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Ravi K, Manoharan TJM, Wang K‐C, Pockaj B, Nikkhah M. Engineered 3D ex vivo models to recapitulate the complex stromal and immune interactions within the tumor microenvironment. Biomaterials. 2024;305:122428. doi: 10.1016/j.biomaterials.2023.122428 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49. Domcke S, Sinha R, Levine DA, Sander C, Schultz N. Evaluating cell lines as tumour models by comparison of genomic profiles. Nat Commun. 2013;4:2126. doi: 10.1038/ncomms3126 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50. Triantafyllidis CP, Barberis A, Hartley F, et al. A machine learning and directed network optimization approach to uncover TP53 regulatory patterns. iScience. 2023;26:108291. doi: 10.1016/j.isci.2023.108291 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Liu Y, Su Z, Tavana O, Gu W. Understanding the complexity of p53 in a new era of tumor suppression. Cancer Cell. 2024;42:946‐967. doi: 10.1016/j.ccell.2024.04.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52. Koyama S, Akbay EA, Li YY, et al. STK11/LKB1 deficiency promotes neutrophil recruitment and proinflammatory cytokine production to suppress T‐cell activity in the lung tumor microenvironment. Cancer Res. 2016;76:999‐1008. doi: 10.1158/0008-5472.CAN-15-1439 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53. Mackenzie NJ, Nicholls C, Templeton AR, et al. Modelling the tumor immune microenvironment for precision immunotherapy. Clin Transl Immunol. 2022;11:e1400. doi: 10.1002/cti2.1400 [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

Figure S1: Classifier evaluation metrics. (A) Calibration curves for Random Forest, SVM, and XGBoost on the test set. (B) Brier scores assessing probabilistic calibration of each classifier. (C) Learning curves showing training and 5‐fold cross‐validation accuracy across incremental training set sizes.

AME2-9999-0-s005.pdf (155.3KB, pdf)

Figure S2: Pathway activity differences between TCGA‐LUAD tumors and CCLE‐LUAD tumors. Comparison of pathway activity (ssGSEA‐derived NES) for the five LASSO‐selected pathways between TCGA tumors (red, n = 513) and CCLE tumors (purple, n = 45). Statistical significance was determined using the Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction.

AME2-9999-0-s004.pdf (6.3MB, pdf)

Figure S3: Differential pathway activity between TCGA‐LUAD tumor–normal paired samples. Comparison of ssGSEA‐derived NES for the five LASSO‐selected Hallmark pathways between TCGA‐LUAD paired tumors (red, n = 58) and their matched adjacent normal tissues (blue, n = 58). Statistical significance was evaluated using the Wilcoxon rank‐sum test with Benjamini–Hochberg false discovery rate (FDR) correction.

AME2-9999-0-s002.pdf (1.5MB, pdf)

Data S1.

AME2-9999-0-s003.pdf (126.1KB, pdf)

Table S1: Gene‐level differential expression of LASSO‐selected Hallmark pathway genes in TCGA‐LUAD tumors versus CCLE‐LUAD cell lines. Differential expression was assessed between 58 TCGA‐LUAD tumors and 45 CCLE‐LUAD cell lines. Rows highlighted in red represent the 459 genes meeting the significance threshold (FDR <0.05 and |log2FC| >1).

AME2-9999-0-s001.xlsx (84.5KB, xlsx)

Data Availability Statement

All data used in this study were obtained from publicly available repositories. Detailed descriptions of the datasets and their sources are provided in the manuscript.

Articles from Animal Models and Experimental Medicine are provided here courtesy of Wiley

RESOURCES