NSCLC accounts for 85% of all lung cancers and most patients are diagnosed at advanced stages. Despite advances in targeted therapy and immunotherapy, five-year survival for advanced NSCLC remains below 15%. Nearly half of patients do not respond to PD-1 and CTLA-4 immune checkpoint inhibitors, and available clinical biomarkers lack sufficient sensitivity and specificity to reliably predict responders from non-responders.
DDR1 is a collagen-binding receptor tyrosine kinase implicated in immune exclusion. Encoded on chromosome 6p21.3, DDR1 promotes tumor progression and immune exclusion by regulating cell motility, extracellular matrix (ECM) remodeling, and collagen alignment that physically restricts immune cell infiltration. Aberrant DDR1 activation has been implicated in pancreatic, cervical, hepatic, and breast malignancies, but its specific role in NSCLC remained unclear.
DDR1's role in the tumor microenvironment is biologically complex and potentially contradictory. In triple-negative breast cancer models, DDR1 knockout enhanced CD8+ and CD4+ T-cell infiltration and suppressed tumor growth, suggesting an immunosuppressive function. However, other work reported that DDR1 inhibition increased tumor burden by promoting a protumorigenic microenvironment, highlighting the need for systematic investigation specific to NSCLC.
This study combined bioinformatics, machine learning, in vitro experiments, and clinical validation. Integrative analysis of TCGA RNA-seq data from 495 LUAD and 481 LUSC tumor samples was performed alongside external validation in two independent GEO datasets, single-cell transcriptomic analysis, cell line knockdown experiments, and immunohistochemistry of 34 clinical NSCLC specimens to comprehensively characterize DDR1's role.
TCGA RNA-seq data from 976 NSCLC tumor samples formed the primary cohort. After quality filtering and removal of samples with missing clinical or staging data, 495 LUAD and 481 LUSC tumor samples were retained alongside 57 LUAD and 48 LUSC normal samples. External validation used GSE30219 (278 NSCLC samples on GPL570) and GSE41271 (274 NSCLC samples on GPL6884), both downloaded from GEO.
Immune landscape characterization used four complementary deconvolution algorithms. CIBERSORT estimated 22 immune cell type proportions, ESTIMATE inferred tumor purity and stromal/immune content, MCPcounter quantified nine immune cell types, and TIMER provided additional immune infiltration estimates. Together, these tools captured different dimensions of the immunological composition of NSCLC tumors stratified by DDR1 expression level.
A 101-combination machine learning framework identified the optimal prognostic model. Ten machine learning algorithms were applied to DEGs with consistent prognostic value across multiple datasets: random survival forest (RSF), elastic net, LASSO, ridge, stepwise Cox, CoxBoost, plsRcox, SuperPC, GBM, and survival-SVM. These were combined into 101 algorithm combinations evaluated by 10-fold cross-validation across TCGA and both GEO validation cohorts, with the highest average C-index used to select the final model.
In vitro functional validation used siRNA knockdown in NSCLC cell lines. HARA, NCI-H292, and CALU-3 cells were screened for DDR1 expression using RT-qPCR, western blotting, and immunofluorescence. CALU-3 and NCI-H292 cells were selected for siRNA knockdown experiments covering cell viability (CCK-8), colony formation, migration (scratch and Transwell), invasion (Matrigel Transwell), adhesion, and apoptosis (Annexin V flow cytometry). IHC of 34 paraffin-embedded clinical NSCLC specimens confirmed DDR1 protein expression in patient tumor tissue.
DDR1 was significantly overexpressed in NSCLC tumor tissue compared to normal lung. Differential expression analysis showed tumor samples had mean DDR1 expression of 7.06 versus 5.95 in normal samples (p less than 0.05). Expression was higher in older patients (over 65 years), in female patients, and in T3-stage tumors, but did not differ significantly across N, M, or pathological stages.
High DDR1 expression was independently associated with shorter progression-free survival. Kaplan-Meier analysis showed significantly shorter PFS in the high DDR1 group (p less than 0.05). Multivariate Cox regression identified DDR1 expression and pathological stage as independent prognostic factors, supporting construction of a predictive nomogram. Calibration curves showed the 1-year nomogram predictions were most consistent with observed outcomes.
DDR1 expression was strongly negatively correlated with DNA methylation. Pearson correlation between DDR1 expression and mean beta value across CpG methylation sites showed R equal to -0.43 (p less than 0.05), suggesting that promoter hypomethylation drives DDR1 upregulation. This epigenetic mechanism may explain DDR1 overexpression in tumors independent of somatic mutation, since DDR1 mutation status did not significantly affect expression or survival.
IHC confirmed DDR1 protein overexpression in 55.88% of clinical NSCLC samples. Of 34 patients, 19 showed positive DDR1 staining, with IHC scores markedly higher in tumor versus adjacent normal tissue (27.84 versus 3.73, p less than 0.0001). DDR1 positivity was substantially higher in LUSC (87.5%, 14 of 16 cases) than in LUAD (29.4%, 5 of 17 cases), and LUSC IHC scores exceeded LUAD scores (30.03 versus 24.02, p equal to 0.0062).
High DDR1 expression was negatively correlated with T-cell and immune cell infiltration across all four algorithms. CIBERSORT showed strong negative correlation with activated CD4 memory T cells (R equal to -0.18) and positive correlation with M0 macrophages (R equal to 0.17). ESTIMATE showed strong negative immune and ESTIMATE scores (R equal to -0.46 and -0.45 respectively). MCPcounter showed negative T-cell and monocytic lineage correlations, while TIMER confirmed negative CD8+ T-cell correlation (R equal to -0.34).
DDR1 expression correlated with 63 immunoregulatory genes, most prominently CD276. The strongest positive correlation was with CD276 (also known as B7-H3, R equal to 0.48), an immune checkpoint molecule. The strongest negative correlation was with HLA-DMB (R equal to -0.37), which encodes an MHC class II chaperone important for antigen presentation. These associations suggest DDR1 co-regulates multiple immune escape mechanisms beyond simple ECM remodeling.
Immunotherapy response predictors were significantly worse in the high DDR1 group. TIDE scores, which predict immune evasion tendency, were significantly elevated in the high DDR1 group (p less than 0.05). CYT scores (reflecting cytotoxic T-cell activity) and TLS scores (reflecting tertiary lymphoid structure formation, a positive prognostic feature) were both significantly lower in the high DDR1 group. These patterns indicate that high DDR1 expression is a marker of an immunotherapy-resistant phenotype.
Single-cell analysis identified macrophages and neutrophils as the primary DDR1-expressing immune cells. UMAP clustering of 11,481 cells from 8 tumor samples revealed 13 cell types. DDR1 expression was highest in macrophages (avg_log2FC equal to 1.87, p_adj less than 0.001) and neutrophils (avg_log2FC equal to 2.74, p_adj less than 0.001). Pseudotime analysis showed DDR1 expression decreased along developmental trajectories, and CellChat analysis identified the NOTCH pathway as a key communication axis between neutrophils (senders) and macrophages (receivers).
High DDR1 expression predicted increased sensitivity to four chemotherapy agents. IC50 analysis showed that patients with high DDR1 expression had significantly lower estimated IC50 values for vinblastine, doxorubicin, cisplatin, and docetaxel (all p less than 0.05). This increased sensitivity to cytotoxic agents in the high DDR1 group may reflect that DDR1-driven ECM remodeling paradoxically increases vulnerability to drugs that disrupt mitotic or DNA repair processes.
High DDR1 expression predicted reduced sensitivity to methotrexate. The methotrexate IC50 was significantly higher in the high DDR1 expression group (p less than 0.05), suggesting that DDR1-mediated collagen barriers or associated drug efflux mechanisms may confer specific resistance to antifolate chemotherapy. No significant difference in gefitinib sensitivity was detected between groups, indicating DDR1 expression does not predict EGFR-targeted therapy response.
These differential drug sensitivities point toward personalized chemotherapy selection based on DDR1 status. The divergent sensitivity patterns suggest that DDR1 expression level could guide chemotherapy decisions: high-DDR1 NSCLC patients may be preferentially treated with platinum-based or taxane regimens while avoiding methotrexate. DDR1 inhibitors have been shown in preclinical models to disrupt DDR1/PYK2/FAK signaling and overcome ECM-mediated drug resistance, potentially enhancing conventional chemotherapy efficacy when combined.
LASSO combined with random survival forest yielded the best-performing prognostic model. Among 101 algorithm combinations evaluated by average C-index across TCGA, GSE30219, and GSE41271, the LASSO plus RSF combination achieved the highest C-index of 0.728 and was selected as the final model. LASSO analysis identified 19 candidate genes, and RSF importance scoring narrowed these to 4 genes with importance scores above 0.01.
The 4-gene risk score formula combines PKP2, DKK1, TEF, and GJB5. The RiskScore was calculated as: (0.135 x PKP2) plus (0.104 x DKK1) plus (-0.118 x TEF) plus (-0.132 x GJB5). Positive coefficients for PKP2 and DKK1 indicate these are risk-promoting genes, while negative coefficients for TEF and GJB5 indicate these are protective. Risk scores ranged from 0.2 to 2.1 with a median cutoff of 1.0 used to classify patients into high- and low-risk groups.
Low-RiskScore patients had significantly better survival in all three validation cohorts. Kaplan-Meier analysis confirmed significant survival differences in TCGA (p less than 0.05), GSE41271 (p less than 0.05), and GSE30219 (p less than 0.05). ROC analysis showed TCGA 1/3/5-year AUCs of 0.600/0.645/0.663, GSE41271 AUCs of 0.716/0.603/0.608, and GSE30219 AUCs of 0.673/0.692/0.662, indicating stable and consistent predictive performance across cohorts and platforms.
The model outperformed the majority of 38 published NSCLC prognostic signatures. Systematic comparison against existing signatures for LUAD, LUSC, and NSCLC showed that the 4-gene model consistently ranked among the top-performing signatures across all three validation cohorts, confirming its competitive prognostic value relative to the current published literature.
siRNA-mediated DDR1 knockdown significantly reduced NSCLC cell proliferative capacity. In CALU-3 and NCI-H292 cell lines, DDR1 siRNA transfection confirmed by RT-qPCR and western blotting led to significant reductions in cell viability measured by CCK-8 assay, colony formation, and increased apoptosis detected by Annexin V flow cytometry (all p less than 0.001).
DDR1 depletion impaired cell migration, invasion, and adhesion. Scratch wound healing assays showed significantly reduced migration, Transwell assays showed decreased cells crossing uncoated and Matrigel-coated membranes, and adhesion assays showed reduced attachment to extracellular matrix. These results confirm that DDR1 actively promotes the invasive and migratory phenotype of NSCLC cells through mechanisms consistent with its known role in ECM remodeling via MMP-2, N-cadherin, and vimentin upregulation.
Enrichment analysis linked high DDR1 expression to cell cycle, NOTCH, and Hedgehog signaling. DEG analysis identified 382 upregulated and 1,749 downregulated genes in the high DDR1 group. KEGG analysis showed associations with cell cycle and PI3K-Akt pathways. GSVA revealed significant enrichment of HALLMARK_NOTCH_SIGNALING and HALLMARK_HEDGEHOG_SIGNALING in high DDR1 tumors, pathways linked to stem cell-like properties, cell fate regulation, and tumor invasiveness.
DDR1's oncogenic activity in NSCLC appears driven more by TME remodeling than intrinsic genomic instability. DDR1 mutation status did not affect expression or survival, and TMB did not differ between DDR1 expression groups. This suggests that DDR1 overexpression driven by DNA hypomethylation promotes tumor progression primarily through collagen alignment that restricts immune infiltration and ECM-mediated macrophage/neutrophil signaling, rather than through mutational dysregulation of cell-intrinsic pathways.
DDR1 is established as an independent prognostic biomarker and functional driver of NSCLC progression. Integrating TCGA bioinformatics, single-cell transcriptomics, machine learning, in vitro experiments, and IHC of clinical specimens, this study demonstrates that DDR1 overexpression (confirmed in 55.88% of patients) promotes tumor progression, immune exclusion, and poor survival outcomes in NSCLC independently of pathological stage.
The DDR1-associated immunosuppressive landscape identifies patients least likely to benefit from ICIs. Elevated TIDE scores, reduced CYT and TLS scores, inverse correlations with T-cell infiltration, and positive correlation with the immune checkpoint molecule CD276 collectively define a DDR1-high immune evasion phenotype. This profile could complement existing ICI response biomarkers such as PD-L1 expression and tumor mutational burden to improve patient selection for immunotherapy.
The 4-gene prognostic model derived from DDR1-associated DEGs offers a validated clinical tool. The PKP2/DKK1/TEF/GJB5 risk score achieved C-index 0.728 with consistent 1-5 year survival stratification across three independent cohorts and outperformed most published NSCLC signatures, representing a ready-for-validation candidate for further prospective clinical utility studies.
Targeting DDR1 with antibodies or small molecule inhibitors warrants further investigation. ECD-neutralizing antibodies that disrupt collagen fiber alignment have shown efficacy in preclinical models by mitigating immune exclusion and inhibiting tumor growth. DDR1 inhibition may also sensitize tumors to conventional chemotherapy by disrupting DDR1/PYK2/FAK ECM-mediated drug resistance. Rigorous prospective clinical validation of DDR1 as a therapeutic target is the essential next step.