Prognostic and Therapeutic Significance of DPP4 in SMARCA4-Deficient Non–Small Cell Lung Cancer

Abstract

Introduction

SMARCA4-deficient NSCLC is an aggressive subtype lacking robust biomarkers and therapeutic targets. To define the expression pattern and clinical significance of DPP4 in SMARCA4-dNSCLC and explore whether DPP4 may represent a therapeutic vulnerability in this subtype.

Materials and Methods

The Cancer Genome Atlas NSCLC cohort was analyzed to compare SMARCA4-low tumors with adjacent normal lung using differential expression, Cox regression, protein–protein interaction networks, and machine-learning ranking. An exploratory immunohistochemistry case series of SMARCA4-dNSCLC from three centers (n=9) quantified DPP4 by percentage of tumor cells; ≥20% cells defined DPP4-high. Clinicopathologic features, RECIST response, and survival were summarized descriptively by DPP4 status. In HCC827 cells, SMARCA4 knockdown and DPP4 inhibition (P32/98) were evaluated alone and in combination using colony-formation, and motility (migration/wound-healing) assays.

Results

DPP4 was downregulated in SMARCA4-low tumors versus adjacent normal lung, yet within tumors higher residual DPP4 expression was associated with worse overall and disease-free survival. In the immunohistochemistry case series, DPP4 protein expression was weaker in SMARCA4-dNSCLC than in adjacent alveolar epithelium. Within this low-expressing background, DPP4-high tumors showed descriptive trends toward inferior survival, lower disease-control rate after first-line therapy, and more extensive baseline organ involvement than DPP4-low tumors. In vitro, SMARCA4 knockdown and P32/98 each reduced clonogenic growth and motility versus control, with the combination producing the greatest inhibition (about 60%–70% reductions).

Conclusions

DPP4 is downregulated at the mRNA and protein levels in SMARCA4-dNSCLC compared with normal lung, yet higher residual expression may mark a more aggressive phenotype. Combined SMARCA4 knockdown and DPP4 inhibition suppress growth and motility in vitro, suggesting DPP4 as a candidate prognostic marker and a candidate therapeutic target that requires orthogonal specificity controls and validation in larger cohorts.

1. Introduction

SMARCA4-deficient non-small cell lung cancer (SMARCA4-dNSCLC) is a subgroup of lung malignancies that has garnered substantial attention in recent years, characterized by atypical pathological features of NSCLC and manifesting as high aggressiveness and refractoriness. ¹ Morphologically, SMARCA4-dNSCLC can present a spectrum ranging from typical adenocarcinoma to poorly differentiated carcinoma, and even high-grade malignancies with rhabdoid phenotypes, exhibiting considerable variability in pathological differentiation, complex morphology, and a lack of specific markers.^2,3 Nicholson et al. ⁴ noted that the fifth edition of the WHO 2021 Thoracic Tumours classification has designated it as an independent entity, distinguishing it from other subtypes. In a cohort of SMARCA4-dNSCLC cases, Zhou et al ⁵ further highlighted its high malignancy and limited yet observable responses to immunotherapy; a retrospective study by Liang et al. ² demonstrated that survival in patients with SMARCA4-dNSCLC was significantly shorter than in those with SMARCA4-intact NSCLC (iNSCLC), supporting its classification as a high-risk subgroup. Collectively, this evidence underscores the necessity for precise subtyping and objective biomarker investigations within this disease spectrum.

From a disease burden perspective, lung cancer remains one of the malignancies imposing the heaviest global incidence and mortality burdens. According to GLOBOCAN 2022 estimates published by Bray et al., ⁶ there were approximately 2.5 million new lung cancer cases in 2022 (accounting for 12.4% of all cancers) and about 1.8 million deaths (18.7%), positioning it as the leading cause of cancer-related mortality. Integrating clinical studies on SMARCA4-related subgroups, Liu et al. ⁷ and other reviews/case series indicate that this subgroup exhibits poor responses to conventional chemotherapy, early recurrence, and limited overall survival. Currently, no established optimal diagnostic and therapeutic pathway is recognized, further underscoring the urgency for patient stratification and the development of verifiable biomarkers.

Dipeptidyl peptidase-4 (DPP4) is a type II transmembrane glycoprotein/serine exopeptidase with dual enzymatic and signal transduction functions, involved in multiple processes including glucose homeostasis, immune regulation, and inflammatory responses. ⁸ Klemann et al. ⁹ and Huang et al. ¹⁰ have both indicated that DPP4 is expressed in immune cells such as T/B cells, NK cells, dendritic cells, and macrophages, where it influences their activation states. In the respiratory system, Koyanagi et al. ¹¹ reported detectable DPP4 expression in type I/II alveolar epithelial cells, alveolar macrophages, and vascular endothelial cells in normal human lungs; Sato et al. ¹² also observed functional associations of DPP4 in alveolar macrophages and endothelial cells within acute inflammation models. These findings provide histological and cellular evidence for DPP4’s involvement in pulmonary immune and inflammatory processes. At the chemokine network level, Casrouge et al. ¹³ discovered that DPP4 cleaves the N-terminus of CXCL10 to generate an antagonistic truncated form, thereby attenuating CXCR3 signaling; Decalf et al. ¹⁴ validated this truncation phenomenon in humans and its restrictive effects on T cell/NK cell migration. This mechanism suggests that DPP4 may reshape the tumor immune microenvironment by modulating effector cell homing, although its directionality is strongly influenced by tissue and inflammatory contexts.

In recent years, the expression and function of DPP4 in various malignancies have garnered increasing attention. DPP4 exhibits pronounced tissue- and microenvironment-dependent “biphasic” characteristics. A pan-cancer omics assessment by Shu et al. ¹⁵ revealed correlations between DPP4 and multiple immune infiltration markers, although the directions of “expression-prognosis” associations were inconsistent across tumor types. Within this framework, high DPP4 expression was more frequently observed in colorectal cancer, malignant mesothelioma, and certain hematological malignancies, often linked to signals associated with tumor progression; conversely, observations in melanoma and ovarian cancer aligned with a suppressive phenotype, indicating context-dependent roles.^16,17 Regarding lung cancer, Kawashima et al. ¹⁸ observed via qPCR and histological comparisons that DPP4 levels were elevated in lung adenocarcinoma (LUAD) relative to paired normal tissues, whereas expression was relatively weak or negligible in lung squamous cell carcinoma (LUSC); Inoue et al. ¹⁹ further reported that DPP4-positive cancer-associated fibroblasts (CAFs) could promote LUAD cell growth through soluble factors, suggesting that tumor-stroma interactions may influence the directionality of DPP4-related pathways. Concurrently, studies have indicated that DPP4-associated signaling accompanies adverse biological behaviors such as epithelial-mesenchymal transition (EMT) and migration under specific microenvironmental or stress conditions, further underscoring the heterogeneity and context-dependency of this pathway. ²⁰ Because tumor–normal comparisons and within-tumor heterogeneity can yield different biological and clinical signals, we therefore examined both contrasts: DPP4 expression relative to normal lung and the prognostic value of residual DPP4 expression among tumors. In summary, the expression profiles and clinical implications of DPP4 across lung cancer subtypes require redefinition, with SMARCA4-dNSCLC representing a critical subgroup currently characterized by limited evidence and in need of systematic delineation.

2. Materials and Methods

2.1. Study Design and Reporting

This retrospective, multicenter, evidence-integration study comprised 3 analytic streams: ① bioinformatics analyses of public transcriptomic data; ②clinical histopathology and immunohistochemistry (IHC) as an exploratory descriptive case series (hypothesis-generating); and ③in-vitro cellular readouts. These streams were integrated to evaluate DPP4 as a candidate biomarker and therapeutic vulnerability. Reporting followed STROBE and REMARK guidelines for observational biomarker research. The statistical analysis plan prespecified primary and secondary endpoints, thresholds, and model settings. All tests were 2-sided. Given the small IHC sample size, clinical comparisons were descriptive/exploratory and not intended for definitive prognostic or predictive inference.

2.2. Public Data Acquisition, Preprocessing, and Stratification

Raw RNA-sequencing count matrices (STAR/HTSeq counts) and accompanying clinical data for non–small cell lung cancer (NSCLC) were obtained from The Cancer Genome Atlas (TCGA) portal (https://portal.gdc.cancer.gov). Quality control procedures removed samples with atypical sequencing depth and genes with near-zero expression across most samples, retaining only cases with complete clinical follow-up and key covariates. After filtering, 1017 NSCLC tumors with RNA-seq data and clinical information were included for transcriptomic analyses, together with 108 adjacent normal lung tissues available in TCGA.

Differential expression analyses were performed on raw count data using DESeq2 (version 1.34.0). For visualization and cross-sample comparison, expression values were transformed to transcripts per million (TPM) and expressed as log2(TPM + 1).

To approximate transcriptomic features associated with reduced SMARCA4 activity, tumors were stratified according to SMARCA4 mRNA expression levels. The lowest quartile of SMARCA4 expression was defined as SMARCA4-low, while the remaining tumors were categorized as SMARCA4-non-low, with the highest quartile sometimes referred to as SMARCA4-high for tumor-to-tumor comparisons. This transcript-based stratification was used as a biological approximation and exploratory grouping, and it does not necessarily correspond to SMARCA4 protein loss defined by immunohistochemistry (IHC) or to genomic loss-of-function mutations.

To provide orthogonal validation of transcript-based stratification, TCGA somatic mutation annotation files (MAF) were also retrieved. Tumors were classified as SMARCA4-altered when harboring putative loss-of-function events (including truncating, nonsense, frameshift, or splice-site mutations) and/or deep copy-number deletions where available; all other cases were considered SMARCA4-wild-type. These genomic annotations were used to evaluate whether candidate genes identified in the transcript-based screening remained associated with SMARCA4 genomic alteration status.

Because tumor–normal contrasts can reflect tissue composition differences rather than SMARCA4-specific biology, additional tumor-only analyses were prespecified. Specifically, we compared gene expression between SMARCA4-low and SMARCA4-high tumors, and between SMARCA4-mutant and SMARCA4-wild-type tumors, to assess whether prioritized candidates remained associated with SMARCA4-related tumor biology independent of tumor–normal comparisons.

Overall survival (OS) was defined as the interval from diagnosis to death or last follow-up. Primary analyses were conducted within TCGA; ComBat batch correction was applied only when cross-dataset harmonization was required.

2.3. Differential Expression and Construction of the “DE ∩ Prognostic” Candidate Set

SMARCA4-low tumors were contrasted with adjacent normal tissue using prespecified thresholds (false discovery rate [FDR] <0.001 via Benjamini–Hochberg and |log2 fold change| >0.7). Gene-wise Cox proportional hazards models (primary endpoint, OS) were fit across the full NSCLC cohort with Benjamini–Hochberg control (FDR <0.05). Prognostic genes were defined in full NSCLC cohort using these Cox models, yielding 722 genes total. The intersection of differentially expressed genes (DEGs) and prognostic genes formed the candidate set for protein–protein interaction (PPI) and machine-learning screening. Exact counts and gene lists are reported in the Results.

Functional enrichment used clusterProfiler (version 4.2.2; Gene Ontology Biological Process and KEGG), with the expressed gene universe as background; significance was defined as Benjamini–Hochberg–adjusted P <0.05.

2.4. Protein–Protein Interaction Network and Hub Candidates

The candidate set was submitted to STRING v11 (organism, Homo sapiens; default interaction score ≥0.4) to construct a PPI network. Nodes were ranked by degree (connectivity), and top-ranked nodes were designated hub candidates for model interpretation and visualization.

2.5. Machine-Learning Feature Selection and Robustness

Classification models were trained on the candidate set to discriminate SMARCA4-low tumors from normal tissue.

Random forest (RF) used the “randomForest” package with ntree = 1000 and mtry ≈ $\sqrt{\mathrm{p}}$ (p, number of features). Performance was assessed by 10-fold cross-validation (CV) and summarized by ROC AUC. Mean Decrease Gini quantified feature importance. Proximity-based multidimensional scaling (MDS) was performed for visualization. Least absolute shrinkage and selection operator (LASSO) logistic regression used “glmnet” (version 4.1-3) with 10-fold CV to select λ_min (with λ1se for sensitivity analysis). Features with nonzero coefficients were considered selected. k-nearest neighbors (KNN) used “FNN” (version 1.1.3.1), selecting k=3–40 by 10-fold CV. Permutation-based importance was used to estimate feature contribution.

Model robustness was further assessed via permutation tests and intersection analysis to identify overlapping features. To limit overfitting, genes retained by at least 2 algorithms were designated a priori as robust candidates for downstream analyses. Model metrics and overlapping selections are presented in the Results.

2.6. Clinical Cohort, Eligibility, and Data Governance

The clinical cohort comprised consecutive, treatment-naive NSCLC cases from Central Hospital of Guangdong Provincial Nongken, The Affiliated Hospital of Guangdong Medical University and Central People’s Hospital of Zhanjiang within a predefined window (January 2023 to January 2025). A subset meeting criteria for SMARCA4-deficient NSCLC constituted an exploratory IHC case series (n = 9) for descriptive clinicopathologic and outcome summarization.

Inclusion criteria were age ≥18 years; histopathologic confirmation of NSCLC (including adenocarcinoma and squamous cell carcinoma); no systemic or local antineoplastic therapy before sampling used in this study; availability of archived formalin-fixed, paraffin-embedded (FFPE) tumor tissue with paired adjacent normal tissue when feasible; complete baseline and follow-up data; and written informed consent or an institutional review board–approved waiver. Exclusion criteria included prior malignancy; severe comorbidity compromising survival; inadequate tissue quantity/quality; or missing key data precluding analysis. Within this cohort, SMARCA4-deficient cases with adequate FFPE material and follow-up were identified for descriptive IHC-based biomarker analyses; given limited sample size, survival and response summaries were hypothesis-generating.

Data were abstracted into a unified case report form (CRF/EDC) by trained personnel with dual entry and reconciliation. Logical rules were applied for internal consistency and completeness. All records were deidentified and assigned unique study identifiers.

2.7. Immunohistochemistry and Pathology Assessment

2.7.1. Specimen Source and Pre-Analytical Handling

All FFPE specimens were obtained by image-guided core needle biopsy. Cold ischemia time was kept <1 hour. Tissues were fixed in 10% neutral buffered formalin for 6–24 hours, processed routinely, embedded in paraffin, sectioned at 3 μm, and baked at 60 °C for 1 hour. H&E staining was used for morphologic assessment and verification of tumor content, avoiding areas of necrosis or hemorrhage.

2.7.2. IHC Workflow

Staining was performed on the Ventana Benchmark Ultra platform. Antigen retrieval used CC1 (Tris–EDTA, pH ≈8.5) at 95–100 °C for 64 minutes; peroxidase blocking followed the platform’s standard procedure. Primary antibodies were incubated for 32–60 minutes. Detection used OptiView DAB IHC (Roche), followed by hematoxylin counterstain and mounting with neutral resin.

2.7.3. Antibodies

SMARCA4/BRG1 rabbit monoclonal EPNCIR111A (Abcam; 1:100) and DPP4/CD26 rabbit monoclonal D9E8B (Cell Signaling Technology; 1:150).

2.7.4. Quality Control

Each batch included tissue positive controls (e.g., tonsil or kidney) and a negative control (omission of primary antibody). Stromal, endothelial, and inflammatory cells on the same slide served as internal positive controls.

2.7.5. Scoring and Definitions

Two senior pathologists independently reviewed slides under masking; discrepancies were adjudicated by a third.

• SMARCA4 status: In the presence of an internal positive control, absent or only weak nuclear staining in tumor cells involving ≥ the prespecified coverage threshold (e.g., ≥50%) was defined as protein “loss.”

• DPP4 quantification:

① H-score = Σ (staining intensity 0–3 × percentage of tumor cells at that intensity, 0–100); range 0–300. For each case, the mean of five tumor-enriched high-power fields (×400) was recorded.

② Percentage of positive tumor cells was also measured, with 20% used as the primary dichotomous cutoff; median split and the H-score as a continuous variable were used for sensitivity analyses.

2.7.6. Inter-Rater Agreement

Cohen’s κ and the intraclass correlation coefficient (ICC; two-way random, absolute agreement) were reported.

2.7.7. Clinical Variables and Stratification

The variables recorded and analyzed in this IHC case series (descriptive/exploratory) included age, sex, organ involvement (bone, liver, pleura, lymph nodes, etc.), treatment category (chemotherapy, immunotherapy-based regimens, radiotherapy/supportive care, etc.), DPP4 expression (percentage positivity and H-score), and follow-up duration and outcome.

2.8. Cell Lines, Genetic Perturbation, and Functional Assays

2.8.1. Cell lines and Culture Conditions

Human NSCLC cell lines HCC827 and A549 (short tandem repeat–authenticated and routinely mycoplasma-negative) were used. HCC827 cells were maintained in RPMI-1640; A549 cells were cultured in RPMI-1640 or F-12 as appropriate. All media were supplemented with 10% heat-inactivated, single-lot fetal bovine serum and 1% penicillin–streptomycin, and cells were incubated at 37 °C in a humidified atmosphere of 5% CO₂. Experiments were conducted with cells between passages 3 and 20 after thaw and during the logarithmic growth phase; treatments were initiated at ∼70%–90% confluence.

2.8.2. siRNA Transfection and Knockdown Verification (SMARCA4)

Three siRNAs targeting SMARCA4 (designated si-SMARCA4-1, si-SMARCA4-2, si-SMARCA4-3; sequences/cat. Nos. In Supplementary Table S1) and a non-targeting negative control (si-NC) were used. Cells were transfected with 50 nM siRNA using Lipofectamine 3000 per the manufacturer’s instructions; antibiotic-free medium was used during transfection and replaced with complete medium ∼6 h later. Knockdown efficiency was assessed 48–72 h post-transfection by qPCR (and Western blot when applicable). A ≥70% reduction at the mRNA level and/or ≥50% at the protein level was prespecified as adequate knockdown. The most efficient siRNA was advanced to functional assays. Where applicable, a mock control (transfection reagent only) was included. Each condition had ≥3 biological replicates with ≥3 technical replicates per experiment.

2.8.3. DPP4 Inhibition and Experimental Groups

DPP4 was pharmacologically inhibited using the specific inhibitor P32/98. Stock solutions were prepared in dimethyl sulfoxide (DMSO), and working solutions were freshly prepared prior to each experiment to ensure efficacy. A preliminary dose-time optimization assay was performed to determine the optimal experimental conditions, with cell viability as the primary endpoint.

For the optimization assay, HCC827 cells were treated with P32/98 at four gradient concentrations (0 µM, 10 µM, 20 µM, 50 µM) and incubated for three different durations (12 h, 24 h, 48 h). The 0 µM group served as the DMSO vehicle control, and the volume fraction of DMSO was strictly standardized across all experimental conditions (≤0.1%) to eliminate solvent-induced bias. Cell viability was detected by the CCK-8 assay, and the optimization was evaluated based on three key criteria: the degree of cell viability inhibition, the maintenance of sufficient cell activity for subsequent functional experiments, and the reproducibility of experimental data across biological replicates. The formal experimental groups were designed in line with the corresponding figures, comprising four conditions: HCC827 (blank control group), HCC827+P32/98 (P32/98 single treatment group), si-SMARCA4-HCC827 (SMARCA4 knockdown group), and si-SMARCA4-HCC827+P32/98 (SMARCA4 knockdown combined with P32/98 treatment group). No additional pharmacologic agents were administered in the experiment.

2.8.4. Quantitative Real-Time PCR (qPCR) for SMARCA4 and DPP4

Total RNA was extracted with TRIzol (Thermo Fisher Scientific; cat. No. 15596018) and treated with DNase to remove genomic DNA. Samples were required to have an A260/280 ratio of 1.8–2.1. Reverse transcription was performed with commercial kits, and amplification was carried out on an ABI 7500 or equivalent platform using SYBR Green chemistry (Thermo Fisher Scientific; cat. No. A25742) in a 20-µL reaction. Cycling conditions were: 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s; melt-curve analysis was performed to verify a single amplicon. GAPDH served as the internal control, and relative expression was calculated by the 2^-ΔΔCt method. Primer sequences, amplicon lengths, and annealing temperatures are listed (Supplementary Table S2). Primer efficiencies were maintained between 90% and 110%. A no-template control (NTC) was included on each plate, and the SD of Ct among technical replicates was kept <0.5.

2.8.5. Cell Viability (CCK-8)

Cells were seeded in 96-well plates at 2,000–5,000 cells per well (a fixed density within each experiment) and allowed to adhere for 12–16 hours before transfection/drug treatment. At 24, 48, and 72 hours, CCK-8 reagent (Dojindo; no. CK04) was added for 1–2 hours and absorbance was read at 450 nm. Medium-only wells with CCK-8 served for background subtraction. Each condition included ≥3 biological replicates and ≥3 technical replicates. Readouts were normalized to the 0-hour value of the same group or to the HCC827 group, as specified.

2.8.6. Colony Formation Assay

Cells were seeded in 6-well plates at 500–1,000 cells per well and cultured continuously for 10–14 days with medium changes every 3–4 days. Colonies were fixed with 4% paraformaldehyde for 20 minutes, stained with 0.1% crystal violet (Sigma-Aldrich; no. C0775) for 20 minutes, rinsed with PBS, and air-dried. Colonies containing ≥50 cells were counted, and colony numbers and areas were quantified with ImageJ; values from multiple wells were averaged.

2.8.7. Migration Assays

For the wound-healing (scratch) assay, confluent monolayers (90%–100%) were scratched with a sterile 200-µL pipette tip. Suspended cells were removed with PBS, and cultures were maintained in serum-free or low-serum (1% FBS) medium to minimize proliferation-driven closure. Images were captured at 0 and 24 hours, and wound closure was quantified as (initial gap width − residual gap width)/initial gap width.

For Transwell migration assays, inserts with 8-µm pores were used without any extracellular-matrix coating. Cells (2∼5×10⁴ per insert) were seeded in the upper chamber in serum-free medium, and medium containing 10% FBS was placed in the lower chamber as a chemoattractant. After 24 hours of incubation (optimized for HCC827), non-migrated cells on the upper surface of the membrane were gently removed with cotton swabs. Cells on the lower surface were fixed with methanol, stained with crystal violet, and counted in at least five random fields per well at 200× magnification. The migration index was calculated relative to the control group.

2.8.8. Experimental Setup, Masking, and Quality Control

Control and treatment conditions were set up concurrently within each batch. Image acquisition and cell counting were performed under masking. Solvent content (DMSO) was matched across groups at ≤0.1% (vol/vol). Results from batches with mycoplasma contamination, abnormal detachment/contamination, qPCR melt curves indicating double or nonspecific peaks, or inadequate siRNA efficiency (qPCR reduction <50%) were excluded a priori. Key experiments were independently repeated at least three times, with ≥3 technical replicates, and intra- and interbatch consistency was monitored.

2.9. Statistical Analysis

R (4.2.2) and GraphPad Prism (9.5.1) were used. DESeq2 (1.34.0) identified differentially expressed genes (BH-FDR < .001; |log₂FC| > 0.7); GO/KEGG enrichment (clusterProfiler 4.2.2) used adjusted P < 0.05. Gene-wise Cox models for OS were fit across the full NSCLC cohort with FDR < 0.05; TCGA Kaplan–Meier curves used median TPM splits and log-rank tests. Random forest, LASSO, and k-nearest neighbors classified SMARCA4-low tumors vs adjacent normals via 10-fold cross-validation (ROC AUC); features selected by ≥2 methods were considered robust. In the IHC case series (n = 9), analyses were exploratory and primarily descriptive. Where formal comparisons were reported, Mann–Whitney U and Fisher exact tests were used; DPP4 was analyzed by a 20% positivity dichotomy and by H-score (per SD), with Kaplan–Meier/log-rank used only for exploratory visualization of survival trends given limited power. Cellular assays used Shapiro–Wilk/Levene pretests, t tests/ANOVA with Tukey or Mann–Whitney/Kruskal–Wallis as appropriate, with BH-FDR where relevant. Data are mean (SEM) or mean (SD) as indicated; tests were 2-sided (α = 0.05); complete-case analyses with sensitivity checks; random seeds were fixed.

2.10. Ethics

The protocol was approved by the Institutional Review Board of Central Hospital of Guangdong Provincial Nongken (Approval No. 25124), with an informed-consent waiver for this retrospective, deidentified analysis.

3. Results

3.1. Bioinformatics Screening for Differential Expression and Prognostic Genes in SMARCA4-Deficient NSCLC

From the TCGA NSCLC dataset, 1017 cases with complete RNA-sequencing and clinical data were retained after quality control. To approximate SMARCA4-deficient–like biology at the transcript level, tumors in the lowest quartile of SMARCA4 mRNA expression were classified as SMARCA4-low (n = 225), while 108 adjacent normal lung tissues served as controls. Differential expression analysis using DESeq2 (FDR < 0.001 and |log2 fold change| > 0.7) identified 4898 differentially expressed genes (DEGs), including 1540 upregulated and 3358 downregulated genes (Figure 1A; Supplementary Table S3). Among these, DPP4 was significantly downregulated in SMARCA4-low tumors relative to adjacent normal lung tissue (log2 fold change ≈ −0.73; FDR ≈ 2.6 × 10^-5) and appeared on the left side of the volcano plot (Figure 1A).

Figure 1.

Bioinformatic discovery and orthogonal validation of DPP4 in SMARCA4-related NSCLC. A: Volcano plot of differential expression between SMARCA4-low tumors and adjacent normal lung tissues in TCGA (DESeq2; FDR <0.001, |log2FC| >0.7). DPP4 is annotated. B: Overlap between differentially expressed genes (DEGs) and overall survival–associated prognostic genes, yielding 245 intersecting candidates. C: Landscape of somatic mutations in the TCGA NSCLC cohort with available mutation annotations (n = 575). D: Oncoprint showing alteration status of SMARCA4 and DPP4. E: DPP4 mutation frequency stratified by SMARCA4 mutation status. F:Comparison of DPP4 mRNA expression between SMARCA4-mutant and SMARCA4-wild-type tumors. G: Tumor-to-tumor comparison showing DPP4 expression between SMARCA4-low and SMARCA4-high tumors

To prioritize candidates with both biological and clinical relevance, gene-wise Cox proportional hazards analyses were performed across the full NSCLC cohort, identifying 722 genes associated with overall survival (FDR < 0.05; Supplementary Table S4). Intersecting these prognostic genes with the 4898 DEGs yielded 245 overlapping candidates with both differential-expression and prognostic significance (Figure 1B). Thus, DPP4 was retained within the joint differential-expression/prognostic candidate space for downstream prioritization.

Because tumor–normal contrasts may be influenced by tissue composition and generic tumor biology, we next performed orthogonal validation using tumor-only genomic and transcriptomic comparisons. In the TCGA subset with available somatic mutation annotations (n = 575), the overall mutation landscape is shown, and SMARCA4 alterations were identified in 55 of 575 tumors (9.57%) (Figures 1C–D). Although DPP4 mutation events were infrequent and showed no major enrichment by SMARCA4 mutation status (Figure 1E), DPP4 mRNA expression was significantly lower in SMARCA4-mutant tumors than in SMARCA4-wild-type tumors (Figure 1F). In addition, a direct tumor-to-tumor comparison demonstrated that DPP4 expression remained significantly lower in SMARCA4-low tumors than in SMARCA4-high tumors (Figure 1G). Taken together, these orthogonal analyses indicate that reduced DPP4 expression is linked to SMARCA4-related tumor biology and is not solely driven by tumor–normal differences.

3.2. Protein-Protein Interaction Network Analysis

Using STRING, we constructed a protein–protein interaction network from the 245 intersecting genes, yielding a single connected network (245 nodes; mean degree, 5.2). Nodes ranked by degree centrality identified 15 hubs (eg, GAPDH, YWHAZ, and DPP4). Functional annotation indicated predominant enrichment of metabolic and cell-cycle pathways, consistent with a tumor-regulatory module (Supplementary Figure 1A–1B).

3.3. Machine-Learning Feature Selection

We next applied three complementary algorithms—random forest (RF), LASSO logistic regression, and k-nearest neighbors (KNN)—with 10-fold cross-validation to the candidate gene set to discriminate SMARCA4-low tumors from adjacent normal tissue. In RF (Figure 2A–C), proximity-based multidimensional scaling showed clear class separation (Figure 2A), and the receiver operating characteristic (ROC) curve indicated excellent performance (AUC of 1.00; Figure 2B). Feature importance ranked by mean decrease in Gini highlighted heterogeneous gene contributions, with DPP4 and YWHAZ among the top variables (Figure 2C). Results were concordant in LASSO (Figure 2D–F): coefficients shrank with increasing penalty, leaving a small set of nonzero features at the optimal λ_min (Figure 2D), and cross-validated discrimination remained near-perfect (AUC of 0.99958; Figure 2E), with DPP4 and YWHAZ again prominent by absolute weight (Figure 2F). Finally, KNN (Figure 2G–I) identified a data-driven optimum across a grid of k values from 3 to 40 (Figure 2G) and achieved a cross-validated AUC of 0.999 (Figure 2H); permutation-based and approximate SHAP importance estimates pointed to a similar high-contribution set (Figure 2I). Across methods, the intersection of selected features yielded two robust candidates—DPP4 and YWHAZ—with DPP4 consistently carrying high weight, indicating a reproducible and strong contribution to class separation (Figure 2J). Importantly, DPP4 was prioritized based on convergent evidence across prespecified analytic streams (DE, prognostic association, PPI hub status, and orthogonal ML feature selection), rather than reliance on a single threshold or a single model. A concise multi-criterion prioritization and robustness summary for DPP4 and YWHAZ is provided in Supplementary Table S5.

Figure 2.

Machine-learning models distinguish SMARCA4-low tumors from adjacent normal tissue and converge on DPP4/YWHAZ. A: Random-forest (RF) proximity–based multidimensional scaling showing clear class separation. B: RF receiver operating characteristic (ROC) curve (AUC, 1.00). C: RF feature importance for top genes (mean decrease in Gini; radar plot). D: LASSO coefficient path with cross-validated error across log(λ), illustrating sparse selection at the optimal penalty. E: LASSO ROC curve (AUC, 0.99958). F: LASSO regression weights for selected features (absolute values ranked). G: k-nearest neighbors (KNN) cross-validation accuracy across k (grid 3–40) identifying the optimal k. H: KNN ROC curve (AUC, 0.999). I: KNN feature importance (permutation/SHAP-like estimate). J: Venn diagram of genes selected by RF, LASSO, and KNN; the intersection yields 2 robust candidates (DPP4 and YWHAZ)

3.4. Survival Analyses

In survival analyses, patients were stratified by the median tumor DPP4 mRNA level. Kaplan–Meier curves separated early and remained clearly divergent: high DPP4 expression was associated with significantly worse overall survival (log-rank P = 2.4×10^-5; Figure 3A) and shorter disease-free survival (log-rank P = 0.0079; Figure 3B) compared with low DPP4 expression. By contrast, high YWHAZ expression was associated with poorer overall survival (log-rank P = 9.4×10^-4; Figure 3C), whereas disease-free survival did not differ materially between YWHAZ strata (log-rank P = 0.97; Figure 3D). Within the same dataset, DPP4 expression was lower in SMARCA4-low tumors than in adjacent normal tissue (P < 0.001). In multivariable Cox models adjusting for age, sex, stage, histology, and smoking status, DPP4 expression remained independently associated with overall survival (P < 0.05), in line with the Kaplan–Meier analyses and supporting DPP4 as an adverse prognostic marker in SMARCA4-dNSCLC.

Figure 3.

Kaplan–Meier survival by tumor DPP4 and YWHAZ mRNA in the TCGA NSCLC cohort. A: Kaplan–Meier overall survival stratified by low vs high DPP4 TPM (median dichotomy; two-sided log-rank P < 0.001). B: Kaplan–Meier disease-free survival stratified by low vs high DPP4 TPM (median dichotomy; two-sided log-rank P = 0.0079). C: Kaplan–Meier overall survival stratified by low vs high YWHAZ TPM (median dichotomy; two-sided log-rank P < 0.001). D: Kaplan–Meier disease-free survival stratified by low vs high YWHAZ TPM (median dichotomy; two-sided log-rank P = 0.97)

3.5. Exploratory Case Series: Immunohistochemical Validation of DPP4 and Clinicopathologic Correlates in SMARCA4-Deficient NSCLC

We assembled a deidentified, consecutive case series of SMARCA4-deficient NSCLC from three centers (Supplementary Table S6). Tumor DPP4 expression was quantified by immunohistochemistry and prespecified as DPP4-high when ≥20% of tumor cells stained positive and DPP4-low when <20%; the H-score was analyzed as a continuous measure with a median split for sensitivity analyses. Representative H&E and DPP4-IHC images across histologic patterns are provided (Figure 4A–F). Compared with adjacent normal alveolar epithelium, tumor cells in well-differentiated adenocarcinomas with acinar or lepidic growth showed lower DPP4 membranous staining (Figure 4A,B). A similar pattern of reduced or absent DPP4 staining in tumor nests was observed in poorly differentiated solid and micropapillary adenocarcinomas (Figure 4C–F), supporting overall downregulation of DPP4 protein across different histologic grades in SMARCA4-dNSCLC. Kaplan–Meier curves stratified by DPP4 status showed visual separation over follow-up, suggesting a possible survival difference by DPP4 status (Figure 4G); given the small sample size, this observation should be interpreted descriptively. Multivariable Cox models were not fitted given the number of events. Group comparisons used the Mann–Whitney U test or Fisher exact test, as appropriate. Age, sex, and histologic subtype were similarly distributed. After first-line therapy, the distribution of best RECIST responses suggested a lower disease-control rate in the DPP4-high group (Figure 4H). Compared with the DPP4-low group, the DPP4-high group showed a descriptively higher proportion of baseline organ involvement (eg, bone, liver, pleura, mediastinal nodes) (Figure 4I). Taken together with the TCGA transcriptomic analyses, these immunohistochemical findings are consistent with relative DPP4 downregulation being a potential characteristic feature of SMARCA4-deficient NSCLC. Exact values for all comparisons are reported. Given the exploratory nature of this case series, clinical outcome and response findings are presented as descriptive/hypothesis-generating. Accordingly, no predictive claims are made from this cohort, and effect estimates should be interpreted with caution pending independent validation.

Figure 4.

Immunohistochemical DPP4 expression and clinicopathologic correlates in SMARCA4-deficient NSCLC representative hematoxylin–eosin (H&E) and immunohistochemistry (IHC) images and clinical correlates stratified by tumor DPP4 status (primary dichotomy: DPP4-high ≥20% positive tumor cells vs DPP4-low <20%, per Methods). A: Well-differentiated adenocarcinoma (acinar/lepidic patterns). B: DPP4 expression in well-differentiated (acinar/lepidic) adenocarcinoma. C: Poorly differentiated solid adenocarcinoma. D: DPP4 expression in poorly differentiated solid adenocarcinoma. E: Poorly differentiated micropapillary adenocarcinoma. F: DPP4 expression in poorly differentiated micropapillary adenocarcinoma. G: Kaplan–Meier survival curves comparing high vs low DPP4 expression. H: χ² analysis of post-treatment response by DPP4 status (high vs low). I: Analysis of organ involvement by DPP4 status (high vs low)

3.6. Selection of a DPP4-High Cell Line and SMARCA4 siRNA

Across HepG2, A549, and HCC827 cells, qPCR (GAPDH-normalized, 2^−ΔΔCt; n=3) showed significant between-group differences in DPP4 mRNA (Supplementary Figure 2A; P<0.001). Mean (SD) relative expression was 1.014 (0.162) in HepG2 (CV 15.9%), 6.633 (0.982) in A549 (CV 14.8%), and 38.994 (2.876) in HCC827 (CV 7.4%) (Supplementary Table S7). Pairwise tests indicated HCC827>A549>HepG2 (all P<0.01). Melt curves were single-peaked with T_m variability <1%, supporting assay specificity (Supplementary Figures 3A–3B). On this basis, HCC827 was selected as the primary model.

In HCC827, three candidate siRNAs (si-SMARCA4-1/si-SMARCA4-2/si-SMARCA4-3) targeting SMARCA4 were screened. qPCR quality control was acceptable (sigmoidal amplification, single-peak melt; T_m≈84–86°C; Supplementary Figures 4A-4B). Relative expression (2^−ΔΔCt; n=3) was 1.002 (0.074) for the non-targeting control (NC), 0.065 (0.076) for si-SMARCA4-1 (≈93.5% reduction), 0.262 (0.047) for si-SMARCA4-2 (≈73.8% reduction), and 0.121 (0.006) for si-SMARCA4-3 (≈87.9% reduction). Group differences were significant (ANOVA, F=48.36, P<0.001). Tukey post hoc tests showed all siRNAs reduced SMARCA4 vs NC (all P<0.001); si-SMARCA4-1 outperformed si-SMARCA4-2 and si-SMARCA4-3 (both P<0.001), whereas si-SMARCA4-2 vs si-SMARCA4-3 was not significantly different (P>0.05). Accordingly, si-SMARCA4-1 was selected for downstream functional assays (Supplementary Figure 2B; Supplementary Table S8).

3.7. Dose–Time Optimization of the DPP4 Inhibitor P32/98

Across all time points, cell viability differed significantly among concentrations (12 hours: F=28.143, P<0.001; 24 hours: F=96.321, P<0.001; 48 hours: F=54.274, P<0.001).

At 12 hours: 0 µM (control) 1.000±0.028; 10 µM 0.721±0.048 (vs control, P<0.01); 20 µM 0.623±0.057 (vs 0 µM, P<0.001; vs 10 µM, P<0.01); 50 µM 0.298±0.036 (vs 20 µM, P<0.01).

At 24 hours: 0 µM 1.000±0.021; 10 µM 0.897±0.056 (vs control, P>0.05); 20 µM 0.702±0.068 (vs control, P<0.01); 50 µM 0.103±0.029 (≈89.7% inhibition), showing a near–floor effect vs all other groups (all P<0.001).

At 48 hours: 0 µM 1.000±0.031; 10 µM 0.732±0.049; 20 µM 0.661±0.052; 50 µM 0.384±0.047; all three concentrations were lower than control (all P<0.001), and the difference between 10 and 20 µM approached significance (P=0.082), suggesting convergence with prolonged exposure.

Integrating effect size, assay window, and reproducibility, 20 µM for 12 hours in HCC827 cells produced an early ≈37.7% reduction in viability while preserving sufficient residual viability for wound-healing, Transwell, and colony assays; within-batch variability was acceptable (CV≈9.1%). Accordingly, 20 µM × 12 hours was adopted as the standard P32/98 condition for downstream experiments (Supplementary Figure 5; Supplementary Table S9).

3.8. Effects of SMARCA4 Knockdown and/or P32/98 on DPP4/SMARCA4 Transcripts

To ensure assay specificity and stability, qPCR for DPP4 underwent quality control: amplification curves were sigmoidal, within-plate Ct variability was <0.5, and melt curves showed a single sharp peak (T_m ≈ 84–86 °C) without nonspecific products or primer–dimers (Supplementary Figures 6A–6B). On this basis, DPP4 and SMARCA4 mRNA were quantified in HCC827 cells after four treatments (Control, si-SMARCA4, P32/98, si-SMARCA4+P32/98) (n=3 biological replicates). DPP4 expression differed significantly among groups (one-way ANOVA: F=138.62, P<0.001; Supplementary Figure 7): Control, 1.000 ± 0.013; si-SMARCA4, 0.748 ± 0.010 (−25.2% vs Control; P<0.001); P32/98, 0.597 ± 0.008 (−40.3%; P<0.001), lower than si-SMARCA4 (P<0.01); and si-SMARCA4+P32/98, 0.497 ± 0.008 (−50.3%), lower than si-SMARCA4 (P<0.001) and P32/98 (P<0.01), indicating an additive suppressive effect.

3.9. Functional Phenotype Analysis: Combined Effects of SMARCA4 Knockdown and P32/98 on Proliferation, Migration, and Invasion of HCC827 Cells

3.9.1. Additive Inhibition of Colony-Forming Ability

Representative colony-formation images showed many dense colonies in parental HCC827 cells, fewer and smaller colonies after SMARCA4 knockdown or P32/98 monotherapy, and the sparsest colonies in the combined-treatment group (Figure 5A). Quantitatively, the bar chart (Figure 5B) demonstrated marked between-group differences in colony-forming efficiency normalized to the HCC827 control (one-way ANOVA, F = 217.356, P < 0.001). Mean (SD) relative colony-forming efficiency was 1.562 ± 0.026 for the control group, 1.317 ± 0.023 for the si-SMARCA4 group (−15.7% vs control; P < 0.01), 1.228 ± 0.020 for the P32/98 monotherapy group (−21.3% vs control; P < 0.01), and 0.646 ± 0.012 for the combination group (−58.6% vs control; P < 0.001 vs control and P < 0.01 vs either single arm). These findings indicate that concurrent SMARCA4 knockdown and DPP4 inhibition produced a markedly greater reduction in colony-forming ability than either intervention alone, consistent with an at least additive inhibitory effect.

Figure 5.

Additive inhibition of HCC827 colony-forming ability by SMARCA4 knockdown and DPP4 inhibition

3.9.2. Inhibition of Migratory Capacity in Transwell Assays

Representative Transwell migration images showed abundant migrated HCC827 cells in the control group, fewer migrated cells after SMARCA4 knockdown or P32/98 monotherapy, and the lowest number of migrated cells with the combination treatment (Figure 6A). Quantitatively, the migration index normalized to the control differed significantly across groups (one-way ANOVA, F = 136.13, P < 0.001; Figure 6B): 1.000 ± 0.034 for the control group, 0.797 ± 0.058 for the si-SMARCA4 group (−20.3% vs control; P < 0.01), 0.613 ± 0.045 for the P32/98 group (−38.7% vs control; P < 0.001), and 0.344 ± 0.018 for the combination group (−65.6% vs control; P < 0.001 vs any single treatment). These findings indicate that DPP4 inhibition reduces the migratory capacity of HCC827 cells and that concurrent SMARCA4 knockdown further enhances this inhibitory effect.

Figure 6.

Inhibition of HCC827 cell migration by SMARCA4 knockdown and DPP4 inhibition in transwell assays

3.9.3. Pronounced Inhibition of Wound Closure

Representative wound-healing (scratch) images at 0 h showed a comparable initial gap width across all four groups, whereas at 24 h the control HCC827 monolayers exhibited substantial wound closure, si-SMARCA4 and P32/98 monotherapy produced only partial closure, and the combination group retained the widest residual gap (Supplementary Figure 8A). Quantitatively, wound-healing assays normalized to the initial gap width demonstrated significant between-group differences in closure at 24 h (one-way ANOVA, F = 396.22, P < 0.001; Supplementary Figure 8B). Relative closure was 0.577 ± 0.017 for the control group, 0.417 ± 0.011 for si-SMARCA4 (−27.8% vs control; P < 0.01), 0.385 ± 0.010 for P32/98 (−33.3% vs control; P < 0.01), and 0.250 ± 0.006 for the combination (−56.6% vs control; P < 0.001 vs either single arm). The visual contrast between 0 h and 24 h thus corroborates the quantitative findings, indicating that concurrent SMARCA4 knockdown and DPP4 inhibition most strongly suppress wound closure and lateral migration of HCC827 cells.

All functional assays used n=3 biological replicates with ≥3 technical replicates under blinded counting (colony defined as ≥50 cells; Transwell counts from ≥5 random 200× fields per well; wound closure normalized to baseline width). These phenotypic data align with the observed transcriptional suppression of DPP4 and collectively indicate that SMARCA4 knockdown together with DPP4 inhibition exerts an additive/cooperative restraint on proliferation and migration–invasion behaviors of HCC827 cells.

4. Discussion

SMARCA4-deficient non–small cell lung cancer (SMARCA4-dNSCLC) is an increasingly recognized subtype characterized by high aggressiveness, treatment resistance, and dismal prognosis. ²¹ Loss of SMARCA4, a core subunit of the SWI/SNF chromatin-remodeling complex, is observed in approximately 5%–10% of NSCLC cases and is enriched in pulmonary sarcomatoid carcinoma and undifferentiated/poorly differentiated adenocarcinoma.^22,23 Clinically, SMARCA4-dNSCLC is strongly associated with a history of tobacco exposure (>70% of patients are smokers) and occurs predominantly in men with advanced-stage disease at diagnosis (≈60% stage IV). Histopathologically, these tumors often display poorly differentiated, highly invasive morphology with frequent sarcomatoid or rhabdomyosarcomatous features. Immunohistochemically, they are defined by complete loss of SMARCA4 (BRG1) expression and commonly harbor co-occurring mutations in TP53, KRAS, STK11, and other oncogenic drivers. Systemic chemotherapy remains the mainstay of treatment; however, platinum-based regimens show limited activity, with reported objective response rates (ORR) of only 10%–20% and median progression-free survival (mPFS) of approximately 2–3 months. ²⁴ Responses to immune checkpoint inhibitors (ICIs) targeting PD-1/PD-L1 are similarly modest (ORR <15%), likely reflecting insufficient T-cell infiltration and an intrinsically immunosuppressive tumor microenvironment (TME). Although synthetic-lethal approaches targeting SMARCA4 loss—such as ARID1A inhibition or EZH2 inhibitors (eg, tazemetostat)—have shown promise in preclinical models, ²⁵ clinical trial data remain sparse. At present, loss of BRG1 by IHC is the diagnostic gold standard for SMARCA4-dNSCLC, but robust biomarkers that can predict prognosis or guide therapeutic decision-making in this subtype are lacking. Identifying new molecular markers and actionable vulnerabilities is therefore a critical unmet need.

This study focused on the potential role of dipeptidyl peptidase 4 (DPP4) in SMARCA4-dNSCLC. DPP4, also known as CD26, is a widely expressed 110-kDa type II transmembrane glycoprotein with serine peptidase activity. It is present on the surface of multiple cell types, including melanocytes, epithelial cells, and lymphocytes, and participates in cell adhesion (as a functional receptor for collagen) and immune regulation.^26,27 A soluble form (sCD26) is detectable in plasma, serum, and urine and largely derives from shedding of the membrane-bound protein ²⁸ . Prior studies have reported markedly heterogeneous DPP4 expression patterns across human cancers: DPP4 overexpression has been observed in thyroid, breast, prostate, and ovarian cancers and has been linked to tumor initiation, progression, invasion, and metastasis,^29,30 whereas decreased or absent DPP4 expression has been reported in homogenates of colorectal, renal, lung, and hepatic tumors, as well as in certain transformed cancer cell lines. ³¹ This tumor- and context-dependent expression pattern suggests that the biological role of DPP4 in malignancy is highly contingent on tumor type and microenvironmental context.

Our findings collectively support a tendency toward reduced DPP4 expression in NSCLC, particularly in SMARCA4-related subsets. In the TCGA-based bioinformatic analysis, we used the lowest quartile of SMARCA4 mRNA expression to define a “SMARCA4-low” NSCLC stratum as a transcriptomic approximation of SMARCA4-deficient biology. Within this stratum, DPP4 transcripts were significantly downregulated in tumors compared with matched adjacent normal lung tissue, and DPP4 was retained within the intersection of differentially expressed and prognostic genes. To validate these observations at the protein level, we performed DPP4 IHC in a clinical series of SMARCA4-dNSCLC cases. Across both relatively well-differentiated adenocarcinomas (acinar and lepidic patterns) and poorly differentiated adenocarcinomas (solid and micropapillary patterns), tumor cells showed overall weaker membranous DPP4 staining than adjacent nonneoplastic alveolar epithelium, and some cases were nearly DPP4-negative. Given the small sample size, our data do not support firm conclusions regarding the relationship between histologic grade and DPP4 expression. However, the overall pattern of reduced DPP4 staining in tumor cells is concordant with the transcriptomic downregulation observed in TCGA and is consistent with prior reports of decreased DPP4 expression in lung cancer tissue. ³¹ These observations suggest that relative downregulation of DPP4 at both the mRNA and protein levels may represent one molecular feature of SMARCA4-related NSCLC. Because tumor–normal contrasts and within-tumor heterogeneity can yield different biological and clinical signals, we prespecified both comparisons, namely expression change relative to normal lung and the prognostic value of residual DPP4 expression among tumors.

We also identified a clinically relevant, “tiered” association between DPP4 expression and outcome. In the TCGA NSCLC cohort, stratification by the median tumor DPP4 mRNA level revealed that patients with higher DPP4 expression had significantly worse overall survival and shorter disease-free survival than those with lower DPP4 expression, indicating that among tumors in which DPP4 is already downregulated compared with normal lung, relatively higher residual DPP4 levels are associated with poorer prognosis. Multivariable Cox models supported an association between DPP4 expression and survival after adjustment for key clinical covariates. In our independent IHC case series (n=9, of whom 8 contributed to survival and response analyses), using 20% DPP4-positive tumor cells as a prespecified cutoff, patients in the DPP4-low group showed descriptive trends toward more favorable survival, better first-line disease control, and less extensive baseline organ involvement than those in the DPP4-high group. Although this series is small and the corresponding confidence intervals are wide, the directional concordance with the TCGA survival analysis is notable: within SMARCA4-related NSCLC, relatively higher DPP4 expression may be associated with worse clinical outcomes and more extensive metastatic burden. These observations suggest that DPP4 could serve as a candidate biomarker for further evaluation in risk stratification and prognostic research in this subtype; however, they remain hypothesis-generating and require validation in larger cohorts.

On the basis of these exploratory clinical observations and the TCGA-based analyses, we further hypothesized that DPP4 might constitute a therapeutic vulnerability in SMARCA4-dNSCLC, even if its baseline expression is not high. To explore this possibility, we generated a SMARCA4 knockdown model in HCC827 cells and exposed both parental and SMARCA4-silenced cells to the DPP4 inhibitor P32/98. DPP4 inhibition significantly suppressed cell proliferation in parental HCC827 cells and exerted comparable antiproliferative effects in SMARCA4-knockdown cells, with time-dependent augmentation of growth inhibition. Together with the results of colony-formation and migration/invasion assays, these findings indicate that, in the context of SMARCA4 deficiency, residual DPP4 activity remains functionally relevant for maintaining proliferative and invasive phenotypes, and pharmacologic blockade of DPP4 can exert direct tumor cell–intrinsic antiproliferative effects. In other words, even in a setting where DPP4 is globally downregulated, its residual expression may represent a context-dependent vulnerability that could be exploited therapeutically.

Although P32/98 is a well-established tool compound used to inhibit DPP4/DPP-IV in preclinical studies, conclusions based on a single small-molecule inhibitor should be interpreted with caution because off-target effects or cross-inhibition within the DPP4 enzyme family cannot be fully excluded. Notably, P32/98 (isoleucine-thiazolidide) has been used in the literature as a non-selective “DPP4 family” inhibitor in some experimental settings, highlighting the need to verify that the observed anti-proliferative and anti-migratory phenotypes are DPP4-dependent rather than driven by unintended targets.^32-34 To strengthen causal inference, future work should incorporate orthogonal validation strategies, such as genetic loss-of-function/rescue and direct target-engagement measurements (e.g., CETSA), ideally complemented by structurally distinct and better-characterized DPP4 inhibitors.^35,36

Our data both complement and extend the work of Almagthali et al. ³⁷ In that study, DPP4 inhibition in vivo enhanced the antitumor effects of ICIs by modulating CXCL10-mediated lymphocyte trafficking. The authors proposed that tumor-expressed DPP4 attenuates CXCL10 signaling and thereby impairs effector lymphocyte infiltration into the tumor bed, whereas DPP4 inhibition restores chemokine-driven T-cell recruitment and potentiates ICI responses. The primary mechanistic emphasis was thus on remodeling the TME. Given that lung cancer has long been conceptualized as an “inflammatory” malignancy, ³⁸ this immune-centric mechanism is highly plausible. In contrast, our experiments in SMARCA4-d–like NSCLC models, performed in the absence of immune cells or an intact microenvironment, demonstrate that DPP4 inhibition can directly suppress tumor cell proliferation. These two perspectives are likely complementary rather than mutually exclusive: in immunologically “hot” tumors or models, DPP4 inhibitors may primarily act by enhancing antitumor immunity, whereas in SMARCA4-dNSCLC—where ICI response rates are low and the TME is often immunologically “cold”—their direct tumor cell–intrinsic effects may be particularly relevant. We also observed a trend toward decreased DPP4 mRNA levels after inhibitor treatment, raising the possibility of feedback regulation at the transcriptional or post-transcriptional level; this will require further mechanistic investigation.

5. Limitations

This study has several limitations. First, the clinical cohort was small, and treatment regimens as well as response assessment criteria were heterogeneous across cases, which may have reduced statistical power and introduced bias into the observed associations. This constraint is partly attributable to the relative rarity of SMARCA4-dNSCLC, and our findings should therefore be interpreted as exploratory and hypothesis-generating. Accordingly, larger, multicenter, prospective studies will be required to validate and extend these results. Second, functional inferences based on P32/98 should be interpreted cautiously; potential off-target effects and the need for orthogonal validation are discussed in the Discussion. In addition, mechanistic and translational work using patient-derived organoids and appropriate in vivo models will be essential to more rigorously evaluate the therapeutic efficacy of DPP4 inhibition in SMARCA4-dNSCLC and to inform rational combination strategies.

6. Conclusions

In summary, we combined TCGA transcriptomic analyses with an exploratory multicenter IHC case series and in vitro assays to describe DPP4 patterns in SMARCA4-related NSCLC.Across datasets, DPP4 tended to be lower in SMARCA4-low tumors than in adjacent normal lung, and DPP4 staining in SMARCA4-deficient tumors was generally weaker than adjacent alveolar epithelium. Within tumors, higher residual DPP4 expression was associated with worse survival in TCGA; the small IHC series provided only descriptive, hypothesis-generating clinical trends. In vitro, P32/98 exposure—alone and combined with SMARCA4 knockdown—coincided with reduced growth and motility readouts, but target specificity and generalizability remain uncertain.

Taken together, the clinical and experimental observations should be interpreted as exploratory and do not establish causality or therapeutic utility. Further validation in larger cohorts and orthogonal specificity controls will be required to determine whether DPP4 has independent clinical relevance in SMARCA4-deficient NSCLC.

Supplemental Material

Supplemental Material - Prognostic and Therapeutic Significance of DPP4 in SMARCA4-Deficient Non–Small Cell Lung Cancer

Supplemental Material for Prognostic and Therapeutic Significance of DPP4 in SMARCA4-Deficient Non–Small Cell Lung Cancer by Chaopeng Chen, Zebin Zhong, Xiaoling Luo, Weiping Dai, Lujian Xu, Wenping Cai, Zhibin Xu, Guolian Liang and Lijuan Pang in Technology in Cancer Research & Treatment.