Clinical Implications of Serum Matrix Metalloproteinase-12 and Tumor Cadherin-13 for Molecular Subtyping and Personalized Therapy in Lung Cancer

Summary

This study evaluates serum matrix metalloproteinase-12 (MMP-12) and tumor cadherin-13 (CDH13) as complementary biomarkers in lung cancer. These markers correlate with molecular subtypes, treatment response, and survival, supporting their potential use in subtyping, treatment stratification, and prognostic assessment.

Abstract

Lung cancer remains a leading cause of cancer-related mortality worldwide, largely due to its biological heterogeneity, frequent late-stage diagnosis, and limitations in comprehensive molecular testing. Single biomarkers are often insufficient for guiding clinical decision-making, highlighting the need for complementary and clinically feasible indicators that integrate systemic and tumor-specific biology. In this single-center cohort study, 262 patients with lung cancer were enrolled. Serum matrix metalloproteinase-12 (MMP-12) levels were measured using enzyme-linked immunosorbent assay (ELISA), and tumor cadherin-13 (CDH13) expression was assessed by immunohistochemistry (IHC) using a validated immunoreactive score (IRS). Molecular subtyping was obtained from routine clinical testing, and associations with treatment response and survival outcomes were evaluated using multivariable logistic and Cox regression models. MMP-12 and CDH13 showed inverse distributions across molecular subtypes (P < 0.001), such that epidermal growth factor receptor (EGFR)/anaplastic lymphoma kinase (ALK)-altered tumors exhibited lower MMP-12 and higher CDH13, whereas Kirsten rat sarcoma viral oncogene homolog (KRAS)-mutant and wild-type tumors demonstrated the opposite pattern. Higher programmed death-ligand 1 (PD-L1) expression was consistently associated with increased MMP-12 and reduced CDH13 (all P < 0.05). Elevated MMP-12 was independently associated with reduced objective response rate (ORR) (adjusted odds ratio [OR] = 0.662, P < 0.001) and poorer survival (progression-free survival [PFS] hazard ratio [HR] = 1.382; overall survival [OS] HR = 1.421; both P < 0.001), while higher CDH13 predicted improved response (adjusted OR = 1.201, P < 0.001) and survival (PFS HR = 0.884; OS HR = 0.862; both P < 0.001). The combined high MMP-12/low CDH13 signature identified a subgroup with significantly worse outcomes (PFS HR = 2.281; OS HR = 2.506; P < 0.001). These findings suggest that serum MMP-12 and tumor CDH13 may serve as complementary biomarkers for molecular subtyping, therapeutic stratification, and prognostic assessment in lung cancer.

Introduction

Lung cancer remains a dominant global health burden, accounting for nearly 2.5 million new cases and about 1.8 million deaths worldwide in 2022¹. Despite substantial progress in screening, imaging, and systemic therapy, survival remains limited for many patients because lung cancer is biologically heterogeneous and often diagnosed at an advanced stage^2,3,4,5. Precision oncology has therefore become central to contemporary lung cancer care, particularly for non-small cell lung cancer (NSCLC), where actionable driver alterations define molecular subsets with distinct natural histories and treatment sensitivities⁶. Clinical practice guidelines recommend broad molecular profiling in advanced disease, including EGFR, ALK, KRAS, ROS1, BRAF, MET exon 14 skipping, RET, ERBB2 (HER2), and NTRK fusions, alongside assessment of immune biomarkers such as PD-L1^7,8,9,10. These developments have transformed outcomes for selected populations, yet they have also created new clinical bottlenecks. Tissue may be insufficient for comprehensive testing, turnaround time can delay therapeutic decisions, and access to high-quality next-generation sequencing varies across health systems^7,9. Even when molecular testing is available, biomarker-guided therapy remains imperfect because molecular status does not fully capture tumor microenvironmental states that influence treatment response and resistance. PD-L1 expression is widely used but has well-recognized limitations as a standalone predictor, and tumor mutational burden has substantial variability and uncertainty in its optimal use^6,9,10. These realities highlight a practical need for complementary biomarkers that are scalable, biologically informative, and capable of enriching molecular subtyping and treatment stratification when genomic information is incomplete, delayed, or ambiguous.

Proteolytic remodeling of the extracellular matrix and tumor microenvironment is one such biologically grounded dimension that is increasingly appreciated as both a hallmark of malignancy and a determinant of therapeutic response. Matrix metalloproteinases (MMPs) regulate invasion, angiogenesis, immune cell trafficking, and cytokine signaling, thereby linking tumor-intrinsic programs to host inflammatory and stromal contexts¹¹. Among these enzymes, matrix metalloproteinase-12 (MMP-12), also known as macrophage elastase, has attracted attention in lung cancer because it is expressed across a spectrum of human lung tumors and appears to be associated with aggressive phenotypes and worse outcomes¹². Beyond its tissue expression, MMP-12 is measurable in circulation, making it attractive as a minimally invasive biomarker with potential utility for serial monitoring¹³. Recent multi-cancer analyses have further supported the clinical relevance of MMP-12 in lung adenocarcinoma, suggesting predictive and prognostic roles that warrant deeper exploration in clinically actionable subgroups¹⁴. Importantly, circulating biomarkers are increasingly used to complement tissue-based testing in oncology, and there is growing interest in blood-based indicators that reflect not only tumor burden but also immune and stromal dynamics that shape sensitivity to systemic therapy¹⁵. In lung cancer, where immunotherapy and targeted therapy are frequently deployed, biomarkers that reflect the tumor microenvironment may be particularly valuable for understanding treatment heterogeneity, early resistance, and the need for combination approaches¹⁶. This biological and clinical context motivates a focused evaluation of serum MMP-12 beyond a diagnostic signal, specifically as a candidate marker linked to molecular subtypes and therapeutic trajectories.

In parallel, cadherin 13 (CDH13), also referred to as T cadherin or H cadherin, represents a mechanistically distinct axis relevant to tumor differentiation, adhesion signaling, and epithelial plasticity. CDH13 is an atypical cadherin anchored to the membrane via a glycosylphosphatidylinositol moiety and has been widely described as a tumor suppressor in multiple malignancies¹⁷. In NSCLC, epigenetic silencing of CDH13 through promoter hypermethylation has been repeatedly reported and has been associated with tumor progression and adverse clinicopathologic features^18,19. Meta-analytic and integrative evidence further supports that CDH13 hypermethylation is more frequent in NSCLC than in normal lung tissue and may have diagnostic relevance^20,21. Beyond its value as a marker of malignant transformation, CDH13 status may plausibly intersect with treatment sensitivity. Experimental work has linked CDH13 promoter methylation and altered expression to cisplatin resistance phenotypes in lung cancer models, raising the possibility that CDH13 reflects biologic programs related to therapy response and tumor adaptation under treatment pressure²². These data collectively position CDH13 as a tissue-based biomarker with potential implications for risk biology and therapeutic stratification. Unlike complex multi-omics assays, CDH13 can be evaluated by immunohistochemistry in routine pathology workflows, enabling practical translation if it provides additive information beyond existing clinical and molecular markers.

Taken together, MMP-12 and CDH13 represent complementary biological compartments and mechanisms. MMP-12 is a circulating readout that may capture immune and stromal activation and protease-driven remodeling, whereas CDH13 is a tissue-level marker that reflects tumor suppressive adhesion and epigenetic regulation. We hypothesize that integrating these axes can improve clinical inference in two key domains that define precision oncology in lung cancer. First, a combined serum and tissue biomarker framework may provide complementary information for molecular subtyping, either by associating with actionable driver alterations and immune biomarker states or by identifying biologically coherent phenotypes that track with specific molecular classes when genomic testing is limited^2,6,8,10. Second, the integrated signature may inform personalized therapy by improving risk stratification and by correlating with treatment selection and outcomes across targeted therapy and immunotherapy paradigms, where resistance is multifactorial and single markers are insufficient^9,10,23. In this context, our approach is intentionally pragmatic. Rather than replacing molecular diagnostics, it aims to complement them by providing scalable, clinic-compatible measurements that can be obtained rapidly and interpreted alongside genotype, PD-L1, and stage. This strategy may be particularly useful in real-world settings where tissues are scarce, repeated biopsies are impractical, or the clinical question is not simply whether a driver is present, but how to anticipate treatment sensitivity and progression risk in an individual patient.

Accordingly, in the present cohort study, we evaluate the clinical implications of serum MMP-12 and tumor CDH13 across lung cancer molecular subtypes and treatment pathways. Specifically, we examine whether these biomarkers, individually and in combination, are associated with major molecular classes and therapeutic groupings, and whether they add prognostic information relevant to personalized management. We also position this work within a broader ecosystem of alternative approaches, including circulating tumor DNA, methylation profiling, and multimodal modeling that integrates clinical, imaging, and molecular features^24,25. The novelty of our study lies not in proposing new individual biomarkers but in deliberately integrating a circulating protease signal with a tissue-based adhesion and epigenetic marker to address a clinically actionable question: how to complement molecular subtyping and therapy personalization using measurements feasible in routine practice. By capturing complementary dimensions of systemic tumor–host interactions and tumor-intrinsic biology, this integrative framework extends beyond prior studies that have evaluated MMP-12 or CDH13 in isolation. Such an approach may provide a practical bridge between diagnostic pathology and precision therapeutics, with potential applications in prioritizing molecular testing, contextualizing biomarker discordance, and refining individualized risk-based treatment planning.

Protocol

All procedures complied with institutional and national regulations and were approved by the Ethics Committee of Xingtai People’s Hospital (No. 2024-124). The requirement for written informed consent was waived because routinely collected clinical data and residual specimens were used. All data were anonymized prior to analysis.

Study design, setting, and overall schematic
This study was conducted as a single-center observational cohort designed to evaluate the clinical implications of serum MMP-12 and tumor CDH13, individually and in combination, for lung cancer molecular subtyping and personalized therapy. The study was performed at Xingtai People’s Hospital. Eligible patients managed between January 2021 and June 2022 were identified, and 262 consecutive patients who met the prespecified inclusion and exclusion criteria were enrolled.

The overall study workflow is illustrated in Figure 1. The schematic depicts the full process from patient identification to analysis, including patient screening and enrolment, baseline clinicopathological characterization and molecular profiling derived from routine clinical testing, collection and processing of pre-treatment serum, tumor tissue preparation for CDH13 immunohistochemistry, quantification of serum MMP-12, CDH13 immunoreactive scoring, classification of first-line therapy and subsequent treatment pathways, and longitudinal outcome ascertainment, including treatment response, progression-free survival, and overall survival. The schematic distinguishes diagnostic and molecular subtyping analyses from treatment and outcome analyses and indicates the time points at which biospecimens were collected, molecular results were generated, therapy was initiated, and follow-up assessments were performed.

Participant screening, eligibility criteria, and clinical data abstraction
Potentially eligible patients were identified from the hospital information system and pathology registry during the prespecified enrolment window (January 2021 to June 2022). Primary lung cancer was confirmed using histopathology, supplemented by multidisciplinary clinical review when required. Consecutive adult patients who satisfied all eligibility criteria were enrolled. Inclusion criteria included histologically confirmed primary lung cancer, availability of a pre-treatment serum sample obtained before initiation of any antitumor therapy (and before surgery in patients undergoing upfront resection), availability of tumor tissue sufficient for CDH13 immunohistochemistry, and availability of molecular subtype information generated as part of routine clinical care. Patients with a documented history of another malignant tumor prior to enrolment, patients with inadequate serum volume or compromised sample integrity, patients without evaluable tumor tissue for CDH13 assessment, and patients who received systemic anticancer therapy before baseline blood collection were excluded.

Clinicopathological and treatment data were abstracted from electronic medical records using a standardized case report form with predefined variable definitions and coding rules. Variables recorded included demographics (age and sex), smoking exposure, Eastern Cooperative Oncology Group (ECOG) performance status, comorbidities, baseline imaging findings, histologic subtype, tumor-node-metastasis (TNM) stage (according to the staging system used in routine clinical practice during the study period), baseline laboratory parameters, and detailed treatment information (including first-line regimen, start dates, dose modifications, and treatment transitions). Smoking exposure was quantified in pack-years as follows: Pack-years = (cigarettes per day ÷ 20) × years smoked. When smoking history is incomplete or inconsistent across records, the smoking variable was classified as unknown, and the source was documented as a discrepancy in the abstraction log; missingness was handled according to the prespecified statistical analysis plan.

Specimen collection, processing, and storage
Peripheral venous blood was collected at baseline before initiation of any antitumor therapy. Serum was separated by centrifugation at 1,500 × g for 10 min at room temperature. The serum supernatant was transferred into low-binding cryovials without disturbing the cellular layer. Serum was aliquoted into 200–500 µL portions to minimize freeze–thaw cycles, and the aliquots were stored at −80 °C until batch analysis. The interval from blood collection to freezing for each specimen was recorded, and samples with a collection-to-freezing time exceeding 2 h were excluded.

Prespecified pre-analytic quality control procedures were applied by documenting hemolysis and lipemia at processing. Visibly hemolyzed specimens were excluded from ELISA quantification. Serum aliquots were thawed on ice immediately prior to assay, mixed by gentle inversion, and centrifuged at 2,000 × g for 5 min to remove particulates when necessary. Vigorous agitation and vortexing were avoided to prevent bubble formation and optical artifacts during absorbance measurements.

Tumor tissue acquisition and preparation
Tumor tissue from routine diagnostic procedures (biopsy, bronchoscopy, core needle biopsy, or surgical resection) was obtained. Tissue was fixed in neutral buffered formalin according to standard pathology practice and embedded in paraffin. Sections (3–4 µm) were prepared on charged slides for immunohistochemistry. Unstained slides were stored in a dry environment at room temperature and stained within a prespecified interval to minimize antigen degradation. A chain-of-custody log was maintained linking specimen identifiers to de-identified study codes. The pathologist performing CDH13 scoring was blinded to serum MMP-12 values, molecular subtype, treatment, and outcomes.

Serum MMP-12 quantification by enzyme-linked immunosorbent assay (ELISA)
Peripheral venous blood was collected before any antitumor therapy, ideally within 7 days before biopsy or treatment initiation. Serum MMP-12 was measured using a commercial human MMP-12 ELISA kit. Each serum aliquot was thawed once on ice, mixed gently, and assayed in duplicate according to the kit instructions. A standard curve was generated on each plate using serial dilutions of the provided recombinant standard, and the absorbance was read at 450 nm with a reference correction using a microplate reader. Concentrations were calculated by fitting a four-parameter logistic regression standard curve. To ensure analytical robustness, internal quality-control (QC) sera were included on each plate and monitored for intra-assay and inter-assay precision (target coefficients of variation consistent with assay validation). If results fell below the lower limit of quantification, half of that limit was assigned for statistical analysis; if results exceeded the upper limit, the measurement was repeated after appropriate dilution. To minimize batch effects, samples were allocated randomly across plates, and an identical set of QC sera was included on each plate to track inter-plate variability. The laboratory personnel performing ELISA were blinded to clinical grouping and outcomes.

Tumor CDH13 immunohistochemistry and immunoreactive scoring
CDH13 immunohistochemistry was performed on formalin-fixed, paraffin-embedded tumor sections using a validated anti-CDH13 primary antibody. Antigen–antibody complexes were detected with a polymer-based detection system, and staining was visualized with 3,3′-diaminobenzidine (DAB). An identical staining workflow was applied to all study specimens; slides were processed in batches, and appropriate controls were included.

Paraffin blocks were sectioned at 3–4 µm, deparaffinized in xylene, and rehydrated through graded ethanol to distilled water. Heat-induced epitope retrieval was performed in citrate buffer (10 mM, pH 6.0) using a microwave oven at 95–100 °C for 15 min, followed by cooling to room temperature for 20 min. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide for 10 min at room temperature, and protein blocking was performed using normal serum blocking solution for 20 min at room temperature. Sections were incubated with the primary anti-CDH13 antibody overnight at 4 °C. After washing, the polymer-based secondary detection reagent was applied for 20–30 min at room temperature, followed by visualization with 3,3′-DAB chromogen for 3–5 min under microscopic monitoring. The sections were counterstained with hematoxylin, dehydrated through graded ethanol, cleared in xylene, and mounted with permanent mounting medium.

Internal quality controls were included in every staining run. A known CDH13-positive tissue section was processed in parallel as the positive control, and the primary antibody was omitted in the negative control. A run was accepted only when the positive control demonstrated the expected staining pattern with appropriate intensity, and the negative control showed no specific staining. Staining was repeated for runs with excessive background, failed control performance, or with a non-specific signal that compromises interpretation.

CDH13 expression was quantified using an immunoreactive score (IRS) derived from staining intensity and the proportion of positive tumor cells. Two trained pathologists scored all slides independently while blinded to serum biomarker measurements, molecular subtype, treatment, and outcomes. Discrepant scores were resolved by joint review at a multiheaded microscope, and the adjudicated IRS was recorded. Staining intensity was assigned as 0 (no staining), 1 (weak), 2 (moderate), or 3 (strong). The proportion of positive tumor cells was assigned as 0 (0%), 1 (1–10%), 2 (11–50%), 3 (51–80%), or 4 (81–100%). The IRS was calculated as follows: IRS = intensity score × proportion score. IRS values ranged from 0 to 12. Low CDH13 expression was defined as IRS ≤ 3, based on a commonly used threshold for immunoreactive scoring. IRS was retained as both a continuous variable and a dichotomized category for downstream analyses, including subgroup comparisons and combined biomarker modeling. Interobserver agreement was evaluated in a prespecified random sample comprising 30% of cases using a weighted kappa statistic, and the sampling procedure, agreement estimate, and adjudication decisions were documented in an audit trail.

Molecular subtyping and treatment classification
Molecular subtype information was obtained from routine clinical testing performed on diagnostic tumor tissue. Results were generated by clinically validated assays implemented in standard care, including targeted PCR-based testing, targeted next-generation sequencing panels, immunohistochemistry, and fluorescence in situ hybridization. For each patient, the assay type, specimen source, specimen adequacy statement, and report date were recorded. Patients were classified according to actionable driver alteration status, including EGFR activating mutations, ALK rearrangements, ROS1 rearrangements, KRAS mutations, BRAF mutations, MET exon 14 skipping alterations, RET rearrangements, ERBB2 alterations, and NTRK fusions. PD-L1 tumor proportion score was recorded where performed, and PD-L1 expression was categorized according to the strata used in routine practice during the study period. All molecular subtyping categories were prespecified before analysis. When more than one actionable alteration was reported, patients were assigned according to a predefined hierarchical classification rule prioritizing alterations with established genotype-directed therapeutic implications in the following order: EGFR mutations, ALK rearrangements, ROS1 rearrangements, BRAF mutations, MET exon 14 skipping alterations, RET rearrangements, ERBB2 alterations, NTRK fusions, and KRAS mutations.

Treatment information was extracted from medical records, including first-line regimen, start and stop dates, and subsequent lines of therapy. Initial management was classified into mutually exclusive treatment categories defined a priori: genotype-directed targeted therapy, immune checkpoint inhibitor-based therapy, chemotherapy-based therapy, and multimodality management incorporating surgery with perioperative systemic treatment as applicable. The index date was defined as the start date of first-line systemic therapy; for patients managed with upfront resection, the index date was the date of surgery, and subsequent systemic therapy was classified as adjuvant according to the clinical record. For analyses addressing personalized therapy, patients were stratified jointly by molecular subtype and treatment category, and serum MMP-12 concentrations and tumor CDH13 expression were compared between driver-positive and driver-negative disease and between targeted-therapy and immune checkpoint inhibitor-based treatment groups.

Outcome definitions and follow-up
The index date was defined as the date of initiation of first-line antitumor therapy. For patients managed with upfront surgical resection, the index date was defined as the date of surgery; subsequent systemic therapy was classified as adjuvant or palliative according to the medical record. Each patient was followed from the index date until death, the date of last confirmed contact, or the administrative censoring date, whichever occurs first. Follow-up duration was recorded in months, and for each endpoint, the event indicator and the corresponding event or censoring date were retained. Outcomes were ascertained through structured abstraction of the electronic medical record, including radiology reports, oncology clinic documentation, inpatient records, and telephone follow-up when required. A prespecified case report form was used, and uniform source hierarchy rules were applied. When dates or classifications differed across records, discrepancies were resolved by prioritizing contemporaneous radiology reports and treatment decision notes; the final adjudicated value and its supporting sources were documented. A follow-up log capturing imaging dates, treatment transitions, hospitalizations, and contact attempts was maintained to minimize loss to follow-up and to ensure traceability.

Tumor response was assessed in patients with measurable disease according to RECIST version 1.1. Best overall response was classified as complete response, partial response, stable disease, or progressive disease. Objective response rate was defined as complete or partial response, and disease control rate as complete response, partial response, or stable disease. Response and progression dates were assigned based on the earliest documented evidence.

Progression-free survival (PFS) was defined as the interval from the index date to the first occurrence of disease progression or death from any cause, whichever occurs first. Progression was determined primarily from radiology reports that document progression according to RECIST 1.1. When radiology is not available at the time of clinical deterioration, clinician-documented progression was accepted if it is explicitly linked to worsening tumor burden and prompts a change in anticancer management. Patients without an event on the date of the last objective assessment were censored, confirming the absence of progression. If a new line of systemic therapy was initiated without explicit documentation of progression, the imaging was reviewed, and contemporaneous clinical documentation was provided. If progression could not be confirmed, patients were censored at the last radiographic assessment prior to a treatment change, and the decision rule applied was recorded.

Overall survival (OS) was defined as the interval from the index date to death from any cause. Vital status was ascertained using hospital records and outpatient follow-up notes, supplemented by structured telephone confirmation when needed. The date of death was recorded as documented in the medical record or, when unavailable, as the date confirmed by family during follow-up contact, with supporting documentation recorded in the follow-up log. Patients who were alive at the date of last confirmed contact were censored.

A standardized follow-up workflow aligned with routine care pathways was implemented. During active systemic therapy, clinical assessments at each treatment cycle were documented, and radiographic reassessments at regular intervals determined by clinical protocol were recorded, with imaging modality and date recorded for each assessment. After completion of first-line therapy or during postoperative surveillance, follow-up visits and imaging evaluations were recorded according to the institutional schedule. When in-person follow-up could not be completed, a structured telephone follow-up was conducted to confirm continuation of treatment, document interval hospitalizations, confirm disease status as documented by external imaging, where applicable, and confirm survival status. All missed visits, unreachable attempts, and reasons for discontinuation of follow-up were documented. Before classifying a patient as lost to follow-up, contact was attempted at least 2 additional times on different days, and each attempt was recorded.

Procedural separation between biomarker assessment and outcome ascertainment was maintained. Personnel performing serum MMP-12 measurement and CDH13 scoring were blinded to clinical outcomes. Outcome abstraction was performed without access to biomarker results whenever feasible. A two-stage adjudication process was applied for progression and response endpoints. A trained abstractor assigned event classification and dates using the prespecified rules, after which a senior clinician independently reviewed the assignments. Discrepancies were resolved through a joint review of the original radiology report, oncology notes, and treatment timeline, and an audit trail of all adjudication decisions was maintained.

Consistency checks to identify implausible sequences, including events dated before the index date, treatment transitions without corresponding assessments, or death dates without supporting documentation, were performed. Flagged cases were reconciled by returning to source documents and updating the dataset with a dated correction note that specifies the reason for the change and the supporting evidence.

Statistical analysis
Statistical analyses were performed using SPSS (version 26.0) and R (version 4.3.0). Continuous variables were presented as mean ± standard deviation or median (interquartile range), as appropriate, and categorical variables were presented as counts and percentages. Group comparisons used parametric or nonparametric tests as appropriate, with chi-square or Fisher’s exact tests for categorical data. All tests were two-sided, with P < 0.05 considered statistically significant.

Covariates were selected a priori based on clinical relevance and prior evidence, and the same covariate set was applied across related models to support comparability. Core adjustment variables included age, sex, smoking exposure, TNM stage, ECOG performance status, and histologic subtype, with additional adjustment for treatment group when the analysis target is molecular subtype within treated populations. Multicollinearity was assessed using variance inflation factors, and problematic collinearity was addressed using prespecified variable reduction rules. Adjusted odds ratios with 95% confidence intervals and two-sided P values were reported. The number of covariates included in the multivariable logistic and Cox regression models was prespecified and restricted relative to the number of outcome events to minimize the risk of overfitting. The events-per-variable ratio remained acceptable (≥10) for all models. For multi-category molecular subtype outcomes, either multinomial logistic regression models were fitted, or a set of binary logistic comparisons using a prespecified reference category was performed, and the modeling strategy was clearly stated.

Progression-free survival and overall survival were analyzed using Kaplan–Meier estimates, and survival curves were compared using the log-rank test. Associations between biomarkers and outcomes were quantified using Cox proportional hazards regression, reporting hazard ratios with 95% confidence intervals. Multivariable Cox models were constructed using prespecified clinicopathological covariates, including molecular subtype and treatment category, to assess the implications for personalized therapy. The proportional hazards assumption was verified using Schoenfeld residual diagnostics and log-minus-log survival plots. If the proportional hazards assumption was violated for a covariate, the issue was addressed using a prespecified approach such as stratification on that covariate or inclusion of a time-dependent interaction term, and the final model specification was reported.

Missing data were handled using a prespecified strategy. When missingness was minimal, complete-case analyses were performed; otherwise, multiple imputation with chained equations was used, including all variables in the analytical models. Ten imputed datasets were generated, and pooled estimates were calculated using Rubin’s rules. For analyses involving multiple subtype comparisons or multiple endpoints, the false discovery rate was controlled using a prespecified procedure, or analyses were clearly designated as exploratory to ensure transparency regarding multiplicity.

The analysis dataset was locked prior to model fitting, and a comprehensive data dictionary was maintained that defined each variable, unit, coding scheme, and permissible range. Duplicate quality control samples across assay plates were included to quantify inter-plate variability and monitor drift. For immunohistochemistry scoring, the scoring rubric, blinding procedures, and interobserver agreement assessment were documented, and adjudication logs were retained for discrepant cases. Programmed range checks and logical consistency checks across key variables, including index date, sample collection date, treatment start date, and outcome dates, were implemented. All statistical scripts and version-controlled output logs were preserved to enable complete replication of the analytical workflow.

Results

Cohort assembly and molecular subtype distribution
The overall study workflow is summarized in Figure 1. In total, 262 consecutive patients with lung cancer were included and stratified into four clinically relevant molecular subtypes: EGFR (n = 92), ALK (n = 48), KRAS (n = 54), and wild-type (n = 68). Baseline clinicopathologic features differed across subtypes (Table 1). Patients with EGFR and ALK alterations were younger on average than those in the K...

Discussion

This study evaluates the complementary roles of serum MMP-12 and tumor CDH13 as an integrated biomarker approach to support the clinical interpretation of molecular subtypes and therapy stratification in lung cancer. The core hypothesis stems from the complementary biological roles of these two biomarkers: MMP-12, a circulating protease reflecting immune-stromal remodeling and tumor aggressiveness, and CDH13, a tissue-based tumor suppressor linked to epigenetic regulation and epithelial plasticity²⁶

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Antibody, anti-CDH13 (T-cadherin)	Abcam, Cambridge, UK	ab15126	Primary antibody for immunohistochemistry; validated for FFPE tissue
Centrifuge (high-speed refrigerated)	Eppendorf, Hamburg, Germany	5424 R	Used for serum separation and sample clarification
DAB Substrate Kit	ZSGB-Bio, Beijing, China	ZLI-9017	For chromogenic detection in immunohistochemistry
ELISA Microplate Reader	BioTek Instruments, Winooski, USA	ELx800	Used for absorbance reading at 450 nm with reference correction
Formalin (10% neutral buffered)	Sinopharm Chemical Reagent Co., Shanghai, China	10009218	For fixation of biopsy and surgical tissue specimens
Hematoxylin and Eosin Staining Kit	Solarbio, Beijing, China	G1120	For counterstaining and quality control of IHC slides
Human MMP-12 ELISA Kit	Abcam, Cambridge, UK	ab246533	Quantitative measurement of serum matrix metalloproteinase-12
Microscope (Bright-field)	Nikon, Tokyo, Japan	Eclipse E200	Used for evaluation of CDH13 immunostaining and scoring
Microtome	Leica Biosystems, Wetzlar, Germany	RM2235	For sectioning paraffin-embedded tissue at 3–4 µm thickness
Mounting Medium	Solarbio, Beijing, China	S2100	For permanent slide mounting after IHC staining
Paraffin Embedding Station	Leica Biosystems, Wetzlar, Germany	EG1150	For FFPE tissue block preparation
Phosphate Buffered Saline (PBS)	Gibco, Thermo Fisher Scientific, USA	10010023	Used for washing during IHC staining
Polymer Detection System	MXB Biotechnologies, Fuzhou, China	DAB-0031	Secondary detection kit for IHC
R Statistical Software	R Foundation for Statistical Computing, Vienna, Austria	Version 4.3.0	Used for statistical analysis and modeling
Serum Separator Tubes	BD Vacutainer, Franklin Lakes, USA	367988	For collection of peripheral venous blood
SPSS Statistics Software	IBM Corp., Armonk, USA	Version 26.0	Used for statistical analysis of continuous and categorical data
Storage Cryovials	Corning Inc., Corning, USA	430488	Low-binding cryovials for aliquoting serum samples
Water Bath (for antigen retrieval)	Grant Instruments, Cambridge, UK	JBN26	Maintains constant temperature during antigen retrieval