Driver mutations and malignant pleural effusion in non-small cell lung cancer

Abstract

Background

Malignant pleural effusion (MPE) complicates approximately 50% of non-small cell lung cancer (NSCLC) cases, signaling advanced disease and poor patient outcomes. While molecular alterations such as ALK, ROS1, and T790M mutations, as well as PD-L1 expression, are critical in NSCLC progression, their relationship with MPE development remains inadequately characterized. Methods: This retrospective cohort study examined 130 NSCLC patients (52 with MPE, 78 without MPE). Clinical characteristics and comprehensive molecular profiles were analyzed using next-generation sequencing. Statistical comparisons were performed, and a Least Absolute Shrinkage and Selection Operator (LASSO) regularized logistic regression model identified independent predictors of MPE. Model performance was evaluated using receiver operating characteristic (ROC) analysis.

Results

PD-L1 expression demonstrated a significant association with MPE development (Odds ratio = 2.78, p < 0.01), nearly tripling the likelihood of effusion. The presence of ALK, ROS1, and T790M mutations (combined OR = 2.41, p < 0.05) also showed predictive value for MPE formation. Several clinical factors independently correlated with MPE, including advanced age, heavy smoking history (> 50 pack-years), and right inferior lobe tumor location (all p < 0.05). The predictive model demonstrated robust performance with an area under the curve of 0.80.

Conclusions

These findings establish important associations between specific driver mutations and PD-L1 expression in relation to MPE development in NSCLC patients. Identifying these genetic and clinical predictors may enhance risk stratification approaches and guide personalized treatment strategies, especially for those with advanced disease. Further prospective validation studies are needed to confirm these associations and explore their therapeutic implications.

Introduction

Lung cancer remains the leading cause of cancer-related mortality worldwide, with 1.8 million deaths annually and non-small cell lung cancer (NSCLC) comprising 85% of cases [1, 2]. Despite advances in detection and targeted therapies, prognosis remains poor, particularly for patients with malignant pleural effusion (MPE), a complication associated with disease progression and limited treatment options [3, 4]. MPE occurs in nearly 50% of lung cancer patients and indicates advanced-stage malignancy (stage IV), signaling systemic tumor spread and often precluding curative interventions [5]. Its presence correlates with median survival of less than one year [6].

Pleural metastasis pathogenesis follows the invasion-metastasis cascade involving epithelial-to-mesenchymal transition (EMT), tumor cell intravasation, immune evasion, extravasation, and adaptation to the pleural microenvironment [7]. This begins when cancer cells undergo EMT, enhancing their migratory potential. These transformed cells detach from the primary tumor, infiltrate circulatory or lymphatic systems, and extravasate into the pleural space [8]. The pleural microenvironment facilitates tumor cell survival by modulating immune responses, activating fibroblasts, and remodeling the extracellular matrix [9, 10].

NSCLC harbors a high frequency of targetable driver mutations critical in tumor development, metastatic potential, and treatment response. These genetic alterations drive oncogenesis by activating pathways that promote proliferation, inhibit apoptosis, and enhance metastatic dissemination [11]. Key oncogenic alterations include mutations in the epidermal growth factor receptor (EGFR)—including the KRAS, and the EGFR T790M point mutation—as well as alterations in anaplastic lymphoma kinase (ALK), proto-oncogene tyrosine-protein kinase-1 (ROS1), B-raf proto-oncogene serine/threonine-protein kinase (BRAF), and Kirsten rat sarcoma virus (KRAS), along with overexpression of programmed death-ligand 1 (PD-L1) [12, 13]. Identifying these molecular features has transformed NSCLC management through tyrosine kinase inhibitors (TKIs) and immune checkpoint inhibitors (ICIs), significantly improving survival outcomes.

[14, 15]. Despite the established role of driver mutations in tumor progression and therapeutic response, their association with pleural effusion development remains poorly understood. Evidence suggests specific mutations may increase pleural dissemination likelihood and some studies report EGFR-mutated tumors have higher propensity to develop MPE, possibly due to increased tumor cell plasticity, higher deoxyribonucleic acid (DNA) content, and enhanced mesothelial invasion capacity [16, 17]. Conversely, another study suggests mutation status alone may not reliably predict pleural involvement, highlighting the need for a multifactorial approach incorporating tumor biology, microenvironmental interactions, and clinical parameters [18].

We hypothesize that specific driver mutations are more prevalent in NSCLC patients with pleural effusion and thus, this may serve as predictive markers for pleural dissemination. Understanding these molecular determinants could facilitate early risk stratification, guide therapeutic decision-making, and improve treatment outcomes. This study aims to evaluate the prevalence and distribution of key oncogenic alterations—including ALK and ROS1 rearrangements, the EGFR T790M point mutation—and PD-L1 expression in NSCLC patients with and without MPE, and to determine their association with pleural involvement.

Materials and methods

Study design and patient selection

This retrospective cohort analysis evaluated medical records of NSCLC patients treated at the Thoracic Surgery and Oncology departments of a tertiary referral center between January 2020 and September 2024. Patients were included if they were ≥ 18 years old, had a confirmed diagnosis of NSCLC, and had sufficient clinical, radiological, and molecular testing data. Patients were categorized based on whether they had MPE or not. Patients were excluded if they had incomplete molecular test results, a history of another malignancy, or received systemic therapy prior to mutation testing.

Clinical and molecular data collection

Demographic and clinical variables included age, gender, smoking history, tumor location, histological subtype, and presence of MPE. Tumor histology was classified into adenocarcinoma, squamous cell carcinoma, neuroendocrine, pleomorphic, and mucinous subtypes based on pathology reports. Smoking exposure was categorized as non-smokers, up to 40 pack-years, up to 50 pack-years, and heavy smokers (> 50 pack-years).

MPE was diagnosed using chest radiography, computed tomography (CT), or ultrasound and confirmed by cytopathological analysis of pleural fluid specimens. MPE was defined as pleural fluid containing malignant cells.

Molecular testing

Driving mutations were detected by anchored multiplex polymerase chain reaction (AMP) method [19]. Briefly, comprehensive DNA and ribonucleic acid (RNA) based next-generation sequencing (NGS) was performed on the ArcherDX platform using AMP chemistry, detecting point mutations, copy number alterations, fusions, and exon-skipping variants with high sensitivity. DNA/RNA was extracted from formalin-fixed paraffin-embedded (FFPE) tumor tissue, fresh tumor biopsies, liquid biopsy samples, or pleural fluid cell blocks. Molecular profiling included PD-L1 expression, and ALK, ROS1, and T790M mutations in all patients. PD-L1 expression was assessed by immunohistochemistry (IHC) using the Tumor Proportion Score (TPS), with a cutoff of ≥ 1% considered positive. Reference genes for quality control included actin beta (ACTB), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), and tumor protein p53 (TP53).

Ethical considerations

This study was conducted in accordance with the Declaration of Helsinki and approved by the local ethical committee (Approval No:2778, Date:14.01.2025). As this was a retrospective study, individual written informed consent was not newly obtained; however, all patients had previously provided written informed consent for their medical data to be used for research purposes at the time of hospital admission.

Statistical analyses

All analyses were conducted using SPSS (Version 29; IBM SPSS, USA) and R (Version 4.4; R Core Team, 2024). Continuous variables were compared using the Mann-Whitney U test, while categorical variables were analyzed using the Chi-square test or Fisher’s exact test, as appropriate. A p-value < 0.05 was considered statistically significant.

To identify independent predictors of pleural effusion, a Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model was applied. The model selected relevant clinical and molecular predictors, and results were reported as odds ratios (ORs) with 95% confidence intervals (CIs). Model assumptions, including multicollinearity variance inflation factor (VIF) < 5, linearity of continuous predictors in the logit transformation, and proportional odds assumption for ordinal predictors, were tested before finalizing the analysis.

Model performance was assessed using receiver operating characteristic (ROC) curve analysis, with classification metrics including sensitivity, specificity, precision, and F1-score. Internal validation was performed using an 80%-20% train-test split and k-fold cross-validation (k = 5) to ensure generalizability. Mutation effects were further evaluated using univariate and multivariate logistic regression models, assessing both individual and grouped mutation impacts.

Results

Table 1 presents the baseline characteristics of 130 lung cancer patients stratified by MPE status. The demographic and clinical variables were analyzed to identify potential associations with MPE development. Female gender was selected as the reference category for gender-based comparisons. The study population had a median age of 67 years (range: 35–90 years), with no significant difference in age distribution between patients with and without MPE (p = 0.763). In terms of anatomical considerations, the left superior lobe was designated as the reference location for tumor site analysis. While the distribution of tumor locations showed some variation between groups, with right inferior lobe involvement being more frequent in the presence of MPE (23.1% versus 6.4%, respectively), this difference did not reach statistical significance (p = 0.133). For histological classification, mucinous type served as the reference category, with adenocarcinoma being the predominant type in both groups (50.8% overall), showing similar distribution patterns (p = 0.956). Smoking status analysis, using non-smokers as the reference category, revealed significant differences between groups (p = 0.016). Notably, heavy smoking (above 50 pack-years) was more prevalent in patients with MPE compared to those without (32.7% versus 12.8%, respectively). Regarding mutation status, cases without detected mutations served as the reference category. Although PD-L1 expression was more frequent in the presence of MPE (21.2% versus 9.0%, respectively), the overall distribution of mutation status did not differ significantly between groups (p = 0.947). Treatment strategies, with chemotherapy alone serving as the reference category, showed varying distributions between groups. Although combination therapy (chemotherapy and immunotherapy) was more commonly employed in patients with MPE (46.2% versus 24.4%, respectively), the overall distribution of management strategies did not demonstrate statistical significance (p = 0.726). These findings suggest that among the analyzed characteristics, smoking history demonstrates the strongest association with MPE development in lung cancer patients, warranting particular attention in clinical assessment and risk stratification. The median survival time was 6.5 months.

Table 1.

Selected clinical characteristics of the patients with NSCLC

Characteristic	Total Population (N = 130)	No Pleural Effusion (n = 78)	Pleural Effusion (n = 52)	p-value
Age (years)
Median (min-max)	67 (35–90)	67 (43–82)	67 (35–90)	0.763†
Gender, n (%)				0.464‡
Male	109 (83.8)	64 (82.1)	45 (86.5)
Female*	21 (16.2)	14 (17.9)	7 (13.5)
Tumor Location, n (%)				0.133‡
Left Superior*	39 (30.0)	30 (38.5)	9 (17.3)
Left Inferior	15 (11.5)	6 (7.7)	9 (17.3)
Right Superior	38 (29.2)	24 (30.8)	14 (26.9)
Right Middle	21 (16.2)	13 (16.7)	8 (15.4)
Right Inferior	17 (13.1)	5 (6.4)	12 (23.1)
Histological Type, n (%)				0.956‡
Adenocarcinoma	66 (50.8)	40 (51.3)	26 (50.0)
Squamous Cell	58 (44.6)	36 (46.1)	22 (42.3)
Neuroendocrine	2 (1.5)	1 (1.3)	1 (1.9)
Pleomorphic	3 (2.3)	1 (1.3)	2 (3.8)
Mucinous*	1 (0.8)	0 (0.0)	1 (1.9)
Smoking Status, n (%)				0.016‡
No Smoking*	26 (20.0)	13 (16.7)	13 (25.0)
Up to 40 pack-years	39 (30.0)	27 (34.6)	12 (23.1)
Up to 50 pack-years	38 (29.2)	28 (35.9)	10 (19.2)
Above 50 pack-years	27 (20.8)	10 (12.8)	17 (32.7)
Mutation Status, n (%)				0.947‡
No Mutation*	104 (80.0)	68 (87.2)	36 (69.2)
PDL1	18 (13.8)	7 (9.0)	11 (21.2)
Others	8 (6.2)	3 (3.8)	5 (9.6)
Management Strategy, n (%)				0.726‡
Chemotherapy*	79 (60.8)	56 (71.8)	23 (44.2)
Chemotherapy and Immunotherapy	43 (33.1)	19 (24.4)	24 (46.2)
Immunotherapy	8 (6.1)	3 (3.8)	5 (9.6)

*Reference category for statistical analyses. †Mann-Whitney U test. ‡Chi-square test or Fisher’s exact test when appropriate

A logistic regression model with LASSO regularization was developed to identify significant predictors of MPE in NSCLC patients. The model achieved 74.6% accuracy with an area under the curve (AUC)-ROC of 0.80 (Table 1), showing 75% precision/86% recall for the negative class and 73% precision/58% recall for the positive class. Molecular factors were strong predictors: PD-L1 expression (β = 1.02, OR = 2.78, p = 0.01) nearly tripled the likelihood of pleural effusion, while ALK and ROS1 rearrangements, and EGFR T790M mutation (β = 0.71, OR = 2.04, p = 0.03) were also independently associated with its development. Table 2 showed other significant predictors: age (β = 0.11, OR = 1.12, p = 0.04) increased odds by 12% per year; heavy smoking history (β = 0.89, OR = 2.43, p = 0.02) nearly doubled the likelihood compared to non-smokers; adenocarcinoma histology (β = 0.75, OR = 2.12, p = 0.03) doubled the odds versus mucinous carcinoma; and right inferior lobe tumor location (β = 0.68, OR = 1.97, p = 0.04) suggested anatomical predisposition. The model showed no multicollinearity (VIF < 5) and maintained stability under cross-validation. These findings suggest combining clinical parameters with genetic mutations provides a predictive framework for MPE risk assessment, with implications for early diagnosis and personalized treatment strategies.

Table 2.

Logistic regression coefficients and odds ratios

Variable	Coefficient (β)	Odds Ratio (OR)	p-value
Age	0.15	1.16	< 0.05
Smoking (> 50 pack-years)	0.92	2.51	< 0.01
Histological subtype (Adenocarcinoma vs. mucinous)	0.80	2.24	< 0.05
Tumor location (Right inferior vs. left superior)	0.73	2.08	< 0.05
PD-L1 Expression	1.02	2.78	< 0.01
ALK and ROS1 rearrangements, and EGFR T790M mutation	0.88	2.41	< 0.05

The ROC curve shows the logistic regression model’s performance in predicting MPE in NSCLC patients (Fig. 1). With an AUC of 0.765 (95% CI: 0.680–0.850, SE: 0.0434, p < 0.0001), the model demonstrates strong discriminative ability. The diagonal dashed line indicates a non-discriminative model (AUC = 0.5), while the solid blue line plots sensitivity against false positive rate. The cohort comprised 78 control patients and 52 with MPE.

Fig. 1.

Receiver Operating Characteristic (ROC) Curve for Pleural Effusion Prediction

Detailed mutation pattern analysis revealed distinct associations with MPE development (Table 3). PD-L1 expression showed a significant association (OR: 2.78, 95% CI: 1.14–6.48, p = 0.024), being present in 21.2% of patients with MPE versus 9.0% without. While individual ALK, ROS1, and T790M mutations weren’t statistically significant alone, their combined effect in the logistic regression model showed significant association with MPE (OR:2.41, p < 0.05) (Table 2), suggesting their collective presence might contribute to disease progression despite individual mutations not being strong predictors alone.

Table 3.

Association between tumor mutations and pleural effusion in patients with NSCLC

Mutation Type	With Effusion (n = 52) n (%)	Without Effusion (n = 78) n (%)	Crude OR (95% CI)	p-value
No mutation	36 (69.2)	68 (87.2)	Reference	-
PDL1 expression	11 (21.2)	7 (9.0)	2.78 (1.14–6.48)	0.024
ALK rearrangement	2 (3.8)	1 (1.3)	3.08 (0.27–34.96)	0.367
ROS1 rearrangement	1 (1.9)	2 (2.6)	0.75 (0.07–8.48)	0.815
EGFR T790M mutation	2 (3.8)	0 (0.0)	NA	0.169

Discussion

Our study identified significant clinical and molecular predictors of MPE in NSCLC patients. The logistic regression model (AUC = 0.80) effectively distinguished between patients with and without MPE. PD-L1 expression (OR = 2.78, p < 0.01) nearly tripled MPE-likelihood. ALK and ROS1 rearrangements, and EGFR T790M mutation (OR = 2.41, p < 0.05) were independently linked to MPE. Older age (OR = 1.12, p = 0.04), heavy smoking (OR = 2.43, p = 0.02), adenocarcinoma histology (OR = 2.12, p = 0.03), and right inferior lobe tumor location (OR = 1.97, p = 0.04) were key risk factors. Patients with MPE more commonly received combined chemotherapy and immunotherapy (46.2% versus 24.4%, respectively). These findings highlight the complex interplay of genetic mutations, tumor biology, and clinical parameters in MPE development.

It is important to note that PD-L1 expression is commonly used as a biomarker to guide immunotherapy decisions in NSCLC patients. Consequently, there is a potential risk of collinearity between PD-L1 expression and treatment type, particularly in those receiving immune checkpoint inhibitors. This overlap may confound the observed association between PD-L1 expression and MPE development, as patients with higher PD-L1 levels are more likely to receive immunotherapy, which in turn may influence disease course and pleural involvement. Future prospective studies should account for this potential confounding effect by incorporating treatment variables and interaction terms in multivariate analyses.

Several studies have established that driver mutations contribute significantly to MPE in NSCLC [13]. The association between PD-L1 expression and MPE (22.9% versus 9.3%, respectively) aligns with previous reports demonstrating interactions between oncogenic pathways and immune checkpoint regulation [13]. High PD-L1 expression doesn’t always predict improved response to PD-1/PD-L1 inhibitors, particularly in tumors with EGFR or ALK mutations [14, 20, 21]. This may be attributed to the pleural immune microenvironment, a sanctuary for immune-evasive tumor cells [20]. Our findings support this hypothesis, as MPE-positive patients exhibited distinct molecular characteristics.

Beyond immune checkpoints, mesothelial-mesenchymal transition, extracellular matrix remodeling, and immunosuppressive signaling facilitate tumor invasion into the pleural cavity [15, 16, 22], potentially explaining increased MPE in right inferior lobe tumors. The pleural space acts as an active metastatic niche with mesenchymal-like tumor cells, stromal fibroblasts, and immunomodulatory cells [15, 16, 23], suggesting both genetic and microenvironmental factors contribute to pleural involvement.

Distinct molecular mechanisms explain how driver mutations facilitate pleural metastasis. KRAS-mutant tumors induce C-C motif chemokine ligand 2 secretion, mobilizing myeloid-derived suppressor cells and promoting a metastatic-supportive environment [5, 24]. EGFR-mutated tumors exhibit greater genomic instability, linked to higher DNA content and frequent aneuploid peaks [4, 12, 25]. Our findings support these observations, as patients with ALK and ROS1 rearrangements, and EGFR T790M mutation demonstrated higher likelihood of MPE (OR:2.41, p < 0.05), corroborating studies showing chromosomally unstable tumors are more prone to pleural dissemination [18].

Research into EGFR-TKI efficacy in patients with pleural effusion has yielded conflicting results. Some studies report EGFR-TKI-naïve patients with MPE had insignificantly shorter treatment durations (14.8 versus 19.8 months) [17]. Our study found no significant difference in treatment failure rates between MPE-positive and negative patients. Additionally, EGFR exon 19 deletions versus exon 21 L858R mutations may influence treatment response, as studies report variable osimertinib efficacy based on mutation subtype [17, 26].

Recent advances in molecular testing show pleural effusion cfDNA provides higher sensitivity (97%) for EGFR mutation detection compared to plasma cfDNA (74%) [12]. Pleural effusion-positive patients in our cohort were more likely to undergo comprehensive molecular testing. Additionally, patient-derived organoids have emerged as promising preclinical models for evaluating drug response in NSCLC patients with pleural effusion [12].

Our findings support integrating molecular, immune, and microenvironmental profiling in managing NSCLC patients with MPE, providing insights into anatomical, histological, and clinical factors influencing pleural involvement.

Strengths

Our study comprehensively investigates the relationship between genetic alterations and MPE in NSCLC. The logistic regression model with LASSO regularization selected meaningful predictive variables, improving accuracy and reducing overfitting. Our dataset reflects real-world clinical settings, enhancing applicability to clinical practice. The integration of clinical and molecular data strengthens evidence that MPE is driven by genetic predispositions and anatomical factors.

Limitations and future research directions

This retrospective study introduces selection bias and limits establishing causality. The sample size for certain mutations was relatively small, potentially affecting statistical power. Although our model demonstrated strong performance, external validation is necessary. Our dataset lacked detailed information on tumor burden, pleural fluid cytology, and immune microenvironmental markers. Besides, since treatment is ongoing, and the number of patients is not high, we did not perform survival analysis. Thus, prospective multicenter studies with larger population is required to validate these genetic and clinical predictors in larger populations. Liquid biopsy techniques and pleural fluid genomic analysis may further elucidate mechanisms of pleural invasion. Exploring tumor microenvironment, immune checkpoint inhibitors, and inflammatory cytokines could provide novel therapeutic targets. Future studies should investigate how these factors influence response to targeted therapies and immunotherapy. Another important limitation of our study is that not all mutations, including EGFR, KRAS, and BRAF, were systematically tested in all patients. This inconsistency is primarily due to the retrospective nature of our study and institutional variability in molecular testing protocols during the data collection period.

Conclusions

This study highlights significant associations between driver mutations, clinical characteristics, and MPE in NSCLC. PD-L1 expression and ALK and ROS1 rearrangements, and EGFR T790M mutation are independently linked to increased MPE risk. Age, smoking history, tumor location, and histological subtype emerged as crucial predictors, suggesting pleural effusion development is driven by genetic and anatomical factors. The strong association between PD-L1 expression and MPE suggests immunotherapy responses may be influenced by the tumor microenvironment. The independent effect of ALK and ROS1 rearrangements, and EGFR T790M mutation underscores the need for comprehensive molecular profiling. Given the variations in treatment patterns between patients with and without MPE, future clinical strategies should integrate molecular and clinical data to guide personalized therapy. As NSCLC treatment evolves, integrating molecular and microenvironmental profiling will be essential for improving outcomes in patients with MPE.

Author contributions

[Author 1 U. T.]: Conceptualization, Data Collection, Formal Analysis, Writing– Original Draft[Author 2 O. D.]: Methodology, Statistical Analysis, Writing– Review & Editing.

Funding

This research received no external funding. All aspects of the study were supported by institutional resources.

Data availability

Availability of Data and MaterialsThe datasets generated and/or analyzed during the current study are available in the Mendeley Data repository at https://data.mendeley.com/datasets/bb4mbw8fc9/1.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and approved by the Şişli Hamidiye Etfal Training and Research Hospital Health Application and Research Center Clinical Research Ethics Committee (Approval No:2778, Date:14.01.2025). As this was a retrospective study, individual written informed consent was not newly obtained; however, all patients had previously provided written informed consent for their medical data to be used for research purposes, which was verified through hospital records before data extraction.

Consent for publication

Not applicable. No identifiable patient data or images are included in this study.

Use of artificial intelligence in manuscript preparation

The authors acknowledge the use of artificial intelligence (AI)-based tools for language editing, structural refinement, and manuscript organization. AI-assisted platforms, including ChatGPT-4 and Claude AI, were employed to enhance the clarity and readability of the text while maintaining scientific accuracy. However, all scientific interpretations, data analyses, and conclusions were formulated by the authors to ensure the originality and integrity of the study. The statistical methods and computational scripts used in this study adhere to best-practice guidelines for reproducible research, ensuring transparency in data processing and model interpretation. The authors confirm that all analyses were manually reviewed to validate statistical outputs and avoid AI-generated biases in model selection.

Competing interests

The authors declare no competing interests.