Malignant pleural effusion (MPE) is a frequent complication of advanced NSCLC and often provides the first or only accessible specimen for diagnosis. Although demonstration of malignant cells remains essential, the advent of targeted therapies and immunotherapy has made molecular profiling equally critical for treatment decision-making. Identification of oncogenic driver alterations, as well as assessment of programmed death ligand-1 (PD-L1) expression, may directly influence therapeutic selection [1].

Pleural fluid is particularly suitable for this purpose. Cell blocks permit immunohistochemical assessment, whereas analysis of cfDNA from the supernatant enables NGS to detect both therapeutically relevant driver mutations and resistance-associated mutations [2]. Molecular profiling of PF has shown high concordance with tumor tissue and may be performed when tissue sampling is limited, technically challenging, or contraindicated [3,4]. Moreover, emerging evidence suggests that cfDNA analysis may identify clinically relevant mutations even in cytology-negative effusions in selected cases, although current data remain limited and heterogeneous [4,5]. This raises the possibility of expanding molecular characterization without requiring additional invasive procedures; however, the diagnostic yield and clinical implications of this approach remain incompletely defined.

Plasma cfDNA is widely used for molecular profiling in advanced NSCLC. However, its performance relative to PF, particularly regarding mutation detection rates, VAF, and concordance, remains incompletely established.

This study aimed to compare PF and plasma cfDNA for detection of actionable mutations in lung cancer-associated MPE using a comprehensive NGS panel and to determine whether one specimen type is superior or whether both sources provide complementary molecular information.

In this study, we demonstrated that PF cfDNA analysis detects clinically relevant mutations more frequently than plasma cfDNA in patients with lung cancer-associated MPE and yields higher VAFs, indicating a greater fraction of tumor-derived DNA within the pleural compartment. Concordance between PF and plasma was limited, indicating that both specimens are complementary rather than interchangeable. Importantly, clinically relevant mutations were identified in a substantial proportion of patients with probable MPE, underscoring the potential value of cfDNA analysis in this setting. Finally, although actionable alterations were frequently detected, only a limited proportion of patients received targeted therapy in routine clinical practice, revealing a gap between molecular detection and clinical implementation.

The proportion of patients in whom clinically relevant mutations were detected in PF cfDNA in our study (57% among confirmed MPEs) was lower than that reported in several previous series, in which detection rates approaching 80–95% have been described [3–7]. This discrepancy is likely explained by methodological rather than biological factors. Many prior studies reported all somatic alterations detected by NGS, including variants of uncertain significance, or applied sample-level selection criteria based on cfDNA yield, tumor fraction, or sequencing adequacy. In contrast, our analysis was performed on a per-patient basis and restricted to clinically relevant driver and resistance mutations, applying a more stringent and clinically oriented definition.

Despite differences in mutation yield, the degree of concordance between PF and plasma cfDNA observed in our cohort is consistent with previous reports [3–7]. Comparative studies have repeatedly shown incomplete overlap between specimens, with mutations detected exclusively in PF or plasma. In our study, only 26% of clinically relevant mutations were identified in both PF and plasma among patients with definite MPE, and no concordant mutations were observed in probable MPEs. These findings indicate that PF and plasma capture partially distinct molecular information. In a subset of patients with available tissue samples, concordance with FFPE-based molecular testing was higher for PF than for plasma, further supporting the greater sensitivity of PF cfDNA. In cases of discordant molecular findings between PF and other tumor sites, results should be interpreted within the context of spatial tumor heterogeneity and integrated into clinical decision-making rather than considered mutually exclusive. A minority of cases in our cohort reflected prior exposure to targeted therapies, as evidenced by the detection of resistance-associated mutations.

Evidence supporting the use of PF cfDNA in patients with cytology-negative pleural effusions remains limited and is derived mainly from small series. In most published studies, this subgroup comprised fewer than 10–15 patients and was included as an exploratory component within broader MPE cohorts [4,5,8,9]. In the study by Tong et al. [5,6], pleural effusions without detectable tumor cells on cytology were analyzed; PF cfDNA identified most tumor-derived mutations present in matched tissue, whereas pleural sediment DNA detected fewer than 10% of these alterations. Similarly, Mahmood et al. [4] reported oncogenic mutations in 7 cytology-negative effusions using PF cfDNA, highlighting that molecular alterations may be present despite the absence of cytologically identifiable malignant cells. More recently, Thakur et al. [10] demonstrated that cfDNA extracted from pleural effusion supernatant can reliably detect oncogenic mutations, including in cases with negative cytology. Against this background, our finding that clinically relevant mutations were identified in 11 of 21 patients (52%) with probable MPE provides additional quantitative support for the diagnostic value of PF cfDNA in this setting and represents one of the largest series focusing on clinically probable MPE.

Although clinically relevant mutations were detected in a substantial proportion of patients, only 16% received targeted therapy in our cohort, and all treated patients harbored sensitizing EGFR mutations. In contemporary series of advanced stage IV NSCLC, approximately 40–50% of patients are considered candidates for targeted therapies based on actionable oncogenic driver alterations [11]. This proportion is even higher in MPE secondary to NSCLC, with molecular analyses of pleural specimens consistently reporting actionable mutation rates ranging from 60% to 68%, largely driven by the high prevalence of EGFR mutations [12]. This range is consistent with our 57% detection rate among patients with pathologically confirmed MPE. The discrepancy between molecular eligibility and actual treatment in our cohort is explained mainly by the clinical context of the study period (2018–2022), during which comprehensive NGS-based profiling was not routinely implemented, analyses were frequently performed in research settings without real-time availability, and treatment decisions relied on limited EGFR testing.

Importantly, this finding should not be interpreted as a limitation of PF cfDNA performance. In the prospective study by Chang et al. [9], PF molecular analysis identified oncogenic driver mutations in 92.3% of patients, compared with 51.3% using standard diagnostic specimens, and cfDNA results were prospectively integrated into treatment selection, supporting the interpretation that the gap observed in our study reflects differences in clinical implementation rather than intrinsic limitations of PF-based cfDNA analysis. Identification of actionable mutations through liquid biopsy has clear therapeutic implications in advanced NSCLC. Current clinical guidelines support plasma-based genotyping when tissue is unavailable or insufficient, and our findings suggest that PF may represent an even more informative substrate in patients with MPE. In this context, liquid biopsy may facilitate timely initiation of targeted therapy and may be particularly valuable in patients who are not candidates for invasive procedures. However, negative results do not exclude targetable alterations, especially in settings of low tumor burden, and potential confounders such as clonal hematopoiesis and the need for appropriate variant filtering should be acknowledged when interpreting cfDNA results.

This study has several limitations. Although the cohort was prospectively enrolled, the analysis was conducted at a single center with a relatively limited sample size, which may affect generalizability. In addition, a proportion of pleural effusions were classified as probably malignant according to predefined clinical criteria rather than pathological confirmation, a strategy that reflects real-world clinical practice but may allow some degree of misclassification. Molecular analyses were performed in a research setting and were not systematically available to treating physicians at the time of clinical decision-making, precluding assessment of their direct impact on treatment selection. Finally, tissue-based molecular data were available only in a subset of patients and were analyzed in an exploratory manner; therefore, concordance results should be interpreted cautiously.

In patients with lung cancer-associated pleural effusions, PF cfDNA analysis detects clinically relevant driver and resistance mutations more frequently than plasma and provides complementary molecular information. Clinically relevant mutations can be identified even in probable MPEs, underscoring the diagnostic value of PF-based molecular profiling. Systematic integration of PF cfDNA into routine clinical workflows has the potential to expand access to targeted therapies in advanced lung cancer [13].

José M. Porcel: Conceptualization, study design, data interpretation, and manuscript drafting. I. Hidalgo, M. Marqués, E. Parisi, A. Esquerda, and M.A. Sorolla: Sample processing and laboratory analysis. C. Farré and A. Salud: Clinical data collection. M.A. Sorolla: Data analysis. All authors reviewed and approved the final manuscript.

The authors used Paperpal (Cactus Communications) as an AI-assisted tool for language editing and typographical corrections. In addition, generative artificial intelligence (ChatGPT, OpenAI) was used to assist in the design and visual development of the graphical abstract. All scientific content, study design, data analysis, interpretation of results, conclusions, and final approval of the manuscript and graphical abstract were performed exclusively by the authors.

This study was supported by grants from the Spanish Society of Pulmonology and Thoracic Surgery (SEPAR; Ref. 678/2018), Instituto de Salud Carlos III (PI20/01500 and PI23/00926), and the Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR; 2021SGR00781).

None declared.