Early Assessment of Circulating Tumor Cells and Their PD-L1 Expression as Predictive Biomarkers in Metastatic Non-Small Cell Lung Cancer Patients Treated with Immunotherapy: An Exploratory Study

Abstract

Introduction

The prognosis of metastatic non-small cell lung cancer (mNSCLC) has improved in recent years but therapeutic benefit remains heterogeneous, underscoring the need for reliable biomarkers. Liquid biopsy techniques, including the analysis of circulating tumor cells (CTCs) and Programmed death-ligand 1 (PD-L1) expression on CTCs (PD-L1-CTCs), have emerged as promising tools.

Materials and Methods

We conducted a retrospective study in a cohort of mNSCLC patients treated with Immune Checkpoint Inhibitors (ICIs) to evaluate the predictive value of these biomarkers at baseline and at an early time point during treatment. CTCs were captured with the ClearCell^® FX platform and then assessed for PD-L1 by immunostaining. Therapeutic drug monitoring of nivolumab and pembrolizumab was performed.

Results

Thirty-seven mNSCLC patients were enrolled. The median CTCs count was 35 CTCs per 7.5 mL of whole blood (standard deviation ± 939). PD-L1-CTCs were detected in 51% of patients. No correlation was found between PD-L1-CTCs and PD-L1 on tumor Formalin-Fixed, Paraffin-Embedded (FFPE) tissue. Higher-than-median CTCs count was associated with a trend toward worse outcomes. In contrast, PD-L1 expression on CTCs did not predict treatment response. Most patients presented a trough concentration of ICI at cure 3 above 10 µg/mL, regardless clinical response.

Conclusion

Baseline CTCs levels tended to be associated with clinical outcomes, whereas PD-L1 expression on CTCs was not. Early evaluation of changes in these two parameters did not improve outcome prediction under ICI treatment. Larger studies with longer follow-up and more in-depth analyses are warranted to further define the prognostic and predictive value of CTC-based biomarkers.

Introduction

Lung Cancer is the leading cause of cancer-related death worldwide.¹ Unfortunately, most of lung cancer patients are diagnosed at a metastatic stage, immediately limiting their chances of recovery.² Currently, the combination of platinum-based chemotherapy with immune checkpoint inhibitors (ICIs) is the front line gold standard treatment in the metastatic setting, in the absence of oncogenic addiction, as it demonstrated its superiority against chemotherapy alone.^3,4 For the tumors with high expression of Programmed Death-Ligand 1 (PD-L1) the question of the use of ICIs alone or in association to chemotherapy is still open.⁵ Accordingly, first-line immunotherapy is given to most patients, but not all are responders, and some gain little or no benefit. Overall, lung cancer prognosis improved over the past decades, owing to the development of innovative drugs such as ICIs and targeted therapies.⁶ Different pathways could be used to continue the improvement of Non-Small Cell Lung Cancer (NSCLC) prognosis. Biomarkers represent one of these, as they provide crucial information for establishing the most appropriate treatment option for each patient. Liquid biopsy—a minimally invasive method for detecting and analyzing circulating biomarkers—offers a readily accessible snapshot of tumor heterogeneity from a simple blood draw and is a promising alternative to tissue biopsy for early detection, prognostication, and personalized treatment strategies. Among liquid biopsy subtypes, circulating tumor cells (CTCs) analysis has demonstrated its utility in NSCLC, regarding molecular characterization and prognosis.^7–10 The main strength of CTCs rely on their versatility, and their ability to provide a real-time monitoring, especially under ICI therapy. Another advantage lies on the non-invasive procedure to collect CTCs, which is crucial when considering that 25% to 67% of patients with NSCLC cannot benefit from a re-biopsy analysis during first progression.¹⁰ Lately, several components are emerging in the field of liquid biopsy, such as exosomes, microRNA, circulating RNA, tumor platelets, and tumor endothelial cells, raising hopes that lung cancer can be characterized in even greater detail using non-invasive methods in the future.^11,12 However, because CTCs are rare in circulation, their isolation and enrichment remain challenging, whether based on differences in size and deformability between tumor and immune cells or on surface marker expression.¹³ Overall, its position in NSCLC patient management remains unclear.^14,15 Considering that PD-L1 staining on the primary tumor is helpful but imperfect—limited by spatial and temporal heterogeneity—researchers have developed assays for circulating PD-L1 and PD-L1 immunostaining on CTCs.^16–18 The first NSCLC study to report PD-L1 expression on CTCs was published in 2016 by Nicolazzo et al, and many studies followed in this topic.¹⁹ However, the heterogeneity of populations, cancers and techniques used for CTCs isolation and labeling currently limits the conclusions we can draw. Besides, early assessment of CTCs and PD-L1 evolution might be very useful to improve treatment allocation and personalization in the context of metastatic NSCLC.

We present here the results of a study conducted at Lyon University Hospital (LUH) in a cohort of patients with advanced NSCLC treated with standard-of-care therapy (first or second treatment line, ICIs with or without chemotherapy). This cohort included an extensive set of clinical and biological data. We explored the associations between baseline CTCs levels, PD-L1 expression on CTCs (PD-L1-CTCs), ICIs exposition and patient outcomes (treatment response and survival). We also assessed the prognostic value of early changes in CTCs levels and PD-L1-CTCs expression between baseline and the first oncological evaluation. Finally, we investigated the concordance between PD-L1-CTCs and PD-L1 expression in tumor tissue.

Materials and Methods

Population

The study population consisted of 37 consecutive adult patients (≥ 18 years) with histologically proven advanced non-small cell lung cancer (NSCLC) who were enrolled in the MADMAS cohort, at LUH. All participants provided written informed consent before inclusion. Inclusion criteria specified first or second line systemic therapy with immune checkpoint inhibition, optionally combined with platinum chemotherapy. Tumor PD-L1 status on the baseline FFPE primary lesion had to be available, and all participants were covered by the French national social security system.

Ethic

The privacy policy of LUH has been respected, and informed consent has been obtained. According to the French law, and as the present study is classified as “RIPH2”, ie, an interventional study with minimal risks and constraints, this work was registered at the Consultative Committee for the Protection of Persons (CPP Ile de France II) under the umber 2021-A01482-39, and at the «Commission Nationale de l’Informatique et des Libertés» (CNIL) of LUH (Hospices Civils de Lyon, HCL) under the number 21–078. The first six patients included in the present study were part of a previous study, “CIRCAN-ALL”, which was approved by the South-East IV Ethics Committee (Comité de Protection des Personnes Sud-Est IV) on September 16, 2016, under reference number L16-160. This study has been carried out in accordance with the World Medical Association Declaration of Helsinki.

Study Period

Sample collection for this study was performed between March 10, 2021 and November 01, 2023.

Sample Collection

30 mL peripheral blood was collected into ethylenediaminetetraacetic acid (EDTA)-anticoagulated tubes at two predefined time-points: (i) immediately before the first infusion of systemic therapy (baseline, T0) and (ii) six weeks later, on the day of the third administration (T3). Because the treatment schedule called for dosing every 21 days, the T3 sampling therefore captured the biological impact after exactly three cycles, and the radiological evaluation was performed in the three weeks following T3. 5 mL of blood were also collected just before the infusion of ICIs and one hour after the end of the perfusion in order to perform therapeutic drug dosing to measure the blood concentrations of nivolumab and pembrolizumab.

CTCs Enrichment and Cytospin Preparation

For each patient, 7.5 mL of peripheral blood was processed within 2 h. Most of the red blood cells were selectively lysed by incubating the sample for 10 min at room temperature with 22.5 mL of RBC Lysis Buffer (G-Biosciences, #786-650), yielding a suspension of leukocytes and putative CTCs. After centrifugation at 500 × g for 10 minutes, the erythrocyte-rich supernatant was removed, and the nucleated-cell pellet was resuspended in 4 mL of Resuspension Buffer (Biolidics, CBB-F016003). The cell suspension was then loaded onto the ClearCell^® FX1 platform equipped with a CTChip^® FR1S microfluidic cartridge (Biolidics, CBB-FR001002) to enrich CTCs by inertial microfluidics. Following microfluidic enrichment, the eluent was centrifuged at 500 × g for 5 min to obtain a nucleated-cell pellet containing both residual leukocytes and CTCs. Approximately 100 µL of the supernatant was retained and used to resuspend the pellet to a homogeneous final volume (V_after-centri). An aliquot of 12 µL from this suspension was sampled to determine the absolute cell concentration (cells µL⁻¹) on Kovaslide (Garden Grove, 87157) and calculate the V_cytospin, the volume required to deposit 100000 cells on glass slide, the threshold to avoid overcrowding and guarantee interpretable cytospin smears. The required V_cytospin was then adjusted to a total of 200 µL with pluronic 0.2%, and transferred to a sampling chamber (Epredia, A78710003) for cytocentrifugation onto a poly-L-lysine–coated glass slide (ThermoFisher, J2800AMNZ) at 400 × g for 4 min. An adhesive silicone isolator (Grace Bio-Labs, 664270) was affixed around the cell spot, after which the slide was immediately subjected to manual immunofluorescence staining using a protocol adapted from Garcia et al, with modifications to the antibody panel and fluorescence-channel allocation as detailed below.¹⁰

PD-L1 Staining on CTCs and Enumeration

Cellular smears were fixed for 10 min at room temperature in 4% paraformaldehyde (Invitrogen, FB002) and permeabilized with 200 µL of 0.1% Triton X-100 (Sigma-Aldrich, X100-1L). After four washes with phosphate-buffered saline (PBS; 200 µL each, Lonza, 17–512F), non-specific binding sites were blocked for 30 min in 200 µL of blocking buffer containing 5% fetal goat serum (Invitrogen, 31873), 5% Fc-receptor blocking reagent (Miltenyi Biotec, 130–059-901) and 1% bovine serum albumin (Sigma-Aldrich, A8412) prepared in sterile PBS. Then, slides were incubated overnight at 4 °C with primary antibodies diluted in the same buffer: anti-PD-L1 (1:200, Cell Signaling, 86744S), anti-CD15-Alexa Fluor 488 (1:250, BD Biosciences, 560172), anti-CD41-Alexa Fluor 488 (1:40, BioLegend, 303723) and anti-CD45 (1:500, ThermoFisher, MA5-17687). Following three PBS rinses, nuclei were counterstained for 1 h at room temperature with DAPI (1:1000, BioLegend, 422801) together with secondary antibodies—goat anti-rat Alexa Fluor 488 (1:500, ThermoFisher, A-11006) and goat anti-rabbit Alexa Fluor 647 (1:500, ThermoFisher, VB296618). All cells negative for CD45 CD41 and CD15 markers were counted as CTCs, and among them, PDL1 positive cells were enumerated (Supplemental - Figure S1).

The examined fraction of the enriched sample was calculated as V_cytospin/V_after-centri; this percentage was recorded and used to normalize the manual CTCs semi-quantitative count back to the original 7.5 mL whole-blood sample by simple cross-multiplication.

Lung Cancer PD-L1 Immunostaining

Tissue specimens were fixed in 10% neutral-buffered formalin, embedded in paraffin, and processed using standard histological procedures. Tumor diagnoses were established based on light microscopic examination of conventionally prepared sections stained with hematoxylin-phloxine-saffron (HPS), following the latest WHO classification criteria of lung tumors.²⁰ Immunohistochemistry (IHC) was performed on 4-μm-thick sections cut from paraffin blocks using a fully automated system (Benchmark XT System, Ventana Medical Systems Inc., Tucson, AZ, USA). PD-L1 staining was performed using the 22C3 pharmDx assay (Dako North America, Carpinteria, CA, USA) according to the manufacturer’s instructions, on freshly cut or formalin-fixed paraffin-embedded (FFPE) tumor tissue sections less than 6 months old, in line with slide stability recommendations.²¹ Tumor PD-L1 expression was evaluated using the Tumor Proportion Score (TPS), defined as the percentage of viable tumor cells exhibiting partial or complete membranous staining of any intensity. Based on TPS, tumors were classified as PD-L1 negative (TPS <1%) or PD-L1 positive (TPS ≥1%), with the exact TPS percentage recorded for each case.

Cell-Free DNA Extraction and Sequencing

Extraction of cfDNA, library preparation from at least 10 ng of cfDNA with our custom 77-genes panel based on capture technology, sequencing and bioinformatics analyses were performed as previously described.²²

ICIs Dosages

Quantification of nivolumab (NIVO) and pembrolizumab (PBZ) was performed with liquid-chromatography mass-spectrometry methods using mAbXmise^® kits (Promise Proteomics), as previously described.^23,24 These methods, fully validated, present a limit of quantification at 2 µg/mL, suitable for therapeutic drug monitoring. Trough and peak concentrations were measured for the three first administrations.

Statistical Analysis

All statistical analyses were performed using R, version “2022.02.1 + 461” (The R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were reported as mean with standard deviation (Std Dev) and categorical variables as count with percentages and their 95% confidence intervals (CI95%). Missing data were reported as such. PFS starting point is the date of first ICI infusion.

Results

Description of our population is available in the Table 1. Median follow-up time was 672 days (Std Dev ± 356 days). There was a majority of men, with a median age of 64 years-old, and good performance status. Adenocarcinoma was the predominant histology (68%); 95% had stage IV disease. Most patients (91%) were on first-line therapy, typically platinum-based chemotherapy combined with ICIs. Over half of patients (54%) received radiotherapy for indications such as pain relief, bone consolidation or brain metastases. Median treatment course between first treatment injection and first radiological assessment was 7.9 weeks (Std Dev ± 2).

Table 1.

Patient Demographics and Baseline Clinical Characteristics (N=37)

Characteristic	Value
Sex (male), n (%)	26 (70.3)
Age, mean (Std Dev), years	64 (9.4)
Ever-smoker, n (%)	37 (100)
ECOG Performance Status, n (%)
≤1	34 (91.9)
≥2	3 (8.1)
Histology, n (%)
Adenocarcinoma	25 (67.6)
SCC	5 (13.5)
NSCLC (NOS)	6 (16.2)
Other	1 (2.7)
Stage, n (%)
IIIB–IIIC	2 (5.4)
IV	35 (94.6)
Metastatic sites, n (%) *
Brain metastasis	14 (37.8)
Liver metastasis	6 (16.2)
Other sites	15 (40.5)
PD-L1+ on tumor FFPE tissue, n (%)	21 (56.8)
PD-L1+ on tumor FFPE tissue, median (Std Dev)	2 (33)
Oncogenic driver alteration (tumor FFPE), n (%)	13 (35.1)
EGFR	2 (5.4)
ALK	0 (0)
ROS1	0 (0)
KRAS non-G12C	6 (16.2)
KRAS G12C	5 (13.5)

Notes: * Metastatic sites can overlap (patients may have >1 site), so totals need not equal 37.

Abbreviations: SCC, Squamous Cell carcinoma; NSCLC, Non-Small Cell Lung Cancer; NOS, Not Otherwise Specified; PD-L1, Programmed Death-Ligand 1; CTCs count, per 7.5 mL of whole blood; FFPE, Formalin-Fixed Paraffin Embedded; Std Dev, Standard Deviation; EGFR, Epithelial Growth Factor Receptor; ALK, Anaplastic Lymphoma Kinase; ROS1, Proto-oncogene tyrosine-protein kinase ROS type 1; KRAS, Kirsten Rat Sarcoma.

Median PBZ dosage at first oncological evaluation is reported in the Table 2. We did not observe any statistically significant difference across the Progressive Disease (PD), Stable Disease (SD) and Partial Response (PR) groups, with dosage of 26.0 (Std Dev ± 14.6), 22.6 (Std Dev ±8) and 22.2 (Std Dev ± 7.9) µg/mL respectively. Only two patients treated with nivolumab were included, receiving second line therapy, and classified in the PD group. Moreover, we had two missing data regarding PBZ dosage in the PD group.

Table 2.

Treatment Exposure, Response, and Translational Measures

Characteristics	Value
Line of treatment, n (%)
Front line	34 (91.9)
≥2nd line	3 (8.1)
Regimen, n (%)
Carbo–Pemetrexed–Pembrolizumab	28 (75.7)
Carbo–Paclitaxel–Pembrolizumab	6 (16.2)
Pembrolizumab	1 (2.7)
Nivolumab	2 (5.4)
Platinum cycles, median (Std Dev)	4 (0.8)
Pembrolizumab cycles, median (Std Dev)	7 (6.7)
Best response at 1st evaluation*, n (%)
PD	6 (16.2)
SD	16 (43.2)
PR	15 (40.5)
Radiotherapy, n (%)	20 (54.1)
CTCs, median (Std Dev; min–max)
Baseline CTCs count	35 (938.8; 0–5625)
CTCs count at 1st evaluation	31.5 (628; 0–3466)
Baseline PD-L1 expression on CTCs**, median (Std Dev)	7 (31.9)
PD-L1–positive CTCs at baseline, n (%)	19 (51.1)
Pembrolizumab concentration at 1st evaluation (µg/mL), median (Std Dev; min–max)	22.5 (7.8; 2.4–39.6)

Notes: *According to RECIST v1.1 criteria. **After removal of baseline CTCs negative patients.

Abbreviations: Carbo, Carboplatine; PD, progressive disease; SD, stable disease; PR, partial response; Std Dev, Standard Deviation.

Considering only the patients with initial CTCs detection (n=32), we found no correlation between the PD-L1 expression on CTCs and tumor tissue, with a Spearman correlation test of −0,28 (p = 0.13).

Median baseline CTCs count was 35 cells (per 7.5 mL of whole blood, Std Dev ± 938.8). The 10 patients with a higher baseline CTCs burden (>35/7.5 mL) showed poorer progression free survival (PFS) than those with ≤35, with early and persistent curve separation, but the difference was not statistically significant (log-rank p = 0.14) (Figure 1A). At first clinical evaluation, the median CTCs count was 31.5 (per 7.5 mL of whole blood, Std Dev ± 628).

Figure 1.

Patient outcomes according to circulating tumor cell detection and PD-L1 expression. (A) Survival probability by baseline CTCs count; (B) Paired CTCs counts at baseline and first evaluation by RECIST response; (C) Survival probability by baseline PDL-1 expression (%) on CTCs; (D) Paired PDL-1 expression at baseline and first evaluation by RECIST response; Time is expressed in days.

Abbreviations: PD-L1, Programmed death-ligand 1; PD, Progressive Disease; SD, Stable Disease; PR, Partial Response; CTC, Circulating Tumor Cells; RECIST, Response Evaluation Criteria in Solid Tumors v1.1.

Visually, it is difficult to establish a trend, although we might think that the CTCs count show a more sharply decrease among PR and SD patients in comparison to the PD group (Figure 1B). We observed no statistically significant correlation over time: indeed, median CTCs count among the progressive disease, stable disease, and partial response groups at first evaluation under treatment were 43 (Std Dev ± 659), 31 (Std Dev ± 859), and 36 (Std Dev ± 138) respectively (Kruskal–Wallis chi-squared, p = 0.76).

Among patients with detectable CTCs, the median PD-L1 CTCs expression at baseline was 7%. Dichotomizing PD-L1 expression at 7% does not stratify survival (log-rank p = 0.70) (Figure 1C). No significant difference in PD-L1 CTCs expression was found between the three RECIST evaluation groups. Given our limited sample size, we were not able to draw any conclusion regarding the impact of PD-L1 expression on CTCs. Baseline PD-L1 expression (%) on CTCs was 17.1 (Std Dev ±14.2) among patients with PD, 5.7 (Std Dev ± 31.6) in the SD group, and 13 (Std Dev ± 37.8) in the PR group.

Longitudinal PD-L1 changes were heterogeneous with increases observed across outcomes (including some responders) (Figure 1D). Regarding the evolution of PD-L1 expression on CTCs under oncological treatment, we did not observe any statistically significant correlation with PFS or RECIST categories (Figures 1C and D). The median evolution of PD-L1 expression (expressed as the median difference in number of PD-L1 positive CTCs for 7.5 mL of whole-blood between first evaluation and baseline) was 13 (Std Dev ± 28.6) in the PD group, 0 (Std Dev ± 47.0) in the SD group and 0 (Std Dev ± 47.4) in the PR group (Kruskal–Wallis chi-squared = 1.9983, df = 2, p = 0.3682).

We also explored the relationship between baseline CTCs counts and mutational status, using our routine NGS gene panel for genomic profiling on cell-free DNA (cfDNA): there was no statistically significant CTCs count difference between the mutated and wild type tumor groups (Wilcoxon test, p = 0.51) (Supplemental - Figure S2).

In Figure 2, we represented a patient-level swimmer plot. For each participant, time on study (blue bar) with censoring (dot), progression (cross) and death (triangle), alongside baseline biomarker status in the left columns (tissue PD-L1, CTCs PD-L1, and CTCs count).

Figure 2.

Swimmer plot representing patient’ outcomes. X‑axis represents time. Y‑axis represents patients. Symbols represents: Progression (+), Death (▲), Censored (●).

Abbreviations: PD-L1 Tissue, Programmed Death-Ligand 1 staining on tumor tissue; PD-L1 CTC, Programmed Death-Ligand 1 staining on CTCs; CTC, CTCs count per 7.5 mL of whole.

Patients with shorter follow-up and early events cluster in the lower part of the plot frequently carry at least one unfavorable marker (tissue PD-L1 ≤2% and/or CTCs PD-L1 ≥1% and/or CTCs count ≥35/7.5 mL). The longest-followed patients (upper rows) more often show CTCs <35 at baseline, and many remain alive/censored without early progression. PD-L1 status (on CTCs or tissue) appears heterogeneous and discordant across individuals and does not display a clear monotonic relationship with outcomes in this cohort (both PD-L1–positive and –negative cases are seen among long and short survivors). Visually, baseline CTCs burden is the marker that most plausibly aligns with worse outcomes, whereas PD-L1 (CTCs or tissue) shows no consistent prognostic separation here. This complements the Kaplan–Meier results and suggests that CTCs quantity may be more informative than PD-L1 status in this dataset.

Discussion

In this observational study, our population was representative of non-oncogene addicted NSCLC, with an overall response rate to chemo-immunotherapy comparable to the keynote 189.²⁵ Although it was not statistically significant, we observed a trend for a prognostic value of the baseline CTCs level, with a worse prognosis associated with a high level of circulating tumor cells. We found no correlation between PD-L1 expression in tumor tissue and on CTCs. PD-L1 CTCs expression was not predictive of patient’s outcome under ICIs ± platinum‑based chemotherapy. CTCs level and PD-L1 expression evolution over time under treatment did not appear to be statistically associated with patient’s outcome. No association was found between mutational status and CTCs count, and ICIs dosages did not differ between responder and non-responders.

Regarding the prognosis impact of CTCs count, the correlation was previously demonstrated in different studies, in the non-metastatic and metastatic setting among NSCLC patients.^26–28 The limited statistical power of our study could probably explain our results. Surprisingly, at the first clinical evaluation (median 7.9 weeks ± 2 Std Dev), CTCs levels did not differ among progressive, stable, and responder groups—unlike some prior reports in different settings that assessed patients earlier in the treatment course.^8,29 The high disease burden, considerable heterogeneity within the patient population, and our isolation method could, at least in part, account for these findings. Moreover, a delayed clearance of CTCs over time could partly explain these findings, as suggested by Su et al³⁰ As perioperative strategies evolve and therapeutic options proliferate for stage IV NSCLC, there is a pressing need for biomarkers to better stratify patients and tailor treatment.³¹

Tumor PD-L1 immunostaining is informative yet limited as a predictor of response to ICIs.³² We hypothesized here that PD-L1 expression on CTCs could be an interesting surrogate to tumor staining. First, we studied the association of PD-L1 expression on tumor tissue and CTCs: we found a weak and negative correlation, suggesting no – or a complex - relationship between the two biomarkers. Previous study explored this relation: a recent meta-analysis synthetized the existing literature and pooled the results through a random effect model to obtain an overall concordance of 63% (95% CI, 44.0–80.0), and up to 74% [95% CI, 38.0–98.0] among studies with homogeneity between PD-L1 CTCs and tumor staining, assuming a wide heterogeneity.¹⁸ In our study, as in previously published works, one of the main limitations lie in the heterogeneity of the PD-L1 antibodies used for staining and in the CTCs isolation methods. Second, we explored the impact of PD-L1 CTCs expression on prognosis. No prognosis impact was observed at the cut-off of 7% (corresponding to the median in our population), and no difference was observed either when comparing the median baseline PD-L1 expression in the three RECIST groups. PD-L1 changes over time were heterogeneous with increases observed across outcomes (including some responders), suggesting treatment-related adaptive modulations rather than a simple monotonic relationship with response. Existing evidence concerning the prognosis impact of PD-L1 expression on CTCs are represented by numerous studies, with the first published in 2016 by Nicolazzo et al¹⁹ A wide heterogeneity exists, summarized in a large meta-analysis conducted by Acheampong et al in 2024, with the conclusion of a lack of statistically significant differences regarding OS and PFS, but a trend toward shorter PFS in the CTCs-PD-L1 positive group not treated with immune checkpoint inhibitors, suggesting a potential regulation of the interactions between the immune system and the CTCs by ICIs.^18,33 Further studies, with larger sample and longer follow-up are needed. In gastric cancer, Liu et al reported that PD-L1 expression on CTCs could be in relation with epithelial to mesenchymal (EMT) transition, suggested by a strong correlation between PD-L1 mRNA and EMT status, migratory and invasive capacities, based on published gene expression datasets and in vitro data.³⁴ Further studies are needed to figure out the signification of PD-L1 expression on CTCs, and the related interactions offered by such expression with the immune system. Mechanistic exploration with advanced single-cell transcriptomic analysis could also help clarify how CTCs evolve and how their PD-L1 expression changes during ICI therapy. However, a recent study offers a comprehensive overview in the non-metastatic NSCLC setting: Purcell et al explored the impact of CTCs, PD-L1 expression on CTCs and differential genes expression through microarray analysis, concluding to a better prognosis of stage III NSCLC patients with a clearance of CTCs after treatment, and a worse prognosis for those who have sustained PD-L1 expression on CTCs. They identified upregulated genes corresponding to an invasive, pro-proliferative and potentially immune-evasive phenotype, reinforcing the hypothesis that CTCs could correspond to tumor cells able to escape chemo and/or radiative cell death. Interestingly there was no difference in the PD-L1 gene expression between the different prognosis groups.⁸ The variability of the outcomes regarding the prognosis impact of PD-L1 CTCs between our work and the existing literature might be attributable to the impact of chemotherapy associated with ICI, biological variability across the cohorts, technical considerations, especially in the case of the manual CTCs count we used.

We performed pembrolizumab and nivolumab blood dosages through the methods validated by Millet et al.²⁴ There was no difference between the PD, SD and PR groups, probably indicating that patients have reached the minimum blood level required to achieve efficacy, which is dependent on many other factors. However, it is interesting to note that one patient in the PR group presented a residual exposure at T3 measured at 2.4 µg/mL, significantly below the threshold of 10 µg/mL defined to saturate the target.³⁵ Therapeutic drug monitoring could be a relevant option to personalize more cancer treatments, in using the minimum effective concentration or the ideal delay between two administrations.

The main strength and originality of our work lies on the representativity and homogeneity of our cohort, and the focus on advanced NSLC patients receiving first line ICI. Indeed, we enrolled 37 NSCLC patients with usual characteristics for this disease, consecutively, and following the standard process for diagnosis and treatment, with a wide majority receiving ICIs associated with platinum based chemotherapy, according to current guidelines.^3,4 Previous studies focused on mono-immunotherapy, frequently among heavily pre-treated patients. Besides, we decided to perform early evaluation of the biomarker of interest, because of the necessity to personalize patient’s treatment as soon as possible in the treatment course, to prevent consequent tumor progression which might have an impact on patient prognosis. Different factors are influencing patient’s outcome under treatment and probably that a combination of biomarkers could help to better predict it. In this cohort, we observed that patients with detectable ctDNA at diagnosis, particularly those with KRAS mutations, may have a poorer prognosis compared to those without detectable ctDNA (data not shown). Moreover, our study is one of the first to describe these parameters in a European population.

A major limitation of our assay was the semi-quantitative manual scoring of CTCs and PD-L1 staining—labor-intensive and prone to inter-operator variability—which would need to be addressed for clinical implementation. In addition, heterogeneity of PD-L1 staining further constrains interpretation. Although our cohort is relatively large for this field, the sample size and its heterogeneity in some regards may justify the low statistical power, limiting the possibility of multivariable analysis to refine our conclusions. Indeed, regarding the low number of events occurred during follow-up, no multivariable Cox regression analysis was performed. We considered that the limited number of events would result in substantial overfitting and unstable estimates, thereby compromising the reliability of the adjusted hazard ratios. The heterogeneity also lies on the potential comorbidities of the patients, that could potentially have influenced their outcomes. Fifty-four percent of the patients included in our study underwent radiotherapy, which may have influenced CTCs shedding. The limited sample size precluded a sensitivity analysis comparing irradiated and non-irradiated patients, representing an important limitation of our study. In addition, a few patients included in this study received different treatment regimens with immunotherapy alone or in combination, making it more difficult to draw conclusions whether the biomarker results are driven by the disease biology or the variable treatment effects. Longer follow-up and additional sampling time points might have revealed CTCs and/or PD-L1 clearance and clarified their prognostic relevance.

Conclusion

This exploratory study adds to the liquid biopsy literature by (i) focusing on advanced NSCLC patients treated with ICI, (ii) showing a trend toward poorer outcomes in patients with higher baseline CTCs counts, while PD-L1 expression on CTCs did not correlate with prognosis, and (iii) showing that early on-treatment reassessment of these two biomarkers did not provide additional prognostic information. Future work should prioritize larger, prospectively designed cohorts with longer follow-up and standardized, automated CTCs/PD-L1 assays to reduce variability and increase sensitivity. Integrating longitudinal sampling with multi-omic profiling—including transcriptomic analysis of CTCs—could illuminate mechanisms of primary and acquired resistance to ICIs and identify actionable pathways. Combination of other relevant biomarkers such as exosome or H3K27Me3-nucleosome is also likely a promising avenue to explore. Finally, evaluating PD-L1 expression on CTCs in earlier-stage disease may help refine selection for adjuvant or neoadjuvant immunotherapy, potentially broadening the clinical utility of liquid biopsy-based biomarkers.

Funding Statement

This project, accredited by Lyonbiopôle, has been carried out thanks to the support of the Auvergne-Rhône-Alpes region and the Cancéropôle CLARA as part of the Proof of Concept programme.

CRediT

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. Emmanuel Grolleau, David Barthelemy, Sebastien Couraud, and Lea Payen are co-first and last authors.

Data Sharing Statement

The data that support the findings of this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request.

Ethic Statement

The privacy policy of LHU has been respected, and informed consent has been obtained. According to the French law, and as the present study is classified as “RIPH2”, ie, an interventional study with minimal risks and constraints, this work was registered at the Consultative Committee for the Protection of Persons (CPP Ile de France II) under the umber 2021-A01482-39, and at the «Commission Nationale de l’Informatique et des Libertés» (CNIL) of LUH (HCL) under the number 21-078. This study has been carried out in accordance with the World Medical Association Declaration of Helsinki.

Disclosure

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:

Dr Emmanuel Grolleau reports Meeting/travel support from Johnson&Johnson, outside the submitted work. Ms Gaelle Lescuyer reports Honoraria from Astra Zeneca, outside the submitted work. Dr Marie Piecyk reports Honoraria from AstraZeneca; Meeting/travel support from Volition, outside the submitted work. Dr Léa Payen reports honoraria from Astrazeneca; research and or training contract from Aima, AstraZeneca, Biodena, Beckton Dickinson, Volition, outside the submitted work. Dorothée Lebert works for the company Promise Proteomics. This paper and the abstract were not presented at any conference.

This paper was previously uploaded as a preprint on SSRN at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6042939.