Baseline levels and dynamic changes of cfDNA, tumor fraction and mutations to anticipate the clinical course of small cell lung cancer (SCLC) patients treated with first-line atezolizumab and chemotherapy: an hypothesis generating study (CATS/ML43257)

Abstract

Background

Atezolizumab (A) plus carboplatin-etoposide (CE) represents the new first-line treatment in extensive stage (ES)-Small Cell Lung Cancer (SCLC) patients. This study aims at identifying the association of baseline and dynamic changes of cfDNA, Tumor Fraction (TF) and variant allele frequency (VAF) of tumor-related mutations with median (m) overall (OS) and progression free survival (PFS) in SCLC patients treated with ACE.

Materials and methods

This is a single-center prospective exploratory study including treatment-naive ES-SCLC patients eligible to first-line ACE. Liquid biopsies were longitudinally collected at baseline (T0), after cycle 1 (T1) and 2 (T2), at disease progression (T3). cfDNA Next Generation Sequencing (NGS) analysis was performed; genomic profiles and TF were inferred from shallow WGS (sWGS).

Results

Thirty-two patients were included; mPFS and mOS were 5.19 and 7.96 months, respectively. Higher T0 cfDNA (HR 1.44, 95% CI 1.17–1.77, p = 0.0006) and VAF (HR 2.6, 95% CI 1.36–4.93, p = 0.0039) were associated with risk of death; higher T0 cfDNA (HR 1.29, 95% CI 1.08–1.54, p = 0.0049), TF (HR 1.97, 95% CI 1.02–3.82, p = 0.044) and VAF (HR 2.32, 95% CI 1.22–4.42, p = 0.01) were predictors of risk of PD. Among the dynamic changes in the biomarkers under investigation, the association of 10-unit increase of VAF T0-T1 and T0-T2 with OS (HR 1.38, 95% CI 1.01–1.88, p = 0.043; HR 1.56, 95% CI 1.21–2.16, p = 0.008) and PFS (HR 1.69, 95% CI 1.18–2.43, p = 0.004; HR 1.81, 95% CI 1.22–2.70, p = 0.003) was estimated.

Conclusion

T0 and dynamic changes of cfDNA, TF and VAF may help physicians to stratify ES-SCLC patients receiving first-line ACE and to anticipate the clinical course of the disease.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13046-025-03434-3.

Background

Small cell lung cancer (SCLC) is a high-grade neuroendocrine disease which occurs in approximately 15% of lung cancer patients [1]. At diagnosis, about 70% of SCLC patients present with extensive-stage (ES) disease [2]characterized by an aggressive phenotype and a dismal prognosis. Platinum-based chemotherapy has represented the standard of care for many years. Despite an initial high responsiveness to the treatment, the sensitivity is temporary, with a median progression-free survival (PFS) of 5–6 months and a median overall survival (OS) of 9–10 months [3]. Recently, immune checkpoint inhibitors combined with platinum-based chemotherapy emerged as the new first-line treatment in ES-SCLC patients [4–8]. In the IMpower133 trial, the addition of atezolizumab to platinum-based chemotherapy for 4 cycles followed by atezolizumab maintenance, showed to prolong the median overall survival (mOS) to 12.3 versus 10.3 months (hazard ratio [HR] 0.70; 95% confidence interval [CI] 0.54–0.91) with chemotherapy alone [5]. Similarly, the CASPIAN trial demonstrated the benefit of the addition of durvalumab to platinum-based chemotherapy, achieving a mOS of 12.9 versus 10.5 months (HR 0.71; 95% CI, 0.60–0.86) with chemotherapy alone [4].

Despite this promising data, the long-term benefit from the addition of immunotherapy to platinum-based chemotherapy is still an unmet medical need, with only 12% (95% CI, 7–17%) of patients alive at 5 years in the IMpower133/IMbrella A trial [9].

Furthermore, in some patients there is a lack of benefit from the addition of immunotherapy [5]. In this scenario, many efforts have been made by exploring novel potential biomarkers capable of predicting the outcome of patients and helping physicians in patients’ selection. However, finding new biomarkers in SCLC may be challenging due to the aggressive nature of the disease that often precludes the availability of adequate and sequential tissue specimens for in-depth studies [10].

Moreover, classical predictive factors of response to immunotherapy, such as programmed death-ligand 1 (PD-L1) and tumor mutational burden (MTB), do not appear to play a significant role in patients’ stratification [7, 11–16].

Conversely, the gene expression profiling (GEP) and the tumor microenvironment (TME) characterization may contribute to the identification of patients who could benefit from immunotherapy, as emerged from recent studies and explored by our group [17, 18].

With the aim of overcoming the issue of lacking tissue, in the last few years there has been an increasing development in liquid biopsy, now representing a reliable, accessible and non-invasive tool applicable in many settings. A relevant advantage of liquid biopsy is also its ability to catch tumor heterogeneity and its applicability in longitudinal studies. In this setting, serial blood samples could catch the plasticity and dynamic changes of SCLC, thus providing both qualitative and quantitative data on circulating tumor DNA [19]. To date, however, only few data are available on the predictive role of liquid biopsy in SCLC [20–22].

Based on these premises, we designed and conducted the CATS/ML43257 prospective study to identify circulating predictive biomarkers of response to chemoimmunotherapy in ES-SCLC patients, detected through longitudinal liquid biopsies, and potentially useful for patient stratification and predicting long-term benefit from immunotherapy.

Methods

Study design and eligibility criteria

This is a single-center exploratory observational translational study including treatment-naive patients affected by ES-SCLC eligible to first line treatment with atezolizumab plus carboplatin and etoposide (ACE). The study was conducted in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki and approved by the Institutional Ethical Committee. All patients signed an informed consent form before enrollment. Patients showing symptomatic, actively progressing CNS metastases or leptomeningeal disease and those with contraindications to immune checkpoint inhibitors have been excluded. The study included 30 months of enrollment and at least 12 months of follow-up. Study design is depicted in Fig. 1.

Fig. 1.

Study design: T0: baseline; T1 after first cycle; T2: after second cycle; T3 at the time of disease progression

Treatment was performed according to standard clinical practice and was based on induction phase with intravenous administration of carboplatin AUC 5 on day 1 every 21 days plus etoposide 100 mg/sm on days 1, 2, 3 every 21 days plus atezolizumab 1200 mg flat dose on day 1 every 21 days for a maximum of 4 planned cycles. Following the induction phase, each patient under disease control received atezolizumab maintenance until unacceptable toxicity, disease progression or loss of clinical benefit according to clinical practice.

Brain, Chest and Abdomen CT-scan for tumor assessment was performed at the baseline and every three cycles, according to clinical practice. Response rate was assessed according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. Safety was collected according to Common Toxicity Criteria for Adverse Events (CTCAE) version 5.0.

According to the study protocol, the primary endpoint of the study was the association between baseline cfDNA and its variations at different timepoints and progression free survival (PFS). The secondary endpoints were: (i) the association between baseline cfDNA and its dynamic variation and overall survival (OS); (ii) the association between baseline TF and VAF and their dynamic variation with PFS and OS; (iii) the characterization of tumor-related mutations in the study population through liquid biopsy, both at baseline and at the time of disease progression.

Plasma collection and cfDNA extraction

Liquid biopsies were collected longitudinally at key timepoints: T0 (baseline evaluation, corresponding to the first day of the first treatment cycle), T1 (after the first cycle of therapy, corresponding to the first day of the second treatment cycle), T2 (after the second cycle of therapy, corresponding to the first day of the third treatment cycle), and T3 (at disease progression). For each collection, approximately 20 mL of peripheral blood was drawn from each patient into two Cell-Free DNA BCT tubes (Streck Corporate, La Vista, NE, USA) and processed within 24 to 72 h. Blood samples underwent a two-step centrifugation process: first at 2000×g for 10 min at 4 °C, followed by 20,000 g for 10 min. Plasma aliquots were stored at − 80 °C until analysis, as detailed previously [23].

Cell-free DNA (cfDNA) was extracted from 2 to 5 mL of plasma using the AVENIO cfDNA Isolation Kit (Roche Sequencing Solutions, Pleasanton, CA) according to the manufacturer’s protocol. The extracted cfDNA was then quantified fluorometrically using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA).

cfDNA next generation sequencing analysis

Next Generation Sequencing (NGS) libraries were prepared from 10 to 50 ng of cfDNA in a total volume of 50 µL using the AVENIO ctDNA Expanded kit v2 (77 genes; Roche Diagnostics Spa) according to the manufacturer’s instructions as previously described [23]. Individual enriched libraries were quantified using the QuBit dsDNA HS Assay kit (Thermo Fisher Scientific), and their profile was assessed using the Agilent High Sensitivity kit on the Agilent 2100 Bioanalyzer. Each sequencing run comprised a pooled set of 8 samples, processed on an Illumina NextSeq 500/550 platform using the 300-cycle NextSeq High Output kit in paired-end mode (2 × 150 cycles). The sequencing-generated FastQ files were analyzed using the AVENIO ctDNA analysis software (Roche Diagnostics) with default parameter settings for the Expanded Panel V2. Following the manufacturer’s guidelines, the percentage of reads aligned to the human reference genome (hg38) within the targeted region (unique depth) should exceed 40%. Additionally, the median unique depth across bases in the targeted region should be at least 2500×. Variant allele frequency (VAF) was calculated as the ratio of mutated reads to the total sequence reads at a given genomic position.

All identified variants were manually reviewed. Synonymous, intronic variants and those found in population databases such as ExAC and dbSNP as polymorphisms were excluded. Coding protein and splicing variants were evaluated for pathogenicity using Varsome Premium (Saphetor SA, Lausanne, Switzerland). Only variants annotated as pathogenic, likely pathogenic or with uncertain/unknown significance (VUS) were considered as trackable mutations in plasma samples for longitudinal analyses.

Shallow whole genome sequencing (sWGS) analysis

Whole genome libraries were prepared using 10 to 50 ng of cfDNA with the KAPA HyperPrep kit (Roche), following the manufacturer’s protocol, and sequenced on the Illumina NextSeq 500/550 platform (Illumina Inc, San Diego, CA, USA) using 75 bp single-end reads.

Genomic profiles and tumor fraction (TF) were inferred using both the SAMURAI and the ichorCNA algorithms, achieving a sensitivity of 0.95 and a specificity of 0.91 for tumor detection, with a TF cutoff of 0.03 (Limit of Detection, LOD = 3%) [24, 25].

Statistical analysis

Continuous variables were summarized as medians and interquartile ranges (IQRs), while categorical variables were presented as frequencies and percentages.

Survival curves for overall survival (OS) and progression-free survival (PFS) were estimated using the Kaplan–Meier method, and median survival times were reported along with the corresponding 95% confidence intervals (CIs). Time-to-event was defined as the interval from the date of first treatment to the occurrence of death (OS and PFS) and/or disease progression (PFS), whichever occurs first. The median follow-up time was estimated using the reverse Kaplan-Meier method. The Mantel-Cox test was used to compare survival distributions between groups of patients with and without TF clearance at T2.

To assess the univariable association between baseline clinical variables, biomarkers, and survival outcomes, Cox proportional hazards (PH) models were fitted, and hazard ratios (HRs) with 95% CIs were reported. For continuous variables, HRs were computed per IQR increase to facilitate interpretation. The proportional hazards assumption was assessed using Schoenfeld residuals and graphical methods.

A dynamic analysis was conducted by modeling the change (deltas) in biomarkers levels over time, adjusting for baseline biomarkers levels. The deltas considered were the differences in biomarkers levels between T1 and T0, as well as between T2 and T0. Initially, restricted cubic splines were used to flexibly model the relationship between biomarker changes and hazards, accounting for potential non-linear effects. However, as a consistent linear relationship was observed—precluding the identification of meaningful cut-off values for the outcomes prediction—a simplified Cox PH model was subsequently used, estimating the HR with 95% CIs per 10-unit increase in biomarkers deltas (Supplementary Fig. 1).

As an exploratory analysis, a random forest algorithm was implemented to assess the relative importance of clinical and biomarker variables. Variable importance was evaluated using the permutation importance measure (VIMP), without further model selection or inference.

The required number of events was estimated considering cfDNA as a continuous covariate in a univariable Cox model, with PFS as the primary endpoint [26]. Assuming a hazard ratio (HR) of 1.5 for a 50-unit difference in cfDNA, corresponding to the interquartile range (Q1-Q3), given a standard deviation of 70, an expected event rate of 80%, 80% power, and a two-sided α = 0.05, the required number of events is approximately 24. This translates to a total sample size of ~ 30 participants (N = E/p).

All statistical analyses were performed using R version 4.4.2. A two-sided p-value < 0.05 was considered statistically significant.

Results

Patients’ characteristics

Overall, 33 patients were included in the CATS/ML43257 study between October 2021 and March 2024.

One patient was finally excluded from data analyses because of rapid clinical worsening and early start of best supportive care, without receiving chemo-immunotherapy. Clinical features of the 32 eligible patients are summarized in Table 1. In summary, patients were mostly males (N = 19, 59.4%), with an ECOG PS of 0 or 1 (N = 28, 87.5%) and a median age of 65 years. All patients reported a previous/current smoking history, with a median of 40 pack/years; 22 (68.8%) patients had less than three metastatic sites, and 7 (21.9%) had treated and/or clinically stable brain metastases at the diagnosis. Liver metastases were present at baseline in 6 (18.8%) patients and bone metastasis in 13 (40.6%) cases.

Table 1.

Patients’ characteristics

	Overall (N = 32)
Gender
Female	13 (40.6%)
Male	19 (59.4%)
Smoking status
Current Smokers	20 (62.5%)
Former Smokers	12 (37.5%)
Charlson Comorbidity Index
≤ 7	6 (18.8%)
> 7	26 (81.2%)
Performance status
0	11 (34.4%)
1	17 (53.1%)
2	4 (12.5%)
Age at diagnosis (years)
Median [IQR]	65 [57, 73]
Stage at diagnosis
Limited	4 (12.5%)
Extended	28 (87.5%)
Number of metastasis
< 3	22 (68.8%)
≥ 3	10 (31.2%)
Extrathoracic metastasis
Absent	8 (25.0%)
Present	24 (75.0%)
Liver Metastasis
Absent	26 (81.2%)
Present	6 (18.8%)
Bone metastasis
Absent	19 (59.4%)
Present	13 (40.6%)
Central nervous system metastasis
Absent	25 (78.1%)
Present	7 (21.9%)
Molecular profile
NEUROD1+	2 (6.0%)
ASCL1+	3 (10.0%)
TRIPLE NEGATIVE	10 (31.0%)
Unknown	17 (53.0%)

Gene expression profiling on tumor tissue was previously performed in 15 (47%) patients: ten harbored an inflamed (triple negative, I), three an ASCL1+ (A) and two a neuroendocrine (NEUROD1+, N) molecular profile, as previously reported [17].

Median number of induction cycles were 4 (1–6), while patients received a median of 3 (0–43) cycles of maintenance. With a median follow-up of 20 months (95% CI 19.0 - not reached), the intention-to-treat (ITT) population showed an objective response rate of 75% and a disease control rate of 84.4%. The median progression-free survival (PFS) was 5.19 months (95% CI 4.74–6.55), and the median overall survival (OS) 7.96 months (95% CI 6.91–14.64) (Supplementary Fig. 2). Among the categorial clinical features, gender (males HR 2.44, 95% CI 1.01–5.88, p = 0.047) and the number of metastatic sites (ms) (≥ 3 h 2.90, 95% CI 1.18–7.09, p = 0.020) significantly associated with OS, while the number of ms was also associated with PFS (≥ 3 h 3.44, 95% CI 1.44–8.23, p = 0.005) (Table 2).

Table 2.

Association of categorial clinical features with PFS and OS

Cox Proportional Hazard Model
		Progression Free Survival			Overall Survival
Sex		HR	95% CI	p-value	HR	95% CI	p-value
	Female	-	-		-	-
	Male	1.26	0.58, 2.73	0.6	2.44	1.01, 5.88	0.047
Smoking Status
	Former smoker	-	-		-	-
	Current smoker	1.02	0.47, 2.20	> 0.9	1.27	0.56, 2.90	0.6
Charlson Comorbidity Index
	≤ 7	-	-		-	-
	> 7	1.26	0.48, 3.35	0.6	1.45	0.50, 4.26	0.5
Performance Status
	≤ 1	-	-		-	-
	> 1	1.41	0.48, 4.19	0.5	1.38	0.39, 4.92	0.6
Number of metastasis
	< 3	-	-		-	-
	≥ 3	3.44	1.44, 8.23	0.005	2.90	1.18, 7.09	0.020
Extra-thoracic metastasis
	No	-	-		-	-
	Yes	1.37	0.58, 3.27	0.5	0.93	0.38, 2.29	0.9
Liver metastasis
	No	-	-		-	-
	Yes	2.11	0.84, 5.33	0.11	1.91	0.70, 5.22	0.2
Bone metastasis
	No	-	-		-	-
	Yes	1.65	0.77, 3.50	0.2	1.07	0.46, 2.44	0.9
Brain metastasis
	No	-	-		-	-
	Yes	1.20	0.48, 2.98	0.7	1.28	0.48, 3.45	0.6

No significant difference in terms of baseline cfDNA (p = 0.12), TF (p = 0.18) and VAF (p = 0.14) was observed between the inflamed and the neuroendocrine plus ASCL1 + phenotype (data not shown), probably because of the small sample size.

cfDNA analysis

We collected 32 samples at T0 and T1, 31 samples at T2 (reduction due to patient’s death), and 21 samples at T3. Specifically, at the data cut-off (31st October 2024), 6 patients had died before T3 sampling, and 5 patients had not yet progressed. cfDNA was obtained in all 116 plasma samples collected across the timepoints. At diagnosis (T0), the median cfDNA level was 24.9 ng/mL of plasma, respectively, ranging from 3.7 to 704.3 ng/mL.

Baseline cfDNA was considered as a continuous variable ranging from 10.16 ng/mL (Q1) to 86.86 ng/mL (Q3) in association with PFS and OS. Higher levels of baseline cfDNA were significantly associated with both disease progression (HR 1.29, 95% CI 1.08–1.54, p = 0.0049) and higher risk of death (HR 1.44, 95% CI 1.17–1.77, p = 0.0006) (Fig. 2).

Fig. 2.

Swimmer plot showing association between baseline values of cfDNA, TF and VAF with PFS (A–C) and OS (D–F). The values reported on the Y axis between the patients’ ID and the plots are the T0 values of cfDNA (A, D), TF (B, E) and VAF (C, F)

Furthermore, a significant reduction in cfDNA levels was observed between T0 and T1 (median 24.9 ng/mL, Q1-Q3: 10.2–86.9 vs. median 19.1 ng/mL, Q1-Q3: 11.3–31.9; p = 0.03) and between T0 and T2 (median 24.9 ng/mL, Q1-Q3: 10.2–86.9 vs. median 14.2 ng/mL, Q1-Q3: 9.9–28.8; p = 0.03). At the time of disease progression (T3), cfDNA levels showed a significant increase relative to T2 (median 14.2 ng/mL, Q1-Q3: 9.9–28.8 vs. median 40.5 ng/mL, Q1-Q3: 12.5–59.3; p = 0.01), sometimes reaching values like those observed at baseline (Fig. 3a). The increase of cfDNA levels at different timepoints was not significantly associated with PFS, whereas a higher risk of death was estimated in presence of a dynamic 10-unit increase of the change in cfDNA between T0-T2 (HR 1.17, 95% CI 1.04–1.32, p = 0.009).

Fig. 3.

Spaghetti plot showing dynamic changes of cfDNA (A), VAF (B) and TF (C) along different timepoints in each patient

Characterization of tumor-related mutations

Tumor-related mutations were identified in 31 out of 32 patients at T0 (96.8%). Among these, 19 patients (61.2%) exhibited two or more pathogenic or likely pathogenic genetic alterations. TP53 and RB1 were the most frequently altered genes. Specifically, 23 (74%) patients exhibited TP53 alterations, and 17 (54.8%) had RB1 mutations. TP53 mutations predominantly involved missense alterations in the DNA-binding domain, while RB1 mutations were primarily nonsense mutations. Other frequently altered genes included PTEN (9.3%), PIK3CA (9.3%) and KRAS (6.2%). A list of mutated genes with specific variant allele fraction (VAF) is shown in Fig. 4. The gene with the highest VAF was used to evaluate clearance (a 99–100% reduction in VAF) during treatment. VAF clearance was observed in 12 (38.7%) patients between T0 and T1 and in 21(67.7%) patients between T0 and T2. Conversely, at the time of disease progression (T3), an increase of the VAF of tumor-related mutations, whose values returned similar to those at the baseline, was observed in 16 patients (50%). Additionally, new pathogenic mutations emerged in 8 (25%) patients, each with a VAF < 10% (Table 3). Moreover, no significant differences in terms of median PFS and OS were observed between patients who, at progression, exhibited new pathogenic variants compared to those at diagnosis (data not shown).

Fig. 4.

Heatmap showing tumor-related mutations and specific VAF detected at T0

Table 3.

Pathogenic mutations detected at progression with their variant allele frequency (VAF). Variants shown in bold emerged at progression, whereas those in regular black font were already identifiable at the time of diagnosis

Patient	Pathogenic mutation	VAF (%)
CATS 1	N.A.
CATS 2	RB1 c.751 C > T; TP53 c.785G > T	78.3; 79.6
CATS 3	N.A.
CATS 4	ALK c.1414 + 1G > A; PTCH1 c.2560 + 1G > A; PTEN c.375 A > T; TP53 c.375 G > T	0.27; 0.9; 42.7; 45.0
CATS 5	JAK c.1744 C > T; TP53 c.578 A > G	0.36; 53.5
CATS 6	RB1 c.1498 + 2T > G; SMO c.1265 G > T; TP53 c.329 G > T	11.8; 9.4; 13.1
CATS 7	TP53 c.427 G > C; TP53 c.848 G > C	0.25; 15
CATS 8	ERBB2 c.932 G >A; TP53 c.542 G > C	0.14; 70.8
CATS 9	RB1 c.717T >A; TP53 c.797 G >T	42; 46.8
CATS 10	N.A.
CATS 11	TP53 c.707 A >T; RB1 c.1694 C >G	0.12; 55
CATS 12	N.A.
CATS 13	N.A.
CATS 14	MTOR c.5930 C > G; RB1 c.1422–2 A >T; TP53 c.476 C > T	34.3; 77; 70.5
CATS 15	RB1c.1422–2 AC >T; RB1 c.1333 C > T; TP53 c.482 C > A	0.7; 52.4; 51.1
CATS 16	PTEN c.319G > T; RB1 c.1215 + 1G > T; TP53 c.737 T > G	0.2; 0.54; 0.5
CATS 17	TP53 c.976G > T	49.9
CATS 18	SMAD4c.1487 G >T; KIT c.1255 G > C; RB1 c.1960 + 1G >A; TP53 c.743 G > T	0.4; 9.6; 71.8; 73.9
CATS 19	N.A.
CATS 20	N.A.
CATS 21	APCc.4648G >T; TP53 c.586 C >T;	8.6; 12.3
CATS 22	BRCA1 c.4148 C >A; FGFR1 c.1591 C >G; KRAS c.29G > C; ALK c.2725 G >A	20.6; 0.2; 20.2; 0.8
CATS 23	RB1 c.2212-1G>  A; TP53 c.314G >T	60.6; 61.5
CATS 24	N.A.
CATS 25	GNAQc.196G >T; RB1 c.2027T > G	0.2; 4.3
CATS 26	N.A.
CATS 27	N.A.
CATS 29	No variants identified
CATS 30	RB1 c.2158 A > T; TP53 c.701 A > G	6.7; 6.6
CATS 31	TP53 c.578 A > G	3.6
CATS 32	No variants identified
CATS 33	N.A.

N. A: Not Available

Ranging from the first quartile (8.33%) to the third quartile (49.08%), higher baseline VAF was significantly associated with higher risk of disease progression (HR 2.32, 95% CI 1.22–4.42, p = 0.01) and death (HR 2.59, 95% CI 1.36–4.93, p = 0.0039) (Fig. 2).

The dynamic trend of VAF was observed through longitudinal analysis; the median VAF at T0, T1, T2, and T3 was 27.0% (Q1–Q3: 8.3–49.1), 0.9% (Q1–Q3: 0–6.0), 0.2% (Q1–Q3: 0–1.6), and 46.8% (Q1–Q3: 12.3–55.0), respectively. A statistically significant VAF clearance was observed between T0 and T1 (p = 0.0002) and between T0 and T2 (p = 0.0001). Conversely, a statistically significant increase in VAF was detected between T1 and T3 (p = 0.0004) and between T2 and T3 (p = 0.0002) (Fig. 3b).

A higher risk of disease progression was estimated in presence of a dynamic 10-unit increase of change in VAF between T0-T1 (HR 1.69, 95% CI 1.18–2.43, p = 0.004) and T0-T2 (HR 1.81, 95% CI 1.22–2.70, p = 0.003), and a higher risk of death with a 10-unit increase of change in VAF between T0-T1 (HR 1.38, 95% CI 1.01–1.88, p = 0.043) and T0-T2 (HR 1.56, 95% CI 1.21–2.16, p = 0.008) was observed.

Characterization of TF and genomic profiles by shallow whole-genome sequencing (sWGS)

sWGS revealed altered genomic profiles with a positive tumor fraction (TF ≥ 3%) in most of the patients both at baseline (T0) and disease progression (T3): among the samples collected, 27 out of 32 (84.4%) and 19 out of 21 samples (90.5%) exhibited a positive TF at T0 and T3 respectively. At interim timepoints T1 and T2, altered genomic profiles with positive TF were observed for 13 out of 32 samples (40.6%) and in 7 out of 31 samples (22.6%) at T1 and T2. Detailed TF values for each timepoint are provided in Supplementary Table.

The median TF at T0 was 29.8% (Q1-Q3: 12.5–48%); the median TF dropped to 0.0% (Q1-Q3: 0.0-11.6%) at T1 and at T2 (Q1-Q3: 0.0–0.0%), and rose again to 33.6 (Q1-Q3:14.5–51.6%) (Fig. 3c).

The only cases with negative TF at T3 were CATS 16 and CATS 31.

Ranging from the first (12.46%) to the third (48.02%) quartile, higher baseline TF was a predictor of higher risk of disease progression (HR 1.97, 95% CI 1.02–3.82, p = 0.044) (Fig. 2).

Among the 27 patients with positive TF at T0, two distinct patterns emerged: 20 patients (20 out of 27, 74%), hereunder referred to as first group, showed normalization of genomic profiles accompanied by complete clearance at T2 (100% TF reduction). In contrast, seven patients (7 out of 27, 26%), hereunder referred to as second group, retained altered genomic profiles and showed incomplete TF clearance, with statistically significant TF reductions from 3.3 to 81.2% at T2 or, in one case (CATS 9), even a TF increase (Fig. 3; Supplementary Fig. 3).

In the first group, the median TF at T0 was 32.39% (Q1-Q3: 20.2–47.8%); at T1 it set down significantly to 0.00% (Q1-Q3: 0.00–1.3%), as well as at T2 (median 0.00); while at progression (T3) the median TF rose again to 25.41% (Q1-Q3: 14.5–51.3%). However, in this group TF levels increased again at T3 for all but two patients (CATS 16 and CATS 31).

In the second group, the median TF at T0 was 48.97% (Q1-Q3: 29.4–56.9%); at T1 it decreased slightly in six patients (CATS 2, CATS 8, CATS 11, CATS 23, CATS 27, CATS 32), with reductions ranging from 6.42 to 68.37%, while it increased in one patient (CATS 9) by 29.76%. The median TF at T1 was 28.11% (Q1-Q3: 19-44.2%); at T2 it decreased in five patients (CATS 2, CATS 9, CATS 11, CATS 23, CATS32) by 3.21–43.05% but increased in the remaining two patients by 0.29–20%, with a median TF of 23.66% (Q1-Q3: 17.9–38.3%). By T3, all patients of this group exhibited rising TF, with a median of 52.96% (Q1-Q3: 41.8–63.2%).

The 20 patients in the first group, showing complete TF clearance at T2, had notably longer PFS (p = 0.0009) compared to the 7 patients in the second group with incomplete TF clearance.

Finally, a third group is composed of five patients (CATS 12, CATS 19, CATS 30, CATS 31, CATS 33), who displayed a TF of 0.00% and normal genomic profiles at T0. Apart from CATS 19, who displayed a TF of 5.36% at T1, all these patients consistently showed negative TF values at both T1 and T2. In this small group of patients, T3 samples were available only for CATS 30 and CATS 31, with TF values of 12.97% and 0.00%, respectively.

A higher risk of disease progression in presence of a dynamic 10-unit increase in change of TF between T0-T1 (HR 1.49, 95% CI 1.1–2.01, p = 0.01) and T0-T2 (HR 1.86, 95% CI 1.25–2.75, p = 0.002) was observed, as well as a higher risk of death in presence a 10-unit increase of TF between T0-T2 (HR 1.56, 95% CI 1.09–2.21, p = 0.014), was estimated.

Lastly, for exploratory purposes, we implemented a random forest algorithm that included all the covariates. This analysis identified baseline cfDNA, VAF and the number of metastases as important predictors of both OS and PFS (Fig. 5).

Fig. 5.

Random forest of covariates impact on OS (A) and PFS (B) and their relevance

Discussion

Small cell lung cancer is biologically characterized by high tumor heterogeneity and plasticity, thus explaining the clinical aggressiveness and dismal outcome from available systemic treatments [27]. Even though a new standard of care has been introduced as first line treatment, long-term benefit from chemoimmunotherapy is reserved to a small patients’ subgroup, which is tough to identify from a clinical perspective [28]. The modest progression free and overall survival shown by our real-world cohort of patients confirmed these data and are in line with available literature from clinical trials.

Recently, transcriptomic profiling on tumor samples of SCLC has defined 4 different subtypes, characterized by positivity to ASCL1 (A), NEUROD1 (N), POU2F3 (P) transcription factor signatures or by the lack or low expression of these three signatures and associated with an inflammatory gene expression signature (SCLC-I) [18]. Although available data show a differential benefit from chemoimmunotherapy according to the different subgroups [28, 29]this classification is not yet currently used in clinical practice.

In this clinical setting, the search for tissue predictive biomarkers is difficult to pursue because of the paucity of tumor tissue and the dynamic evolution of the disease [10]whereas liquid biopsy approaches are valuable also considering that SCLC is an highly tumor DNA shedding cancer, as demonstrated by the higher concentration of ctDNA compared with plasma samples of Non-Small Cell Lung Cancer and other tumor types [22, 30, 31]. Tumor DNA shedding is determined by cancer type and other biological and clinical features, such as the number and location of the metastatic site [32, 33].

In line with previous findings, we detected a median cfDNA concentration of 24.9 ng/mL of plasma (range 3.7-704.3 ng/mL) and a median TF of 29.8% (Q1-Q3: 12.5–48%) in our case series, whereas in NSCLC only 20% of cases have TF exceeding 10%³⁰. These data reinforce the potential usefulness of liquid biopsy to detect and monitor cfDNA during systemic SCLC treatment. The prognostic role of cfDNA in cancer has been reported in many studies [34]and our data confirm this in SCLC unveiling an association of increasing cfDNA concentration with the risk of death ([p = 0.0006], with an estimated higher risk of death in presence of a dynamic 10-unit increase of the change in cfDNA between T0-T2 [p = 0.009]), probably because of a direct correlation with the number of metastases. Additionally, our data showed an association of cfDNA levels with the risk of disease progression to first line chemoimmunotherapy.

The attitude of SCLC to release ctDNA enabled the successful measurement of the TF by sWGS, a well-established method in the field to quantify the proportion of cancer derived cfDNA in plasma samples [35].

In our study, higher baseline TF and a 10-unit increase in TF levels between T0 and T1 or T0 and T2 were associated with an elevated risk of disease progression (p = 0.044; p = 0.01; p = 0.002), while a 10-unit TF increase between T0 and T2 correlated with a higher risk of death (p = 0.014). Notably, TF clearance at T2 in our cohort was linked to improved PFS (p = 0.0009). These findings are consistent with previous research on different cancer types, suggesting that baseline TF and its dynamic changes can serve as prognostic and predictive biomarkers [36, 37].

Notably, a small subgroup of 5 patients showed absence of TF in the baseline plasma samples, along with a lower cfDNA concentration compared to the rest of the cohort (median 9.3 ng/mL of plasma versus 28.95 ng/mL of plasma). In these patients, however, somatic point mutations were detected by the Avenio NGS test, with a VAF ranging from 0.21 to 2.6%. These VAF values fall below the limit of detection of the algorithms used for TF calculation, which requires a TF of at least 3%^24,25, confirming the low level of ctDNA in these plasma samples. Considering the undetectable TF, the low plasmatic cfDNA yield and the low VAF of the driver mutations observed in this cohort, we considered these patients to be low ctDNA shedders.

In this respect, a previous large study by Husain et al. reported a median TF of 17.7% in liquid biopsies of SCLC patients. Notably, out of 115 patients, TF was undetectable for 12% of the samples analyzed. These results align with our findings, where TF was undetectable in 5 out of 32 samples (15.6%). Low or undetectable TF has been attributed to insufficient ctDNA shedding or limitations in assay sensitivity [30, 38].

Notably, all these 5 SCLC patients had less than 3 metastatic sites, bone metastases were present in 2 (40%) patients, brain metastasis in 1 (20%) patient and no one had liver metastasis at baseline. It is possible to hypothesize that TF-negative SCLC may exhibit a distinct biological tumor behavior, characterized by a lower propensity for metastasis. However, the lack of available tissue prevented further investigation of this hypothesis.

In addition to the prognostic value of static values of these biomarkers, the clinical value of dynamic monitoring of cfDNA is even more attractive from a clinical perspective, because the change in concentration over time may be useful especially in the first cycles of treatment, because it may help physicians to anticipate the radiological detection of disease progression. The analysis of the relationship between changes in cfDNA levels across different time points and survival outcomes, conducted using restricted cubic splines, revealed an approximately linear association with the log hazard ratio. This finding supported our decision to treat cfDNA clearance as a continuous variable in survival analyses, rather than defining an arbitrary cut-off for patient stratification.

As in other settings [39]our data showed a significant reduction in cfDNA levels between T0 and T1 (median 24.9 ng/mL vs. 19.1 ng/mL) and between T0 and T2 (median 24.9 ng/mL vs. 14.2 ng/mL), while at T3 cfDNA levels showed a significant increase compared to T2 (median 14.2 ng/mL vs. 40.5 ng/mL). These data acquire even more clinical relevance when we consider the rapid disease progression and clinical worsening of SCLC patients; in our series, 11 patients (34%) were not eligible to the blood collection at the T3 time point, because of rapid PS deterioration and death.

In our case series, we identify two patients without TF detected in the plasma samples collected at the time of disease progression. One patient was classified as a low shedder patient, with undetectable TF at all timepoints; while the other showed initially a positive TF at T0 (12.8%), that dropped to 0 at subsequent timepoints.

Both patients had less than 3 metastatic sites at diagnosis with respectively lymph nodes and pleural involvement, without bone, brain or liver metastasis: at T3 they both experienced systemic disease progression with brain metastasis in one case and lymph nodes disease in the other case.

Our study reports frequent TP53 and RB1 gene mutations at the baseline, thus allowing their monitoring along treatment cycles; significant association of both baseline values and dynamic deltas were shown with OS and PFS. Previous studies from our group [40, 41] and others [42, 43] underlined the clinical applicability of VAF detection; the identification of a small number of specific mutations and their adoption as sentinel of clinical cancer course could hopefully make this tool accessible and cost-saving compared with repeated wide-genome sequencing [44]. Interestingly, PTEN mutations have been detected in three patients, and only in one patient with a VAF of 20–50%. Interestingly, this patient developed brain metastases at the time of disease progression, in line with findings of a previous study [45]suggesting that specific pathway alterations might correlate with progression patterns.

Finally, an additional insight offered by the dynamic monitoring of gene mutations in liquid biopsies of SCLC patients is the identification of acquired resistance mutations at the time of disease progression which could be hypothesis-generating for further studies on resistance mechanisms and signature after immune checkpoint inhibitors. Acquired mutation of the APC gene has been observed at high VAF in one patient, who experienced a clinical course of aggressiveness and early death. In general, however, the genomic profile of relapsed SCLC was very similar to that found before treatment, suggesting non-genetic mechanisms of acquired resistance, although the limited number of genes interrogated by our NGS panel does not allow us to draw firm conclusions.

The main limitation of the present study is the small sample size, which makes it difficult to deepen some biological issues, such as the correlation of circulating biomarkers with tissue genomic profiling and respective molecular subgroups. However, the study was exploratory in nature, and the prospective nature of the data and samples collection lays the groundwork for a validation multicenter prospective study in a real-world population receiving first line chemoimmunotherapy, as well as in randomized clinical trials where cfDNA, TF and tumor-related VAF may be considered as stratification biomarkers for intensification strategies.

Conclusions

Baseline and dynamic changes of cfDNA, TF and VAF may help physicians to stratify ES-SCLC patients receiving first-line ACE and to anticipate the clinical course of the disease. Monitoring these biomarkers throughout therapy could not only provide useful insights into treatment response but also enable an early assessment of disease progression. Specifically, analyzing their temporal variations may enhance the ability to predict the clinical course of the disease, facilitating more personalized therapeutic decisions and potentially improving treatment outcomes.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

A

Atezolizumab

A+

ASCL

AUC

Intravenous Administration of Carboplatin

CE

Carboplatin-Etoposide

cfDNA

cell-free DNA

CI

Confidence Interval

CNS

Central Nervous System

CTCAE

Common Toxicity Criteria for Adverse Events

ctDNA

circulating tumor DNA

ES

Extensive Stage

GEP

Gene Expression Profiling

HR

Hazard Ratios

I

Inflamed

IQRs

Interquartile Ranges

ITT

Intention-To-Treat

M

Median

MTB

Tumor Mutational Burden

N

Neuroendocrine

NGS

Next Generation Sequencing

OS

Overall Survival

PD-L1

Programmed Death-Ligand 1

PFS

Progression Free Survival

PH

Proportional Hazards

RECIST

Response Evaluation Criteria in Solid Tumors

SCLC

Small Cell Lung Cancer

TF

Tumor Fraction

TME

Tumor Microenvironment

VAF

Variant Allele Frequency

VIMP

Variable Permutation Importance Measure

VUS

Variant of Uncertain/Unknown Significance

sWGS

shallow Whole-Genome Sequencing

Author contributions

GP³, EL performed experiments, analyzed and interpreted data, wrote, and reviewed the manuscript; SP and CR conducted the bioinformatics analysis of genomic data; DS and MVR performed patients’ data collection, interpreted data, wrote, and reviewed the manuscript; SL performed the statistical analysis, wrote, and reviewed the manuscript; AR, SF, AF, ADM, LB collaborated in patients’ enrollment, VG supervised and reviewed the manuscript; SI supervised the study, interpreted data and reviewed the manuscript; GP^1,2 conceptualized the study, interpreted data, wrote, and reviewed the manuscript and supervised the study.All authors read and approved the final manuscript.

Funding

Open access funding provided by Università degli Studi di Padova. This work was unconditionally supported by Roche (ML43257) and DOR by the Department of Surgery, Oncology and Gastroenterology (DiSCOG), University of Padova; Funding Ricerca Corrente IOV IRCCS 2025/ Institutional funding of the Veneto Institute of Oncology IOV-IRCCS intramural research grant 5 × 1000 2019 (TESLA, to E.L.)

Data availability

The datasets generated and during the current study are available in the Sequence Read Archive (SRA) repository under accession number PRJNA1260986.

Declarations

Ethics approval and consent to participate

This study has been approved by Istituto Oncologico Veneto Ethical Committee (Cod. Int CESC IOV: 2021-85). All patients signed an informed consent form before enrollment.

Consent for publication

All patient information has been anonymised and consent for publication of the current study was not required. All co-authors agreed for publication.

Competing interests

The authors declare no competing interests.