Multiplexed quantitative proteomics in small-cell lung cancer: stage-linked DLL3 distribution and an independent prognostic signal of thymidine phosphorylase

Introduction

Small-cell lung cancer (SCLC) is the most aggressive subtype of lung neoplasm, accounting for approximately 13–15% of all lung cancers (1). It is characterized by rapid clinical progression, early metastatic spread, and a strong but transient initial chemo-response that gives way to broad cross-resistance and short-lived remissions (1,2). Disease is staged using the Veterans Administration Lung Study Group system as limited-stage (confined to one hemithorax and an irradiable mediastinal field) or extensive-stage (distant spread or malignant pleural/pericardial effusion) (3,4); approximately two-thirds of newly diagnosed patients present with extensive-stage disease (2,4,5) and median overall survival (OS) remains less than one year despite contemporary first-line therapy (2,5,6).

Platinum/etoposide doublet chemotherapy has been the first-line standard for several decades; the addition of an anti-PD-L1 antibody (atezolizumab or durvalumab) extends OS modestly in extensive-stage disease (5,7-9), but responses remain short-lived (7). For relapsed disease, topotecan was for many years the only approved second-line agent (3,7); other later-line options include irinotecan, taxanes, temozolomide (TMZ), capecitabine or 5-fluorouracil (5-FU)-based regimens, and gemcitabine (7,10). Two of these later-line agents have well-characterized tumor-intrinsic predictive biomarkers that are particularly relevant to SCLC management: O⁶-methylguanine-DNA methyltransferase (MGMT) directly repairs the alkylation lesions induced by TMZ and is a recognized predictor of TMZ responsiveness (11-13); thymidine phosphorylase (TYMP) catalyses the terminal activation of capecitabine to 5-FU and is a predictive biomarker for capecitabine efficacy (14,15). The capecitabine-temozolomide (CAPTEM) combination is a category 2A National Comprehensive Cancer Network (NCCN)-recommended regimen in advanced pulmonary neuroendocrine neoplasms (14). For these reasons, MGMT and TYMP were prioritized in our panel as candidate predictors of TMZ and capecitabine/5-FU sensitivity in SCLC.

Beyond cytotoxic predictors, a second class of biologically informative proteins reflects the cell-of-origin biology of SCLC. Four major SCLC molecular subtypes (SCLC-A, -N, -P, -Y) have been established, with the ASCL1-driven SCLC-A neuroendocrine-high subtype representing approximately 70% of cases, while the smaller non-neuroendocrine fraction has comparatively indolent biology and may be more relevant in the immune-checkpoint era (16-18). Delta-like protein 3 (DLL3), a Notch ligand aberrantly expressed on the SCLC cell surface, is a hallmark of the SCLC-A subtype (19) and the target of DLL3-directed agents, including the recently approved bispecific T-cell engager tarlatamab (20). Additional therapy-relevant proteins included in this study reflect neuroendocrine differentiation and predictive sensitivity to topoisomerase-I (Topo-I) inhibitors, taxanes, and DNA-damaging agents: CD56 (4), EZH2 and SLFN11 (5,21-24), TOPO1, and class III β-tubulin (TUBB3).

These markers are typically measured one at a time and by heterogeneous methods [immunohistochemistry (IHC), promoter methylation, or transcript-based assays], and are seldom integrated into a single quantitative readout. Selected reaction monitoring-mass spectrometry (SRM-MS) enables reproducible absolute quantification of multiple predefined proteins from formalin-fixed paraffin-embedded (FFPE) tumor tissue in molar units (amol/µg) that are independent of antibody lot, scoring convention, or laboratory (25-29).

In the present study, we assessed the feasibility of multiplexed SRM-MS quantification of these nine therapy-relevant proteins in archival FFPE samples from a real-world SCLC cohort treated with platinum/etoposide, and explored associations of the resulting proteomic profile with clinicopathologic variables, platinum-sensitivity status, and OS. We present this article in accordance with the STROBE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0513/rc).

Methods

Study design and population

We conducted a retrospective two-center cohort study of patients with histologically confirmed SCLC treated with platinum/etoposide-based first-line chemotherapy between September 2005 and September 2015 at two affiliated teaching hospitals of The Catholic University of Korea—Seoul St. Mary’s Hospital (n=77) and Yeouido St. Mary’s Hospital (n=23). Eligible patients had histopathologically confirmed SCLC, sufficient archival FFPE tumor tissue for laser microdissection and SRM-MS, and complete clinical follow-up. Clinical and pathological variables were abstracted from electronic medical records, including biological sex, age at diagnosis, smoking history (pack-years), tumor stage (limited vs. extensive, per the Veterans Administration Lung Study Group classification), Eastern Cooperative Oncology Group (ECOG) performance status, first-line chemotherapy regimen, best response per RECIST v1.1 [complete response (CR)/partial response (PR)/stable disease (SD)/progressive disease (PD)], date of progression, date of last follow-up, and date of death where applicable. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of The Catholic University of Korea (approval number KC15SISE0263); informed consent was obtained from all patients.

Treatment classification and platinum-sensitivity status

All patients received etoposide-based first-line chemotherapy, either etoposide combined with cisplatin (EP) or with carboplatin (EC). For analyses focused on later-line treatment, patients were grouped into mutually exclusive categories: (I) first-line only (no documented later-line cytotoxic therapy), and (II) those who received a Topo-I inhibitor (topotecan, irinotecan, or belotecan) at any later line. Patients in this cohort were treated before the regulatory approval of immune checkpoint inhibitors and tarlatamab in extensive-stage SCLC; no patient received a checkpoint inhibitor, DLL3-directed agent, TMZ-based regimen, or capecitabine-based regimen during follow-up.

For patients with a documented date of first-line completion and a documented date of radiologic progression, the relapse-free interval (RFI) was defined as the number of days from the last day of first-line chemotherapy to the date of progression, in accordance with prior multicentre conventions (30,31). Negative RFI values reflect radiologic progression documented before protocol-defined completion of first-line therapy (e.g., progression on a scheduled mid-course restaging scan that triggered early discontinuation); such cases were retained with their actual (negative) RFI values and classified as platinum-refractory by convention. Patients were classified as platinum-sensitive [RFI ≥90 days, consistent with the Genestreti multicentre definition (31)], platinum-resistant (60≤ RFI <90 days), or platinum-refractory [RFI <60 days, per Ardizzoni et al. (30)]. Patients without computable RFI (because of absent date of progression, ongoing treatment at last follow-up, or treatment in the adjuvant setting after surgery) constituted an “unknown” stratum and were excluded from platinum-sensitivity analyses. OS was calculated from the date of diagnostic biopsy to death from any cause, with patients alive at last follow-up censored at that date.

Sample preparation, laser microdissection and protein quantitation

Archival FFPE tumor tissue was profiled by mass-spectrometry-based selected reaction monitoring (SRM), as previously described (32). For each tumor, one section was cut and stained with hematoxylin and eosin (H&E), and additional serial sections were placed onto Director^TM microdissection slides (OncoPlex Diagnostics, Rockville, MD, USA) and stained with eosin. A board-certified pathologist marked tumor regions on the H&E-stained slide; corresponding regions of the eosin-stained sections were laser-microdissected using a modified MMI dissection instrument (Molecular Machines & Industries, Zurich, Switzerland). Microdissected tumor tissue was solubilized using Liquid Tissue^® (OncoPlex Diagnostics) reagent according to the manufacturer’s instructions (32), and total protein concentration was measured using a micro bicinchoninic acid assay (Thermo Fisher Scientific Inc., Waltham, MA, USA).

A mixture of stable-isotope-labeled synthetic peptide standards corresponding to the proteotypic peptides of the nine target proteins (DLL3, ASCL1, EZH2, SLFN11, TYMP, MGMT, CD56, TUBB3 and TOPO1) was added to each solubilized tumor sample as internal standards. Samples were analyzed in technical triplicate on a TSQ Quantiva^TM triple-quadrupole mass spectrometer interfaced with a nanoACQUITY liquid chromatography system (Thermo Fisher Scientific; Waters Corp., Massachusetts, USA); peak-area ratios of endogenous to heavy-labeled peptides were calculated using PinPoint 1.3, and absolute protein quantities were reported in attomoles per microgram of total protein (amol/µg) and compared with pre-defined clinical or assay-based thresholds derived from previously published clinical SRM studies (32,33). SRM-MS assays were performed at OncoPlex Diagnostics and all clinical-pathological correlations and statistical analyses reported here were performed independently by the authors. Assay performance characteristics were established as previously described (32). Values measured at or above the lower limit of quantification (LLOQ) were reported as quantitative absolute molar quantities; values between the lower limit of detection (LLOD) and the LLOQ were reported as detected but below the LLOQ; and values below the LLOD were reported as 0 amol/µg (“non-detectable”). The inter-assay coefficient of variation across technical triplicates was below 20% for all nine targets across the clinically relevant dynamic range. For the present analyses, “detectable” denotes any value >0 amol/µg.

Biomarker thresholds and dichotomization

Three complementary thresholding strategies were pre-specified. (I) For markers with a published clinical decision threshold, predicted-responder status was defined using the established cut-off: MGMT <200 amol/µg (predicted TMZ-sensitive) and TYMP >1,335 amol/µg (predicted 5-FU-sensitive), as previously validated in clinical SRM studies. (II) For markers without a published clinical threshold, “detectable” was defined as expression above the assay LLOQ (>0 amol/µg); this dichotomization was applied to DLL3, CD56, SLFN11, ASCL1, EZH2, TUBB3 and TOPO1. (III) For exploratory continuous-marker analyses, the cohort median was used as a data-driven cut-off (e.g., TOPO1 above vs. below median). All three thresholding strategies were defined a priori and applied uniformly to the entire cohort. As a quantitative subtype proxy, tumors were classified into four mutually exclusive groups based on the joint detectability of DLL3 and ASCL1: DLL3+/ASCL1+ (“A-high”), ASCL1+/DLL3−, DLL3+/ASCL1−, and DLL3−/ASCL1−. TYMP was analyzed at two complementary thresholds with distinct biological interpretations: (I) detectable TYMP (>0 amol/µg, used in prognostic analyses) reflects any tumor-cell TYMP expression, which is biologically linked to both fluoropyrimidine activation and angiogenic/proliferative activity (34); (II) TYMP >1,335 amol/µg (used in candidate-responder analyses) corresponds to the published clinical threshold for predicting 5-FU/capecitabine sensitivity in prior SRM-MS studies (18). Sensitivity analyses with TYMP modelled as a continuous variable (per log10 unit) are reported in Results.

Statistical analysis

Continuous variables are summarized as mean ± standard deviation or median [interquartile range (IQR)]; categorical variables as counts and percentages. Protein expression values were analyzed both as continuous variables [after log₁₀(x+1) transformation] and as dichotomized categories. Spearman rank correlation quantified between-protein associations; Mann-Whitney U tests compared independent groups; Fisher’s exact or χ² tests assessed categorical associations; OS was estimated by Kaplan-Meier with 95% confidence bands; and univariable and multivariable Cox proportional hazards models estimated hazard ratios (HRs) with 95% confidence intervals (CIs). The pre-specified multivariable model included a priori clinical confounders (age ≥65 years, stage, ECOG) together with proteomic variables reaching P<0.05 in univariable analysis (detectable DLL3, ASCL1, TYMP, above-median TOPO1). Detectable DLL3 and ASCL1 were expected a priori to be strongly collinear with extensive-stage disease, both being markers of the SCLC-A neuroendocrine-high lineage; the multivariable model was therefore designed to assess whether any proteomic variable retained an independent prognostic effect after adjustment for stage and ECOG. Pre-specified subgroup analyses by platinum-sensitivity status combined the resistant (n=9) and refractory (n=31) strata into a single resistant/refractory group (n=40) for biomarker comparisons with the platinum-sensitive stratum (n=35), because of small individual stratum sizes.

Multiplicity and exploratory analysis policy. Given the hypothesis-generating nature of this retrospective biomarker discovery study and the breadth of the panel (nine proteins × multiple dichotomization strategies × several clinical endpoints), no formal adjustment for multiple comparisons [Bonferroni or false discovery rate (FDR) correction] was applied to the primary analyses. All findings—including biomarker-clinical associations, receiver operating characteristic (ROC) analyses, univariable Cox effects, platinum-sensitivity comparisons, and multivariable Cox results—are accordingly reported as exploratory and require prospective validation. To reduce dependence on any single threshold, DLL3, ASCL1 and TYMP were additionally modelled as continuous (per log₁₀ unit) Cox predictors as a sensitivity analysis. Findings with nominal P values in the 0.01–0.05 range are explicitly labelled as exploratory in the Results and Discussion sections. As a sensitivity analysis, Benjamini-Hochberg FDR (BH-FDR)-adjusted q-values for all Cox regression P values are reported alongside the unadjusted P values in Tables 1,2; q<0.05 was treated as a stricter threshold for FDR-adjusted significance.

All P values are two-sided. Sex was included as a covariate in exploratory Cox models and did not independently predict survival (HR 0.93, P=0.81). Analyses were performed in Python 3.11 (lifelines v0.30, SciPy v1.11) with cross-validation in IBM SPSS Statistics v21.

Results

Patient characteristics

A total of 100 patients with histologically confirmed SCLC were included (Table 3). The median age at diagnosis was 68 years (range, 46–85 years), and 87% were male. Sixty-four patients (64%) had a documented smoking history. Limited-stage disease was present in 41 patients (41%) and extensive-stage disease in 59 (59%); 69 patients (69%) had ECOG performance status 0–1 and 31 (31%) had ECOG ≥2. All patients received etoposide-based first-line chemotherapy, most commonly EP. Of the 94 patients with available best-response data, 66 (70%) achieved a CR or PR (CR 5, PR 61) and 28 (30%) were classified as non-responders (SD 15, PD 13). At the time of analysis, 93 patients (93%) had died; the median OS for the entire cohort was 11.5 months (95% CI: 10.2–13.0), with 1-year and 2-year OS probabilities of 48% and 23%, respectively (Table 3).

Quantitative distribution of nine biomarkers in archival SCLC tissue

All 100 tumors yielded interpretable proteomic measurements for the nine target proteins. The distribution of quantifiable expression among detectable samples is shown in Figure 1, and the prevalence of each biomarker at its pre-defined threshold is summarized in Table 4. MGMT was low (<200 amol/µg) in 49 of 100 tumors (49%), supporting potential sensitivity to TMZ; TYMP was high (>1,335 amol/µg) in 24 of 100 tumors (24%), the threshold associated with 5-FU/capecitabine responsiveness. DLL3 was detectable in 57 of 100 tumors (57%), with substantial inter-tumor heterogeneity (median 338 amol/µg, IQR, 200–478 amol/µg among detectable samples; range, 97–1,200 amol/µg). CD56 was the most broadly expressed neuroendocrine marker (detectable in 83% of tumors; median 1,543 amol/µg). SLFN11 was detectable in 74%, ASCL1 in 84%, EZH2 in 84%, TUBB3 in 83% and TOPO1 in 87% of tumors (Table 4).

Figure 1 Distribution of biomarker expression in detectable samples (N=100 SCLC tumors). Box plots show log₁₀-transformed expression for samples with detectable signal (>0 amol/µg); the number of detectable samples (n₊) per biomarker is shown beneath the x-axis. Boxes span the interquartile range, the horizontal line marks the median, whiskers extend to 1.5× IQR, and individual samples are overlaid as jittered points. ASCL1, achaete-scute homologue 1; CD56, neural cell adhesion molecule 1 (NCAM1); DLL3, delta-like ligand 3; ECOG, Eastern Cooperative Oncology Group; EZH2, enhancer of zeste homologue 2; IQR, interquartile range; MGMT, O6-methylguanine-DNA methyltransferase; SCLC, small-cell lung cancer; SLFN11, schlafen-11; TOPO1, topoisomerase I; TUBB3, class III β-tubulin; TYMP, thymidine phosphorylase.

The per-patient pattern of biomarker expression was markedly heterogeneous (Figure 2): every patient had at least five detectable biomarkers, and 67% of patients expressed seven or more of the nine proteins. A subset of patients clustered with high DLL3/ASCL1/EZH2 z-scores; a separate group showed higher TYMP/TOPO1, supporting the technical feasibility of a single multiplexed assay to interrogate the full panel (Figure 2).

Figure 2 Patient-level heatmap of biomarker expression in SCLC tumors. Log-transformed, z-scored expression of the nine biomarkers across all 100 SCLC samples. Patients (columns) are ordered by ascending overall survival (left to right). The five annotation tracks above the expression matrix encode, from top to bottom: sex (light blue, male; pink, female), age group (dark green, ≥65 years; light green, <65 years), ECOG performance status (white-to-purple gradient, 0–4), tumor stage (teal, limited; orange, extensive), and death event (black, event; white, censored). ASCL1, achaete-scute homologue 1; CD56, neural cell adhesion molecule 1 (NCAM1); DLL3, delta-like ligand 3; ECOG, Eastern Cooperative Oncology Group; EZH2, enhancer of zeste homologue 2; MGMT, O6-methylguanine-DNA methyltransferase; SCLC, small-cell lung cancer; SLFN11, schlafen-11; TOPO1, topoisomerase I; TUBB3, class III β-tubulin; TYMP, thymidine phosphorylase.

Stage-linked enrichment of DLL3 and a four-class DLL3/ASCL1 model

Quantitative DLL3 expression discriminated extensive- from limited-stage disease with high accuracy [area under the receiver operating characteristic curve (AUC) 0.78]. DLL3 was detectable in 45 of 59 (76.3%) extensive-stage tumors compared with 12 of 41 (29.3%) limited-stage tumors (Fisher’s exact test P<0.001; Table 5, Figure 3). Median DLL3 expression in extensive-stage tumors was 284 amol/µg compared with non-detectable in limited-stage tumors (Mann-Whitney U test P<0.001; Figure 3B). ASCL1 detectability showed a similar but weaker stage-enrichment (AUC 0.67; 91.5% in extensive-stage vs. 73.2% in limited-stage tumors; P=0.02; Mann-Whitney P=0.005). Among the remaining markers, only above-median TOPO1 was significantly more frequent in extensive-stage disease (Mann-Whitney P=0.01, AUC 0.65), whereas MGMT, TYMP, CD56, EZH2, SLFN11 and TUBB3 showed no significant stage-association (Table 5). High TYMP (>1,335 amol/µg) was associated with older age (≥65 years; P=0.03), suggesting that the 5-FU-responsive subset may be enriched among elderly patients in whom oral capecitabine is often a more practicable second-line option.

Figure 3 Quantitative DLL3 expression discriminates extensive- from limited-stage SCLC. (A) ROC curves for individual biomarkers as quantitative predictors of extensive-stage disease. DLL3 has the largest area under the ROC curve (AUC 0.78), substantially exceeding ASCL1, TOPO1, SLFN11 and EZH2. (B) Continuous DLL3 expression by tumor stage. The distribution is markedly shifted toward higher expression in extensive-stage disease (Mann-Whitney P<0.001). ASCL1, achaete-scute homologue 1; AUC, area under the receiver operating characteristic curve; DLL3, delta-like ligand 3; EZH2, enhancer of zeste homologue 2; ROC, receiver operating characteristic; SLFN11, schlafen-11; TOPO1, topoisomerase I.

To convert these dichotomized observations into a continuous discriminator, we computed ROC curves for each biomarker as a quantitative predictor of extensive-stage disease (Figure 3A). DLL3 was the strongest single discriminator (AUC 0.78), followed by ASCL1 (AUC 0.67), TOPO1 (AUC 0.65) and SLFN11 (AUC 0.59); EZH2, CD56 and the cytotoxic predictors MGMT and TYMP did not discriminate stage. We then defined a four-class DLL3/ASCL1 model (hereafter, the four-class model; Figure 4A): half of the cohort (50/100) was DLL3+/ASCL1+ (the “A-high” class), 34/100 (34%) was ASCL1+/DLL3−, 7/100 (7%) was DLL3+/ASCL1−, and 9/100 (9%) was DLL3−/ASCL1−. The A-high class was strongly enriched in extensive-stage disease (40/50, 80%), whereas the DLL3−/ASCL1− class was almost exclusively limited-stage (8/9, 89%; χ² P<0.001).

Figure 4 The four-class DLL3/ASCL1 quantitative subtype proxy and its association with overall survival. (A) Distribution of patients across the four DLL3/ASCL1 quantitative subtype proxy classes, stacked by tumor stage. The DLL3+/ASCL1+ “A-high” class is strongly enriched in extensive-stage disease, whereas the DLL3−/ASCL1− class is almost exclusively limited-stage (χ² P<0.001). (B) Kaplan-Meier overall survival estimates for the four DLL3/ASCL1 quantitative subtype classes. Four-group log-rank P=0.004; the DLL3−/ASCL1− class shows markedly longer survival than the A-high class (median 25.8 vs. 10.7 months). ASCL1, achaete-scute homologue 1; DLL3, delta-like ligand 3.

Stage-stratified analysis of the four-class model

To assess whether the apparent prognostic separation of the four-class model (Figure 4B) reflected an independent prognostic effect or stage confounding, we performed stage-stratified analyses (Figure S1 and Appendix 1). The DLL3−/ASCL1− subgroup (n=9), which carried the longest median OS overall (25.8 months), was exclusively limited-stage, and DLL3+/ASCL1+ tumors were strongly enriched in extensive-stage disease (40/50). Within limited-stage disease (n=41) the four-class model continued to stratify OS (4-group log-rank P=0.008), with median OS 12.5 months (95% CI: 4.3–14.5) in A-high (n=10), 21.8 months (95% CI: 7.8–40.2) in ASCL1+ only (n=20), 41.7 months (95% CI: 41.7–not reached) in DLL3+ only (n=2; small subgroup), and 25.8 months (95% CI: 6.5–not reached) in DLL3−/ASCL1− (n=9). Within extensive-stage disease (n=59) the four-class model did not stratify OS (4-group log-rank P=0.85). Cox models incorporating stage × DLL3 and stage × ASCL1 interaction terms did not detect significant interactions (P=0.30 and P=0.09–0.39), but the formal interaction test is underpowered because the DLL3−/ASCL1− subgroup contributes only nine patients (all limited-stage). The prognostic information conveyed by the four-class model is therefore concentrated within limited-stage disease and is largely driven by the stage-restricted distribution of the DLL3−/ASCL1− subgroup.

Non-overlapping predicted sensitivity to TMZ and 5-FU

Patients were classified into four mutually exclusive predicted-chemosensitivity groups based on the joint MGMT/TYMP threshold pattern: predicted TMZ-sensitive only (MGMT <200, TYMP ≤1,335; n=37, 37%), predicted 5-FU-sensitive only (MGMT ≥200, TYMP >1,335; n=12, 12%), predicted sensitive to both agents (n=12, 12%), and predicted sensitive to neither (n=39, 39%). The complete 2×2 cross-classification of MGMT-low and TYMP-high status is shown in Table S1. The proportion of patients meeting at least one cytotoxic predictor was 61%, of whom only 12 (20%) met both, indicating that the MGMT-low and TYMP-high candidate populations are largely non-overlapping. OS did not differ significantly between these four predicted-chemosensitivity strata (4-group log-rank P=0.43; data not shown), consistent with these markers being predictive of agent-specific response rather than prognostic of outcome under platinum/etoposide.

Univariable and multivariable survival analyses

Kaplan-Meier estimates of OS by clinical and proteomic groupings are shown in Figure 5. Extensive-stage disease was strongly associated with shorter survival (median OS 9.5 vs. 18.7 months for limited-stage disease; log-rank P<0.001; Figure 5A), as was ECOG ≥2 (Figure 5B). Among the proteomic markers, detectable DLL3 (median OS 10.8 vs. 18.4 months; P=0.002; Figure 5C), detectable ASCL1 (median OS 11.0 vs. 18.6 months; P=0.04; Figure 5D), detectable TYMP (median OS 10.8 vs. 21.7 months; P=0.01; Figure 5E) and above-median TOPO1 (median OS 10.9 vs. 12.2 months; P=0.01; Figure 5F) were all associated with shorter OS, whereas MGMT-low and TYMP >1,335 amol/µg were not individually prognostic.

Figure 5 Kaplan-Meier estimates of overall survival by clinical and proteomic groupings in SCLC, with 95% confidence bands. (A) Overall survival by tumor stage (limited vs. extensive). (B) Overall survival by ECOG performance status (0–1 vs. ≥2). (C) Overall survival by DLL3 detectability (detectable >0 vs. not detected =0 amol/µg). (D) Overall survival by ASCL1 detectability. (E) Overall survival by TYMP detectability. (F) Overall survival by TOPO1 expression (above vs. below the cohort median). Two-sided log-rank P values are annotated in each panel. ASCL1, achaete-scute homologue 1; DLL3, delta-like ligand 3; ECOG, Eastern Cooperative Oncology Group; SCLC, small-cell lung cancer; TOPO1, topoisomerase I; TYMP, thymidine phosphorylase.

Five univariable predictors of shorter OS were identified (Figure 6A, Table 1): extensive stage (HR 2.47, 95% CI: 1.60–3.83; P<0.001), ECOG ≥2 (HR 1.68, 95% CI: 1.07–2.62; P=0.02), detectable DLL3 (HR 1.96, 95% CI: 1.26–3.04; P=0.003), detectable ASCL1 (HR 1.89, 95% CI: 1.03–3.47; P=0.04), detectable TYMP (HR 1.92, 95% CI: 1.16–3.18; P=0.01) and above-median TOPO1 (HR 1.72, 95% CI: 1.12–2.64; P=0.01). When DLL3 was modelled as a continuous variable (per log unit increase), each unit of log₁₀-DLL3 was associated with HR 1.11 (95% CI: 1.03–1.20, P=0.005); similar continuous-effect estimates were obtained for ASCL1 (HR 1.11 per log unit, P=0.04) and TYMP (HR 1.12, P=0.002), supporting the categorical findings as a sensitivity analysis.

Figure 6 Forest plots of (A) univariable and (B) multivariable Cox regression for overall survival. Hazard ratios are plotted on a log scale; the vertical reference line at HR =1 indicates no effect. Filled red squares mark variables with two-sided P<0.05. After multivariable adjustment, extensive stage and detectable TYMP remain the only independent predictors of shorter OS. aHR, adjusted hazard ratio; ASCL1, achaete-scute homologue 1; CD56, neural cell adhesion molecule 1 (NCAM1); CR, complete response; DLL3, delta-like ligand 3; ECOG, Eastern Cooperative Oncology Group; EZH2, enhancer of zeste homologue 2; HR, hazard ratio; MGMT, O6-methylguanine-DNA methyltransferase; OS, overall survival; PR, partial response; SLFN11, schlafen-11; TOPO1, topoisomerase I; TYMP, thymidine phosphorylase.

Other clinical and proteomic variables were not significantly associated with OS in univariable analysis
(Table 1): age ≥65 years (HR 1.43, P=0.10), sex (HR 0.93, P=0.81), smoking status (HR 0.93, P=0.75), best response (CR/PR vs. SD/PD; HR 0.71, P=0.12), MGMT-low (HR 0.84, P=0.41), CD56 (HR 0.72, P=0.22), SLFN11 (HR 1.09, P=0.71) and EZH2 (HR 0.96, P=0.88).

In the pre-specified multivariable Cox model adjusted for age ≥65 years, stage, ECOG, detectable DLL3, ASCL1, TYMP and above-median TOPO1 (Figure 6B, Table 2), only extensive stage (adjusted HR 1.84, 95% CI: 1.02–3.31; P=0.04) and detectable TYMP (aHR 2.06, 95% CI: 1.20–3.54; P=0.009) remained independent predictors of shorter OS, whereas the univariable effects of DLL3 (aHR 1.68, P=0.07) and ASCL1 (aHR 1.19, P=0.60) were attenuated, consistent with their strong collinearity with extensive-stage disease (Table 5). Applying Benjamini-Hochberg FDR adjustment as a sensitivity analysis (Tables 1,2), detectable DLL3, detectable TYMP, above-median TOPO1 and extensive stage retained significance at q<0.05 in the univariable analysis; in the multivariable model, the strongest signal was detectable TYMP (q=0.06, borderline at q<0.05), supporting the principal findings as exploratory but not as definitive after correction for multiple testing. The independent effect of TYMP after adjustment for stage is a notable finding that has not, to our knowledge, been previously reported in SCLC.

To address the concern that TYMP’s independent prognostic effect might reflect aggressive metastatic-site biology rather than a tumor-intrinsic effect, we performed an exploratory extended multivariable Cox analysis incorporating metastatic-site variables from the medical record: liver metastasis (23/100), brain metastasis (14/100), and multi-site involvement (≥2 metastatic organs; 37/100). Detectable TYMP was modestly more frequent in patients with bone metastasis (89.7% vs. 70.4%; P=0.04) and multi-site involvement (89.2% vs. 68.3%; P=0.03), consistent with TYMP’s known role in tumor proliferation and angiogenesis (34). In the extended Cox model simultaneously adjusting for stage, age ≥65 years, ECOG ≥2, the four pre-specified proteomic variables, and the three metastatic-site indicators, detectable TYMP retained an independent association with shorter OS (adjusted HR 1.88, 95% CI: 1.08–3.26; P=0.03), modestly attenuated from aHR 2.06 but still significant. Detectable DLL3 also became independently significant in the extended model (aHR 2.04, P=0.02), whereas extensive-stage disease was substantially attenuated (aHR 1.01, P=0.98), reflecting overlap between staging and metastatic-site variables.

Quantitative biomarker discriminators of extensive-stage disease

To complement the categorical analysis above, we examined each biomarker as a continuous quantitative discriminator of extensive-stage disease using ROC analysis (Figure 3). DLL3 had the largest area under the ROC curve (AUC 0.78, 95% CI: 0.69–0.86), substantially exceeding ASCL1 (AUC 0.67), TOPO1 (AUC 0.65) and SLFN11 (AUC 0.59); EZH2 and CD56 were not discriminative (AUC ≈0.5). Median DLL3 expression in extensive-stage tumors was 284 amol/µg (IQR, 99–422 amol/µg) compared with non-detectable in limited-stage tumors (Mann-Whitney P<0.001; Figure 3B). The quantitative DLL3 distribution in extensive-stage tumors was right-skewed, with 27% of extensive-stage tumors expressing DLL3 above
400 amol/µg, defining a candidate “DLL3-high” subgroup that may be enriched for the SCLC-A subtype and that could be of interest for prospective evaluation as a quantitative companion-diagnostic threshold for tarlatamab.

Treatment patterns and platinum-sensitivity analysis

Among the 100 patients, 65 received only first-line etoposide-based chemotherapy without a documented later-line cytotoxic agent, while 35 received a Topo-I inhibitor (topotecan, irinotecan, or belotecan) at any later line (see the Methods section). Median OS from biopsy was 10.8 months in the first-line-only group and 13.6 months in patients who received a Topo-I inhibitor; the between-group difference was not statistically significant (log-rank P=0.81; Figure 7A), consistent with selection effects inherent to a retrospective cohort in which later-line therapy is preferentially offered to fitter patients with longer disease control.

Figure 7 Overall survival by later-line treatment received and by platinum-sensitivity status. (A) Kaplan-Meier estimates of OS from biopsy stratified by later-line treatment: first-line only and second-line topoisomerase-I inhibitor. (B) Kaplan-Meier estimates of OS from biopsy stratified by platinum-sensitivity status defined from the RFI: sensitive (RFI ≥90 days, Genestreti definition), resistant (60–89 days), and refractory (RFI <60 days, Ardizzoni cut-off). Two-sided log-rank P values are annotated. 1L, first-line; 2L, second line; OS, overall survival; RFI, relapse-free interval.

Of the 100 patients, 75 had a computable RFI (median 72 days; range, −131 to 1,136 days). Applying the cut-offs defined in the Methods section, 35 patients (47% of those with computable RFI) were classified as platinum-sensitive (RFI ≥90 days), 9 (12%) as resistant (RFI 60–89 days), and 31 (41%) as refractory (RFI <60 days). The remaining 25 patients (without computable RFI, e.g., those still on first-line therapy at last follow-up, those with documented adjuvant intent after surgical resection, or those who died before documentation of progression) constituted the “unknown” stratum and were excluded from platinum-sensitivity analyses. Platinum-sensitivity status was a strong prognostic factor: median OS was 16.8 months for platinum-sensitive patients, 12.0 months for the resistant subgroup and 7.6 months for refractory patients (3-way log-rank P<0.001; pairwise log-rank, sensitive vs. refractory P<0.001; Figure 7B).

Continuous biomarker expression did not differ significantly between platinum-sensitivity strata. When biomarker expression was compared as a continuous variable between sensitive (n=35) and combined resistant/refractory (n=40) tumors (Table 6, Figure 8A), no marker reached the conventional significance threshold on the Mann-Whitney U test, although TOPO1 showed a directional trend (median 1,133 vs. 1,458 amol/µg; P=0.10).

Figure 8 Biomarker expression by platinum-sensitivity status. (A) Continuous biomarker expression (log₁₀-scaled) in platinum-sensitive (n=35; green) vs. resistant/refractory (n=40; red) tumors for the six clinically most relevant biomarkers. Violin plots show the full expression distribution; horizontal lines mark medians. None of the continuous comparisons reached statistical significance (Mann-Whitney all P>0.10). (B) Categorical biomarker positivity rates between platinum-sensitivity strata. Above-median TOPO1 was significantly more frequent in resistant/refractory tumors (Fisher P=0.04, denoted by an asterisk). ASCL1, achaete-scute homologue 1; DLL3, delta-like ligand 3; MGMT, O6-methylguanine-DNA methyltransferase; Res/Refr, resistant/refractory; SLFN11, schlafen-11; TOPO1, topoisomerase I; TYMP, thymidine phosphorylase.

In categorical positivity analyses, only above-median TOPO1 was significantly more frequent in resistant/refractory than in sensitive tumors (28/40, 70% vs. 16/35, 46%; Fisher’s exact P=0.04; Figure 8B), and detectable DLL3 showed a directional trend (29/40, 72% vs. 19/35, 54%; P=0.15). MGMT-low, TYMP-high, CD56, SLFN11, ASCL1 and EZH2 status did not differ between strata. ROC analyses for predicting refractory status from individual continuous biomarker values yielded AUCs of 0.50–0.60 across all nine markers; a composite z-score sum of DLL3 + TOPO1 yielded AUC 0.60. Per the multiplicity policy stated in the Methods section, these P=0.04 and trend-level findings should be interpreted with caution and not as clinically actionable predictors of platinum sensitivity.

Discussion

In this retrospective two-center study of 100 patients with SCLC treated with platinum/etoposide between 2005 and 2015—before immune-checkpoint inhibitors and tarlatamab were available in routine practice—multiplexed targeted proteomics by SRM-MS reproducibly quantified nine therapy-relevant proteins on archival FFPE tumor tissue in a single integrated assay. The findings provide a real-world quantitative reference for these markers under uniform platinum/etoposide first-line therapy and frame hypotheses for prospective evaluation in cohorts treated with the newer modalities.

Across the two participating centers, SRM-MS reproducibly quantified all nine proteins in absolute molar units (amol/µg) on routinely archived FFPE samples up to approximately 8 years old, demonstrating analytical feasibility on real-world archival material.

Several biological and prognostic patterns emerged from the quantitative profiles. Quantitative DLL3 expression was tightly linked to tumor stage (AUC 0.78; detectable in 76% of extensive-stage vs. 29% of limited-stage tumors; P<0.001). A four-class DLL3/ASCL1 model stratified OS (4-group log-rank P=0.004), but stage-stratified analyses showed that the prognostic information is concentrated within limited-stage disease (within-limited P=0.008; within-extensive P=0.85) and is largely driven by the small DLL3−/ASCL1− subgroup (n=9, exclusively limited-stage, median OS 25.8 months, 95% CI: 6.5–not reached); the model therefore tracks SCLC-A neuroendocrine biology in a stage-dependent manner rather than functioning as a stage-independent prognostic classifier. Detectable TYMP was independently associated with shorter OS in multivariable Cox regression (adjusted HR 2.06, 95% CI: 1.20–3.54; P=0.009) after adjustment for age, stage, ECOG and other proteomic markers—an independent prognostic effect of TYMP not, to our knowledge, previously reported in SCLC.

Beyond the stage- and TYMP-linked observations, the panel also informed exploratory identification of candidate populations for later-line cytotoxic therapy. No single proteomic measurement quantitatively predicted platinum-refractory disease (AUCs 0.50–0.60), although categorical TOPO1 above the cohort median was more frequent in resistant/refractory than in sensitive tumors (70% vs. 46%; P=0.04, hypothesis-generating). MGMT-low (<200 amol/µg; 49%) and TYMP-high (>1,335 amol/µg; 24%) populations were largely non-overlapping (12% co-positive), defining complementary candidate subsets for hypothesis-generating evaluation in TMZ- or capecitabine-based later-line therapy.

The principal strengths of this study are the use of an absolute-quantitation, laboratory-portable platform (SRM-MS) on routinely archived FFPE tissue, the breadth of the multiplexed nine-protein panel, the uniform first-line treatment landscape (all patients received platinum/etoposide, with no immune-checkpoint inhibitor or DLL3-directed therapy administered) that simplifies survival interpretation, and the long-term follow-up with 93% events captured. Reporting absolute molar quantities (amol/µg total protein) sidesteps the antibody-clone, lot-to-lot, and scoring-convention heterogeneity that complicates cross-cohort comparison of IHC-based DLL3 measurements (20,35-39).

Several limitations warrant acknowledgement. First, the sample size of 100 limited statistical power for subgroup analyses; the TOPO1-refractoriness association (Fisher P=0.04) is exploratory and requires replication. Second, the cohort was treated between 2005 and 2015, before the regulatory approval of immune-checkpoint inhibitors and tarlatamab; no patient received either, and the dataset cannot address the predictive value of DLL3 quantitation for tarlatamab response. Third, paired DLL3 IHC validation was not performed in the present manuscript; the technical reconciliation of SRM-MS detectability and IHC positivity therefore relies on previously published IHC literature (19,39-41) and general assay-comparison considerations (27,28). Fourth, the cohort is exclusively Korean, predominantly male (87%), and predominantly smoker (64%), reflecting the demographic profile of SCLC in Korea (2,10); generalizability to other populations requires confirmation. Fifth, residual unmeasured selection bias inherent to retrospective archival cohorts cannot be excluded. Sixth, the first-line CR/PR rate of 70% is at the upper end of published SCLC response rates, likely reflecting referral and biopsy-availability selection at tertiary academic centers. Seventh, although the extended Cox analysis incorporating liver, brain and multi-site metastasis did not eliminate the independent prognostic effect of TYMP, additional confounders—serum LDH, NSE, thoracic radiotherapy, prophylactic cranial irradiation, and subsequent-line therapy—were not abstracted at the granularity required for multivariable adjustment; the residual TYMP signal cannot be unambiguously attributed to a tumor-intrinsic effect.

Our DLL3 detectability rate of 57% by SRM-MS is lower than the IHC positivity rates reported in landmark studies. Saunders et al. reported DLL3 protein expression in approximately 80% of SCLC tumors by IHC (19), and the recent multicenter DELFI study using a clinically validated DLL3 IHC assay reported any-positive staining in 85.0% of SCLC tumors and high expression (≥50%) in 67.6% (39). The recent Korean retrospective analysis by Nam et al. (40) examined DLL3 expression by IHC in extensive-stage SCLC and identified high DLL3 expression as an independent adverse prognostic factor for OS, broadly consistent with our quantitative findings.

The apparent gap between IHC-based positivity (≈80%) and SRM-MS detectability (57%) is central to interpreting SRM-MS DLL3 as a potential companion-diagnostic readout, and two non-mutually-exclusive technical considerations established in the literature should be considered. First, IHC and SRM-MS use detection thresholds that differ in kind. Clinical IHC commonly defines positivity as >1% of tumor cells with any membranous staining (a categorical cell-level call), whereas SRM-MS reports a continuous absolute molar quantity averaged over the bulk microdissected tumor lysate. A tumor with DLL3 expressed at low per-cell abundance across many cells, or at high per-cell abundance in a small minority of cells, may therefore be IHC-positive yet fall below the SRM-MS LLOQ—a discordance pattern documented for several membrane proteins (27,28). Second, IHC and SRM-MS sample different sub-cellular compartments: several DLL3 IHC clones, including SP347 (the Ventana clone used in DeLLphi-301), detect both cytoplasmic and membranous DLL3, whereas peptide-based SRM measures total extractable DLL3 from solubilized FFPE without compartment information (41). Because tarlatamab activity depends on cell-surface (membranous) DLL3 accessible to the engager, and the clinical relevance of cytoplasmic versus membranous DLL3 remains incompletely defined, these considerations motivate paired IHC validation in future SCLC cohorts to establish cross-platform threshold equivalence.

Regarding the SCLC-A subtype framework, our four-class model aligns with the IHC- and transcriptome-based classification proposed by Rudin et al. (16) and refined by Gay et al. (17) and Baine et al. (4), in which ASCL1-driven SCLC-A is the largest subtype (≈70% by transcriptome). The 50% A-high prevalence in our cohort is somewhat lower than expected, reflecting our threshold-based detectability definition. The recently established association between DLL3 and shorter OS in IHC-based multinational (39), Japanese (42) and Korean (40) cohorts further supports the reproducibility of this finding across diverse populations. As noted above, the four-class model carries no detectable prognostic information within extensive-stage disease; the DLL3−/ASCL1− subgroup is best interpreted as a low-neuroendocrine, predominantly limited-stage subset, biologically consistent with the SCLC-I/SCLC-Y subtype described in recent transcriptomic studies (17,21).

The independent prognostic association of detectable TYMP in our multivariable model (aHR 2.06; P=0.009) is, to our knowledge, not previously reported in SCLC. TYMP is best known as the activator of capecitabine to 5-FU and is therefore considered a predictive (rather than prognostic) biomarker (14,15); in colorectal cancer and other malignancies, however, high tumoral TYMP has also been linked to angiogenic activity and worse outcome (43). The dual potential of TYMP as both a prognostic factor and a predictive biomarker for capecitabine-based regimens—used clinically in pulmonary neuroendocrine neoplasms in the CAPTEM combination (14)—makes TYMP a particularly attractive candidate for prospective evaluation in SCLC.

The lack of a quantitative biomarker that predicts platinum sensitivity at the single-patient level is biologically expected: molecular determinants of acute platinum response include diverse and partially redundant pathways (DNA damage repair, apoptosis regulation, tumor heterogeneity) that are unlikely to be captured by any single protein measurement. The absence of an independent prognostic effect of SLFN11 is also expected: SLFN11 is best characterized as a predictive biomarker of PARP inhibitor sensitivity (5,19,24), and no patient in this pre-2015 cohort received a PARP inhibitor; SLFN11 was included for completeness and to provide reference SRM-MS quantitation for future PARP-inhibitor studies. The TOPO1-refractoriness directional signal is consistent with the published association between higher tumoral TOPO1 expression and topoisomerase-inhibitor activity, though the magnitude in this cohort is modest. The independent prognostic effect of TYMP may reflect TYMP-associated tumor angiogenesis and proliferation (44) rather than a 5-FU-pathway effect, given that no patient in this pre-CAPTEM cohort received capecitabine.

From an assay-development perspective, three observations have practical implications. First, the high inter-patient heterogeneity in DLL3 expression among detectable tumors (Figure 1; 200–1,200 amol/µg among DLL3+ samples, with a long right tail) means that a binary “DLL3-positive” designation discards considerable quantitative information. Continuous quantitation may therefore allow finer stratification—for example, separating “DLL3-high” and “DLL3-low-positive” tumors—that may matter for response to a DLL3-CD3 bispecific T-cell engager whose pharmacology is plausibly threshold-dependent. Second, the strong co-detection of DLL3 with ASCL1 (87.7% of DLL3+ tumors are also ASCL1+) and with TOPO1 (94.7%) and EZH2 (91.2%) supports the existence of a coordinated SCLC-A transcriptional program that is detectable at the protein level by SRM-MS. Third, the assay performed reproducibly on routinely archived FFPE samples up to approximately 8 years old (biopsies 2009–2015; SRM-MS run 2017). Tissue age was not associated with DLL3 detectability (median 4.0 years among DLL3+ versus 4.4 years among DLL3− samples; Mann-Whitney P=0.51) and did not correlate with quantitative DLL3 signal among detectable tumors (Spearman ρ=0.09, P=0.50). DLL3 detectability also did not differ between the two participating centers [Seoul St. Mary’s Hospital, 43/77 (55.8%); Yeouido St. Mary’s Hospital, 14/23 (60.9%); Fisher’s exact P=0.81], nor did the median quantitative DLL3 signal (339 vs. 287 amol/µg). These observations support the analytical robustness of SRM-MS DLL3 quantitation across routine archival timeframes and across collection sites.

Our findings have several implications. For clinical practice, the present results support continued use of stage and ECOG performance status as the principal prognostic anchors in SCLC (Tables 1,2). Quantitative DLL3 and ASCL1 provide a biologically informative readout of the SCLC-A neuroendocrine-high phenotype and may help identify the small subset of patients with non-neuroendocrine, comparatively indolent disease. However, because no patient in this cohort received tarlatamab or an immune-checkpoint inhibitor, the predictive value of these proteomic measurements for either modality cannot be inferred from the present dataset and requires prospective evaluation in treated cohorts.

The complementary, largely non-overlapping MGMT-low and TYMP-high candidate-responder populations (61% of the cohort meeting at least one criterion; Table S1) provide a framework for biomarker-guided selection of TMZ- versus capecitabine-based later-line cytotoxic therapy. The clinical utility of these proteomic thresholds in SCLC has not yet been demonstrated and should be evaluated in cohorts actually treated with TMZ- and capecitabine-based regimens.

Whether quantitative DLL3 will substantially refine tarlatamab patient selection beyond IHC remains uncertain: the phase 3 Rova-T trials selected DLL3-high patients by IHC yet failed to improve OS (35-37); DeLLphi-301 did not require DLL3 positivity for entry; and DeLLphi-304 reported tarlatamab benefit across a broad range of DLL3 IHC strata (20,38). Continuous DLL3 quantitation may therefore be more useful for mechanistic understanding—identification of low-neuroendocrine responders or DLL3-negative responders to bispecific T-cell engagers—than for refining tarlatamab candidacy at the bulk-IHC level. Prospective paired SRM-MS/IHC validation in independent tarlatamab-treated SCLC cohorts is required to address this question.

Conclusions

Multiplexed SRM-MS reproducibly quantified nine therapy-relevant proteins on archival FFPE SCLC tissue in a pre-checkpoint-inhibitor, pre-tarlatamab Korean cohort. Quantitative DLL3 expression discriminated extensive- from limited-stage disease (AUC 0.78); a four-class DLL3/ASCL1 model stratified OS (P=0.004) primarily within limited-stage disease; and detectable TYMP was an independent predictor of shorter OS in multivariable Cox regression (aHR 2.06, P=0.009). MGMT-low and TYMP-high candidate-responder populations were largely non-overlapping. Because no patient received tarlatamab, immune-checkpoint inhibitors, TMZ or capecitabine, predictive utility for these modalities cannot be inferred from the present data; prospective evaluation in treated cohorts, ideally with paired immunohistochemistry, is required.

Acknowledgments

The authors used a large-language-model assistant (Claude, Anthropic) for language editing and for assistance with figure-rendering code; all scientific content, study design, statistical analysis, interpretation, and conclusions are the responsibility of the authors. No AI-generated text was incorporated without manual review and revision.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0513/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0513/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0513/prf

Funding: This work was supported by the Korean Ministry of Trade, Industry and Resources (Industrial Technology Innovation Program) (grant No. RS-2024-00451980).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0513/coif). All authors report that this work was supported by the Korean Ministry of Trade, Industry and Resources (Industrial Technology Innovation Program; grant No. RS-2024-00451980). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of The Catholic University of Korea (Approval No. KC15SISE0263), and informed consent was obtained from all individual participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Rudin CM, Brambilla E, Faivre-Finn C, et al. Small-cell lung cancer. Nat Rev Dis Primers 2021;7:3. [Crossref] [PubMed]
2. Pezzuto F, Fortarezza F, Lunardi F, Calabrese F. Are there any theranostic biomarkers in small cell lung carcinoma? J Thorac Dis 2019;11:S102-12. [Crossref] [PubMed]
3. Dingemans AMC, Früh M, Ardizzoni A, et al. Small-cell lung cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2021;32:839-53.
4. Baine MK, Hsieh MS, Lai WV, et al. SCLC Subtypes Defined by ASCL1, NEUROD1, POU2F3, and YAP1: A Comprehensive Immunohistochemical and Histopathologic Characterization. J Thorac Oncol 2020;15:1823-35. [Crossref] [PubMed]
5. Sabari JK, Lok BH, Laird JH, et al. Unravelling the biology of SCLC: implications for therapy. Nat Rev Clin Oncol 2017;14:549-61. [Crossref] [PubMed]
6. Subbiah S, Nam A, Garg N, et al. Small Cell Lung Cancer from Traditional to Innovative Therapeutics: Building a Comprehensive Network to Optimize Clinical and Translational Research. J Clin Med 2020;9:2433. [Crossref] [PubMed]
7. Farago AF, Keane FK. Current standards for clinical management of small cell lung cancer. Transl Lung Cancer Res 2018;7:69-79. [Crossref] [PubMed]
8. Horn L, Mansfield AS, Szczęsna A, et al. First-Line Atezolizumab plus Chemotherapy in Extensive-Stage Small-Cell Lung Cancer. N Engl J Med 2018;379:2220-9. [Crossref] [PubMed]
9. Paz-Ares L, Dvorkin M, Chen Y, et al. Durvalumab plus platinum-etoposide versus platinum-etoposide in first-line treatment of extensive-stage small-cell lung cancer (CASPIAN): a randomised, controlled, open-label, phase 3 trial. Lancet 2019;394:1929-39. [Crossref] [PubMed]
10. Yang S, Zhang Z, Wang Q. Emerging therapies for small cell lung cancer. J Hematol Oncol 2019;12:47. [Crossref] [PubMed]
11. Hiddinga BI, Pauwels P, Janssens A, et al. O(6)-Methylguanine-DNA methyltransferase (MGMT): A drugable target in lung cancer? Lung Cancer 2017;107:91-9. [Crossref] [PubMed]
12. Lazzari C, Gregorc V, Bulotta A, et al. Temozolomide in combination with either veliparib or placebo in patients with relapsed-sensitive or refractory small-cell lung cancer. Transl Lung Cancer Res 2018;7:S329-33. [Crossref] [PubMed]
13. Pietanza MC, Waqar SN, Krug LM, et al. Randomized, Double-Blind, Phase II Study of Temozolomide in Combination With Either Veliparib or Placebo in Patients With Relapsed-Sensitive or Refractory Small-Cell Lung Cancer. J Clin Oncol 2018;36:2386-94. [Crossref] [PubMed]
14. Al-Toubah T, Morse B, Strosberg J. Capecitabine and Temozolomide in Advanced Lung Neuroendocrine Neoplasms. Oncologist 2020;25:e48-52. [Crossref] [PubMed]
15. Jonna S, Reuss JE, Kim C, et al. Oral Chemotherapy for Treatment of Lung Cancer. Front Oncol 2020;10:793. [Crossref] [PubMed]
16. Rudin CM, Poirier JT, Byers LA, et al. Molecular subtypes of small cell lung cancer: a synthesis of human and mouse model data. Nat Rev Cancer 2019;19:289-97. [Crossref] [PubMed]
17. Gay CM, Stewart CA, Park EM, et al. Patterns of transcription factor programs and immune pathway activation define four major subtypes of SCLC with distinct therapeutic vulnerabilities. Cancer Cell 2021;39:346-360.
18. Redin E, Quintanal-Villalonga Á, Rudin CM. Small cell lung cancer profiling: an updated synthesis of subtypes, vulnerabilities, and plasticity. Trends Cancer 2024;10:935-46. [Crossref] [PubMed]
19. Saunders LR, Bankovich AJ, Anderson WC, et al. A DLL3-targeted antibody-drug conjugate eradicates high-grade pulmonary neuroendocrine tumor-initiating cells in vivo. Sci Transl Med 2015;7:302ra136. [Crossref] [PubMed]
20. Mountzios G, Sun L, Cho BC, et al. Tarlatamab in Small-Cell Lung Cancer after Platinum-Based Chemotherapy. N Engl J Med 2025;393:349-61. [Crossref] [PubMed]
21. Deneka AY, Boumber Y, Beck T, et al. Tumor-Targeted Drug Conjugates as an Emerging Novel Therapeutic Approach in Small Cell Lung Cancer (SCLC). Cancers (Basel) 2019;11:1297. [Crossref] [PubMed]
22. Coleman N, Zhang B, Byers LA, et al. The role of Schlafen 11 (SLFN11) as a predictive biomarker for targeting the DNA damage response. Br J Cancer 2021;124:857-9. [Crossref] [PubMed]
23. Jiang T, Wang Y, Zhou F, et al. Prognostic value of high EZH2 expression in patients with different types of cancer: a systematic review with meta-analysis. Oncotarget 2016;7:4584-97. [Crossref] [PubMed]
24. Zoppoli G, Regairaz M, Leo E, et al. Putative DNA/RNA helicase Schlafen-11 (SLFN11) sensitizes cancer cells to DNA-damaging agents. Proc Natl Acad Sci U S A 2012;109:15030-5. [Crossref] [PubMed]
25. Steiner C, Ducret A, Tille JC, et al. Applications of mass spectrometry for quantitative protein analysis in formalin-fixed paraffin-embedded tissues. Proteomics 2014;14:441-51. [Crossref] [PubMed]
26. Coscia F, Doll S, Bech JM, et al. A streamlined mass spectrometry-based proteomics workflow for large-scale FFPE tissue analysis. J Pathol 2020;251:100-12. [Crossref] [PubMed]
27. Picotti P, Aebersold R. Selected reaction monitoring-based proteomics: workflows, potential, pitfalls and future directions. Nat Methods 2012;9:555-66. [Crossref] [PubMed]
28. Carr SA, Abbatiello SE, Ackermann BL, et al. Targeted peptide measurements in biology and medicine: best practices for mass spectrometry-based assay development using a fit-for-purpose approach. Mol Cell Proteomics 2014;13:907-17. [Crossref] [PubMed]
29. Guo T, Steen JA, Mann M. Mass-spectrometry-based proteomics: from single cells to clinical applications. Nature 2025;638:901-11. [Crossref] [PubMed]
30. Ardizzoni A, Tiseo M, Boni L. Validation of standard definition of sensitive versus refractory relapsed small cell lung cancer: a pooled analysis of topotecan second-line trials. Eur J Cancer 2014;50:2211-8. [Crossref] [PubMed]
31. Genestreti G, Tiseo M, Kenmotsu H, et al. Outcomes of Platinum-Sensitive Small-Cell Lung Cancer Patients Treated With Platinum/Etoposide Rechallenge: A Multi-Institutional Retrospective Analysis. Clin Lung Cancer 2015;16:e223-8. [Crossref] [PubMed]
32. Hembrough T, Thyparambil S, Liao WL, et al. Selected Reaction Monitoring (SRM) Analysis of Epidermal Growth Factor Receptor (EGFR) in Formalin Fixed Tumor Tissue. Clin Proteomics 2012;9:5. [Crossref] [PubMed]
33. Hembrough T, Thyparambil S, Liao WL, et al. Application of selected reaction monitoring for multiplex quantification of clinically validated biomarkers in formalin-fixed, paraffin-embedded tumor tissue. J Mol Diagn 2013;15:454-65. [Crossref] [PubMed]
34. Owonikoko TK, Park K, Govindan R, et al. Nivolumab and Ipilimumab as Maintenance Therapy in Extensive-Disease Small-Cell Lung Cancer: CheckMate 451. J Clin Oncol 2021;39:1349-59. [Crossref] [PubMed]
35. Rudin CM, Pietanza MC, Bauer TM, et al. Rovalpituzumab tesirine, a DLL3-targeted antibody-drug conjugate, in recurrent small-cell lung cancer: a first-in-human, first-in-class, open-label, phase 1 study. Lancet Oncol 2017;18:42-51. [Crossref] [PubMed]
36. Blackhall F, Jao K, Greillier L, et al. Efficacy and Safety of Rovalpituzumab Tesirine Compared With Topotecan as Second-Line Therapy in DLL3-High SCLC: Results From the Phase 3 TAHOE Study. J Thorac Oncol 2021;16:1547-58. [Crossref] [PubMed]
37. Johnson ML, Zvirbule Z, Laktionov K, et al. Rovalpituzumab Tesirine as a Maintenance Therapy After First-Line Platinum-Based Chemotherapy in Patients With Extensive-Stage-SCLC: Results From the Phase 3 MERU Study. J Thorac Oncol 2021;16:1570-81. [Crossref] [PubMed]
38. Ahn MJ, Cho BC, Felip E, et al. Tarlatamab for Patients with Previously Treated Small-Cell Lung Cancer. N Engl J Med 2023;389:2063-75. [Crossref] [PubMed]
39. Rojo F, Corassa M, Mavroudis D, et al. International real-world study of DLL3 expression in patients with small cell lung cancer. Lung Cancer 2020;147:237-43. [Crossref] [PubMed]
40. Nam H, Jung SH, Lee JB, et al. Impact of DLL3 Expression as Prognostic Factor in Extensive Stage of Small Cell Lung Cancer Treated With First-Line Chemotherapy. Thorac Cancer 2025;16:e15522. [Crossref] [PubMed]
41. Aredo JV, Singhal S, Berry GJ, et al. DLL3 Immunohistochemical Expression in Neuroendocrine-Transformed EGFR-Mutant Lung Cancer and Two Cases of Tarlatamab Therapy. JTO Clin Res Rep 2025;6:100913. [Crossref] [PubMed]
42. Furuta M, Sakakibara-Konishi J, Kikuchi H, et al. Analysis of DLL3 and ASCL1 in Surgically Resected Small Cell Lung Cancer (HOT1702). Oncologist 2019;24:e1172-9. [Crossref] [PubMed]
43. Bronckaers A, Gago F, Balzarini J, et al. The dual role of thymidine phosphorylase in cancer development and chemotherapy. Med Res Rev 2009;29:903-53. [Crossref] [PubMed]
44. Huang B, Yuan Q, Sun J, et al. Thymidine phosphorylase in nucleotide metabolism: physiological functions and its implications in tumorigenesis and anti-cancer therapy. Front Immunol 2025;16:1561560. [Crossref] [PubMed]