1 Introduction

Small cell lung cancer (SCLC) is a highly heterogeneous and aggressive malignancy that accounts for about 15% of lung cancers [1]. Due to its propensity for metastasis, concurrent chemotherapy and thoracic radiation therapy are considered the standard care for limited-stage SCLC [2]. Although patients with SCLC often exhibit sensitivity to initial chemotherapy, their prognosis remains poor due to the rapid development of acquired resistance [3]. Recently, several retrospective studies have demonstrated that surgical resection improves survival in patients with limited-stage SCLC [4,5,6,7,8]. Some reports have also shown that adjuvant chemotherapy (ACT) following surgery further improves survival in patients with limited-stage SCLC [9,10,11]. Immune checkpoint inhibitors (ICIs) targeting the programmed death ligand 1 (PD-L1) have recently been used to treat SCLC [12]. Despite these advancements, disease-free survival (DFS) and overall survival (OS) for limited-stage SCLC remain relatively suboptimal. The accurate identification of prognostic biomarkers for patients with limited-stage SCLC is therefore urgently needed to predict survival outcomes and treatment response.

Recently, there has been rapid advancement in the development of biomarkers based on cancer genomics. The expression levels of PD-L1, tumor mutation burden (TMB), and tumor immune microenvironment (TIME) have demonstrated their effectiveness as predictive biomarkers for the outcome of ICIs across various cancers [13,14,15,16]. Additionally, m6A regulator expression score has been reported as a prognostic biomarker for guiding chemotherapy and immunotherapy in SCLC [17]. However, little is known about the TIME of SCLC due to the paucity of resected tumors. Comprehensive genomic and immune profiles of limited-stage SCLC and their prognostic implications are also unknown.

Here, we investigated the genomic and immunological profiling using whole exome sequencing (WES) and multiplex immunofluorescence (MIF). Our findings revealed potential prognostic biomarkers for limited-stage SCLC. And we developed a predictive prognostic model based on PD-L1 expression, CD8-positive tumor-infiltrating lymphocytes (TILs), and PI3K pathway mutation for patients with limited-stage SCLC. The present model has the capability to predict the prognosis of patients with limited-stage SCLC and determine the effectiveness of ACT.

2 Materials and methods

2.1 Patients and samples

We enrolled 38 SCLC patients undergoing surgical resection of the lung at the General Hospital of Chinese PLA from 2015 to 2018. The diagnosis of SCLC was confirmed by two experienced pathologists through comprehensive evaluation. Tumor tissues and corresponding paracancerous tissues, preserved in formalin-fixed paraffin-embedded (FFPE) samples, were subjected to WES and MIF analysis.

2.2 WES

WES was performed as previously described [18]. Cancer-related genes included those classified as “Tumor suppressor gene” and “Oncogene” in the OncoKB database. TMB was calculated with the formula: the number of nonsynonymous somatic mutations * 1,000,000/Panel exonic base number. Mutant-allele tumor heterogeneity (MATH) score was calculated using the formula: 100 × median absolute deviation/median of the variant allele frequency. The copy number instability (CNI) score was determined according to a previously established method [19]. Patients were stratified into high and low groups based on the median values of TMB, MATH, and CNI, respectively.

2.3 MIF

MIF was performed as described previously [18]. In brief, 5 μm FFPE sections were deparaffinized, rehydrated, and subjected to epitope retrieval in Tris-EDTA buffer (pH 9) via microwave treatment. Endogenous peroxidase and nonspecific proteins were blocked using Antibody Diluent/Block (72424205; PerkinElmer, MA, USA), followed by sequential antibody labeling involving primary antibodies, secondary antibody incubation, and TSA visualization with the Opal seven-color IHC Kit (NEL797B001KT; PerkinElmer, MA, USA). Finally, the slides were scanned using the Vectra 3.0.5 imaging system (PerkinElmer, MA, USA) according to the manufacturer’s protocol. The multispectral images were spectrally unmixed employing a reference library constructed from single-stained tissue sections using inForm Advanced Image Analysis software (v2.3.0; PerkinElmer, MA, USA). For algorithm training, 5–10 representative multispectral images were utilized to optimize tissue segmentation, cellular segmentation, phenotyping parameters, and positivity thresholds. Quantitative analysis was performed by calculating the percentage of positive cells relative to the total nucleated cell population. For each slide, we evaluated 15 randomly selected fields of view to ensure representative sampling.

The primary antibodies included: LAG3 antibody (1:200, ab209236, abcam), TIM3 antibody (1:200, ab241332, abcam), CD8 antibody (1:100, ZA0508, Zsbio), PD-1 antibody (1:50, ZM0381, Zsbio), and PD-L1 antibody (1:100, #13684, Cell Signaling Technology). The percentage of positively stained cells was calculated in the tumor, stroma, and total region, respectively. A cut-off value of 1% was used to determine positivity for PD-L1 expression, PD-1 expression and TIM3 expression. The first 1/3rd quantile was used as the cut-off value for low CD8 expression. The first 1/3rd quantile was also used as the cut-off value for high LAG3 expression.

2.4 Calculation of the risk score

The risk score for each patient was calculated based on the following criteria. Firstly, PD-L1 positivity was assigned a value of 0, while PD-L1 negativity was assigned a value of 1. Secondly, a high percentage of CD8-positive TILs received a value of 0, whereas a low percentage of CD8-positive TILs received a value of 1. Lastly, the wild-type status in the PI3K pathway obtained a value of 0, while the presence of any mutation in the PI3K pathway obtained a value of 1. The sum of the values of these three markers determined the patient’s risk score. A risk score of ≥ 2 was classified as high risk, while a score of < 2 was classified as low risk.

2.5 Statistical analysis

Statistical analysis was performed using R software (version 4.1.0). Fisher’s exact test was used to compare differences in categorical variables. Survival analysis was conducted using Kaplan-Meier and Cox proportional hazards models. Factors with P values less than 0.05 in the univariate analyses were used in the multivariate analyses. The Fisher’s exact was used to examine correlations between the factors. DFS was defined as the period from the initial surgical resection to recurrence, while OS was defined as the period from the date of diagnosis to death. The statistical tests were two-sided and a significance level of P < 0.05 was considered statistically significant.

3 Results

3.1 Clinical and molecular characteristics of SCLC patients

The clinical characteristics of SCLC patients are shown in Table 1. The cohort consisted of 38 patients aged between 38 and 75, with a male predominance of 84.21% and a smoking history observed in 68.42% cases. Approximately 68.4% of patients were classified as stage I/II, while the remaining 31.6% were at stage III. ACT was administered to 71% of patients, with carboplatin or cisplatin plus etoposide being the primary chemotherapy regimen for a duration of 4 to 6 cycles. The median follow-up time was 36.0 months (range: 3.3 to 88 months).

Table 1 Clinical characteristics of SCLC
Full size table

The most frequently altered cancer-related genes were RB1 (73.68%), TP53 (63.16%), KDM6A (15.79%), and EP300 (10.53%, Fig. S1, Table S1). A co-occurrence of mutations in RB1 and TP53 was observed in approximately 55.26% of patients. POLE mutations were exhibited by approximately 7.89% of patients, while around 26.32% had mutations in the PI3K pathway. The predominant mutation type observed was C > A transition, which has been linked to smoking. The mean TMB value for these patients was 7.1.

The MIF analysis revealed that about 52.63% (20/38) of the cases exhibited PD-L1 expression (combined positive score (CPS) > 1) in the total region (Table S2). The proportion of CD8-positive TILs in the total region varied from 0.04 to 18.96%, with a mean value of 3.27% (Table S2). The percentage of LAG3-positive cells in the overall region ranged from 0 to 9.46%, with a mean of 0.11% (Table S2). TIM3-positive cell percentage ranged from 0 to 41.59%, with an average of 2.84% across the entire region (Table S2). The percentage of PD-1-positive cells in the total region ranged from 0 to 8.64%, with a mean of 1.92% (Table S2). Representative MIF image is shown in Fig. 1A. Heatmap depicting immune cell infiltration across the total region were generated and clustered (Fig. 1B). Strong positive correlations were observed between the following: CD8 + and TIM3 + cells; PD-L1 + and CD8 + PD-L1 + cells; PD-1 + and CD8 + PD-1 + cells; LAG3 + and CD8 + LAG3 + cells; and TIM3 + and CD8 + TIM3 + cells (Fig. 1B–C).

Fig. 1
figure 1

Multiplex immunohistochemical analysis of SCLC immune composition. A Representative multiplex immunofluorescence images. B Heatmaps of immune cell infiltration in the total region. C Pearson correlation analysis of immune cell infiltration in the total region. Blue dots: P < 0.05

Full size image

Correlation analysis demonstrated a positive association between PD-L1 and smoking status (P = 0.049), while no significant correlations were observed between other molecular features and clinical characteristics, including gender, age, ACT, location, smoking, and stage (Table 2). Furthermore, there was no significant correlation found between TMB and the expression of PD-1, PD-L1, CD8, TIM3, or LAG3 (Table S3).

Table 2 Correlation of clinical features with TMB, PD-L1, CD8 + TILs, and PI3K pathway mutations
Full size table

3.2 Clinical and molecular features associated with survival

Subsequently, we utilized the Cox proportional hazards model and Kaplan–Meier curve to examine the correlation between clinical or molecular characteristics and survival outcomes. Apart from stage, no significant associations were found between other clinical factors and DFS or OS. Notably, although not statistically significant, patients who underwent ACT unexpectedly had worse DFS (HR = 1.958, P = 0.165) and OS (HR = 2.762, P = 0.081, Table 3; Fig. 2A). The median DFS for patients who received ACT was 14.8 months (95% CI 9.33-NA), while the median DFS for patients who did not receive ACT was not reached (95% CI 23.1-NA). To eliminate the influence of stage, we stratified patients according to stage. Within the same stage, patients who received ACT also had relatively poorer prognosis (Fig. 2B).

Table 3 Univariate and multivariate analyses of clinical features on survival
Full size table
Fig. 2
figure 2

Kaplan-Meier curves for overall survival (OS) by adjuvant chemotherapy (ACT) and TNM stage

Full size image

Univariate analysis revealed significant associations between DFS and molecular features, including the expression level of PD-L1 and mutation in the PI3K pathway (Table 3). Molecular features significantly associated with OS included the expression level of PD-L1 and the proportion of CD8-positive TILs (Table 3). No significant correlations were found between cancer-related gene mutations, such as TP53 and RB1, and prognosis (Table 3). Additionally, there were no significant correlations observed between TMB, CNI, MATH, or other classical cancer pathways and prognosis (Table 3). Similarly, the expression levels of proteins including PD-1, LAG3, and TIM3 were not found to be associated with prognosis (Table 3). Fisher’s exact test showed that none of the four factors associated with prognosis (stage, PD-L1, CD8, and the PI3K pathway) were significantly correlated with each other (PD-L1 vs. CD8, P = 0.815; PD-L1 vs. PI3K, P = 1; PI3K vs. CD8, P = 0.106; PD-L1 vs. stage, P = 0.204; CD8 vs. stage, P = 0.078; PI3K vs. stage, P = 0.063). Moreover, multivariate analysis also confirmed that PD-L1 expression level, proportion of CD8-positive TILs, and PI3K pathway mutation had prognostic values (Table 3).

Kaplan–Meier curves clearly showed that patients with PD-L1 positivity had significantly improved DFS (P = 0.02) and OS (P = 0.023, Fig. 3A, B). Patients with a high proportion of CD8-positive TILs had significantly improved OS (P = 0.008) and a favorable trend in DFS (P = 0.083, Fig. 3C, D). Additionally, patients with wild-type PI3K pathway had significantly improved DFS (P = 0.04) and a favorable trend in OS (P = 0.058, Fig. 3E, F). It is noteworthy that while the level of PD-L1 expression in the total region was significantly correlated with prognosis, this major contribution primarily originated from PD-L1-positive cells in the stromal region rather than the tumor region (Fig. S2).

Fig. 3
figure 3

Kaplan-Meier curves for disease-free survival (DFS) and overall survival (OS) by PD-L1 (A, B), CD8 (C, D), and PI3K pathway (E, F). Neg negative, Pos positive, WT wild type, MUT mutant

Full size image

We further developed a prognostic prediction model based on these three biomarkers, which enables more significant stratification of patients in terms of DFS (HR = 2.020, P < 0.001) and OS (HR = 2.344, P < 0.001, Fig. 4A, B). Moreover, the risk score derived from this model exhibited a positive correlation with TNM stage (P = 0.042, Fig. 4C). This model can further stratify patients with stage III SCLC for DFS (P = 0.039), but not patients with stage I/II SCLC (P = 0.11, Fig. 4D). In addition, we explored the clinical outcomes of ACT stratified by this model. For low-risk patients, there was no difference in DFS and OS whether they received ACT or not (Fig. 4E, F). However, for high-risk patients, those who received ACT tended to have worse DFS (P = 0.049, Fig. 4E) and OS (P = 0.047, Fig. 4F). Among the subgroup of patients receiving ACT, the DFS (P = 0.022) and OS (P = 0.015) of low-risk patients were also significantly better than those of high-risk patients (Fig. 4E, F).

Fig. 4
figure 4

A prognostic prediction model based on PD-L1, CD8, and PI3K pathway. Kaplan-Meier curves for disease-free survival (DFS, A) and overall survival (OS, B) by the prognostic prediction model. Correlation analysis between risk score and TNM stage (C). Kaplan-Meier curves for DFS by the prognostic prediction model and TNM stage (D). Kaplan-Meier curves for DFS (E) and OS (F) by the prognostic prediction model and adjuvant chemotherapy (ACT)

Full size image

We further conducted an analysis on the impact of these three biomarkers on ACT and found that the expression level of PD-L1 was the main factor affecting ACT outcomes. Among patients who received ACT, both DFS (P = 0.004, Fig. 5A) and OS (P = 0.005, Fig. 5B) were significantly lower for PD-L1 negative patients compared to those PD-L1 positive patients. For PD-L1 positive patients, there was no difference in DFS and OS regardless of whether they received ACT or not. However, in PD-L1 negative patients, those who did not receive ACT had a notably improved OS (P = 0.029) and displayed a favorable trend in terms of DFS (P = 0.053) when compared to their counterparts who received ACT (Fig. 5).

Fig. 5
figure 5

Kaplan-Meier curves for disease-free survival (DFS, A) and overall survival (OS, B) by PD-L1 and adjuvant chemotherapy (ACT). Neg negative, Pos positive

Full size image

4 Discussion

Accurate prognosis prediction will undoubtedly contribute to the optimization of personalized clinical management of SCLC. In this study, we explored the genomic and immune microenvironmental characteristics of SCLC by using WES and MIF analysis, with a specific focus on identifying novel prognostic biomarkers. TP53 and RB1 are the most frequently mutated genes in SCLC [20, 21], but our findings indicate that these mutations are not associated with prognosis (including OS and DFS), which is consistent with previous reports [22]. Additionally, genomic features such as TMB, CNI, and MATH were also not found to be associated with prognosis. However, we have identified three novel prognostic biomarkers, including PD-L1 expression, percentage of CD8-positive TILs, and PI3K pathway mutation.

The MIF analysis showed that despite the considerable heterogeneity, the overall proportion of CD8-positive TILs and the expression levels of immune checkpoint such as PD-L1 were low in SCLC, aligning with the recognized “immune-cold” phenotype of SCLC [23]. The strong correlations between CD8 + and co-expression markers (TIM3/LAG3/PD-1/PD-L1) may indicate T-cell exhaustion or adaptive immune resistance [24, 25], revealing possible mechanisms of immune regulation within the tumor microenvironment.PD-L1 expression is a well-established biomarker for predicting the efficacy of ICIs therapy in lung cancer [14, 16], but the prognostic significance of PD-L1 expression in SCLC remains underreported and controversial [26]. Some studies have shown that high PD-L1 expression in SCLC is associated with better OS or DFS [27,28,29], while other studies have shown that SCLC patients with high PD-L1 expression have worse outcomes [30, 31]. One possible reason for this contradiction lies in the assessment of PD-L1 positivity, such as the variations between CPS and tumor proportion score (TPS). Our results showed that PD-L1 positivity was an independent favorable factor affecting DFS and OS in SCLC patients. Further analysis showed that it was specifically the presence of PD-L1 positive cells in the stromal region that had a significant correlation with prognosis, rather than those in the tumor region. This result aligns with previous reports by Rivalland et al. [32]. The stromal region exhibited a higher prevalence of TILs rather than tumor cells in terms of PD-L1 positive cell distribution. Notably, several other studies have also reported a favorable prognosis associated with elevated PD-L1 expression on TILs [33,34,35]. In addition, we found that PD-L1 expression was significantly and positively associated with smoking status, but not with clinical factors such as age, sex, stage, or location.

We also found that PI3K pathway mutation was significantly associated with DFS, while the proportion of CD8-positive TILs was significantly associated with OS. The PI3K pathway is frequently mutated in SCLC and may contribute to brain metastases [36, 37]. Activation of the PI3K pathway could lead to chemoresistance in SCLC [38]. Consequently, our findings suggest that patients with mutations in the PI3K pathway have an unfavorable prognosis. In SCLC, a high percentage of TILs has been consistently associated with limited stage and improved outcomes [31, 33, 35, 39, 40]. Furthermore, our study provides evidence supporting better survival among patients with a higher proportion of CD8-positive TILs.

Based on these three biomarkers - PD-L1 expression level, proportion of CD8-positive TILs, and PI3K pathway mutations - a model was developed to forecast the prognosis of patients with limited-stage SCLC. This model demonstrates a significant distinction between DFS and OS levels in these patients, and it can also differentiate DFS in patients with stage III disease. An additional advantage of this model is its ability to predict the efficacy of ACT. While previous studies have suggested that ACT may improve survival rates in SCLC patients [10, 41, 42], our findings indicate that ACT does not confer a survival advantage in all patients and indeed diminishes survival in PD-L1 negative patients. Although higher PD-L1 expression generally indicates a better response to immunotherapy, its influence on chemotherapy is more complex. Further research is needed to fully comprehend this relationship.

The study has several limitations that should be considered when interpreting the results. First, the relatively small sample size may introduce selection bias, limiting the generalizability of the findings. Second, this study lacked an independent cohort to validate this prognostic model.

In conclusion, we have identified three biomarkers significantly associated with the prognosis of limited-stage SCLC, despite the relatively small sample size. The risk model based on PD-L1 expression, CD8-positive TILs, and PI3K pathway mutation could potentially provide a viable option for the clinical management of these patients. However, further validation of the prognostic value of this model in larger cohorts is necessary.