From 2014 to 2019, a total of 143 eligible patients with SCLC were enrolled in the study. We gathered tumor tissues and blood samples of patients with SCLC before any clinical therapy. We collected and reviewed complete electronic records of all enrolled cases. The SCLC staging were determined by the tumor-node-metastasis (TNM) system. With the approval of the ethics committee of Shanghai Pulmonary Hospital, China (ethical numbers: K20-022), we conducted the study. All participants signed informed consents at the start of study.

We measured the expression of OX40 and OX40L in 102 SCLC specimens by means of IHC. Under the strict asepsis procedure, samples were acquired and diagnosed as SCLC by pathology. After formalin fixation and paraffin embedment, pathology slides were made and prepared for further staining. We used the routine method of dewaxing. Then, specimens were soaked in citrate buffer for antigen recovery. The usage of 3% hydrogen peroxide helped block the activity of endogenous peroxidase. Later, fetal bovine serum was used as blocking reagent. Primary antibodies (OX40, 61637, Rabbit mAb, Cell Signaling Technology; OX40L, ab211287, Rabbit mAb, Abcam) and secondary antibody were applied standardly. In the end, the commonly used diaminobenzidine colorimetry and digital microscope (IX73, OLYMPUS) were applied to quantify protein expression levels of OX40 and OX40L. Three microscope fields at 20× magnification was selected randomly for expression evaluation. The expression of OX40 and OX40L on cancer cells and tumor-infiltrating lymphocytes (TILs) was scored by two pathologists independently.

The survival analysis was the approach to determine the best cutoff point of OX40 and OX40L (27). According to survival analysis results, the p value of the best cutoff value was the minimum. For OX40, more than 20% staining on cancer cells was defined as positive expression, while any staining on TILs was deemed positive expression. The scope of positive OX40L expression on tumor cells (TCs) and TILs was more than 80% staining and 10% staining, respectively.

We performed WES on 41 SCLC cases. After the standard process of sample collection, the DNA characteristics of SCLC tissues and blood samples were analyzed by WES. For DNA extraction and quantification, we used QIAamp DNA Tissue Kit (Qiagen, Valencia, USA), RelaxGene Blood DNA system (Tiangen, Beijing, China), Fluorometer (Qubit 2.0), and the Qubit dsDNA HS assay kit (Thermo Fisher Scientific, CA, USA). Once we obtained fragmented genomic DNA by particular instrument and reagents (28, 29), we constructed DNA library by Kapa Biossystems (MA, USA). Then, fragments with poor quality were excluded, while remaining reads were amplified by a certain circulation of polymerase chain reaction. The Illumina Novaseq 6000 platform was used for DNA sequencing.

With the help of Genome Analysis ToolKit (V 4.1) and Mutect2, DNA fragments were aligned to hg19 reference genome (GRch37), thus detecting somatic mutation, single nucleotide variants, and insertion-deletion mutations. Eventually, according to conventional formula for TMB calculation (28), the somatic TMB value of each case was obtained.

We aimed at further investigating the clinical significance of OX40 and OX40L in the SCLC public cohorts. We queried the Cancer Genome Atlas (TCGA) Database (https://portal.gdc.cancer.gov) and the Gene Expression Omnibus (GEO) Database (https://www.ncbi.nlm.nih.gov/geo/) to get datasets that match the inclusive criteria. All enrolled datasets must own complete RNA sequencing (RNA-seq) and clinicopathologic data for human SCLC specimens. In addition, we surfed a website, named the Cancer Cell Line Encyclopedia (CCLE, https://portals.broadinstitute.org/ccle) Database (30) to verify the expression of OX40 and OX40L in SCLC cell lines. R Studio software (V4.0.1) and several R packages were installed for exporting data from public databases and conducting subsequent bioinformatics analysis, such as differently expressed genes (DEGs) identification, functional analysis, etc.

The biological functions and pathway enrichment of DEGs between two groups were explored by the Gene Ontology (GO) analysis (http://www.geneontology.org/) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis (https://www.kegg.jp/). The GO bar charts consisted of three sections, namely, cellular components, biological processes, and molecular functions. The KEGG plot demonstrated significantly DEGs-related pathways.

In the current study, we conducted the Estimation of STromal and Immune cells in MAlignant Tumours using Expression data (ESTIMATE) algorithm and CIBERSORTx algorithm on the global gene expression profiles of SCLC patients. The ESTIMATE algorithm provided the overall abundance of immune cells and stromal cells of each clinical sample (31). Apart from the immune score and the stromal score, the tumor purity of each SCLC case was also deduced by the ESTIMATE algorithm (31). As one of deep deconvolution and machine-learning tools, the CIBERSORTx used the linear support vector regression and a set of leukocyte genes called LM22 (32, 33). For CIBERSORTx algorithm, it can statistically infer the particular infiltration percentage of 22 kinds of TILs in SCCL TME, such as subgroups of T cells, B cells, natural killer (NK) cells, etc.

The chi-square tests and Pearson correlation tests were used for detecting relationship among OX40/OX40L, PD-1, PD-L1, other biomarkers, and clinical factors. In addition, the Mann-Whitney U test was applied to compare continuous variables between two groups. For survival analysis, the Kaplan-Meier approach and the log-rank test were utilized. By means of univariate and multivariate logistic regression method, factors that affected the expression status of OX40 and OX40L were found. Through univariate and multivariate Cox regression method, independent prognostic factors in SCLC were also identified. Two-sided p value less than 0.05 was deemed to be statistically significant. Two statistical tools were used in the research, namely, the SPSS software (V 22.0) and the RStudio software (V4.0.1).

Totally, the whole cohort contained 143 patients with SCLC ( Table 1 and Table S1 ). The age of participants ranged from 38 to 84 years old. A high percentage of patients were males (121/143, 84.6%) and smokers (77/143, 53.8%). The baseline characteristics of the IHC group and the WES group were separately detailed in Tables S2 and S3 . The sample size of the IHC cohort and the WES cohort was 102 and 41, respectively.

Through IHC, the protein expressions of OX40 and OX40L were found on both TCs and TILs ( Figure 1 ). For OX40, the positive rate on cancer cells and TILs was 7.8% (8/102) and 72.5% (74/102), respectively. Of these 102 tissues, 22 samples exhibited positive OX40L staining on TILs (21.6%), while only two samples showed positive OX40L staining on TCs (2.0%). We then explored the correlation among OX40, OX40L, clinical factors, and eight conventional IHC markers ( Table S4 ). The OX40 protein expression status on TILs had contacts with TNM staging (p=0.044), synaptophysin (p=0.009), and P40 (p=0.013). However, no significant relationship was found between TCs’ OX40 expression and clinical factors. On TILs, the expression degree of OX40 was not significantly linked to enrolled binary markers.

In the study, we also investigated relevance between OX40/OX40L and immune markers ( Figure S1A and Table S5 ). The expression level of OX40 on TILs was significantly correlated with OX40L expression on TILs (p=0.048), PD-1 expression on TILs (p=0.003), PD-L1 on TILs (p=0.019), CD3 (p<0.001), CD4 (p=0.001), and CD8 (p=0.023). OX40L on TILs also had widespread connection with PD-1, PD-L1, and several major markers of T cells (both p<0.05). There was a distinct correlation between OX40 and OX40L expression on TCs (p=0.001). Except for OX40L on TCs, the negative correlation between malignant cells’ OX40 expression and other markers was indicated.

The genetic expressions of OX40 and OX40L in SCLC were verified in 22 SCLC human samples that were extracted from GSE43346 and 54 cell lines that were collected in the public CCLE Database ( Figure S1 ). As shown in Figure S1B , mRNA expressions of OX40 and OX40L in SCLC samples were higher than them in normal tissues. When compared with NSCLC cell lines, the relative low expression levels of OX40 and OX40L in SCLC cell lines were displayed ( Figure S1C ).

The logistic regression analysis was used to study factors predicting expression of OX40 and OX40L in SCLC. Regarding the limited number of cases that showed positive OX40 or OX40L expression on TCs, we separately developed the logistic regression models for OX40 and OX40L on TILs ( Tables S6 , S7 ). We calculated odds ratio (OR) and 95% confidence interval (CI) of each factor for quantitative analysis. On TILs, eight variables were identified as underlying predictors in OX40 expression status by univariate logistic regression analysis, while none of these was deemed to be significantly predictive factor by multivariate logistic regression analysis ( Table S6 ). Similar negative results were found in OX40L expression ( Table S7 ).

RFS data of 102 patients with SCLC was collected in the IHC cohort. Higher recurrence rate was found in patients with smoking history (31/44, 70.5%) when compared with non-smokers (33/58, 56.9%). The Kaplan-Meier curves reflected prognostic differences between different expression status groups ( Figure 2 ). On TCs, OX40 expression status had no marked relation with RFS (p=0.333, Figure 2A ). The overexpression of OX40 on TILs significantly elevated clinical profits in SCLC (vs low OX40 expression: RFS 34.7 months, 95% CI 27.8-41.5 vs 16.3 months, 95% CI 11.5-21.1, p=0.009; Figure 2B ). Regretfully, there was no statistical significance in the survival analysis of OX40L in SCLC ( Figures 2C, D ).

On the basis of OX40 and OX40L expression in SCLC, we conducted the subgroup analysis in the primary cohort ( Figure S2 ). We combined OX40 and OX40L with each other and tested their influences on outcomes ( Figure S2 ). TILs’ OX40 in combination with TCs’ OX40L (p=0.015) or TILs’ OX40L (p=0.012) could effectively distinguish patients with different prognosis.

Given the meaningful findings in the IHC cohort, we further verified the clinical values of OX40 and OX40L in the public TCGA cohort ( Figures 2E, F ). The OS data of 77 SCLC patients in the TCGA cohort were downloaded online (34). The survival analysis demonstrated OS was better in SCLC patients with higher expression levels of OX40 (p<0.001) and OX40L (p=0.019), which had high consistency with our IHC results.

Sixteen clinicopathological traits were enrolled in the univariate and multivariate Cox regression analyses for RFS ( Table 2 ). All cox regression outcomes were analyzed by using hazard ratio (HR) and 95% CI. By means of univariate Cox regression analysis, nine potential prognostic indicators were found, including age (HR=1.744, p=0.045), smoking status (HR=1.612, p=0.064), TNM staging (HR=1.854, p=0.014), OX40 on TILs (HR=0.502, p=0.011), PD-L1 on TILs (HR=0.604, p=0.068), CD3 (HR=0.613, p=0.052), CD4 (HR=0.627, p=0.089), FOXP3 (HR=0.564, p=0.049) and CD8 (HR=0.577, p=0.058). Then, we performed multivariable regression analysis on the above nine factors. After adjustment of confounding characteristics, multivariate results suggested that smoking status (HR=1.915, p=0.029) and OX40 on TILs (HR=0.523, p=0.032) were significant prediction factors in patients with SCLC.

The statistics analysis implied that the mutation frequency of OX40 was 4.9% (2/41), while OX40L did not show any mutation in SCLC tissues. Concurrent mutations were found in OX40 and 14 genes, including ADGRB3, BBX, FRMPD3, LAMA5, MUC16, MUC5B, OR5L2, OTOG, PCDH10, PGP, TAAR8, THSD4, and TP53. Then, the somatic TMB values of 41 SCLC samples were calculated for further analysis. As shown in Figure S3A , the OX40 mutation group represented higher TMB values (vs. the OX40 wild-type group: TMB 7.545 vs. 7.214), but the difference was not significant (p>0.05).

In the whole WES cohort, the progression-free survival (PFS) information was collected in 36 patients (87.8%). In addition, the OS data were obtained in 31 subjects (75.6%). In both PFS and OS, instead of having a perfect separation between two survival curves, the Kaplan-Meier curve of OX40 mutation cases met the curve of OX40 wild-type cases, indicating few influences of OX40 mutation status on prognoses of patients with SCLC ( Figures S3B, C ).

In order to explore the biological behaviors between different OX40 and OX40L expression patterns, we performed GO and KEGG enrichment analyses on the TCGA dataset. For OX40, the high-expression group contained 41 subjects, while the rest of the subjects were allocated to the low-expression group (n=40). Of all DEGs between the two groups, a large proportion of DEGs downregulated in the high-OX40-expression group (2,719/4,405, 61.7%; Figure 3A ). As shown in Figures 3B, C , high OX40 expression was markedly enriched in activation pathways of immune cells, such as neutrophil activation (GO:0042119, p=2.16E-39), T cell activation (GO:0042110, p=5.92E-39), leukocyte proliferation (GO:0070661, p=4.50E-32), and cytokine–cytokine receptor interaction (hsa04060, p=5.66E-15). Concrete top 10 OX40-related functions, biological processes, and pathways are summarized ( Tables S8 , S9 ).

For OX40L, similar grouping method was adopted. The volcano plot demonstrated the enrichment or depletion of 3,048 DEGs in SCLC patients with high OX40L expression level ( Figure 4A ). Specifically, the ratio of downregulated genes to upregulated genes was closed to 2:1 (1,941:1,107). As the ligand of OX40, OX40L also principally related to immune-related functions and pathways. There was a huge overlap between the GO and KEGG enrichment analyses results of OX40 and OX40L in SCLC ( Figures 4B, C and Tables S10 , S11 ). Therefore, we hypothesized that OX40 along with OX40L might have important functions in antitumor immunity of SCLC.

To test the hypothesis, we compared the TME landscapes between the high- and low-expression groups by the ESTIMATE and CIBERSORTx methods ( Figures 5 – 7 ). The overall abundance of immune cells and stromal cells was higher in the high-OX40- or high-OX40L-expression group ( Figure 5 ). Relative low tumor purity was found in tissues with high OX40 or OX40L expression. Moreover, the CIBERSORTx results were consistent with above findings and further confirmed our hypothesis ( Figures 6 , 7 ). The TME with high OX40 expression significantly existed increased immune infiltration of various types of immune cells, including memory B cells (p=0.0018), resting dendritic cells (p=4.3e-05), M1 macrophages (p=1.6e-05), activated memory CD4+ T cells (p=0.0024), and CD8+ T cells (p=0.014; Figure 6 ). For the high-OX40L-expression group, the enrichment of TME resting dendritic cells (p=0.00018) and activated memory CD4+ T cells (p=0.0018) infiltration was also found ( Figure 7 ). To sum up, OX40 and OX40L acted as vital immunoregulators in SCLC.

For the sake of illustrating distinct expression patterns among OX40, OX40L, and immune genes (IGs), we developed the OX40/OX40L-based immune network. The gene set of 2,518 IGs was available in the ImmPort Database (35). The correlation analysis was firstly implemented to quantify the relationship between OX40/OX40L and 1,347 IGs, which were tested in the public TCGA cohort. Correlation coefficients and p value were calculated. The number of significantly OX40-related IGs and OX40L-related IGs was 471 and 360, respectively. A small portion of gene expression levels were inversely associated with OX40 expression (25/471, 5.3%). Then, 50 significantly OX40/OX40L-related IGs with correlation coefficient more than 0.70 were selected for subsequently analysis. For survival analysis, subjects of the TCGA cohort were split evenly on the basis of expression level of target gene. Eventually, eight IGs with the p value equal to or less than 0.1 in the survival analysis were chosen for the construction of the OX40/OX40L-based immune interaction network ( Figure S4 ). The interactions among OX40, OX40L, NCR3, CSF1, TGFB1, RELB, TNFRSF8, HCST, TNFRSF1B, and CARD11 were extensive ( Figure S5 ).