We collected two published SCLC cohorts from the gene expression omnibus (GEO) database (Clough and Barrett, 2016), namely SCLC (George et al., 2015) and SCLC (Jiang et al., 2016). In this study, the inclusion criteria for SCLC patients were that the included SCLC patients must have survival data, expression data, and mutation data. According to the above inclusion criteria, a total of 68 SCLC patients were recorded from paper 1 (Supplementary Table S1) and 48 SCLC patients from paper 2 (Supplementary Table S2). We filtered the mutation data of the two SCLC cohorts according to the definition and type of non-synonymous mutation in a maftools R package (Mayakonda et al., 2018). The non-synonymous mutation data obtained after filtering was then used for subsequent analyses. The analysis process of this study is detailed in Figure 1.

We used the pRRophetic algorithm (Geeleher et al., 2014) to predict the IC50 value of cisplatin by constructing a relevant ridge regression model with a GDSC cell line expression profile as a training set, and the SCLC cohort as validation set. According to the median value of MMP9 expression, SCLC patients were divided into MMP9-High and MMP9-Low categories. The IC50 values of cisplatin between the two groups were analyzed using the Mann Whitney U test. Further, a Spearman correlation test was used to analyze the expression of MMP9 and the related IC50 value of cisplatin.

MCPcounter, quanTIseq and xCell algorithms were used to analyze the expression profile data of SCLC and obtain the overall score of immune cells (Becht et al., 2016; Aran et al., 2017; Plattner et al., 2020). GSEA was used to analyze the difference of signal pathway activity between MMP9-High and MMP9-Low classes (Reimand et al., 2019). ssGSEA (Hänzelmann et al., 2013) was used to evaluate the signal pathway activity of each SCLC patient in the c2 and c5 pathway sets according to the MsigDB database (Liberzon et al., 2011).

We verified the prognostic value of MMP9 expression in the NSCLC cohort receiving ICIs treatment using a CAMOIP webpage tool (Lin et al., 2021a).

The mutation data and expression data of the TCGA cohort were downloaded using the TCGAbiolinks R package. The data of tumor mutation burden (TMB) and neoantigen loads (NALs) of the TCGA cohort were obtained from published literature (Thorsson et al., 2018).

Tissue samples were deparaffinized and rehydrated. After treatment with endogenous peroxidase blocking solution, they were treated with specific antibodies against MMP9 (ab76003, Abcam), overnight at 4°C. After they were washed with PBS, the samples were treated with horseradish peroxidase-conjugated anti-rabbit IgG (SV0002, Boster, Wuhan, China) and then stained with diaminobenzidine (DAB). All results were assessed by two pathologists. Expression levels were scored by multiplying the percentage of positive cells by the staining intensity. The positivity percentage was scored as 0 if <5% (negative), 1 if 5–30% (sporadic), 2 if 30–70% (focal) and 3 if 70% (diffuse) of the cells were stained; and staining intensity was scored as 0 for no staining, 1 for weak to moderate staining and 2 for strong staining. A score of ≥2 was regarded as ‘high’ and the score of <2 is regarded as “low” in immunohistochemical staining.

Cells were cultured at 5 × 10³ cells per well in a 96-well plate with cytotoxic drugs for 24 h. Cytotoxic drugs (cisplatin and etoposide) were diluted to obtain different concentration gradients. Absorbance was detected at 450 nm after treatment with 10 μl CCK-8 reagent (Dojindo, Kumamoto) for 4 h. The experiments were performed in five replicate wells per sample and the assays were conducted in triplicate.

Total RNA was isolated using Trizol reagent (Invitrogen, United States), according to the manufacturer’s instructions. The quantity and purity of the total RNA was measured using the Nanodrop^® ND1000 (Thermo Fisher) and the Agilent Bioanalyzer. Reverse transcription was performed with 2 μg of total RNA using M-MLV reverse transcriptase (Accurate Biology, AG11728) according to the manufacturer’s recommendations. Quantitative PCR was performed using CFX96 Touch Real-Time PCR Detection Instrument (BioRad, United States). Reactions were performed using SYBR^® Green Premix Pro Taq HS qPCR Kit (ROX Plus) (AG11718). Values were normalized to GAPDH via the 2^−ΔΔCt method.

The statistical results of KM survival analysis were obtained by a log rank test, and the visual results of KM survival analyses were obtained by survivor and survminer. Univariate COX and multivariate COX models were used to verify whether MMP9 can be used as an independent predictor of the prognosis of SCLC patients treated with cisplatin. The different analysis of continuity variables between MMP9-High and MMP9-Low groups was completed by the Mann Whitney U test. In this study, all the analysis is based on R software. The p value is bilateral, and p < 0.05 is regarded as statistically significant.

In order to explore the influence of MMP9 on the prognosis of patients with SCLC treated with cisplatin, we used the univariate COX regression model and multivariate COX regression model to evaluate the SCLC cohort, including SCLC (George et al.) and SCLC (Jiang et al.). In SCLC (George et al.), we found that only MMP9-High was related to an improved clinical prognosis of patients with SCLC, while common clinical factors were not relevant in the prognosis of patients (Figure 2A; HR = 0.425; p = 0.0085). Multivariate results showed that MMP-High could be used as an independent predictor of prognosis of SCLC after cisplatin treatment (Figure 2A; HR = 0.216; p = 0.00153). Then, univariate COX and multivariate COX regression models were also applied to SCLC (Jiang et al.), and the results showed that MMP9-High was not only related to significantly prolonged clinical prognosis time (Figure 2B; HR = 0.365; p = 0.0219) but it can also be used as an independent predictor (Figure 2B; HR = 0.352; p = 0.0199). The KM curve shows that MMP9-High is related to significantly prolonged OS in patients with SCLC (Figure 2C: log-rank p = 0.007; HR = 0.44; Figure 2D: log-rank p = 0.017; HR = 0.39). We used a Sankey diagram to visualize the clinical features of two SCLC cohorts one by one (Figures 2E,F).

We employed the pRRophetic algorithm to predict the cisplatin sensitivity of each SCLC patient to obtain an IC50 value. We found that MMP9-High had significantly lower IC50 values of cisplatin than MMP9-Low SCLC (Figure 3A; p < 0.05). Similarly, the expression of MMP9 was negatively correlated with the IC50 value of cisplatin (Figure 3B, p = 0.001; R = −0.38; method: spearman). We found that MMP9-High had significantly higher TMB than MMP9-Low (Figure 3C; All p < 0.05) and NALs (Figure 3D; all p < 0.05). Figure 3E shows that for SCLC (George et al.), the type and mutation frequency of the driver genes in the top 20 mutation frequencies in the cohort. The results showed that there was no significant difference between the driver genes in the top 20 mutation frequencies of MMP9-High and MMP9-Low. In SCLC (Jiang et al.), the mutation frequency of MMP9-High is significantly higher than that of MMP9-Low in PTPRB and NTRK (Figure 3F). Supplementary Figures S1A,B shows the mutual exclusion and co-occurrence of the top 20 driving mutations in MMP9-High and MMP9-Low, respectively.

Under the MCPcounter algorithm, whether in SCLC (George et al.) or SCLC (Jiang et al.), we found that the TIME when MMP9-High was higher than MMP9-Low was significantly infiltrated with t-cells, cytoxic lymphocytes, B-cells, NK cells, and dense cells (DCS) (Figures 4A,B; all p < 0.05). Under the quanTIseq algorithm, we found that MMP9-High has significantly increased B cells and m1 macrophages when compared with MMP9-Low (Figure 4C: George et al.; Figure 4D: Jiang et al.). With the xCell algorithm, we found that, compared with MMP9-Low, MMP9-High has significant DCs and NKT (Figure 4E: George et al.; Figure 4F: Jiang et al.).

In order to further explore the difference of signal pathway activity between MMP9-High and MMP9-Low, we used GSEA and ssGSEA to evaluate and calculate the signal pathway activity. Figures 5A,B shows that in SCLC (George et al.) and SCLC (Jiang et al.), MMP9-High has significantly increased activity of immune activation related pathways, such as cytokine binding and immune response mediated by B cells and NK cells when compared with MMP9-Low. In addition, the results of ssGSEA analysis showed that the activity of MMP9-High in B-cell, T-cell and NK cell activation, cytokine secretion, chemokine binding, and antigen presentation was significantly higher than that of MMP9-Low (Figures 5C,D).

In order to explore the role of MMP9 in the prognosis of patients receiving immune checkpoint inhibitors, we used CAMOIP as a web tool to verify the relationship between MMP9 and the prognosis of immunotherapy. We found that in the NSCLC (Kim et al.) cohort, the PFS time of MMP9-High was significantly longer than that of MMP9-Low (Figure 6A; log-rank p = 0.026; HR = 0.4). In another NSCLC (Hwang et al.) cohort, the PFS time of MMP9-High was significantly longer than that of MMP9-Low (Figure 6B; log-rank p = 0.029; HR = 0.37). These results suggest that MMP9 may be an important biomarker for the prognosis of immunotherapy.

In this study, the IC50 values of two pairs of chemotherapy-sensitive and drug-resistant small cell lung cancer cells to the first-line chemotherapy drug (cisplatin) were detected using the CCK8 method. The results showed that the IC50 of chemotherapy-sensitive H69/H446 small cell lung cancer cells to cisplatin was significantly lower than that of the corresponding chemotherapy-resistant cells (H69AR/H446DDP) (Figures 7A,B). To detect the effect of MMP9 on the chemotherapy resistance of small cell lung cancer cells, this study used the CCK8 method to detect the IC50 value of chemotherapy-resistant small cell lung cancer cells with upregulated expression of MMP9 to first-line chemotherapy drugs (cisplatin and etoposide). The results showed that upregulating the expression level of MMP9 (H69-MMP9, H446-MMP9) in chemotherapy-resistant small cell lung cancer cells significantly reduced the IC50 value of SCLC cells to chemotherapy drugs (Figures 7C,D), and the subsequent chemotherapy resistance of cells decreased.

Firstly, the basic expression levels of MMP9 mRNA and MMP9 protein in small cell lung cancer cells were detected by real-time fluorescence quantitative PCR and immunohistochemistry. The results of the real-time quantitative PCR showed that the expression level of MMP9 mRNA in chemotherapy-resistant H69AR cells and H446DDP cells was significantly lower than that in chemotherapy-sensitive H69 cells and H446 cells (Figure 7E) (p < 0.001). Immunohistochemically, the results also showed that the expression level of CDYL protein in chemotherapy-resistant H69AR cells and H446DDP cells was lower than that of their parents’ chemotherapy-sensitive H69 cells and H446 cells (Figure 7F). Based on the basic expression level of MMP9 in the above two pairs of chemotherapy-sensitive and chemotherapy-resistant small cell lung cancer cells, this study further uses lentivirus-mediated LV5-MMP9 to up-regulate the expression level of MMP9 in chemotherapy-resistant H69AR and H446DDP small cell lung cancer cells. The verification results of quantitative PCR, and immunohistochemistry showed that we successfully constructed the small cell lung cancer cell model with an up-regulated expression of CDYL (H69AR-MMP9, H446DDP-MMP9). (Figures 7G,H). (p < 0.001).