Introduction

Small-cell lung cancer (SCLC), constituting around 15% of all lung cancers, is distinguished by its exceptionally high proliferative rate, strong propensity for early metastasis, and poor prognosis1. Conventionally, etoposide combined with platinum-based chemotherapy agents has been the first-line treatment for SCLC2. Yet, drug resistance inevitably emerges, leaving resistant patients with limited effective treatment options and resulting in a dismally low 5-year overall survival (OS) rate of less than 7%3. The advent of immune checkpoint inhibitors (ICI) has revolutionized SCLC treatment4. The combination of atezolizumab with chemotherapy as the first-line treatment has shown a significant survival advantage, extending median OS by 2 months and reducing the risk of death by 30%5. Similarly, durvalumab combined with chemotherapy has been associated with a median OS of 13.0 months and a 27% reduction in the risk of death compared to standard care6. Consequently, the U.S. Food and Drug Administration (FDA) approved atezolizumab and durvalumab in combination with chemotherapy for first-line SCLC treatment in 2019 and 2020, respectively7. However, not all SCLC patients respond well to ICI combination chemotherapy, highlighting the need to understand the impact of this regimen on the tumor ecosystem.

SCLC has been classified based on the expression levels of four transcription factors: ASCL1, NEUROD1, POU2F3, and YAP18. This classification aims to distinguish molecular subtypes but currently has limitations in accurately predicting prognosis and treatment responses. Recently, spatial transcriptomics technology has emerged as a powerful tool with unique advantages in analyzing the tumor microenvironment9. Unlike traditional bulk-sequencing methods that average gene expression, spatial transcriptomics enables visualization of gene expression patterns in a spatially resolved manner, allowing precise mapping of gene expression within tumor tissue and providing insights into cell-cell interactions, cellular communication, and functional heterogeneity at the spatial level.

In this study, we utilized spatial single-cell transcriptome to analyze baseline samples from SCLC patients. We characterized the subtypes and diverse cell states of SCLC epithelial cells and explored the correlation between epithelial cell subtypes and treatment efficacy. Notably, we found that domesticated M2 cells could activate epithelial cell proliferation, potentially explaining the poor treatment efficacy in patients. Our research provides a detailed map of the spatial heterogeneity of cells at the baseline of SCLC and offers a framework for systematically interpreting the evolution of the tumor ecosystem in a spatial context. This could potentially lead to a better understanding of SCLC and the development of more targeted therapeutic strategies.

Results

Spatial single-cell atlas of SCLC patients with distinct treatment efficacies

Eighteen extensive SCLC patients who received a combination of immunothearpy with chemotherapy were enrolled. Among them, 8 patients had progressive disease (PD), and 10 patients had partial response (PR). The median age was 61 years (ranging from 43 to 78 years), and males occupied the majority (66.7%). A total of 31 biopsy samples were obtained and subjected to single-cell spatial detection based on CosMX Spatial Molecular Imager (SMI) platform (Fig. 1A–C; Supplementary Data 1).

Fig. 1: Spatial transcriptome profile of small cell lung cancer.
figure 1

A Schematic illustration of SCLC patients undergoing immuno-combination chemotherapy, along with the detailed workflow of spatial transcriptomics analysis. B Principal diagram of SMI. C Imaging assessment of patient treatment response. D Spatial single cell atlas of SCLC. E Distribution of classical marker genes. F Dot plot showing marker genes of eight cell types. G In situ visualization of HE, SMI immunofluorescence, and annotated cells. H Total cell numbers of samples. I Cell proportions between efficacy groups. J Principal component dimensionality reduction plots of cell proportions for each patient.

Full size image

Through rigorous quality control, four samples with overexposed immunofluorescence staining, derived from two PD patients (P2 and P3), were excluded from further analysis. A total of 188,970 cells were retained for subsequent analysis (Fig. 1D). We employed an innovative approach that integrates single-cell transcriptome data and spatial-proximity to identify different cell populations10. As a result, eight major cell clusters were obtained, including epithelial cells (EPCAM), basal cells (KRT19 and KRT7), fibroblasts (COL1A1 and FN1), endothelial cells (VIM and IGFBP7), T cells (CD3D and PTPRCAP), plasma/B cells (IGHG2 and IGKC), dendritic cells (CLEC4A and CSF1R) and myeloid cells (CD74 and HLA-DRA) (Fig. 1E, F; Supplementary Data 2). The spatial in situ comparison of cell annotations, hematoxylin and eosin (H&E) staining images, and multiplex immunofluorescence (mIF) results confirmed the accuracy of the cell annotations (Fig. 1G, Fig. S1A). Additionally, there was no bias for treatment efficacy among the SMI samples, and there was no significant difference in the number/proportion of each cell type between the PR and PD groups (Fig. 1H, I). Furthermore, we found that the proportions of various cell types were not preferentially associated with treatment efficacy by using principal component analysis (PCA) (Fig. 1J). Based on a comparative analysis of the major cell types between the PD and PR groups, we found that epithelial cells, fibroblasts, myeloid cells, and endothelial cells showed an upward trend in the PD group (Fig. S1B), yet the difference was not statistically significant, prompting us to perform in-depth exploration into the associations between diverse cellular subtypes or states and treatment efficacy.

Analyze the tumor epithelial cell subtypes associated with therapeutic efficacy

We initially delved into tumor epithelial cell subtypes and different cellular states to unearth associations with treatment efficacy. Based on the cellular molecular expression profiles, two distinct epithelial cell subsets were identified: epithelial type 1 (Epi-I), which exhibited high expression of H4C3, and epithelial type 2 (Epi-II), characterized by high expression of KRT17 (Fig. 2A, Fig. S2; Supplementary Data 3). We found that within the PD group, the ratio of Epi-I to Epi-II was higher compared to that in the PR group (Fig. 2B, C), suggesting that Epi-I cells may be associated with treatment resistance, while Epi-II cells may be indicative of treatment sensitivity. Moreover, to comprehensively understand the transcriptional regulatory mechanisms within epithelial cells and uncovering the crucial molecular signatures underlying the cell states associated with treatment efficacy, we employed the consensus non-negative matrix factorization (cNMF) approach11. The cNMF method enabled us to identify distinct and robust gene modules that govern the functions and states of epithelial cells. Consequently, a total of eight meta-programs (MPs) were identified (Fig. 2D; Supplementary Data 4), which were associated with five different categories spanning a wide range of crucial cellular functions: the Proliferating processes (MP-Proli), manifested in MP1 and MP6; the Immunoglobulin-related activities (MP-IG), represented by MP2; the Epithelial-to-Mesenchymal Transition processes (MP-EMT), represented by MP3, MP4, and MP8; the Hypoxia-related responses (MP-Hypoxia), represented by MP5; and the MHC-II associated immune-regulatory functions (MP-MHC-II), represented by MP7 (Fig. 2E). The Curated Cancer Cell Atlas (3CA) is a comprehensive database that contains a vast amount of MPs related to cancer and serves as a reliable resource for validating findings such as the presence of specific immune-related meta-programs in cancer studies12. The presence of immune-related MPs within epithelial cells also can be found in this database.

Fig. 2: Cellular states of epithelial cells in SCLC.
figure 2

A UMAP showing the epithelial cell subtypes. B Ratio of epithelial cell subtypes in each patient. C Cell proportion of epithelial subtypes in the PR/PD groups. Paired Wilcoxon rank-sum test, no significant differences were observed between the two groups. D Heatmap showing the diverse MPs of epithelial cells. E Validation of the MPs in the 3CA database. F Distribution of MP scores according to efficacy and epithelial subtype. Paired t test, ****p < 0.001. G Proportion of each state in the epithelial cells. Paired Wilcoxon rank-sum test, no significant differences were observed. H Spearman correlation between efficacy and epithelial subtype in cell state. I Differential genes of epithelial cells in terms of efficacy and epithelial subtype. J Protein-protein interaction (PPI) network of significantly altered MP-Proli genes. K Effect of significantly altered MP-Proli genes on the overall survival of lung cancer patients in the TCGA database.

Full size image

Based on the expression levels of genes associated with each MP, we computationally derived scores for different MPs in each epithelial cell. Intriguingly, we found that Epi-I cells predominantly exhibited a cellular state characterized by MP-Proli, while Epi-II cells manifested cell states related to MP-IG and MP-EMT. Furthermore, a comparative analysis between the PD and PR groups revealed that the MP-Proli scores of epithelial cells were significantly higher in the PD group (Fig. 2F). This finding strongly implies a potential role of tumor cells in the MP-Proli state in conferring treatment resistance.

We further represented the state of each epithelial cell by the MP with the highest score. Consistently, the proportion of cells in the MP-Proli state was higher in the PD group than in the PR group (Fig. 2G). Moreover, we observed a significant correlation between cell states and treatment efficacy, with patients exhibiting poor treatment outcomes showing a higher propensity towards MP-Proli state in epithelial cells (Fig. 2H). Further investigation revealed differences in the genes involved in MP across treatment efficacy groups and cell subtypes (Fig. 2I; Supplementary Data 5). The results showed that genes related to cell proliferation were significantly enriched in the Epi-I cells of PD patients, indicating that the tumor epithelial cells in PD patients possess stronger proliferative capabilities. To further explore the biological states or processes involved among malignant cells, we performed protein-protein interaction (PPI) analysis using the STRINGdb website for those genes within MP-Proli (MP1/6). These genes affect three aspects that influence the cell cycle, including energy supply, signaling, and mitosis-related biological processes (Fig. 2J). According to the TCGA database13, patients with lung cancer (lung adenocarcinoma and lung squamous cell carcinoma) who have high expression levels of these genes exhibit significantly shorter OS compared to those with low expression levels (Fig. 2K).

The transitions between Epi-I and Epi-II and their spatial characteristics

To understand the relationship between Epi-I and Epi-II, we performed trajectory analysis of SMI data using Monocle214. The analysis revealed that Epi-II appears earlier than Epi-I (Fig. 3A–C), indicating that Epi-I might come from Epi-II. To further confirm the sequential relationship of the transition between Epi-I and Epi-II, we performed pseudotime analysis by incorporating normal basal cells together with Epi-I and Epi-II cells. The trajectory results from both Monocle2 and Slingshot15 algorithms indicated that Epi-II cells were more similar to normal cells, while Epi-I cells were more likely to be at the endpoint of the transition trajectory (Fig. S3A, B). Moreover, genes associated with the MP-Proli show an increasing trend within pseudo-time (Fig. 3D). In addition, we also observed an increase in macrophage migration inhibition factor (MIF). Genes such as MARCKSL1 and PFN1, which are involved in signal transduction16,17, increase within pseudo-time (Fig. 3D), suggesting that the transition from Epi-II to Epi-I may be influenced by surrounding non-epithelial cells.

Fig. 3: Relationship and spatial distribution of two types of epithelial cells in SCLC.
figure 3

A Pseudo-time analysis of epithelial cells. B Evolutionary relationship between EPI-I/II. C EPI-I/II Pseudo-time distribution. D Genes with significant Pseudo-time changes. E Neighbor enrichment of various cell types. F EPI-I/II cells enriched neighbor cell types. G The average score of the centrality in each sample. H, I PR/PD cells in situ and EPI-I/II co-occurrence probability with other cells. J Statistics of co-occurrence probability of EPI-I/II cells with other cells in each sample. T test, *p < 0.05, **p < 0.01, ***p < 0.005.

Full size image

Through spatial analysis of cell colocalization, we discovered that Epi-II is surrounded by more non-epithelial cells than Epi-I (Fig. 3E). Specifically, Epi-II is predominantly surrounded by fibroblasts and plasma cells, which might play a role in promoting the transformation of Epi-II into Epi-I (Fig. 3F). To explore the roles of different cell types, we utilized centrality values from social network analysis to depict the spatial distribution among cell types. While there was no significant difference in the average clustering among various cell types, epithelial cells exhibited a stronger tendency toward aggregation (Fig. 3G). Compared with Epi-I cells, Epi-II cells have higher closeness centrality and degree centrality (Fig. 3G, Figure S4), indirectly indicating that Epi-II has more interactions with other cells than Epi-I does. Furthermore, we calculated the probabilities of Epi-I and Epi-II co-occurring with other cell types. The results indicated that Epi-I tends to appear more solitarily, showcasing self-aggregation traits, which aligns with our previous finding. In contrast, Epi-II is more likely to co-occur with dendritic cells in PR and fibroblasts in PD (Fig. 3H, I). These observations suggest that Epi-II cells exhibited a significantly closer proximity to TME compared to Epi-I cells. The peripheral localization of Epi-II enables it to have more extensive contact with a wide variety of cells, which might play crucial roles in the modulation of the tumor microenvironment. These findings were further supported by fluorescence staining and in-situ visualization of the cells (Figure S3). Additionally, the therapeutic efficacy may correlate with the types of cells Epi-II encounters: interaction with fibroblasts might facilitate the transition from Epi-II to Epi-I, while contact with dendritic cells could contribute to the responsiveness of PR patients to therapy (Fig. 3J). These results were further validated in subsequent analyses of spatial domains. Specifically, the stLearn algorithm18 delineated cellular trajectories, demonstrating that Epi-II cells can undergo a transition to Epi-I cells. This transition pattern was consistently observed in both PD and PR patient cohorts, thereby providing novel spatial evidence for the phenotypic plasticity between Epi-I and Epi-II cell subtypes (Figure S3C).

Spatial analysis revealed the characteristics of the Epi-II-dominant zone in SCLC

One of the advantages of spatial transcriptomics over single-cell sequencing is the ability to observe the tumor microecology by determining the spatial relationships between various cell types based on their positional distances. SMI technology similarly possesses this advantage. To further investigate the niche differences between Epi-I and Epi-II, we performed a spatial zone clustering analysis on all cells in the SMI data of this study. We identified seven types of zones and defined these zones based on the proportion of each cell type (Fig. 4A). Specifically, the zones include the basal cell zone (basal cell 56.9%), Epi-DCs zone (DCs 21.0%, Epi-II 39.6%), Epi-Fib zone (Fib 27.6%, Epi-II 26.3%), Epi-myeloid zone (Myeloid 27.6%, Epi-II 26.4%), plasma cell zone (Plasma 49.0%, Epi-II 14.0%), Epi-II-dominant zone (Epi-II 67.1%), and stable Epi zone (Epi-I 67.1%)(Fig. 4B). We found that, in most spatial zones, the proportion of Epi-II cells is generally greater than that of Epi-I cells, which corroborates the conclusion mentioned earlier from the side, that is, Epi-II has more connections with other cell types, while Epi-I cells tend to cluster together.

Fig. 4: Spatial zone characteristics of SCLC.
figure 4

A UMAP showing each zone in SCLC. B Proportion of cell types within each zone, as well as the ratio of PR/PD. C Cell cycle MP scores for each zone of each sample. The Wilcoxon test, no significant difference. D Domain-specific gene enrichment result of HALLMARK. E The biological process of zone-specific gene enrichment. F The cell proportion of PD/PR in the Epi-II-dominant zone. G The cell communication of PD/PR in Epi-II-dominant zone. H The total number of interactions in the PR/PD groups within the Epi-II-dominant zone. I PD/PR communication path with significant difference in Epi-II-dominant zone.

Full size image

In addition to the high proportion of Epi-I in stable Epi zone, the MP-Proli (MP1/6) of the stable Epi zone were also higher than other zones (Fig. 4C). To further explore what biological states or processes are involved among individual malignant cells, we performed Gene Set Enrichment Analysis (GSEA) by using the ‘Hallmark’ gene set collections for those genes with high variability (Supplementary Data 6). The differential expressed genes (DEGs) were mainly involved in hallmarks such as the G2M checkpoint, inflammatory response, interferon α response, interferon γ response, TNFα signaling via NF-κB, EMT, KRAS signaling up (Fig. 4D). The features genes in the sensitive epithelial zone were enriched for immune response-related markers, while those in the stable epithelial zone were more associated with cell proliferation. Additionally, the Gene Ontology (GO) enrichment analysis results for Biological Processes (BP) validated the same conclusions (Fig. 4E). However, there was no difference in the proportion of cells in the sensitive epithelial zone between the PR and PD groups (Fig. 4F), suggesting that the difference in treatment efficacy is not due to the number of immune cells but rather may from the interaction patterns between epithelial cells and immune cells. Therefore, we conducted cell communication analysis separately for cells from the sensitive epithelial zone in the PR and PD groups. Although the PR group contained more cells than the PD group (Fig. 4B), the CellChat analysis showed that the frequency of interactions among various cell types was higher in the PD group than in the PR group (Fig. 4G, H). This indicates that in the sensitive epithelial zone of the PD group, cell communication is more frequent. Comparing the most significantly different ligand-receptor pairs between the treatment efficacy groups, we identified 23 receptor-ligand signals, with 15 significantly enriched in the PR group and 8 significantly enriched in the PD group (Fig. 4I).

M2-polarized myeloid cells promote the transition of Epi-II to Epi-I

To better understand the communication patterns among various cells within the sensitive epithelial zone, we conducted an in-depth analysis of the 23 significantly different ligand-receptor signals related to treatment efficacy. We observed that in the PD group, there is evident cell communication between myeloid cells and Epi-II cells, both in the input and output signals (Fig. 5A, B). Compared to the PR group, myeloid cells, basal cells, and endothelial cells in the PD group exhibit more frequent interactions, with the most significant input signal being MIF and the most significant output signal being SPP1 (Fig. 5A, B) Further analysis revealed that, apart from myeloid cells, Epi-I, Epi-II, and fibroblasts are the three cell types that interact most frequently within the sensitive epithelial domain. Notably, the analyses conducted using CellChat19 and CellPhoneDB20 algorithms consistently yielded the same results, frequent interactions involving myeloid cells were exclusively observed within the PD group, whereas absent in the PR group (Fig. 5C; Fig. S5A). This suggests that myeloid cells may influence other cells, particularly Epi-II cells, making patients less responsive to therapy, resulting in poor treatment efficacy and prognosis.

Fig. 5: Epithelial cells domesticate macrophages in the Epi-II-dominant zone.
figure 5

A Overview of significant cell incoming signals. B Overview of significant cell outgoing signals. C Signal incoming and outgoing intensity of cells. D Ligand receptor pairs are exported by macrophages and received by epithelial cells. E Pseudo-bulk of genes in SPP1 signaling pathway. F The Cell cycle MP score of PD/PR Epi-II cells in the Epi-II-dominant zone. G MIF signal strength network. H The signaling patterns of crosstalk between epithelial cells and macrophages. I Ligands secreted by Epi-II cells may modulate target genes expression in myeloid cells. J Ligands secreted by myeloid cells may modulate target genes expression in Epi-II cells.

Full size image

Previous studies on small cell lung cancer have reported that SPP1 signals are primarily received by integrin complexes21. By comparing the differences in this signal between the PD and PR groups, we found that myeloid cells are more likely to communicate with Epi-II cells in the PD group than in the PR group (Fig. 5D). The SPP1 signaling showed consistent results in CellPhoneDB analysis, with significantly stronger activity observed in the PD group compared to the PR group (Figure S5B). When the integrin complex on Epi-II cells binds to SPP1, it activates the PI3K-Akt signaling pathway22, thereby enhancing the proliferative capacity of the cells (Fig. 5E, F). On the other hand, the express of SPP1 by myeloid cells may be induced by malignant tumor cells23. In the PD group, myeloid cells receive more MIF released by epithelial cells (Fig. 5G). MIF can retain myeloid cells near tumor cells and promote their transition to an M2 phenotype, thereby creating an immunosuppressive tumor microenvironment (Fig. 5H).

To further validate the interactions between Epi-II cells and myeloid cells, we employed the NicheNet algorithm, a method published in Nature Methods specifically designed for inferring downstream genes regulated by ligand-receptor interactions24. In the Epi-II-dominant zone, comparative analysis between the PD and PR groups revealed that MIF secreted by Epi-II cells in the PD group was predicted to modulate the expression of SPP1 in myeloid cells (Fig. 5I, Figure S5). The ligand-receptor pairs mediating these interactions were identified as MIF-CXCR2/4, which is consistent with the results of CellChat analysis (Fig. 5G). Conversely, SPP1 expressed by myeloid cells was predicted to promote the expression of genes associated with Epi-II cell proliferation, including TOP2A, MYC, and TUBB, through binding to the ITG complex on Epi-II cells (Fig. 5J, Figure S5). Collectively, these results from NicheNet strongly indicate that the interactions between Epi-II and myeloid cells can exert regulatory effects on downstream genes, ultimately leading to alterations in cellular functions and states.

Discussion

Research into SCLC has been significantly challenged by the disease’s aggressive nature and the scarcity of accessible tumor tissue, which has limited our insights into the TME and potentially contributed to the failure of current therapeutic approaches. Utilizing the CosMx Spatial Molecular Imaging approach, we provide a detailed spatial characterization of this disease. Our findings reveal a complex cellular ecosystem within the SCLC TME, encompassing the spatial organization of malignant cells alongside infiltrating immune and stromal cells. Historically, SCLC has been treated as a monolithic entity, with standard treatments remaining largely unchanged for decades25. Our study uncovers previously unrecognized regional heterogeneities in SCLC at single-cell resolution, highlighting the spatial dynamics that have been overlooked. These discoveries offer fundamental implications for understanding SCLC biology and could pave the way for developing more targeted and effective treatment strategies. By providing unprecedented spatial insight into the heterogeneity of SCLC, our research underscores the importance of considering the TME’s complexity when designing future therapies.

Our analysis uncovered diverse expression profiles within malignant cells of SCLC patients, characterized by distinct programs associated with cell cycle regulation, immune responses, hypoxia, EMT, and other markers. Through the integration of marker information derived from mIF and cell segmentation data, we have confirmed the expression of genes within immune-related MPs in epithelial cells. This finding is further substantiated by data from the comprehensive 3CA database, which provides additional evidence for the presence of immune-related MPs in epithelial cells. Notably, cell cycle activity was prevalent across a substantial proportion of SCLC cases, underscoring why these patients often respond positively to radiation and chemotherapy—therapies that specifically target rapidly dividing cells. The prominence of cell cycle programming in SCLC may also contribute to treatment resistance and tumor recurrence, factors that significantly influence patient survival outcomes. For example, we found that although the Epi-I/Epi-II ratios showed no significant differences between PD and PR groups, the MP-Proli scores demonstrated significant differences in relevant comparisons. Our study also identified potential actionable targets for therapies that could be integrated with current standard-of-care regimens to enhance efficacy and overcome resistance mechanisms. Additionally, we observed dynamic interactions between malignant cells and the changing TME in SCLC patients. Specifically, certain cellular transitions facilitated the progression from Epi-II to Epi-I states, indicating a complex interplay between malignant transformation and TME evolution. This observation suggests that targeting not only the malignant cells but also the evolving TME components could offer novel therapeutic opportunities.

In our study, malignant cells exhibited distinct spatial signatures that underscore their potential to interact with immune components within the tumor microenvironment (TME). To validate the association between this spatial organization and patient outcomes, we analyzed the spatial distance relationships between Epi-I/Epi-II cells and other TME constituents. Consistent with our expectations, over half of the malignant cells in the partial response (PR) dataset were identified as Epi-II, demonstrating strong spatial proximity to immune cells. The spatial properties were calculated based on the coordinates and transcriptomes of all cells within the tumor samples, revealing high levels of immune cell and stromal cell infiltration in these SCLC tumors. Our Spatial Molecular Imaging (SMI) data uncovered distinct “stable” and “sensitive” zones within the SCLC microenvironment, strongly associated with treatment efficacy. “Stable” Zone: Dominated by Epi-I cells, characterized by their high proliferative properties. This dominance aligns with the aggressive nature of these tumors. “Sensitive” Zone: Marked by the presence of Epi-II cells that promote tumor development through interactions with macrophages in the MIF-acclimated region. Specifically, domesticated macrophages stimulate the proliferation of Epi-II cells via SPP1 signaling, potentially explaining suboptimal clinical outcomes in some patients. Our research provides critical insights into spatial heterogeneity and immune interactions within SCLC tumors. Targeting the interactions between Epi-II cells and specific macrophage subtypes could represent a novel therapeutic avenue, enhancing the effectiveness of current immunotherapy regimens and improving patient outcomes.

While our study utilizes spatial omics sequencing technology to characterize SCLC, several limitations are acknowledged. The sample size was relatively small, with 18 patients included, and many samples were collected from small biopsies. This approach may not fully represent the biology of the entire tumor, thereby limiting the scope of our analysis. We have addressed this by integration of public single-cell RNA-seq data26 (Figure S6). Our analysis showed that most patients in our cohort clustered with the SCLC-A subtype, and although one PD patient may belong to the SCLC-Y subtype (Figure S6A, B), this did not disproportionately influence our findings. Through further analysis of MIF expression levels, we found that Epi-I cells, similar to multiple types of lung cancer cells, exhibit high MIF expression, whereas Epi-II cells and normal lung cells show low MIF expression (Figure S6C). Additionally, the MIF signaling interaction network analysis demonstrated that MIF secreted by various lung epithelial cells primarily targeted myeloid cells (Figure S6D). Furthermore, we conducted drug sensitivity analyses of MIF and SPP1 using the Genomics of Drug Sensitivity in Cancer (GDSC) and CellMiner databases27,28. The results indicated that blocking SPP1 signaling might be more effective than targeting MIF signaling (Figure S7). Although our current findings suggest that the identified associations may be applicable across different SCLC subtypes, larger multi-cohort studies, especially those with longitudinal designs, are imperative to comprehensively validate these relationships within diverse SCLC patient populations.

In summary, our study provides a high-precision single-cell resolution spatial omics analysis of the tumor microenvironment in human SCLC. Despite the limitations, these findings serve as a robust basis for guiding therapy design and offer valuable insights into the complex interactions within SCLC tumors. Future research should focus on longitudinal studies, incorporating data from treated patients and conducting more extensive experimental validations using models such as patient-derived organoids or PDX models. Such studies will be crucial for further elucidating the mechanisms underlying SCLC progression and developing more effective therapeutic strategies.

Methods

Patient recruitment and sample collection

Between November 2014 and August 2019, 31 formalin-fixed and paraffin-embedded (FFPE) tissue samples were collected from 18 extensive-stage SCLC patients at Harbin Medical University Cancer Hospital. Of these 18 patients, 8 had progressive disease (PD) and 10 had partial response (PR), and they were included in the analysis based on their treatment responses. (Supplementary Data 1). Patients were excluded if they had a prior history of malignancy, previous chemotherapy, radiotherapy or immunotherapy. This study was approved by the Medical Ethical Committee of Harbin Medical University Cancer Hospital and was performed in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all patients prior to their participation in the study.

CosMx™ spatial molecular imaging (SMI)

CosMxTM technology [CosMxTM Human Universal Cell Characterization RNA Panel (1000-plex) + 20 custom genes; Nanostring, USA] was applied to 9 FFPE samples: 10 PR and 8 PD with a mean of 11 fields of view (FOVs) per sample (197 total FOVs). The mean area explored was 1153 mm2 per sample. The following cell surface markers were used for morphology visualization: PanCK, CD45, CD68 antibodies and DAPI.

CosMx™ spatial molecular imager (SMI) sample preparation

FFPE tissue sections were prepared for CosMxTM SMI profiling as described in ref. 15. Briefly, five-micron tissue sections mounted on VWR Superfrost Plus Micro slides (cat# 48311-703) were baked overnight at 60 °C, followed by preparation for in-situ hybridization (ISH) on the Leica Bond RXm system by deparaffinization and heat-induced epitope retrieval (HIER) at 100 °C for 15 min using ER2 epitope retrieval buffer (Leica Biosystems product, Tris/EDTA-based, pH 9.0). After HIER, tissue sections were digested with 3 µg/ml Proteinase K diluted in ACD Protease Plus at 40 °C for 30 min.

Tissue sections were then washed twice with diethyl pyrocarbonate (DEPC)-treated water (DEPC H2O) and incubated in 0.00075% fiducials (Bangs Laboratory) in 2X saline sodium citrate, 0.001% Tween-20 (SSCT solution) for 5 min at room temperature in the dark. Excess fiducials were rinsed from the slides with 1X PBS, then tissue sections were fixed with 10% neutral buffered formalin (NBF) for 5 min at room temperature. Fixed samples were rinsed twice with Tris-glycine buffer (0.1 M glycine, 0.1 M Tris-base in DEPC H2O) and once with 1X PBS for 5 min each before blocking with 100 mM N-succinimidyl (acetylthio) acetate (NHS-acetate, Thermo Fisher Scientific) in NHS-acetate buffer (0.1 M NaP, 0.1% Tween pH 8 in DEPC H2O) for 15 min at room temperature. The sections were then rinsed with 2X saline sodium citrate (SSC) for 5 min and an Adhesive SecureSeal Hybridization Chamber (Grace Bio-Labs) was placed over the tissue.

NanoString® ISH probes were prepared by incubation at 95 °C for 2 min and placed on ice, and the ISH probe mix (1 nM 980 plex ISH probe, 10 nM Attenuation probes, 1 nM SMI-0006 custom, 1X Buffer R, 0.1 U/μL SUPERase•In™ [Thermo Fisher Scientific] in DEPC H2O) was pipetted into the hybridization chamber. The chamber was sealed to prevent evaporation, and hybridization was performed at 37 °C overnight. Tissue sections were washed twice in 50% formamide (VWR) in 2X SSC at 37 °C for 25 min, washed twice with 2X SSC for 2 min at room temperature, and blocked with 100 mM NHS-acetate in the dark for 15 min. In preparation for loading onto the CosMx SMI instrument, a custom-made flow cell was affixed to the slide.

CosMx spatial molecular imager (SMI) instrument run

RNA target readout on the CosMx SMI instrument was performed as described in Ref. 53. Briefly, the assembled flow cell was loaded onto the instrument and Reporter Wash Buffer was flowed to remove air bubbles. A preview scan of the entire flow cell was taken, and 15–25 fields of view (FOVs) were placed on the tissue to match regions of interest identified by H&E staining of an adjacent serial section. RNA readout began by flowing 100 μl of Reporter Pool 1 into the flow cell and incubation for 15 min. Reporter Wash Buffer (1 mL) was flowed to wash unbound reporter probes, and Imaging Buffer was added to the flow cell for imaging. Nine Z-stack images (0.8 μm step size) for each FOV were acquired, and photocleavable linkers on the fluorophores of the reporter probes were released by UV illumination and washed with Strip Wash buffer. The fluidic and imaging procedure was repeated for the 16 reporter pools, and the 16 rounds of reporter hybridization-imaging were repeated multiple times to increase RNA detection sensitivity.

After RNA readout, the tissue samples were incubated with a 4-fluorophore-conjugated antibody cocktail against PanCK (532 nm), CD45 (594 nm), and CD68 (647 nm) proteins and DAPI stain in the CosMxTM SMI instrument for 2 h. After unbound antibodies and DAPI stain were washed with Reporter Wash Buffer, Imaging Buffer was added to the flow cell and nine Z-stack images for the 5 channels (4 antibodies and DAPI) were captured.

Segmentation and quality control

Images were segmented to delineate cell boundaries, assign transcripts at the cell level, and generate a transcript-by-cell count matrix29,30. Four samples with overexposed immunofluorescence staining, derived from two PD patients (P2 and P3), were excluded from further analysis. Cells with an average negative control count exceeding 0.5 or fewer than 10 detected features were removed. The scDblFinder R package was used to identify and filter out doublets31. Following quality control, 82.3% of the cells across samples, equivalent to 172 FOVs, were retained, averaging 12,598 cells per sample.

Preprocessing and neighboring expression construction

The NormalizeData function from the Seurat R package was used to normalize the data, setting the normalization.method parameter to “RC” and the scale.factor to the median of nFeature_RNA32. To construct the neighboring expression matrix of cells, the Banksy R package was employed10, which enhances each cell’s features by using the average and gradient of spatial neighbor features. Based on the normalized RNA expression, the RunBanksy function was used to build the neighboring expression matrix, calculating the nearest 15 cells’ expression, selecting all features (1,000 genes), and scaling was performed separately for each FOV.

Integration and dimension reduction

Next, Principal Component Analysis (PCA) was performed on both the cell’s own expression matrix and the neighboring expression matrix. By examining the elbow plot (% variance explained vs. number of PCs), we selected the first 20 principal components (PCs) as input for integration using Harmony33, with group.by.vars set to “samples.” Additionally, we used uniform manifold approximation and projection (UMAP)34 for dimensionality reduction.

Cell type annotation

To identify cell types, we ran FindNeighbors to build a network of neighbors between cells based on the Harmony-integrated data. We then used FindClusters to set multiple resolutions (0.1, 0.3, 0.5, 1) and selected the number of clusters at a resolution of 0.5 as the basis for subsequent analysis. The FindAllMarkers function was used to identify marker genes in each cluster, considering only the cell’s own expression and testing for logarithmic changes in positivity. The top 20 highest-ranking genes were selected and compared with reference cell type markers. Cell type annotations were then made by referencing the CellMarker35 and PanglaoDB36 databases based on these marker genes, with additional manual adjustments as necessary. Based on these marker genes, 8 cell types were identified: epithelial cells (EPCAM, H4C3), myeloid cells (LYZ, CD68), plasma cells (IGHG1, IGHA1), fibroblasts (COL1A1, FN1), T cells (CD3D, CCL5), Basal cells (KRT19, KRT7), and dendritic cells (CLEC4A, CSF1R). Epithelial cells were further divided into 2 biological compartments, each with its own unique expression pattern and spatial distribution characteristics.

Identifying cell consistency status

We used the GeneNMF R software package (v0.4.0) for non-negative matrix factorization (NMF) analysis of a subset of epithelial cell11. To determine robust programs, we applied the multimf() function across multiple values of (k) (from 6 to 9) and identified eight metaprograms (MPs). Based on the Jaccard index, we cut the similarity tree at a specified height to identify similar blocks and obtain the consensus gene signature for each MP. To better understand the capabilities of these MPs, we compared them to known signatures in a public database. Summarized MPs were downloaded from the 3CA database for reference12, and the generality and robustness of the MPs identified in this study were further confirmed by calculating the Jaccard index between the public MPs and the identified MPs. The maximum MP ( >75%) was used to define the program in epithelial cells, and the proportion of different program epithelial cells was counted.

Inference of cell state by trajectory analysis

To determine the relationship between the two types of epithelial cells, trajectory analysis was performed using the monocle2 R package (v2.14.0) to reveal cell-state transitions14. The ordering genes for the trajectory analysis were selected based on their expression levels (mean expression >1) and variance (dispersion_empirical >2 × dispersion_fit), as recommended by monocle2. The DDRTree() function in monocle2 was used to reduce dimensions with default settings. The cells selected by the algorithm were set as the root state, and the “order” function in monocle2 was applied. Differentially expressed genes (DEGs) that changed along the pseudotime were identified using the differentialGeneTest() function in monocle2. Slingshot15 was utilized to validate the sequential relationship during the transition between Epi-I and Epi-II. Subsets of basal cells, Epi-I, and Epi-II were extracted, and the data were analyzed following the default workflow and parameters, with basal cells designated as the starting point. stLearn, a Python-based package18, was employed for spatial trajectory analysis. We conducted an analysis of the spatial trajectories of Epi-I and Epi-II at the level of each FOV, using default parameter settings.

Co-localization analysis

To quantify the spatial relationships of cell types within given samples and FOVs, we used the Squidpy Python package to calculate the X&Y coordinates and spatial relationships of each cell type within each FOV37. Within each FOV, the boundary coordinates of cells of a given cell type were estimated on a rectangular grid, and the Neighbors enrichment, the centrality scores, and the co-occurrence probabilities were calculated. The neighbor matrix of cells is constructed using the gr.spatial_neighbors() function, then the corresponding features are calculated using gr.nhood_enrichment(), gr.centrality_scores(), and gr.co_occurrence(), respectively. Only cell types with a cell number greater than 10 were analyzed. Subsequently, the average values of these spatial features were calculated for each cell type in 172 FOVs.

Identification of zones

Compared with single-cell data, SMI data contains coordinate information for each cell, allowing for the analysis of the ecological niche of cells. We used the BANKSY R package for domain identification and set the lambda parameter to 0.8, as recommended by the software10, to improve the coefficient of the neighbor matrix. Like cell type clustering, zone classification was performed through the following steps: PCA was used to reduce the dimensionality of the data. Harmony was applied to correct for batch effects and integrate the data. UMAP was used for further dimensionality reduction and visualization. The number of clusters with a final resolution of 0.5 was used for subsequent analysis. Zones were defined based on the proportion of cell types within each zone, with a threshold of greater than 50%. Seven distinct zones were identified: Basal zone, Epi-DC zone, Epi-Fib zone, Epi-Myeloid zone, Plasma zone, Epi-II-dominant zone and Stable Epi zone.

Identification and functional analysis of differentially expressed genes

The FindAllMarkers() function in Seurat was used to identify zone-specific genes with default parameters. P-values were adjusted using the Bonferroni test. The identified marker genes and differentially expressed genes (DEGs) are provided in the Supplementary Data 6. To better understand the functions of these DEGs, we compared them to known signatures from public databases38,39,40. The runGSEA() function was used to scan the Molecular Signatures Database hallmark gene sets and evaluate the enrichment between the DEGs and the signatures in the database. Additionally, the R package clusterProfiler (version 4.12.6) was applied to perform Gene Ontology (GO) enrichment analysis of the DEGs41.

Cell-cell interaction and ligand-receptor analysis

We investigated the cell-cell interactions among diverse cell types using the R package CellChat (v1.1.3), which could quantitatively infer and analyze intercellular communication networks from scRNA-seq data19,42. We mainly utilized the “netVisual_circle” function of CellChat, and the corresponding results were presented via crosstalk maps and ligand-receptor heatmaps. Additionally, the AggregateExpression() function of Seurat package was used to calculate the pseudo-bulk expression of genes. The CellPhoneDB python package20 was employed to validate the findings from CellChat. All parameters were set to the default values, and the input matrices were kept identical to those used in the CellChat package. NicheNet24 was deployed to infer interactions between ligands, receptors, and target genes within the Epi-II-dominant zone. For ligand–receptor interactions, we considered genes expressed in more than 10% of cluster cells. The top 30 differentially expressed ligands from “sending cells” and their corresponding target genes. Ligands ranked in the top 30 by area under the precision–recall curve (AUPR) were utilized to construct an AUPR heatmap. Heatmaps of differentially expressed ligands, target genes, and receptors were drawn using ligand_target_heatmap and ligand_receptor_heatmap.

Statistical analyses

Details of statistical testing for each analysis are included in the corresponding figure legends. For normally distributed data, the two-sided paired t test was conducted for two paired group comparisons. For not normally distributed data, the two-sided unpaired Wilcoxon test was conducted for two independent group comparisons. The Spearman test was performed for correlation analysis. P < 0.05 was considered significant. Statistical analysis of SMI was described in the corresponding section.