Spatial Atlas of Small Cell Lung Carcinoma Reveals Intratumoral Tumor‐State Heterogeneity and an MMP12 ⁺ Macrophage Niche

ABSTRACT

Small‐cell lung carcinoma (SCLC) is a highly aggressive tumor with a dismal prognosis among lung cancers. Treatment is failing because of primary or rapidly developing secondary resistance. Recent molecular studies have suggested four SCLC subtypes with unique therapeutic vulnerabilities—ASCL1, NEUROD1, POU2F3, and YAP. These subtypes still distinguish themselves by considerable intra‐tumoral heterogeneity and the influence of the microenvironmental context, which are both considered to be responsible for treatment resistance. We integrated in situ hybridization‐based spatial transcriptomics with histological and protein‐expression analyses of 18 specimens, comprising 16 SCLCs, one atypical carcinoid, and one normal lung specimen. Systematic case‐by‐case analysis revealed multiple coexisting tumor states, including neuroendocrine and non‐neuroendocrine components, regional hypoxic states, and case‐specific FABP7‐positive and loss‐of‐chromosome‐Y‐associated populations. Integrative transcriptome clustering, copy number estimation, and exploratory trajectory analysis revealed transcriptional relationships among spatially coexisting tumor states, including phenotypically distinct tumor components with molecular features consistent with close clonal relationships in selected heterogeneous tumors. Furthermore, we identified immune signatures and uncovered a previously unrecognized MMP12‐positive tumor‐associated macrophage population, enriched at the tumor interface and spatially associated with necrotic and hypoxic regions. In vitro co‐culture experiments further showed that direct contact with SCLC cells increased MMP12 expression in macrophages, with hypoxia further enhancing its expression under specific conditions. Together, our study delineates the complex spatial architecture of SCLC by defining specific cellular programs and adjacent immune phenotypes that provide an organizational framework for understanding its marked intratumoral heterogeneity.

Spatial transcriptomic profiling of SCLC revealed marked intratumoral heterogeneity, including coexisting neuroendocrine and non‐neuroendocrine states and spatially heterogeneous hypoxia, and identified an MMP12 ⁺ macrophage population enriched at hypoxic and necrotic tumor interfaces. These findings highlight the spatial organization of tumor cell states and tumor–microenvironment interactions in SCLC.

Abbreviations

CAF

cancer associated fibroblast

CNV

copy number variation

FRC

fibroblastic reticular cell

LUAD

lung adenocarcinoma

LUSC

lung squamous cell carcinoma

NE

neuroendocrine

NNE

non neuroendocrine

NSCLC

non small cell lung carcinoma

PBMo

peripheral blood monocyte

PCA

principal component analysis

SCLC

small cell lung carcinoma

TAM

tumor‐associated macrophage

TME

tumor microenvironment

UMAP

Uniform Manifold Approximation and Projection

1. Introduction

Small‐cell lung cancer (SCLC) accounts for approximately 15% of all lung cancers and is characterized by an aggressive growth pattern, early metastasis, and an extremely poor prognosis [1, 2]. Genome sequencing studies have revealed frequent inactivation of the tumor suppressor genes TP53 and RB1, along with recurrent loss of NOTCH signaling, as key genetic alterations [3]. The absence of key druggable driver alterations has limited the development of effective targeted therapies. Recently, comprehensive transcriptomic analyses of tumor samples, cell lines, and patient‐derived xenografts have established a new SCLC classification comprising four major phenotypes: ASCL1, NEUROD1, POU2F3, and an inflamed subtype (formerly termed YAP‐driven). ASCL1‐ and NEUROD1‐high tumors represent neuroendocrine (NE) subtypes, whereas POU2F3‐driven and inflamed tumors define non‐neuroendocrine (NNE) subtypes [4, 5, 6]. Importantly, each subtype is considered to have unique therapeutic vulnerabilities [7]. Clinical outcomes may also differ among subtypes, although findings have been inconsistent across studies. For instance, POU2F3 expression, a marker of NNE subtypes, has been associated with poor clinical outcomes [8, 9, 10]. Together, these observations illustrate how subtype‐specific transcriptional programs define the biology of SCLC and may inform the development of new therapeutic strategies.

Although genomic profiling and molecular subtyping have improved our biological understanding of SCLC, this knowledge has not yet been fully translated into diagnostic pathology or clinical practice. This problem is illustrated by the fact that the efficacy of DLL3‐targeting bispecific T‐cell engagers is not restricted to the ASCL1 subtype and that their clinical use does not require subtype‐based biomarker selection [11]. Consistent molecular classification of SCLC is complicated by its marked intratumoral heterogeneity and cellular plasticity over time. At the microscopic level, approximately one third of SCLC cases show combined histology (combined SCLC; cSCLC). These tumors display phenotypic heterogeneity and lineage plasticity, as exemplified by distinct patterns of YAP expression [12]. Consistent with these findings, mouse models have further demonstrated that YAP, NOTCH, and MYC signaling can promote NE‐to‐NNE fate conversion and that cooperation between distinct lineage states promotes metastatic progression and chemoresistance [13]. Nevertheless, the spatial organization and transcriptional relationships among coexisting tumor states have not been systematically characterized in human SCLC tissues. Defining these relationships may provide insights into molecular subtype heterogeneity, although the directionality of lineage change cannot be determined from cross‐sectional tissue samples alone.

Extending beyond tumor‐intrinsic heterogeneity, accumulating evidence highlights the essential role of the tumor microenvironment (TME) in shaping SCLC biology and therapeutic vulnerability. Among various immune cell populations in the TME, tumor‐associated macrophages (TAMs) have attracted considerable attention because they contribute to cancer progression in multiple tumor types, including hepatocellular carcinoma, high‐grade serous ovarian carcinoma, and esophageal carcinoma, through diverse modes of interaction with cancer cells [14, 15, 16]. In SCLC, direct interactions between cancer cells and macrophages activate STAT3 signaling and enhance tumor proliferation, invasion, and chemoresistance [17]. Furthermore, SCLC cells can “educate” macrophages toward tumor‐supportive phenotypes, while oncogenic metabolic reprogramming, such as MYC‐driven hypersialylation, further reinforces these interactions [18]. We previously showed that miR‐9‐5p‐high SCLC exhibits a higher stromal area ratio and increased numbers of CD8⁺ T cells and CD163⁻ macrophages in the intratumoral area, indicating subtype‐dependent macrophage architecture [19]. Nevertheless, the phenotypic spectrum and spatial organization of macrophages in SCLC remain incompletely defined.

Single‐cell RNA sequencing (scRNA‐seq) enables detailed characterization of cellular heterogeneity, diverse cell states, and cell–cell interactions within the TME, but its lack of spatial context limits the ability to reconstruct the architectural organization of highly plastic tumors such as SCLC. Spatial transcriptomics overcomes this challenge by capturing gene expression profiles together with spatial coordinates, allowing direct mapping of transcriptional states and cellular interactions in situ. Here, we performed spatial multi‐omics profiling of SCLC and generated a single‐cell spatial atlas that revealed multiple coexisting transcriptional states within individual tumors. By resolving the tumor–immune microenvironment at single‐cell resolution, we also identified a previously unrecognized MMP12 ⁺ macrophage subset enriched at viable tumor interfaces adjacent to necrotic and hypoxic regions. These findings provide a spatial framework for understanding intratumoral tumor‐state heterogeneity and tumor–macrophage interactions in SCLC.

2. Materials and Methods

2.1. Formalin‐Fixed, Paraffin‐Embedded (FFPE) Samples

We collected formalin‐fixed paraffin‐embedded (FFPE) tumor specimens from 48 patients who underwent surgical resection and were diagnosed with SCLC at Kanazawa University Hospital between 2001 and 2021. The histology was confirmed by two independent board‐certified pathologists. Molecular subtyping was performed by immunohistochemistry (IHC) for ASCL1, NEUROD1, POU2F3, and YAP, as previously described [19]. As controls, 24 lung adenocarcinoma (LUAD) and 24 lung squamous cell carcinoma (LUSC) samples were included.

2.2. IHC

IHC was performed as previously described [19]. The primary antibodies are listed in Table S1.

2.3. In Situ Hybridization (ISH) With IHC

RNA ISH combined with IHC was performed using the RNAscope 2.5 HD Reagent Kit‐RED (Advanced Cell Diagnostics, Newark, CA, USA) and the REAL EnVision Detection System (Agilent Technologies, Santa Clara, CA, USA). For ISH, probes targeting human MMP12 (Advanced Cell Diagnostics, #412651) and the positive‐control probe against PPIB (Advanced Cell Diagnostics, #313901) were utilized. For IHC, an anti‐CD68 antibody (R&D Systems, Minneapolis, MN, USA, MAB2040) was applied, followed by an HRP‐conjugated anti‐mouse polymer secondary reagent (Agilent Technologies, K4001). Chromogenic signals for FAST Red and DAB were separated using the Mantra 2 imaging platform (Akoya Biosciences, Marlborough, MA, USA). Five representative fields were captured per case. Images were processed and analyzed using inForm software (Akoya Biosciences), and cells were classified based on DAB (threshold 0.3) and Fast Red (threshold 1.2) intensity levels.

2.4. In situ Gene Expression Profiling

In situ gene expression profiling was performed on 18 FFPE specimens, including 16 SCLC specimens, one atypical carcinoid, and one normal lung specimen, using the Xenium In Situ platform (10× Genomics, Pleasanton, CA, USA) with the commercially available Xenium Prime 5K Human Pan‐Tissue and Pathways Panel as previously described [20]. Downstream analyses were performed using Seurat (v5.1.0) [21]. Spatial visualization was performed using Xenium Explorer (v4.0.0). Normalization, scaling, and variable feature selection were performed using SCTransform. Differentially expressed genes (DEGs) were identified using FindAllMarkers function. To evaluate spatial proximity, the minimum Euclidean distance between each cell and all cells of another cell type within the same tissue section was calculated. Sample‐wise mean distances were then averaged to derive pairwise intercellular distances. For trajectory analysis, relevant cell populations were subset and re‐embedded in UMAP space. UMAP embeddings and gene expression matrices were imported into monocle3 (v1.4.26) for exploratory trajectory analysis [22]. Copy‐number variation patterns were estimated using inferCNV (v1.14.2) [23]. ASCL1‐positive cells were designated as the computational root for pseudotime visualization. Genomic regions exhibiting copy‐number gain or loss were identified for each chromosomal short and long arm. The proportion of chromosomal arms showing significant CNV events was then quantified and defined as the CNV‐score. To investigate transcription factors potentially regulating DEGs, ChIP enrichment analysis was performed using ChEA3 [24], and the mean rank across ChEA3 datasets was used as the enrichment score.

2.5. Tumor Heterogeneity Scoring (H‐Index)

Tumor transcriptomic heterogeneity was quantified using a diversity‐score approach conceptually consistent with the method described by Ma et al. [25] For each sample, the H‐index was defined as the mean Euclidean distance of individual tumor cells from the sample centroid in principal‐component space. The first 50 principal components were used for calculation.

$X_{\mathit{mn}}$: Score of the nth cell on the mth PC

$$\overline{X_m}=\frac{{\sum }_n{X}_{\mathit{mn}}}{n}$$

$$\mathrm{H}\_\text{index}=\frac{{\sum }_n\sqrt[2]{{\sum }_m{\left(X_{\mathit{mn}}-X_m\right)}^2}}{n}$$

Samples with H‐index above the median were classified as the high‐heterogeneity group, and the remainder as the low‐heterogeneity group.

2.6. Cell Culture

Three SCLC cell lines, DMS53, DMS454, and WA‐hT, were obtained from ATCC (Manassas, VA, USA), ECACC (Salisbury, UK), and the RIKEN BioResource Center (Tsukuba, Japan), respectively and maintained as previously described [26]. Peripheral Blood Monocyte (PBMo)‐derived macrophages were generated from healthy volunteers using previously described protocols [27, 28]. Direct and indirect co‐culture experiments were performed as previously described [14]. For hypoxic conditions, the BIONIX hypoxic culture kit (Sugiyama‐Gen, Tokyo, Japan) was applied and the oxygen concentration was maintained at 2% ± 0.5% O₂. After 48 h of culture, cells were collected for subsequent analyses.

2.7. MMP12 Knockdown in the Direct Co‐Culture Model

For the direct co‐culture experiments with MMP12 knockdown, macrophages were transfected with siRNA as previously described [29]. Briefly, following 6 days of differentiation with GM‐CSF and M‐CSF, the culture medium was replaced with serum‐free RPMI‐1640 containing Lipofectamine RNAiMAX (Invitrogen) and either an siRNA targeting human MMP12 (siMMP12; sc‐41557; 20 pmol; Santa Cruz Biotechnology) or MISSION siRNA Universal Negative Control #1 (siNC; SIC001; 20 pmol; Sigma‐Aldrich), according to the manufacturers' instructions. After 12 h of siRNA transfection, the medium was replaced with RPMI‐1640 containing 10% FBS, and the macrophages were directly co‐cultured with DMS53 cells for 24 h. Following co‐culture, DMS53 cells and macrophages were separated as described above and used for subsequent experiments.

2.8. Transwell Invasion Assay

Macrophage invasion was assessed using a Transwell assay as previously described [14]. Briefly, following separation using the autoMACS system, macrophages transfected with siNC or siMMP12 were seeded at (1 × 10⁶) cells per well into the upper compartment of a Matrigel‐coated invasion chamber with an 8‐μm pore‐size membrane (BD Falcon) in 2 mL of serum‐free RPMI‐1640. The lower compartment was filled with 3 mL of RPMI‐1640 containing 1% FBS. After incubation for 24 h, non‐invading cells remaining on the upper surface of the membrane were removed using a cotton swab. Cells that had invaded through the membrane were stained using the Diff‐Quik Kit (Sysmex, Kobe, Japan). Five randomly selected fields were imaged at ×100 magnification, and the number of invading macrophages per field was counted.

2.9. Quantitative Real‐Time PCR (qRT‐PCR)

qRT‐PCR was performed as previously described [30]. Primer sequences were as follows: MMP12: forward, 5′‐GCTGTCACTACCGTGGGAAA‐3′; reverse, 5′‐GCAGCTTCAATGCCAGATGG‐3′; GAPDH: forward, 5′‐TGAGATGTCTTCAGGGAGTG‐3′; reverse, 5′‐AGCCAGGTATTTGTGGTTCG‐3′.

2.10. Statistics

Differences between two groups were evaluated using Student's t‐test, and for multiple comparisons, Benjamini–Hochberg correction was applied. Pearson's correlation coefficient was used to assess correlations between two parameters. Survival differences in disease‐free survival (DFS) were evaluated using the log‐rank test. Statistical significance was defined as p < 0.05.

3. Results

3.1. Spatial Transcriptomic Profiling Reveals Cellular Composition and Molecular Subtype Diversity in SCLC

We performed in situ spatial transcriptomic analyses with the Xenium platform on 18 specimens, including 16 SCLC, one atypical carcinoid, and one normal lung (Figure 1A). The 16 SCLC cases were classified into molecular subtypes based on IHC for ASCL1, NEUROD1, POU2F3, and YAP. This resulted in multiple molecular subtypes, including 5 SCLC‐A, 1 SCLC‐N, 3 SCLC‐A/N, 5 SCLC‐P, 1 SCLC‐PA, and 1 SCLC‐Unclassified (Table S2). After integration of datasets and filtering out low‐quality cells, we obtained 947,230 cells across the 18 samples (Figure 1B). These were classified into four major groups based on canonical marker expression: epithelial cells (n = 685,890), immune cells (n = 122,982), mesenchymal cells (n = 80,299), and necrotic cells (n = 58,059) (Figure 1C,D). To classify tumor subtypes in detail, we extracted epithelial cells and performed reclustering. Using mRNA levels of the four lineage‐defining transcription factors, we assigned SCLC subtypes (Figure 2A), with high concordance with the IHC‐based classification (Table S2). Furthermore, by incorporating genes related to phenotypic shifts, such as members of the MYC family [31], we identified more refined tumor subpopulations (Figures 2B,C and S1A). A MYCN–high subset was observed within both SCLC‐A/N and SCLC‐NNE tumors. Copy‐number variation analysis using inferCNV suggested chromosome 2 amplification—where MYCN is located—in SCLC‐A/N, indicating genomic amplification–driven overexpression (Figure S1B). In contrast, SCLC‐NNE lacked such amplification (Figure S1B), implying that MYCN upregulation in this subtype is driven by transcriptomic or epigenetic regulation rather than genomic copy‐number gain.

FIGURE 1.

Integration of the spatial transcriptomic dataset and global cell‐type annotation using canonical markers. (A) Schematic overview of the present study. (B) UMAP plot of 947,230 cells across the 18 samples. (C) UMAP plot of the cells colored by the four annotated groups: Epithelial cells (n = 685,890), immune cells (n = 122,982), mesenchymal cells (n = 80,299), and necrotic cells (n = 58,059). (D) Representative case. Left: H&E staining. Middle: Multiplex immunofluorescence staining used for segmentation. Right: Cells colored according to the annotations in figure C.

FIGURE 2.

Molecular subtyping by spatial transcriptomics. (A) Representative H&E images, multiplex immunofluorescence images, and gene expression patterns of ASCL1, NEUROD1, POU2F3, YAP1, and ANGPTL3 across each SCLC subtype, atypical carcinoid, and normal lung. Each dot represents the location of an individual transcript. (B) UMAP plot of epithelial cells, colored by cell clusters identified using the four transcription factors and canonical markers. (C) Dot plot showing the expression of the four transcription factors and canonical markers across each cell cluster.

3.2. Systematic Characterization of Intratumoral Transcriptional Heterogeneity in SCLC

We next systematically evaluated intratumoral transcriptional heterogeneity across all 16 SCLC cases. For each case, we summarized the H‐index, number of transcriptionally defined tumor clusters, number and identity of distinct tumor states, and their relationship to molecular subtype composition (Table 1, Figure S2). The H‐index ranged from 9.04 to 22.62. Several patterns of heterogeneity were observed, most prominently the coexistence of NE and NNE tumor states within the same specimen. Spatially distinct VEGFA/NDRG1‐positive hypoxic tumor states were also identified in multiple cases. In addition, a distinct FABP7‐positive tumor state was identified in one case (Figure S2), whereas another case contained a tumor component characterized by marked downregulation of Y chromosome‐associated genes such as KDM5D, SRY, and CD99, consistent with loss of chromosome Y (LOY) (Figure S2). In an exploratory survival analysis, patients in the high H‐index group showed numerically shorter disease‐free survival than those in the low H‐index group; however, the difference was not statistically significant (p = 0.24; Figure S3). Collectively, these findings indicate that intratumoral transcriptional heterogeneity in SCLC is multifactorial, with NE/NNE coexistence representing a prominent pattern alongside regional hypoxia and case‐specific transcriptional alterations.

TABLE 1.

Systematic case‐by‐case characterization of intratumoral transcriptional heterogeneity.

Sample	H‐index	No. of tumor clusters	No. of distinct tumor components	Additional heterogeneous tumor components (proportion)	Coexisting NE and NNE components	Hypoxic tumor state	LOY(+) subpopulation
SCLC‐Unclassified #1	22.62	6	2	NNAT + NE (88.8%), NNE (11.2%)	Yes
SCLC‐A/N #3	16.39	4	2	Hypoxic state (54.0%)		Yes
SCLC‐NNE #6	15.96	5	2	NE (59.0%)	Yes
SCLC‐NNE #1	13.80	5	3	NE (21.3%), LUSC (17.7%)	Yes
SCLC‐A #3	13.65	8	2	Hypoxic state (44.3%)		Yes
SCLC‐NNE #4	13.59	4	3	Hypoxic state (34.2%), FABP7 + NNE (3.5%)		Yes
SCLC‐A #4	13.57	6	2	LOY(−) (18.0%), LOY(+) (82.0%)			Yes
SCLC‐A #2	12.25	6	1
SCLC‐A/N #2	11.65	5	1
SCLC‐NNE #3	11.42	6	2	Hypoxic state (27.1%)		Yes
SCLC‐A #5	11.14	5	2	Hypoxic state (21.7%)		Yes
SCLC‐NNE #5	11.05	4	1
SCLC‐N #1	10.59	4	2	Hypoxic state (20.6%)		Yes
SCLC‐A/N #1	10.36	3	1
SCLC‐A #1	9.20	4	2	Hypoxic state (1.5%)		Yes
SCLC‐NNE #2	9.04	8	2	Hypoxic state (15.0%)		Yes

Note: The H‐index, number of transcriptionally defined tumor clusters, number of distinct tumor components, additional heterogeneous tumor components and their proportions, and the presence of coexisting NE and NNE components, hypoxic tumor states, and LOY‐positive subpopulations are summarized for each case. Cases are listed in descending order of H‐index. Tumor components were defined according to their characteristic transcriptional profiles. In the “Additional heterogeneous tumor components” column, the reference tumor component was generally omitted.

Abbreviations: LOY, loss of chromosome Y; NE, neuroendocrine; NNE, non‐neuroendocrine.

3.3. A Highly Heterogeneous SCLC Harbors Clonally Related NE and NNE Tumor States

To further characterize the coexistence of NE and NNE tumor states, we focused on the case with the highest H‐index (22.62; Figure 3A). This tumor consisted of two transcriptionally distinct components; a POU2F3‐positive/NREP‐high NNE component and another component lacking expression of the conventional subtype markers ASCL1, NEUROD1, POU2F3, and YAP1, which was initially classified as an unclassified transcriptional state. Morphologically, both components appeared similar (Figure 3A), and both showed diffuse p53 overexpression and loss of Rb protein expression (Figure S4A). DEG analysis identified NNAT as a highly upregulated gene (Figure 3B). NNAT‐positive cells expressed both NCAM1 and NKX2‐1 (Figure 3C,D), supporting their NE phenotype. NNAT protein expression was further validated by immunohistochemistry (Figure S4B). Notably, MYCL expression was higher in NNAT‐positive cells, whereas NOTCH3 expression was elevated in POU2F3‐positive cells (Figure 3C,D). CNV analysis demonstrated chromosome 1p amplification, encompassing MYCL, in both populations, supporting a close clonal relationship between the two tumor states. In contrast, the POU2F3‐positive population showed a higher proportion of cells with chromosome 19p amplification, encompassing NOTCH3, than the NNAT‐positive population (0.90 vs. 0.19; Figure 3E). These findings suggest that differential NOTCH3‐associated copy‐number alterations may be associated with the phenotypic divergence of these clonally related NE and NNE tumor states.

FIGURE 3.

Identification of an NNAT‐positive tumor state coexisting with a POU2F3‐positive tumor state in SCLC. (A) Left: Two transcriptomically distinct clusters identified within a single core. Middle: H&E staining. Right: Magnified images of each component. A/N/P/Y, ASCL1/NEUROD1/POU2F3/YAP1. (B) Volcano plot comparing component #1 and component #2. The x‐axis indicates the log₂ fold change, and the y‐axis indicates the –log₁₀ adjusted p‐values. Green dots represent genes with adjusted p‐values < 5 × 10⁻¹⁰ and |log₂ fold change| > 1, and red dots represent genes with adjusted p‐values < 1 × 10⁻¹⁰⁰ and |log₂ fold change| > 1. (C) Spatial plots showing the expression of representative genes identified by DEG analysis. (D) Dot plot showing the expression of representative genes identified by DEG analysis. (E) Estimated copy‐number variation assessed by inferCNV for NNAT‐high cells (red) and POU2F3‐high cells (blue). The reference was set to normal lung epithelium (gray). The region highlighted in green and red indicates chromosome 1 and 19, respectively.

3.4. Spatially Distinct Tumor States With a Common Clonal Relationship in Combined SCLC

Another heterogeneous sample (H‐index = 13.80) contained three spatially distinct tumor populations: an ASCL1‐positive population, a POU2F3‐positive population, and a third population lacking expression of ASCL1, NEUROD1, POU2F3, and YAP1 (Figure 4A). On closer histopathological and molecular examination, the third population showed morphology distinct from conventional SCLC, expressed TP63 and CLDN1 at the mRNA level (Figures 4B,C and S5A), and exhibited diffuse p40 protein expression (Figure S5B). These findings supported squamous differentiation, leading to the classification of this case as cSCLC with a squamous cell carcinoma component. The ASCL1‐positive population corresponded to SCLC‐A with high DLL3 expression, whereas the POU2F3‐positive population expressed NREP. Further differential expression analysis identified integrin β4 (ITGB4) as an additional marker of the POU2F3‐positive population (Figure S5C). All three components showed loss of p53 and Rb protein expression (Figure S5D). In addition, they exhibited similar inferred CNV profiles and comparable CNV scores (ASCL1‐positive, 0.79; POU2F3‐positive, 0.83; squamous, 0.84) (Figure S5E), supporting a common clonal relationship. An exploratory trajectory analysis using Monocle 3 suggested a transcriptional continuum among the three tumor populations, although their cellular origin and lineage directionality remained unresolved (Figure S5F). To further characterize the ASCL1‐positive population in this case, we compared it with ASCL1‐positive cells from other SCLC‐A cases (Figure S5G,H). ChEA3 analysis of genes specifically upregulated in this population identified MYBL2 as the top candidate regulator (Figure S5I), suggesting an association between MYBL2‐regulated transcriptional programs and the distinct phenotype of this component. Collectively, these findings demonstrate marked phenotypic heterogeneity among clonally related tumor components in cSCLC.

FIGURE 4.

Spatially distinct, clonally related tumor states in combined SCLC with a squamous cell carcinoma component. (A) Left: Three transcriptomically distinct clusters identified within a single core. Middle: H&E staining. Right: Magnified images of each component. A/N/P/Y, ASCL1/NEUROD1/POU2F3/YAP1; LUSC, lung squamous cell carcinoma. (B) Spatial plots showing specific genes identified by DEG analysis among the three regions. (C) UMAP plot of the three clusters after reclustering.

3.5. Identification of an MMP12 ‐Positive Macrophage Subset Spatially Associated With SCLC Tumor Cells

We next examined the TME of SCLC. Mesenchymal cells were reclustered (Figure S6A,B), revealing five fibroblast subsets, including three CAF subtypes, along with smooth‐muscle and endothelial cells. Immune cells were reclustered and categorized into monocytes/macrophages, T cells, B cells, and plasma cells (Figure S6C,D). Within the monocytes/macrophages compartment, twelve subsets were identified, including IFN‐TAM, LA‐TAM, Angio‐TAM, alveolar macrophages, and an MMP12 ⁺ macrophage subset that is a major population among TAMs (Figure 5A,B). Among T cells, five subsets and one NK‐cell population were identified (Figure S6E,F). In total, 42 cell types were defined across all samples, comprising 11 epithelial, 20 immune, 9 mesenchymal, 1 unclassified, and 1 necrotic population (Figure S6G). To explore spatial relationships among mesenchymal, immune, and tumor cells, we calculated pairwise distances between all cell types (Figure 5C). In most cases, the nearest neighboring cell was a tumor cell, whereas the second‐closest often belonged to the same lineage (Table S3). Distinct tissue architectures also influenced these patterns; for instance, B cells were often adjacent to CD4⁺ T cells, follicular helper T cells, and fibroblastic reticular cells (FRCs), reflecting tertiary lymphoid structures. Among all stromal and immune populations, MMP12 ⁺ macrophages were the cell type most closely associated with tumor cells, aside from the tumor cells themselves (Figure 5C). They were significantly nearer to tumor cells than other macrophage subsets (Figure 5D,E), with approximately 75% located within 15 μm of tumor cells (Figures 5F and S7A).

FIGURE 5.

Spatial analysis identifies MMP12 ⁺ macrophages as a novel subset in SCLC. (A) UMAP plot of monocyte/macrophage populations, colored by 12 distinct clusters. Angio‐TAM, angiogenesis‐associated TAM; IFN‐TAM, interferon‐associated TAM; LA‐TAM, lipid‐associated TAM; TIM, tumor‐infiltrating monocyte. (B) Dot plot showing the expression of representative genes for each macrophage subset. (C) Heatmap showing the normalized pairwise distance among the annotated cell types. Red indicates shorter distances (closer proximity), whereas blue indicates longer distances (greater separation). The distance between tumor cells and MMP12 ⁺ macrophages is highlighted with a red circle. (D) Boxplots show the mean distance (center line) between tumor cells and each macrophage cluster, the interquartile range (box), and whiskers (1.5 × IQR). Individual data points for each sample are overlaid as jittered dots. (E) Left: Representative H&E staining. Right: The corresponding spatial plot. Red cells, white cells, and gray cells indicate MMP12 ⁺ macrophages, other macrophages, and tumor cells, respectively. Green and blue dots indicate MMP12 and CD68 mRNA transcripts, respectively. (F) Bar chart indicating the proportion of each macrophage subset located within < 15 μm, 15–30 μm, or > 30 μm from the nearest tumor cells. (G) Left: Representative H&E staining. Right: The corresponding spatial plot. Red cells, white cells, and gray cells indicate MMP12 ⁺ macrophages, other macrophages, and tumor cells, respectively. Green and blue dots indicate MMP12 and NDRG1 mRNA transcripts, respectively. (H) Double staining with immunohistochemistry for CD68 (left) and in situ hybridization for MMP12 (right) after color separation using the Mantra2 imaging system. Highlighted with red circles are double‐positive cells, while green circles indicate CD68‐positive, MMP12‐negative cells. (I) Boxplots show the proportion of MMP12 ⁺ macrophages among all CD68⁺ macrophages across lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), and SCLC. The center line indicates the mean, the box indicates the interquartile range (IQR), and the whiskers represent 1.5 × IQR. Individual data points for each sample are overlaid as jittered dots. (J) Quantitative PCR for MMP12 after indirect or direct coculture and/or hypoxic treatment. Bars indicate the mean and SE. *p < 0.05; ***p < 0.001. N.S., not significant.

3.6. Tumor‐Cell Contact and Hypoxic Stress Are Associated With MMP12 Expression in Macrophages

Spatial mapping showed that MMP12 ⁺ macrophages frequently accumulated at interfaces between viable tissue and necrotic regions within tumor nests (Figures 5G and S7B). MMP12 ⁺ macrophages exhibited foam‐cell‐like morphology (Figure S7C) and tended to be larger than other macrophage populations (Figure S7D). Tumor cells adjacent to MMP12‐positive macrophages showed increased expression of NDRG1, VEGFA, and LDHA, indicating that these macrophages were enriched at viable tumor interfaces adjacent to hypoxic and necrotic regions (Figure 5G). To validate these findings, RNAscope ISH for MMP12 combined with CD68 IHC was performed. After chromogenic separation using the Mantra2 imaging system, analysis of 48 SCLC cases confirmed that MMP12 ⁺ macrophages were mainly located near tumor regions (Figure 5H). The proportion of MMP12 ⁺ cells among CD68⁺ macrophages did not differ significantly among molecular subtypes (data not shown), but was significantly higher in SCLC than in LUAD, and higher, though not significantly, than in LUSC (Figure 5I). Some MMP12 ⁺ macrophages expressed lipid‐metabolism genes, such as LIPA, as well as inflammation‐related genes, such as CXCL9 (Figure S7E). Reanalysis of a public SCLC scRNA‐seq dataset (GSE281740 [32]) likewise identified MMP12‐expressing macrophages distributed across multiple macrophage subsets—particularly LA‐TAMs and IFN‐TAMs (Figure S7F,G). These cells similarly expressed lipid‐ and inflammation‐related genes (Figure S7H), consistent with our data. Finally, to assess whether MMP12 ⁺ macrophages could be recapitulated in vitro, we established a direct co‐culture system combining SCLC cell lines with PBMo‐derived macrophages under hypoxic conditions. MMP12 expression was increased in macrophages directly co‐cultured with SCLC cells across all three cell lines tested, and hypoxia further increased MMP12 expression in two cell lines (Figure 5J). To explore the potential functional significance of MMP12, we subsequently performed a macrophage invasion assay following siRNA‐mediated MMP12 knockdown. The mean number of invading macrophages was approximately 30% lower in the siMMP12 group than in the siNC group, but the difference was not statistically significant (p = 0.20; Figure S8). These results indicate that direct tumor‐cell contact and hypoxic stress are associated with increased MMP12 expression in macrophages under the tested conditions.

4. Discussion

In this study, systematic spatial profiling of SCLC revealed substantial intratumoral heterogeneity, including the coexistence of NE and NNE tumor states, regional hypoxic states, and case‐specific transcriptional and genomic alterations. In selected heterogeneous tumors, integrated transcriptomic and inferred CNV analyses were consistent with close clonal relationships among spatially distinct tumor states. We also identified an MMP12‐positive macrophage population enriched at viable tumor interfaces adjacent to necrotic and hypoxic regions. In vitro co‐culture experiments showed that tumor‐cell contact and hypoxic stress were associated with increased MMP12 expression in macrophages. A graphical summary of the major findings of this study is presented in Figure 6.

FIGURE 6.

Graphical summary of the study. Spatial transcriptomic profiling of SCLC revealed marked intratumoral heterogeneity, including coexisting neuroendocrine and non‐neuroendocrine states and spatially heterogeneous hypoxia, and identified an MMP12 ⁺ macrophage population enriched at hypoxic and necrotic tumor interfaces. These findings highlight the spatial organization of tumor cell states and tumor–microenvironment interactions in SCLC.

Intratumoral heterogeneity is a major determinant of therapeutic resistance in SCLC and can arise from genomic, epigenetic, transcriptional, proteomic, and metabolic diversity [33]. Our systematic case‐by‐case analysis revealed several sources of transcriptional heterogeneity, including the coexistence of NE and NNE tumor states, regional hypoxic states, and case‐specific FABP7‐positive and LOY‐associated populations. The H‐index was used in the present study as a descriptive transcriptome‐based measure of heterogeneity. It provided a simple quantitative framework for comparing the degree of transcriptomic heterogeneity across cases and identifying tumors with particularly complex transcriptional architecture. Although further validation in larger cohorts will be required, the H‐index may serve as a useful descriptive metric for case‐level assessment of transcriptomic heterogeneity.

One high‐heterogeneity case contained spatially distinct ASCL1‐positive, POU2F3‐positive, and LUSC components and was consistent with cSCLC with a squamous cell carcinoma component. We previously reported heterogeneous, often mosaic YAP1 expression in combined SCLC, highlighting phenotypic diversity within this pathological entity [12]. Our spatial analysis extends this framework by resolving coexisting ASCL1‐positive, POU2F3‐positive, and squamous tumor components. However, the present data do not establish a YAP1‐mediated mechanism underlying their coexistence. Shared loss of p53 and Rb protein expression together with similar inferred CNV profiles supported a close clonal relationship among these components. However, our cross‐sectional spatial transcriptomic data cannot determine whether these states arose through sequential phenotypic change or parallel divergence from a common precursor. This distinction is particularly relevant in light of Ireland et al., who demonstrated in genetically engineered mouse models that basal cells can generate neuroendocrine and tuft‐like SCLC states along distinct lineage trajectories [34]. Thus, our findings should be interpreted as evidence of clonally related phenotypic heterogeneity rather than a defined directional transition. The ASCL1‐positive component also exhibited increased MYBL2 expression. Given the reported association of MYBL2 with genomic instability in LUAD [35, 36], this finding may reflect a transcriptional program associated with the distinct phenotype of this tumor component, although its functional significance in SCLC remains to be determined.

We also identified a case containing an NNAT‐positive NE component lacking expression of the conventional subtype markers together with a POU2F3‐positive NNE component. In a previous study, NNAT was observed in areas lacking NE morphology and was proposed as a marker differentiating LCNEC from SCLC [37]. In contrast, the NNAT‐positive component in our case expressed NE markers, including NCAM1 and NKX2‐1, despite lacking expression of ASCL1, NEUROD1, POU2F3, and YAP1. These findings suggest that NNAT‐positive SCLC may represent a distinct transcriptional tumor state not captured by the conventional ASCL1/NEUROD1/POU2F3/YAP1 framework. Ireland et al. also identified an Atoh1‐positive state in basal‐derived SCLC [34]. However, the NNAT‐positive tumor population in our study showed no appreciable ATOH1 expression, suggesting that it is distinct from the previously described Atoh1‐positive state. The NNAT‐positive and POU2F3‐positive components shared chromosome 1p amplification encompassing MYCL, supporting a close clonal relationship, whereas chromosome 19p amplification encompassing NOTCH3 was more frequent in the POU2F3‐positive component. These NOTCH3‐associated copy‐number differences may reflect phenotypic divergence between the two tumor states, but their functional and temporal roles remain to be established.

MMP12 ⁺ macrophages have previously been reported in association with liver injury [38, 39] and smoking [40], and have also been observed in NSCLC [41], but have not been studied in SCLC. In the present study, MMP12 ⁺ macrophages were consistently enriched at viable tumor interfaces adjacent to necrotic and hypoxic regions. Their foam‐cell–like morphology and distinct transcriptional features suggest that they represent a macrophage state associated with the local tumor microenvironment. Consistent with this interpretation, direct co‐culture with SCLC cells and hypoxic conditions increased MMP12 expression in macrophages in vitro. Their localization to necrotic and hypoxic tumor interfaces, microenvironments linked to poor prognosis [42] and epithelial–mesenchymal transition [43], suggests that MMP12 ⁺ macrophages are associated with tumor regions exhibiting features relevant to treatment resistance in SCLC. Although MMP12 knockdown reduced macrophage invasion, the effect was not statistically significant, and further studies are needed to establish its functional significance.

The proportion of MMP12 ⁺ macrophages was substantially higher in our spatial dataset than in a published scRNA‐seq dataset (31.9% vs. 5.4% in GSE281740). This discrepancy may, at least in part, reflect methodological differences between spatial transcriptomics and dissociation‐based single‐cell approaches. MMP12 ⁺ macrophages were relatively large and frequently located within extracellular matrix‐rich regions (Figure S7C,D), features that may affect their recovery during enzymatic dissociation and size filtering. In contrast, spatial transcriptomics preserves tissue architecture and enables characterization of macrophage states in their native histological context. Potential technical effects associated with necrotic regions should also be considered when comparing prevalence estimates across platforms. These findings highlight the value of spatial approaches for characterizing the SCLC microenvironment.

This study has several limitations. First, the cohort size was small, with only 16 SCLC cases subjected to spatial profiling. Validation in larger cohorts will be necessary. Second, although we analyzed two cores per case to reduce sampling bias, this approach may not fully capture whole‐tumor spatial heterogeneity. Validation using whole‐section profiling will be required. Third, because the spatial transcriptomic assay used in this study relies on a targeted panel of approximately 5000 genes rather than whole‐transcriptome coverage, transcripts and pathways outside the panel were not assessed. Finally, the functional significance of MMP12‐positive macrophages remains incompletely defined, and additional perturbation studies will be required to determine whether MMP12 itself contributes to macrophage function or SCLC biology.

In conclusion, our spatial transcriptomic analysis revealed extensive intratumoral heterogeneity in SCLC, including phenotypically distinct tumor states, and identified an MMP12‐positive macrophage population spatially associated with hypoxic and necrotic tumor interfaces. These findings provide a spatial framework for understanding tumor‐state heterogeneity and tumor–macrophage interactions in SCLC and establish a foundation for future mechanistic and translational studies.

Author Contributions

Naozane Nomura: methodology, investigation, formal analysis, writing – original draft. Naoe Jimbo: investigation, validation. Takaaki Nakanishi: investigation, validation. Yukinobu Ito: resources. Maki Sakaguchi: investigation, validation. Daichi Maeda: supervision. Hiroko Ikeda: resources. Tsukasa Ueda: resources. Hayato Koba: resources. Seiji Yano: supervision. Isao Matsumoto: resources. Patrick Micke: supervision, writing – review and editing. Takayuki Kodama: supervision. Manabu Shigeoka: supervision. Yu‐ichiro Koma: supervision. Masafumi Horie: conceptualization, methodology, supervision, project administration, writing – review and editing.

Funding

This work was supported by the Japan Agency for Medical Research and Development (AMED) (JP256f0137011); Grants‐in‐Aid for Scientific Research (KAKENHI) from the Japan Society for the Promotion of Science (JSPS) (23K07649); the Takeda Science Foundation; the Japan Lung Cancer Society; and the Kobayashi Foundation for Cancer Research to M.H. This work was also supported by JSPS KAKENHI Grant Numbers 16H06279 (PAGS), JP22H04925 (PAGS), and JP22K06938 (PAGS).

Ethics Statement

Approval of the research protocol by an Institutional Review Board: This study was approved by the Institutional Review Board of Kanazawa University (approval number: 12644‐2) and was conducted in accordance with the Declaration of Helsinki.

Consent

Written informed consent was obtained from all participants prior to inclusion in the study.

Conflicts of Interest

Seiji Yano is an editorial board member of Cancer Science. Other authors do not have COI.

Supporting information

Figure S1: Characterization of tumor cell subsets. (A) Dot plot showing the expression of representative genes for each tumor subset. The dot size indicates the percentage of expressing cells, and the color represents the average expression within each cluster. (B) Estimated copy‐number variation assessed for each tumor subset. The reference was set to normal lung epithelium (green).

Figure S2: Spatial and transcriptional characterization of heterogeneous tumor components in five additional SCLC cases with high H‐index values. Five SCLC cases with high H‐index values (13.57–16.39) that are not presented in detail in Figures 3 and 4 are shown. For each case, the left panel shows the spatial distribution of the annotated tumor components within the tissue, the middle panel shows their distribution on the UMAP, and the right panel shows the expression of representative marker genes. The cases include SCLC‐A/N #3, comprising NE and hypoxic states; SCLC‐NNE #6, comprising NE and NNE components; SCLC‐A #3, comprising NE and hypoxic states; SCLC‐NNE #4, comprising NNE, hypoxic, and FABP7‐positive NNE components; and SCLC‐A #4, comprising LOY‐negative and LOY‐positive subpopulations. In the dot plots, dot size represents the percentage of cells expressing each gene, and color intensity represents the scaled average expression level. The other two cases with high H‐index values are presented in Figures 3 and 4.

Figure S3: Disease‐free survival stratified by H‐index in patients with SCLC Kaplan–Meier analysis of SCLC patients stratified into two groups based on the heterogeneity score (H‐index). DFS, disease‐free survival.

Figure S4: Immunohistochemical validation of p53, Rb, and NNAT expression in the case shown in Figure 3. (A) Representative immunohistochemical images showing p53 and Rb expression. (B) Representative immunohistochemical images showing NNAT expression.

Figure S5: Characterization of the case shown in Figure 4. (A) Volcano plot comparing CLDN1‐positive cells with other SCLC cells. The x‐axis indicates the log₂ fold change, and the y‐axis indicates the –log₁₀ adjusted p‐value. Green dots represent genes with adjusted p‐values > 5 × 10⁻¹⁰⁰ and |log₂ fold change| > 1, and red dots represent genes with adjusted p‐values < 5 × 10⁻¹⁰⁰ and |log₂ fold change| > 1. (B) Representative immunohistochemical images for p40. (C) Representative immunohistochemical images for integrin β4. (D) Representative immunohistochemical images for p53 and Rb. (E) Estimated copy number variation assessed by inferCNV for ASCL1‐high cells (blue), POU2F3‐high cells (green) and LUSC cells (red). The reference was set to normal lung epithelium (gray). (F) Exploratory transcriptional trajectory analysis using Monocle 3. Left: pseudotime visualized on the UMAP; ASCL1‐positive cells were designated as a user‐defined computational root for pseudotime visualization (pseudotime = 0). Right: pseudotime visualized on the tissue section. (G) The three tumor cell clusters from the case shown in Figure 4 mapped onto the UMAP of all 18 samples, shown in red. (H) ASCL1‐high cells from the case shown in Figure 4 (red) and ASCL1‐high cells from other samples (blue) are shown on the UMAP. (I) Volcano plot comparing ASCL1‐high cells in this case with ASCL1‐high cells from other cases. The x‐axis indicates the log₂ fold change, and the y‐axis indicates the –log₁₀ adjusted p‐value. Blue dots represent genes with adjusted p values < 5 × 10⁻¹⁰, and red dots represent genes with adjusted p‐values < 5 × 10⁻¹⁰ and |log₂ fold change| > 1. Genes identified by ChIP enrichment analysis were highlighted.

Figure S6: Annotation of mesenchymal and immune cell populations in the spatial transcriptomic dataset. (A) UMAP plot of mesenchymal cells colored by annotated cell subtype. CAF, cancer‐associated fibroblast; FRC, fibroblastic reticular cell. (B) Dot plot showing the expression of representative marker genes across mesenchymal cell subsets. Dot size indicates the percentage of cells expressing each gene, and color intensity represents the average expression level within each subset. (C) UMAP plot of immune cells colored by annotated cell subtype. (D) Dot plot showing the expression of representative marker genes across immune cell subsets. (E) UMAP plot of T cells colored by annotated T‐cell subtype. (F) Dot plot showing the expression of representative marker genes across T‐cell subsets. (G) UMAP plot of all annotated cells in the dataset.

Figure S7: Spatial and transcriptomic characterization of MMP12 ⁺ macrophages in small‐cell lung carcinoma. (A) Boxplots and violin plots showing the distance to the closest tumor cells for each macrophage cluster. The box indicates the interquartile range (IQR), and the whiskers represent 1.5 × IQR. Individual data points for each sample are overlaid as jittered dots. The violin plot illustrates the full distribution of the data. (B) Left: representative H&E staining. Right: the corresponding spatial plot. Red cells, white cells, and gray cells indicate MMP12 ⁺ macrophages, other macrophages, and tumor cells, respectively. Green and blue dots indicate MMP12 and NDRG1 mRNA transcripts, respectively. (C) Enlarged view of the boxed area in panel B. (D) Boxplots showing the cell size of MMP12 ⁺ macrophages and other macrophages. Each dot represents the average cell size for an individual case. (E) Feature plots showing the expression of MMP12, LIPA, CXCL9, and CXCL10 within macrophage populations. (F) UMAP plot of the macrophage subset from SCLC in a published dataset (GSE281740). (G) UMAP plot of the macrophages in the published SCLC dataset. Red dots represent MMP12‐high macrophages and blue dots represent MMP12‐low macrophages. (H) Volcano plot comparing MMP12 ⁺ macrophages with other macrophages in the published SCLC dataset. The x‐axis indicates the log₂ fold change, and the y‐axis indicates the –log₁₀ adjusted p‐value. Green dots represent genes with adjusted p‐values > 5 × 10⁻¹⁰ and |log₂ fold change| > 1 and red dots represent genes with adjusted p‐values < 5 × 10⁻¹⁰ and |log₂ fold change| > 1.

Figure S8: Effect of siRNA‐mediated MMP12 knockdown on macrophage invasion. (A) Quantification of macrophage invasion following siRNA‐mediated MMP12 knockdown. Macrophages were transfected with 20 pmol of either siRNA Negative Control (siNC) or MMP12 targeting siRNA (siMMP12) and the invasion assay was subsequently performed. The y‐axis indicates the number of invading macrophages per field. Data are presented as the mean ± SD from three independent experiments (n = 3). The difference between the siNC and siMMP12 groups did not reach statistical significance (p = 0.20). (B) Representative microscopic images of invading macrophages in the siNC and siMMP12 groups.

Table S1: Antibodies used in this study.

Table S2: Patient information used in this study.

Table S3: Pairwise distances between all cell types.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.