Characteristic miRNA profiles represent clinicopathological diversity of small cell lung cancer

Abstract

Small cell lung cancer (SCLC) is classified into distinct molecular subtypes based on the expression patterns of four transcription regulators: achaete–scute homolog 1 (ASCL1), neuronal differentiation 1 (NEUROD1), POU class 2 homeobox 3 (POU2F3), and yes‐associated protein 1 (YAP1). MicroRNAs (miRNAs) play critical roles in cancer cellular processes but their subtype‐specific implications in SCLC remain underexplored. Out of 46 surgically resected SCLC samples, miRNA visualization through in situ hybridization identified high expression of miR‐375 in the ASCL1, NEUROD1, and ASCL1/NEUROD1 subtypes, and miR‐9‐5p in the POU2F3 subtype. Comprehensive enhancer profiling using SCLC cell lines indicated that miR‐375 and miR‐9‐5p were regulated by super‐enhancers in a subtype‐specific manner. Multiplex immunohistochemistry by imaging mass cytometry found that the miR‐9‐5p‐high SCLC was characterized by a higher stromal area ratio, increased numbers of CD8⁺ T cells and CD163⁻ macrophages in the intra‐tumoral area, and an increased number of plasma cells in the stromal area, as compared with the miR‐9‐5p‐low SCLC. Finally, clinicopathological analysis revealed that the miR‐375‐high SCLC was associated with YAP1 downregulation, increased serum pro‐gastrin‐releasing peptide levels, and poor prognosis. These findings highlight the critical role of super‐enhancer‐related miRNAs in the diversity of SCLC, and underscore the potential for novel diagnostic and prognostic biomarkers based on these subtype‐specific miRNAs. © 2025 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

Small cell lung cancer (SCLC) is a highly aggressive lung cancer and comprises 10%–17% of all lung cancers [1, 2]. Due to the limited treatment options available, the prognosis for SCLC patients has remained extremely poor over the past decades, with a low 5‐year survival rate of <7% [3]. Since surgical resection is rarely performed and few specimens are available, the molecular characteristics of SCLC have remained unclear for decades. Recent genome sequencing of clinical samples has identified frequent inactivation of TP53 and RB1, along with NOTCH signaling inactivation, as drivers of SCLC [4]. More recent studies that explored the heterogeneity of SCLC have identified distinct molecular subtypes based on the expression patterns of four transcription regulators: achaete–scute complex homolog 1 (ASCL1), neuronal differentiation 1 (NEUROD1), POU class 2 homeobox 3 (POU2F3), and yes‐associated protein 1 (YAP1) [5, 6, 7, 8, 9]. Each SCLC molecular subtype exhibits distinct drug sensitivities: the ASCL1 subtype is responsive to BCL2 inhibitors; the NEUROD1 subtype shows sensitivity to aurora kinase inhibitors; and the POU2F3 subtype may respond to PARP inhibitors [5]. Prognosis also varies by subtype: the ASCL1 and NEUROD1 subtypes are generally associated with better outcomes, while the POU2F3 subtype tends to have an intermediate prognosis [10]. This molecular classification has facilitated the development of novel and promising therapeutic modalities, including a bispecific T‐cell engager immunotherapy targeting delta‐like ligand 3, a protein regulated by ASCL1 [11].

SCLC has historically been treated with chemotherapy and radiation, but recent advances have highlighted the potential of immunotherapy, particularly immune checkpoint inhibitors (ICIs), as a promising strategy. Despite the modest survival benefits shown by anti‐PD‐L1 (programmed death‐ligand 1) agents, such as atezolizumab in combination with platinum‐based chemotherapy [12], durable responses remain limited to a subset of patients. This underscores the need for a deeper understanding of the tumor immune microenvironment to guide more effective therapeutic approaches. Recent immunogenomic and spatial profiling efforts have begun to reveal striking heterogeneity in immune infiltration and activation among molecular subtypes of SCLC, particularly identifying an ‘inflamed’ subtype that may be more amenable to immunotherapy [10]. However, the underlying mechanisms driving this immune heterogeneity, as well as robust predictive biomarkers for ICI response, remain poorly defined and require further investigation.

MicroRNAs (miRNAs) are a class of short, non‐coding RNAs that play a pivotal role in gene regulation by binding to the 3' untranslated regions (3' UTRs) of their target mRNAs [13, 14]. miRNAs are critical in various cellular processes, including cell proliferation, differentiation, and tumorigenesis [15]. In cancer, miRNAs have both tumor‐suppressive and oncogenic roles [16]. Recent discoveries have indicated that a subset of miRNAs regulated by super‐enhancers (SEs) exhibits cell‐type‐specific expression [17]. Our recent comprehensive epigenetic profiling of SCLC cell lines identified candidates of subtype‐specific miRNAs, including miR‐7, miR‐375, miR‐200b‐3p, miR‐429, and miR‐455‐3p, and their potential roles in the molecular heterogeneity of SCLC [18]. However, the expression patterns of these subtype‐specific miRNAs in SCLC tissues and their clinical significance remain to be elucidated.

Herein, we investigated subtype‐specific miRNAs in clinical SCLC samples using the miRNAscope assay, a recently developed technique for visualizing miRNA in situ. We then examined how the expression patterns of subtype‐specific miRNAs influenced immune profiles within the SCLC tumor microenvironment, along with their effects on gene and pathway regulation and clinical parameters.

Materials and methods

Ethics statement

The study received approval from the institutional ethics committee of Kanazawa University (approval number: 12644‐2).

Formalin‐fixed, paraffin‐embedded (FFPE) sample

We collected FFPE samples from 48 patients who underwent surgical procedures and were diagnosed with SCLC at Kanazawa University Hospital between 2001 and 2021, which is a different patient cohort from that previously analyzed by Miyakawa et al [18]. The clinical characteristics of the SCLC cohort are shown in supplementary material, Table S1. Histopathological evaluation was performed independently by two board‐certified pathologists (DM and IY). Two cores were extracted from each sample to create a tissue microarray (TMA) using tissue microarrayers, and molecular subtyping was performed using immunohistochemistry (IHC) for ASCL1, NEUROD1, POU2F3, and YAP1. If a core contained two or more subtypes, or if the subtypes varied between cores, whole sections were stained and evaluated. When staining of the whole section was heterogeneous, the sample was considered a mixed subtype (e.g. SCLC‐A/N, indicating co‐expression of ASCL1 and NEUROD1). After excluding one sample from a patient with a final diagnosis of pulmonary carcinoid tumor, and one sample with low staining quality, 46 samples were used for further analysis. A total of 24 lung adenocarcinoma (LUAD) samples with 24 lung squamous cell carcinoma (LUSC) samples were included as non‐small cell lung cancer (NSCLC) control.

Immunohistochemistry

A representative paraffin block was sectioned into 4‐μm‐thick slices, which were then subjected to IHC staining using the primary antibodies listed in supplementary material, Table S2. Immunostaining was performed either using a Ventana BenchMark ULTRA system (Roche, Basel, Switzerland) and ultraView Universal DAB Detection Kit (Roche, 109431), or manually using the Universal LSAB2 Kit/HRP for Rabbit/Mouse (Dako, Santa Clara, CA, USA; K0675) and Liquid DAB Substrate Chromogen System (Dako, K3468). Appropriate controls were included. Immunostaining was considered positive if more than 10% of the tumor cells were stained. In cases where tumors exhibited positivity for two transcription factors, the samples were classified as double‐positive.

miRNA in situ hybridization

Tissue blocks from TMA or cell blocks from cell lines were sectioned into 4‐μm‐thick slices, and miRNA in situ hybridization was performed using the miRNAscope HD Reagent Kit [Advanced Cell Diagnostics (ACD), Newark, CA, USA], following the manufacturer's protocol. Probes targeting miR‐375, miR‐9‐5p, and miR‐7‐5p (ACD) were hybridized to the target miRNAs using the HybEZ II Hybridization System (ACD). The negative control probe (Scramble‐S1; ACD, 727881‐S1) and the positive control probe (RNU6‐S1; ACD, 727871‐S1) served as negative and positive controls, respectively. Using QuPath (version 0.4.3) software (https://qupath.github.io/, last accessed 1 July 2025), we performed an image‐based analytical evaluation of the miRNAscope signals. After cell segmentation with default parameters, the signal intensity was quantified and categorized into four classes: 0 (intensity < 0.05), 1+ (intensity ≥ 0.05), 2+ (intensity ≥ 0.1), and 3+ (intensity ≥ 1.5). A core was considered positive if ≥10% of tumor cells exhibited at least 1+ signal intensity (supplementary material, Figure S1).

Imaging mass cytometry (IMC)

The tumor microenvironment of a subset of SCLC cases (n = 16) included in the TMA was analyzed by imaging mass cytometry (IMC). IMC of the FFPE section was carried out for immune profiling as previously described [19]. In brief, FFPE tissue sections were stained with a cocktail of metal‐conjugated antibodies using the Human Maxpar Immuno‐Oncology Imaging Mass Cytometry Panel Kit (#201508; Fluidigm, South San Francisco, CA, USA), together with 141Pr‐CD38 (#3141018D; Fluidigm), 143Nd‐Vimentin (#3143027D; Fluidigm), 147Sm‐CD163 (#3147021D; Fluidigm), and 151Eu‐CD31 (#3151025D; Fluidigm). For epithelial cell identification, 148Nd‐Pan‐Cytokeratin (AE1/AE3; #3148022D; Fluidigm) was used instead of the C11 clone included in the panel kit due to its higher sensitivity for SCLC cells. Antibody concentrations are detailed in supplementary material, Table S2. Image processing was performed using HALO software (Indica Labs, Albuquerque, NM, USA) and the following three modules were utilized: Highplex FL, Tissue Classifier, and Spatial Analysis. Cell segmentation was performed using default parameters, and cell annotation was conducted according to the algorithm described in supplementary material, Table S3.

Cell cultures and inhibition of miRNA using miRNA inhibitor

H209 and H526 cells were purchased from the ATCC (Rockville, MD, USA). Lu134A cells and SBC5 cells were obtained from RIKEN BRC (Tsukuba, Japan). The cells were cultured as previously described [8]. H209 and Lu134A cell lines (1 × 10⁷ cells) were mixed with 1 μmol of miRCURY LNA microRNA Power Inhibitor (QIAGEN, Hilden, Germany; Negative Control B, hsa‐miR‐375 or hsa‐miR‐9‐5p) and electroporation was conducted using the Gene Pulser Xcell Electroporation Systems (Bio‐Rad, Hercules, CA, USA) as previously described [20]. H526 cell lines (3 × 10⁶ cells) were mixed with 400 pmol of miRCURY LNA microRNA Power Inhibitor (QIAGEN; Negative Control B or hsa‐miR‐9‐5p) and electroporated using the Invitrogen Neon Transfection System (Thermo Fisher Scientific, Waltham, MA, USA) with settings of 1,600 V, 20 ms, and 1 pulse.

RNA sequencing

RNA was isolated using the RNeasy Mini Kit (QIAGEN). RNA sequencing (RNA‐seq) was performed on an Illumina NovaSeq 6000 system (Illumina, San Diego, CA, USA) with paired‐end reads of 150 base pairs. Quality control of the RNA‐seq data was performed using FastQC (version 0.12.1, https://www.bioinformatics.babraham.ac.uk/projects/fastqc, last accessed 1 July 2025). Subsequently, trimming was conducted using fastp (version 0.23.4, https://github.com/OpenGene/fastp, last accessed 1 July 2025) to remove low‐quality reads and adapter sequences. The reads were then mapped to the reference genome using STAR (version 2.7.11b, https://github.com/alexdobin/STAR, last accessed 1 July 2025). The resulting BAM file was indexed using Samtools (version 1.20, https://www.htslib.org/, last accessed 1 July 2025), and gene expression levels were quantified using RSEM (version 1.3.3, https://github.com/deweylab/RSEM, last accessed 1 July 2025). Predicted targets for each miRNA were obtained from TargetScan Release 8.0 (https://www.targetscan.org/vert_80/, last accessed 1 July 2025). Gene ontology analysis was performed using the Enrichr webtool [21].

Comprehensive enhancer profiling

Library preparation and data processing for the Cleavage Under Targets and Tagmentation (CUT&Tag) assay – a recently developed technique for genome‐wide chromatin profiling that enables high‐resolution mapping of protein–DNA interactions using low‐input material and producing minimal background noise – were carried out using the CUT&Tag‐IT Assay Kit (Active Motif, Carlsbad, CA, USA), following the manufacturer's protocol. H209, Lu134A, H526, and SBC5 cells (1 × 10⁵) were used for library preparation. Details of the antibodies used are provided in supplementary material, Table S2. The pooled library was sequenced on the NovaSeq X Plus system (Illumina) using 150 bp paired‐end reads. Alignment, peak calling, and the definition of super‐enhancers (SEs) and typical enhancers (TEs) were performed as previously described [18, 22]. Mapped sequence data were visualized using the Integrative Genomics Viewer (https://igv.org/, last accessed 1 July 2025). Raw sequencing data have been deposited in the DNA Data Bank of Japan (accession number DRA012871, https://www.ddbj.nig.ac.jp, last accessed 1 July 2025).

Public databases

Gene expression and matched miRNA expression datasets for 63 SCLC cell lines were obtained from GSE73160 and GSE73161, respectively [23]. ChIP‐seq data for 15 SCLC cell lines (7 SCLC‐A, 4 SCLC‐N, and 4 SCLC‐P) were obtained from GSE151002 and GSE115124 [9, 24].

Statistical analyses

The Wilcoxon rank‐sum test was used to compare different groups. Fisher's exact test was used for contingency table analyses. The significance level was set at 0.05. To account for multiple comparisons, the Bonferroni correction was applied.

Results

Molecular subtypes of SCLC

Forty‐six surgically resected SCLC samples were classified as the ASCL1 subtype (SCLC‐A) in 17 (37%), the NEUROD1 subtype (SCLC‐N) in four (9%), the POU2F3 subtype (SCLC‐P) in nine (20%), and the YAP1 subtype (SCLC‐Y) in two (4%) based on the results of IHC (Figure 1A). Six cases (13%) revealed double positivity for ASCL1 and NEUROD1 (SCLC‐A/N), five (11%) for POU2F3 and YAP1 (SCLC‐P/Y), one (2%) for ASCL1 and YAP1 (SCLC‐A/Y), and two (4%) remained unclassified. Representative images of these SCLC subtypes are shown in supplementary material, Figure S2. Most SCLC‐A and SCLC‐N were positive for neuroendocrine (NE) markers (CD56, synaptophysin, and chromogranin A) and thyroid transcription factor‐1 (TTF‐1) [20]. The expression of intact Rb protein was significantly more common in POU2F3‐positive SCLC (SCLC‐P and SCLC‐P/Y) than in POU2F3‐negative SCLC (p = 0.04; 36% versus 9%) (supplementary material, Table S4).

Figure 1.

Expression patterns of miR‐375, miR‐7‐5p, and miR‐9‐5p in different molecular subtypes of SCLC. (A) Panel showing the expression profiles of miRNAs in different molecular subtypes of SCLC based on the expression of ASCL1, NEUROD1, POU2F3, and YAP1 (SCLC‐A, SCLC‐N, SCLC‐P, and SCLC‐Y), along with the immunohistochemical expression patterns of p53/Rb, NE markers (CD56, SYP, and CHGA), and TTF‐1. The columns marked with * and ** correspond to the sample shown in Figure 4 and supplementary material, Figure S6, respectively. (B) Representative H&E staining, miRNAscope images of miR‐375, miR‐7‐5p, and miR‐9‐5p, and immunohistochemical expression of transcription factor (TF) in SCLC‐A, SCLC‐N, SCLC‐P, SCLC‐Y, LUAD, and LUSC. Scale bar, 50 μm. ASCL1, achaete–scute complex homolog 1; CHGA, chromogranin A; NE, neuroendocrine; NEUROD1, neuronal differentiation 1; POU2F3, POU class 2 homeobox 3; SYP, synaptophysin; TTF‐1, thyroid transcription factor 1; YAP1, yes‐associated protein 1; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; TF, transcription factor.

Subtype‐specific miRNA expression in SCLC

To identify robust subtype‐specific miRNAs, we first screened miRNAs that were highly correlated with gene expression for each transcription factor (Pearson's R > 0.4) using a comprehensive gene expression and miRNA dataset derived from SCLC cell lines (GSE73160 and GSE73161). Among SE‐related miRNAs identified by H3K27ac ChIP‐seq of SCLC cell lines in the datasets GSE151002 and GSE115124 (supplementary material, Table S5), we chose miRNAs with the highest signal intensities in each subtype, miR‐375 and miR‐9‐5p, as subtype‐specific miRNAs for SCLC‐A and SCLC‐P, respectively, and miR‐7‐5p as a subtype‐specific miRNA for SCLC‐A [18]. The miRNAscope images showed prominent expression of miR‐375 in most SCLC samples with NE phenotypes (SCLC‐A, SCLC‐A/N, and SCLC‐N) (Figure 1A,B). The expression of miR‐7‐5p in NE phenotypes was not as prominent as that of miR‐375, likely due to differences in absolute expression levels. Notably, miR‐375 was expressed in non‐NE and unclassified SCLC samples with CD56 positivity. Conversely, miR‐9‐5p was specifically expressed in SCLC‐P and SCLC‐P/Y, suggesting its involvement in the POU2F3‐regulated transcriptional network. Among the NSCLC control samples, miR‐375 was expressed in 8 out of 24 LUAD samples (33%) and 1 out of 24 LUSC samples (4%); miR‐9‐5p was expressed in one LUAD sample (4%) but was not expressed in LUSC samples; miR‐7 was not expressed in LUAD or LUSC samples (Figure 1B).

SE‐mediated regulation of subtype‐specific miRNAs

The miRNAscope images of the SCLC cell lines identified miR‐375 expression in H209 (SCLC‐A) and Lu134A (SCLC‐A/N) and exclusive miR‐9‐5p expression in H526 (SCLC‐P) (Figure 2A). miR‐7‐5p was weakly expressed in Lu134A (SCLC‐AN) (supplementary material, Figure S3A). SBC5 (SCLC‐Y) expressed none of the three miRNAs. These findings align with the miRNA expression patterns observed in the SCLC samples (Figure 1A). The CUT&Tag assay for H3K27ac, a recently developed technique for genome‐wide chromatin profiling, enables high‐resolution mapping of protein–DNA interactions using low‐input material and generating minimal background noise [25, 26]. Using this method, we identified 29,606, 20,910, 23,434, and 18,077 enhancer regions in H209, Lu134A, H526, and SBC5, respectively. Using the ROSE algorithm, we defined 1,428, 1,290, 939, and 459 SEs for each cell line, respectively (Figure 2B), and identified miR‐375 as an SE‐associated miRNA in H209 and Lu134A (Figure 2B,C), miR‐9‐5p encoded by the miR‐9‐1 gene as an SE‐associated miRNA in H526 (Figure 2B,D), and miR‐7‐5p encoded by the miR‐7‐3 gene as a TE‐associated miRNA in H209 and Lu134A (supplementary material, Figure S3B). All SE‐ and TE‐associated miRNAs in the four cell lines are listed in supplementary material, Table S6. These findings in SCLC cell lines indicate that miR‐375 and miR‐9‐5p are epigenetically regulated by SEs in a subtype‐specific manner.

Figure 2.

Expression patterns of miRNAs and enhancer profiling in SCLC cell lines representing molecular subtypes of SCLC. (A) Representative miRNAscope images of miR‐375 and miR‐9‐5p in SCLC cell lines: H209 (SCLC‐A), Lu134A (SCLC‐A/N), H526 (SCLC‐P), and SBC5 (SCLC‐Y). (B) Enhancer ranking based on H3K27ac signal in each SCLC cell, showing SE‐associated miRNAs (red) and TE‐associated miRNAs (blue). (C) Enhancer profiling using the CUT&Tag assay (Active Motif) for H3K27ac in each cell line, focusing on the genomic regions of ASCL1, NEUROD1, and miR‐375. Red and blue bars indicate genomic regions of SEs and TEs, respectively. The height of the plot represents the H3K27ac signal intensity. (D) Enhancer profiling using the CUT&Tag assay (Active Motif) for H3K27ac in each cell line, targeting the genomic regions of POU2F3, YAP1, and miR‐9‐1. ASCL1, achaete–scute complex homolog 1; NEUROD1, neuronal differentiation 1; POU2F3, POU class 2 homeobox 3; SCLC‐A/N, double positive for ASCL1 and NEUROD1; SE, super‐enhancer; TE, typical enhancer; YAP1, yes‐associated protein 1.

Specific immune profiles in SCLC samples with high and low miR‐9‐5p expression

Based on the miRNA expression pattern, SCLC can be roughly divided into miR‐9‐5p‐high/miR‐375‐low and miR‐9‐5p‐low/miR‐375‐high subtypes. To investigate the immunologic and microenvironmental differences between these subtypes, we assessed the immune profiles in SCLC samples with high and low miR‐9‐5p expression by IMC with 14 antibodies (Figure 3A and supplementary material, Table S2). We annotated the tumor/stroma/necrosis areas using machine learning and identified 11 cell types based on staining patterns. The proportion of stromal area was significantly higher in the miR‐9‐5p‐high SCLC than in the miR‐9‐5p‐low SCLC (Figure 3B). The miR‐9‐5p‐high SCLC had a significantly higher proportion of CD8⁺ T cells and CD163⁻ macrophages in the intra‐tumor area and a significantly higher proportion of plasma cells in the stromal area (Figure 3C–E). No significant difference was observed in the proportion of stromal area or the immune cell profiles between samples with high and low miR‐375 expression, except for the proportion of plasma cells in the stromal area (supplementary material, Figures S4 and S5). These findings indicate that miR‐9‐5p may alter immune profiles in the tumor microenvironment.

Figure 3.

Immune cell profiling in SCLC samples with high and low miR‐9‐5p expression. (A) Representative H&E staining, IMC signals, tissue classification via machine learning, and cell annotation in SCLC samples with high and low miR‐9‐5p expression. (B) Box plot illustrating the proportion of stromal area in miR‐9‐5p‐high SCLC (red) and miR‐9‐5p‐low SCLC (green). The top and bottom ends of the box represent the 75th and 25th percentiles, and the line in the middle of the box indicates the median. (C) Heatmap showing the proportion of each cell component in SCLC samples in the intra‐tumoral and stromal areas. (D) Box plots illustrating the proportion of CD8⁺ T cells and CD163⁻ macrophages in the intra‐tumoral area and plasma cells in the stromal area. The top and bottom ends of the box represent the 75th and 25th percentiles, and the line in the middle of the box indicates the median. (E) Representative images of CD8⁺ T cells, CD163⁺/CD163⁻ macrophages, and plasma cells in miR‐9‐5p‐high SCLC. *p < 0.05. IMC, imaging mass cytometry.

Intra‐tumoral heterogeneity and subtype‐specific miRNA expression patterns

SCLC often exhibits intra‐tumoral heterogeneity, which can contribute to drug resistance and metastasis [27, 28]. In one SCLC‐P sample (asterisk in Figure 1A), part of the SCLC showed the mutually exclusive expression of miR‐375 and miR‐9‐5p (Figure 4A). In IHC analysis, an miR‐375‐high component was positive for TTF‐1 and CD56 and negative for ASCL1, NEUROD1, POU2F3, and YAP1, while an miR‐9‐5p‐high component was positive only for POU2F3 and negative for TTF‐1 and CD56 (Figure 4B). An miR‐9‐5p‐high component in this sample showed an enlarged stromal area and increased numbers of intra‐tumoral CD8⁺ T cells and CD163⁻ macrophages (Figure 4C,D), as shown in miR‐9‐5p‐high SCLC (Figure 3C–E). In one SCLC‐A/Y sample (double asterisk in Figure 1A), an miR‐375‐high component was positive for ASCL1 and CD56, while an miR‐375‐low component was positive for YAP1 (supplementary material, Figure S6). These findings indicate that the expression patterns of subtype‐specific miRNAs may contribute to subtype switching in SCLC.

Figure 4.

Heterogeneous expression of miRNAs in the intra‐tumoral area. (A) Representative images showing the mutually exclusive expression of miR‐375 and miR‐9‐5p in an SCLC sample, quantified using QuPath analysis. (B) Representative H&E staining and IHC staining for TTF‐1, POU2F3, and CD56 in miR‐375‐high and miR‐9‐5p‐high components. Scale bar, 50 μm. (C) Representative images showing tissue classification via machine learning and cell annotation in miR‐375‐high and miR‐9‐5p‐high components. (D) Representative images showing the distribution of CD8⁺ T cells and CD163⁺/CD163⁻ macrophages in miR‐375‐high and miR‐9‐5p‐high components. IHC, immunohistochemistry; POU2F3, POU class 2 homeobox 3; TTF‐1, thyroid transcription factor 1.

Genes and pathways regulated by subtype‐specific miRNAs

To elucidate the functional role of miR‐9‐5p and miR‐375 in SCLC, we explored their direct targets using a combination of in vitro and in silico analyses. At 48 h after electroporation with miRNA inhibitors, the transcriptome was analyzed by RNA‐seq. Filtering by a fold‐change greater than 1.33 and TPM greater than 0.5, we identified more than 1,000 genes upregulated by each miRNA inhibitor (Figure 5A). By integrating with the target genes predicted by the TargetScan, 60 and 21 overlapping genes were identified as robust targets of miR‐9‐5p and miR‐375, respectively. Gene ontology analysis found that miR‐9‐5p regulated ECM–receptor interactions and focal adhesion, whereas miR‐375 regulated pathways such as neuroactive ligand–receptor interaction and the Hippo signaling pathway (Figure 5B). Notably, the miR‐375 inhibitor upregulated YAP1 and its direct targets, including CCN1 and CCN2, associated with an active Hippo signaling pathway (supplementary material, Figure S7A). There was a clear negative correlation between miR‐375 expression and YAP1 expression (R = −0.73, p < 0.05) in the SCLC cell lines (Figure 5C) and a significantly lower YAP positive rate (i.e. fewer cases of SCLC‐Y) in miR‐375‐high SCLC (p < 0.05) (Figure 5D), indicating that miR‐375 may downregulate YAP1. Additionally, HES1, a canonical Notch target and transcriptional repressor [29, 30], and PKIB, an activating kinase of the PI3K/Akt pathway, were likely to be regulated by miR‐9‐5p (supplementary material, Figure S7B). This was supported by a moderate negative correlation between miR‐9‐5p expression and HES1 expression (R = −0.16, p < 0.05), as well as between miR‐9‐5p expression and PKIB expression (R = −0.19, p < 0.05) (supplementary material, Figure S7C).

Figure 5.

Genes and pathways regulated by miR‐375 and miR‐9‐5p. (A) Venn diagrams showing the overlap between the genes upregulated by miRNA inhibitors (fold‐change > 1.33 and TPM > 0.5) and the target genes of miRNAs predicted by the TargetScan. (B) Pathways identified by KEGG pathway analysis of the genes upregulated by miRNA inhibitors. The x‐axis indicates the log₂ p value, which represents the level of significance of each pathway. (C) Correlation between the expression of miR‐375 and the YAP1 gene in SCLC cell lines (GSE73160/73161). (D) The YAP1‐positive rate in miR‐375‐high and miR‐375‐low SCLC. *p < 0.05. IHC, immunohistochemistry; KEGG, Kyoto Encyclopedia of Genes and Genomes; TPM, transcripts per kilobase million; YAP1, yes‐associated protein 1.

Clinical relevance of miR‐375 and miR‐9‐5p expression in SCLC

Finally, we performed survival analysis to assess the impact of miRNA expression on survival in SCLC using miRNAscope assay data. Patients with SCLC with high miR‐375 or miR‐7‐5p expression had a worse overall survival than those with low miR‐375 or miR‐7‐5p expression, with p values of 0.032 and 0.0077, respectively (Figure 6A and supplementary material, Figure S8A). Conversely, miR‐9‐5p expression did not affect overall survival. Furthermore, univariate Cox regression analyses using the following variables revealed that only high miR‐375 expression was significantly associated with a poor prognosis (p = 0.040; HR = 3.99; 95% CI 1.06–15.00) (supplementary material, Figure S8B). Serum pro‐gastrin‐releasing peptide (proGRP) levels were significantly higher in patients with miR‐375‐high SCLC and significantly lower in those with miR‐375‐low SCLC, while serum neuron‐specific enolase (NSE) levels did not correlate with the expression of miR‐375 or miR‐9‐5p (Figure 6B). The miR‐7‐5p expression did not affect serum proGRP or NSE levels (supplementary material, Figure S8C).

Figure 6.

Clinical relevance of miR‐375 and miR‐9‐5p expression in SCLC. (A) Kaplan–Meier curves showing the estimated probability of survival over time in patients with high and low expression of each miRNA. (B) Dot plots illustrating serum proGRP and NSE levels in patients with high and low expression of each miRNA. The top and bottom lines represent the 75th and 25th percentiles, and the dot in the middle indicates the median. N.S., not significant; NSE, neuron‐specific enolase; proGRP, pro‐gastrin‐releasing peptide.

Discussion

In this study, we performed comprehensive histopathological profiling of miRNAs in a large series of surgically resected SCLCs and found that miR‐375 and miR‐7‐5p were predominantly expressed in SCLC subtypes with an NE phenotype, whereas miR‐9‐5p was predominantly expressed in SCLC subtypes with non‐NE phenotypes. Comprehensive enhancer profiling using the H3K27ac CUT&Tag assay showed that these subtype‐specific miRNAs were regulated by SEs. IMC demonstrated that miR‐9‐5p‐high SCLC had a distinct immune profile in the tumor microenvironment, characterized by a larger stromal area and increased numbers of CD8⁺ T cells and CD163⁻ macrophages in the intra‐tumoral area. A combination of in vitro and in silico analyses demonstrated that high miR‐375 expression in SCLC was associated with YAP1 downregulation. Patients with miR‐375‐high SCLC had increased serum pro‐gastrin‐releasing peptide levels and a poor prognosis.

miR‐375 has been shown to act as a downstream effector of ASCL1 in lung cancer cells with NE features, with YAP1 identified as a direct target of miR‐375 [7, 8, 31]. In this study, we demonstrated that miR‐375 expression is not restricted to SCLC‐A but is broadly expressed across subtypes with NE phenotypes, suggesting that miR‐375 is a robust NE‐associated miRNA. Considering that miR‐9‐5p is specifically expressed in SCLC‐P, histological evaluation of miR‐375 and miR‐9‐5p expression may contribute to a more refined molecular classification of SCLC, particularly by distinguishing between NE and non‐NE phenotypes. Furthermore, if present at detectable levels, miRNAs may serve as potential serum biomarkers. Combining the serum levels of these miRNAs with conventional markers – particularly proGRP, which is strongly regulated by ASCL1 – may enhance the accuracy of molecular subclassification in SCLC and facilitate the selection of targeted therapies, eliminating the need for biopsy, which is often challenging in clinical settings for patients with SCLC. Additionally, in certain SCLC cases, miR‐375 expression exhibited heterogeneity, even in the absence of ASCL1 or NEUROD1 expression (Figure 4). This suggests that miR‐375 may play a key role in regulating NE characteristics and transitions between NE and non‐NE phenotypes, partly cooperatively with miR‐9‐5p and potentially independent of ASCL1 or NEUROD1. Further studies are necessary to explore the role of miR‐375 in NE differentiation in greater detail.

miR‐9‐5p plays diverse roles across various cancers, functioning as both a tumor suppressor and promoter by regulating target genes [32, 33, 34]. In NSCLC, miR‐9‐5p is associated with increased tumor progression and poor prognosis and serves as a potential biomarker for aggressive disease [35, 36]. The present study found that miR‐9‐5p was strongly associated with the SCLC‐P subtype and might regulate HES1 and PKIB. In NSCLC [37], PKIB promotes tumorigenesis by activating the PI3K/Akt pathway. Given that the PI3K/AKT/mTOR pathway is activated by genomic alterations in 36% of SCLC cases [38], miR‐9‐5p may play a crucial role in controlling tumor growth in SCLC‐P by suppressing PKIB. Furthermore, Notch signaling is known to modulate tumor‐mediated immune suppression through its effects on T cells [39] and immunosuppressive tumor‐associated macrophages [40]. Thus, miR‐9‐5p may regulate the tumor immune phenotype via the HES1–Notch signaling axis.

SCLC tumors with non‐NE phenotypes have been shown to harbor more CD8⁺ T cells and exhibit an ‘inflamed’ immunophenotype [41]. In patients with SCLC, the intra‐tumoral abundance of CD163/CD204⁺ M2 macrophages has been linked to poor prognosis, indicating the association between the composition of the inflammatory microenvironment and overall survival. We found that miR‐9‐5p was associated with a distinct tumor microenvironment characterized by an increased stromal area, higher CD8⁺ T‐cell infiltration, and CD163⁻ (i.e. M1) macrophages. Given that miR‐9‐5p regulated ECM‐related pathways, stromal changes, such as the appearance of cancer‐associated fibroblasts, may play a key role in shaping the tumor microenvironment in miR‐9‐5p‐high SCLC. Further studies are warranted to explore the function of miR‐9‐5p within the tumor microenvironment and its potential as a marker for molecular subtyping based on miRNA profiles.

There are several limitations to the present study. First, the sample size is relatively small and limited to a single institution, which may affect the generalizability of the findings. Larger, multi‐center cohorts are needed to validate our results. Second, all the samples used were from surgically resected tumors obtained prior to chemotherapy. Thus, further investigation of post‐treatment samples is warranted to assess therapy‐induced changes. Third, none of the patients received immune checkpoint inhibitors, which may limit interpretation of the immune microenvironment in the context of current immunotherapy approaches.

In conclusion, using a novel miRNA in situ hybridization technique that has enabled clear detection of miRNAs, we identified subtype‐specific miRNAs in SCLC samples and confirmed their clinical implications. Our findings indicate that subtype‐specific miRNAs have potential as novel diagnostic and prognostic biomarkers and may also serve as therapeutic targets in SCLC.

Author contributions statement

MH, AS and DM contributed to conceptualization. MH contributed to analysis, methodology, and writing. HTak contributed to data acquisition. YI and AI contributed to pathological analysis. HK, TU, IM and SY contributed to the acquisition of clinical data. HTan, MS and MK contributed to library preparation. HIS, AS and DM contributed to supervision.

Supporting information

Figure S1. Image‐based analytical evaluation of miRNAscope signals by QuPath

Figure S2. Molecular subtyping of SCLC by immunohistochemistry of ASCL1, NEUROD1, POU2F3, and YAP1

Figure S3. Expression patterns of miR‐7‐5p and enhancer profiling in SCLC cell lines representing four molecular subtypes of SCLC

Figure S4. Box plot illustrating the proportion of stromal area in SCLC samples with high and low miR‐375 expression

Figure S5. Box plots showing the proportion of various immune cells in the intra‐tumoral and stromal areas in SCLC samples with high and low miRNA expression

Figure S6. Representative images showing heterogeneous expression of ASCL1, YAP1, and CD56 in the intra‐tumoral area in miR‐375‐high and miR‐375‐low components

Figure S7. Genes associated with miR‐9‐5p and miR‐375

Figure S8. Clinical relevance of miR‐7‐5p expression in SCLC

Table S1. Clinical characteristics of the SCLC cohort

Table S2. Antibodies used in this study

Table S3. Algorithm for cell annotation in imaging mass cytometry

Table S4. Molecular subtyping and clinical parameters of 46 SCLC samples

Table S5. List of SCLC subtype‐specific miRNAs

Table S6. SE‐ and TE‐miRNAs in SCLC cell lines (H209, Lu134A, H526, and SBC5)

Data availability statement

All raw sequencing data generated in this study have been deposited in the DNA Data Bank of Japan (DDBJ) under accession number DRA012871 and are publicly available at https://ddbj.nig.ac.jp/public/ddbj_database/dra/fastq/DRA012/DRA012871/ (accessed 1 July 2025). All other supporting data are available from the corresponding author upon reasonable request.