MicroRNA clustering in pediatric asthma reveals differential corticosteroid responsiveness Authors: Minsuk Yang, Upasna Srivastava, Alvin Kho, Anshul Tiwari, Alberta Wang, Shraddha Piparia, Scott T Weiss, Michael McGeachie, Kelan Tantisira Published: 2026 ABSTRACT AbstractBackgroundAsthma is a heterogeneous disease with variable treatment responses. miRNAs play a critical role in asthma pathogenesis and may serve as biomarkers for distinct asthma endotypes. We tried to identify possible molecular asthma clusters based on miRNA expression profiles in pediatric patients from the Childhood Asthma Management Program (CAMP) cohort.MethodsWe performed miRNA profiling and k‐means clustering on 462 CAMP participants. Differential miRNA expression and pathway enrichment analyses were conducted to characterize cluster‐associated molecular features. Clinical characteristics and budesonide‐related changes in lung function were compared across clusters, including analyses stratified by peripheral blood eosinophil counts.ResultsThree possible molecular clusters were identified, each with distinct miRNA expression patterns and enriched biological pathways. All clusters showed enrichment in IL‐4/IL‐13 signaling and PIP3/AKT pathways. Cluster 1 was characterized by Toll‐like receptor‐related signaling pathways, Cluster 2 by FOXO‐mediated transcription pathways, and Cluster 3 by pathways associated with SCF‐KIT signaling. Baseline demographic and clinical characteristics were largely similar across clusters. However, patterns of budesonide‐related lung function change differed. Clusters 1 and 3 showed more favorable lung function profiles after budesonide treatment than Cluster 2. In Cluster 1, eosinophilic participants had a higher proportion of moderate asthma, whereas those without eosinophilia were more often classified as having mild asthma. In Cluster 3, eosinophilic participants had greater baseline airway hyperresponsiveness and lower baseline FEV1/FVC but showed greater improvement in FEV1/FVC after budesonide.ConclusionmiRNA profiling identified possible molecular asthma clusters with different clinical, physiologic, and treatment‐related patterns.Keywords: asthma, cluster analysis, microRNA Key message Blood microRNA profiling identified three molecular clusters in pediatric asthma that were not clearly distinguishable by baseline clinical characteristics alone. These clusters differed in enriched biologic pathways as well as in clinical, physiologic, and treatment‐related patterns, supporting the potential value of molecular profiling for understanding asthma heterogeneity. INTRODUCTION Asthma is characterized by significant heterogeneity, consisting of various conditions with different pathological mechanisms but similar observable clinical characteristics. 1 Despite the availability of effective treatments, such as inhaled corticosteroids (ICS), there is variability in responsiveness to standard treatment among patients. 2 This variability in treatment response, including the responsiveness to ICS and biologics, underscores the need for a deeper understanding of the underlying mechanisms driving different asthma endotypes. To address this heterogeneity, cluster analysis has been utilized to categorize asthma into distinct subgroups based on various demographic, clinical, and inflammatory features in both children and adults. 1 , 3 , 4 , 5 , 6 This approach has proven useful in identifying specific asthma subgroups that respond differently to treatments and have unique underlying biological mechanisms. 7 , 8 However, there are still limitations in utilizing clinically derived asthma endotypes in actual practice. This is due to the unclear biological processes underlying asthma endotypes and their causation, as well as the absence of readily usable biomarkers to confirm them. Furthermore, specific treatments targeting these biological processes are lacking, thereby impeding their clinical application. 9 Inhaled corticosteroids exert their anti‐inflammatory effects primarily through binding to the glucocorticoid receptor, which modulates transcription of inflammatory genes through transactivation and transrepression mechanisms, resulting in suppression of key pro‐inflammatory cytokines including IL‐4, IL‐5, and IL‐13, and inhibition of eosinophilic airway inflammation. 10 Despite their efficacy, a substantial proportion of patients exhibit reduced responsiveness to ICS, particularly those with non‐eosinophilic or neutrophilic phenotypes. 11 Several molecular pathways have been implicated in corticosteroid resistance, including upregulation of Toll‐like receptor (TLR) signaling and PI3K/AKT pathway activation, both of which can impair glucocorticoid receptor function or promote alternative inflammatory cascades less amenable to steroid suppression. 11 MicroRNAs (miRNAs), small non‐coding RNAs that regulate gene expression at the post‐transcriptional level, have emerged as critical players in various biological processes and diseases, including asthma. 12 , 13 They are involved in immune response, inflammation, and cellular processes, all of which are pivotal in asthma pathogenesis. 12 , 13 Recent studies have suggested that miRNA expression profiles could provide valuable insights into the molecular mechanisms underlying asthma heterogeneity. 5 , 14 , 15 Extracellular miRNAs are known for their stability, making them easier to measure compared to mRNA and ideal biomarker candidates. Moreover, due to their direct involvement in biological processes, they could potentially serve as targets for drug therapy. 16 , 17 Based on this background, we utilized miRNA expression profiles to cluster asthma subjects in the Childhood Asthma Management Program (CAMP) cohort. By employing this approach, we intended to identify possible molecular asthma clusters and examine the clinical characteristics of each cluster. METHODS Study populations The CAMP cohort was a multi‐center, randomized, placebo‐controlled trial involving 1041 children aged 5–12 years. The trial investigated the effects of budesonide, nedocromil, or placebo in the treatment of childhood asthma. For the present analysis, we selected 462 participants with available baseline serum miRNA sequencing data from the CAMP Genetic Ancillary Study (ClinicalTrials.gov Identifier: NCT00000575), including 188 from the budesonide arm and 274 from the placebo arm. 18 Participants randomized to nedocromil (n = 30) were excluded a priori due to the small sample size and the distinct non‐steroidal mechanism of action of nedocromil, which could introduce heterogeneity in analyses focused on corticosteroid response. In the CAMP, data were collected on various demographic characteristics such as the participant's height, weight, onset of asthma, severity of asthma, and presence of atopy, among others. Following randomization, the participants were followed up for a period of 4–6 years. During this period, data on asthma symptoms, airway hyperresponsiveness, pulmonary function tests, bronchodilator responses, exacerbations, and treatment responses were collected. 19 miRNA profiling Baseline serum samples from 462 participants in the CAMP cohort were profiled using small RNA sequencing. Small RNA‐seq libraries were prepared using the Small RNA Library Prep Kit from Norgen Biotek Corp. (Thorold, ON, Canada) and sequenced on the NextSeq 500 platform from Illumina (San Diego, CA). The exceRpt pipeline was utilized for quality control of the RNA‐seq data. 20 The CAMP Genetics Ancillary Study received approval from the Brigham and Women's Hospital Internal Review Board, under protocol #2015P001622/BWH. Informed consent and assent were obtained from the parents and participants, respectively. Cluster analysis In this study, we performed an unsupervised cluster analysis on pediatric asthma patients from the CAMP cohort, using their blood miRNA expression profiles. Out of the total miRNAs, we selected 256 that had a read count of more than 5 in over 50% of the 462 included participants as variables for the cluster analysis. The read counts were normalized using the DESeq2 package in R (version 4.3.1) and then log transformed. To determine the appropriate number of clusters, we utilized hierarchical cluster analysis using Ward's method and generated a dendrogram (Figure S1). Additionally, before performing k‐means clustering, we reduced the dimensionality of the data using principal component analysis. Based on these analyses, participants were divided into three clusters using k‐means cluster analysis. Differential miRNA expression and functional analysis in each cluster To identify differences in miRNA expression across clusters, we utilized the Kruskal–Wallis test for analysis. We then conducted post hoc analysis using the Dunn test, which allowed us to identify the clusters with the highest and lowest expression for each miRNA. In the Kruskal–Wallis test, we considered results to be statistically significant if the Benjamini‐Hochberg procedure yielded a value of 0.05 or less. The cluster with the highest expression of a specific miRNA is defined as the cluster with high expression, while the cluster with the lowest expression of that specific miRNA is defined as the cluster with low expression. If another cluster that did not belong to either “high” or “low” showed no statistically significant difference from the high or low expression clusters, that cluster is considered to fall within that expression category. This process resulted in the identification of sets of miRNAs with high expression in each of the three clusters. To investigate the functional roles of miRNAs with high expression in each cluster, we first identified their target genes using the multiMiR package in R. The “multiMiR” package searches for target genes using multiple databases, including miRecords, miRTarBase, and TarBase. Among these, we limited our analysis to genes with experimentally validated functions, as revealed by assays such as luciferase reporter assay, qRT‐PCR, and western blot. Following the identification of target genes, we performed pathway enrichment analysis on these gene lists using the ReactomePA package in R. For this analysis, we used a false discovery rate threshold of 0.05 to ensure statistical significance. Each pathway required a minimum of 5 genes to be considered in the analysis. The background set of genes used for this analysis was based on the comprehensive gene set from the Reactome database. Comparison of phenotype and treatment responsiveness among clusters We compared the phenotypes of participants within each of the three clusters. Pulmonary function parameters such as Forced Vital Capacity (FVC) and Forced Expiratory Volume in 1 s (FEV1) were expressed as a percentage of the predicted value. Changes in pulmonary function before and after treatment were expressed in percentage change in the predicted value, calculated by dividing the percentage change before and after treatment by the predicted percentage of pulmonary function before treatment ((PostFEV1% − PreFEV1%)/PreFEV1%). To assess treatment efficacy across clusters, we primarily examined changes in lung function over 12 months among participants assigned to the budesonide arm and compared these changes across the three clusters. Within the budesonide group, 12‐month changes in lung function were further examined according to peripheral blood eosinophil status within each cluster. The criterion for peripheral blood eosinophilia was set at 300 cells/mm3 or greater. Statistical analysis Normally distributed variables were presented as mean ± standard deviation (SD), while non‐normal variables were presented as median (interquartile range (IQR)). Categorical variables were shown as frequencies and percentages. Comparisons between groups for continuous variables were conducted using either ANOVA, Welch, or the Kruskal–Wallis test as appropriate. For categorical variables, comparisons between groups were made using the Chi‐square or Fisher's exact test as appropriate. The analysis was conducted excluding missing data. Statistical significance was determined at a p‐value threshold of .05 unless otherwise specified. RESULTS Participants were divided into three clusters using k‐means cluster analysis (Figure 1). Clusters 1, 2, and 3 included 63, 62, and 337 participants, respectively. To investigate differences in miRNA expression among the clusters, we performed a Kruskal–Wallis test on 256 miRNAs. Of these, 246 were found to be significantly differentially expressed across the 3 clusters (Table S1). When comparing the expression levels of these 246 miRNAs among the clusters, we identified that Clusters 1, 2, and 3 had 44, 96, and 164 highly expressed miRNAs, respectively (Tables S2 and S3). Subsequently, we identified the target genes of the highly expressed miRNAs in each cluster. The number of target genes for the highly expressed miRNAs in Clusters 1, 2, and 3 was 1132, 464, and 1024, respectively (Table S4). Results from k‐means clustering K‐means cluster analysis was used to classify 462 pediatric asthma patients from the CAMP cohort into three distinct clusters based on miRNA expression profiles. Cluster 1 (red) contains 63 patients, Cluster 2 (green) contains 62 patients, and Cluster 3 (blue) contains 337 patients. Each dot represents an individual patient, with the x and y axes representing the first and second principal components obtained from principal component analysis (PCA). Pathway analysis was performed using the set of target genes to identify significantly enriched pathways in each of the three clusters (Table S5). The “interleukin‐4 and interleukin‐13 signaling” and “PIP3 activates AKT signaling” pathways were highly enriched across all three clusters. In Cluster 1, pathways related to Toll‐Like Receptor (TLR), such as “MyD88:MAL (TIRAP) cascade initiated on plasma membrane”, “Toll Like Receptor TLR6:TLR2 Cascade”, “Toll Like Receptor TLR1:TLR2 Cascade”, “Toll Like Receptor 2 (TLR2) Cascade”, and “Toll Like Receptor 4 (TLR4) Cascade” were significantly enriched. Clusters 2 and 3 exhibited similar patterns, with frequent enrichment in pathways primarily involving VEGF and NTRK signaling. Notably, Cluster 2 uniquely exhibited enrichment in the “FOXO‐mediated transcription” pathway, while Cluster 3 showed enrichment in the “Signaling by SCF‐KIT” pathway, which was absent in Cluster 2. We examined the clinical characteristics of participants within each of the three clusters. There were no significant differences among the clusters in terms of age, sex, race, body mass index, age of asthma onset, asthma severity, lung function parameters at the time of enrollment, atopy, comorbid allergic rhinitis, peripheral blood eosinophil (PBE) counts, and total IgE levels (Table 1). TABLE 1Baseline characteristics of participants in each cluster.Cluster 1 (n = 63)Cluster 2 (n = 62)Cluster 3 (n = 337) p ValueAge (y)8.1 (2.9)8.3 (3.7)8.8 (3.6).45Age at first asthma symptoms (y)3.0 (2.5)2.0 (4.0)2.2 (3.1).97Gender (%)Male38 (60.3%)34 (54.8%)199 (59.1%).79Female25 (39.7%)28 (45.2%)138 (40.9%)EthnicityWhite52 (82.5%)46 (74.2%)264 (78.3%).11Black11 (17.5%)15 (24.2%)54 (16.0%)Hispanic0 (0%)1 (1.6%)19 (5.6%)Body mass index (kg/m2)16.9 (2.5)17.3 (3.2)17.3 (4.1).52Exposure to home tobacco smoke21 (34.4)29 (46.8)143 (42.6).36Baseline asthma severityMild31 (49.2)23 (37.1)158 (46.9).31Moderate32 (50.8)39 (62.9)179 (53.1)Treatment groupBudesonide35 (55.6%)20 (32.3%)133 (39.5%).02Placebo28 (44.4%)42 (67.7%)204 (60.5%)Hay fever31 (49.2%)29 (48.3%)178 (53.0%).72Atopic dermatitis18 (28.6%)13 (21.0%)104 (30.9%).28Aeroallergen sensitization on skin testing53 (84.1%)58 (93.5%)291 (86.4%).23Baseline PC20 (mg/mL)1.2 (2.0)1.1 (1.6)1.1 (2.1).85Baseline FVC (%pred)102.8 ± 12.8105.8 ± 13.0105.0 ± 12.6.37Baseline FEV1 (%pred)92.1 ± 13.593.7 ± 12.994.9 ± 13.5.29Baseline FEV1/FVC ratio (%)79.0 (14.5)78.5 (10.0)81.0 (10.0).28Baseline bronchodilator response (%)10.0 (12.0)8.5 (9.0)9.0 (10.0).98Peripheral blood neutrophil count (cells/mm3)2834.0 (2865.5)3234.0 (1940.0)3125.5 (1737.0).70Peripheral blood lymphocyte count (cells/mm3)3174.0 (935.5)3008.0 (1084.0)2808.0 (1084.5).01Peripheral blood monocyte count (cells/mm3)456.0 (437.0)448.0 (352.0)434.5 (297.2).99Peripheral blood eosinophil count (cells/mm3)404.0 (455.5)360.0 (416.0)406.0 (479.0).69Peripheral blood basophil count (cells/mm3)0.0 (72.5)0.0 (70.0)0.0 (75.0).61Total IgE517.0 (1051.5)376.5 (979.5)465.0 (964.0).59Abbreviations: FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; PC20, provocative concentration of methacholine that causes a 20% fall in FEV1. Baseline characteristics of participants in each cluster. Abbreviations: FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; PC20, provocative concentration of methacholine that causes a 20% fall in FEV1. To assess treatment efficacy in each cluster, we analyzed data specifically from participants assigned to the budesonide arm. Baseline lung function and lung function parameters at 12 months were compared across clusters. At baseline, there were no significant differences in FEV1, FVC, or FEV1/FVC ratio among the clusters. While the magnitude of change in lung function over 12 months was not significantly different among clusters, lung function at 12 months differed significantly across clusters. In particular, FEV1 (Cluster 1 vs. Cluster 2 vs. Cluster 3 = 99.2% ± 13.6% vs. 91.5% ± 11.6% vs. 100.6% ± 13.1%, p = .02) and FEV1/FVC (Cluster 1 vs. Cluster 2 vs. Cluster 3 = 81.9% ± 8.1% vs. 77.4% ± 6.6% vs. 81.8% ± 6.9%, p = .04) ratio at 12 months were significantly higher in Cluster 1 and Cluster 3 compared with Cluster 2 (Table 2). However, when examining changes over 12 months of treatment, Cluster 1 and Cluster 3 showed statistically significant increases in both FEV1 (Cluster 1 vs. Cluster 2 vs. Cluster 3 = 5.9% ± 12.3% vs. 1.1% ± 10.1% vs. 6.8% ± 16.0%, p = .02) and FEV1/FVC ratio (Cluster 1 vs. Cluster 2 vs. Cluster 3 = 5.5% ± 18.1% vs. −3.0% ± 13.3% vs. 2.4% ± 16.8%, p = .02) compared to Cluster 2. In contrast, the placebo group did not show significant changes in lung function across all clusters (Figure 2 and Figure S2). TABLE 2Changes in lung function parameters after budesonide treatment in each cluster.Cluster 1 (n = 35)Cluster 2 (n = 20)Cluster 3 (n = 133) p ValueFEV1 Baseline (%pred)92.5 ± 14.590.4 ± 12.193.5 ± 13.0.63At 12‐month follow‐up (%pred)99.2 ± 13.691.5 ± 11.6100.6 ± 13.1.02The percentage change during 12 months5.9 (12.3)1.1 (8.5)6.6 (15.9).06FVCBaseline (%pred)102.0 (18.5)106.0 (19.0)104.0 (15.0).70At 12‐month follow‐up (%pred)106.7 ± 13.7104.4 ± 13.0108.1 ± 12.2.46The percentage change during 12 months3.6 (11.4)−0.5 (7.5)2.8 (11.6).24FEV1/FVC ratio (%)Baseline (%pred)80.0 (15.0)77.5 (7.8)79.0 (9.0).42At 12‐month follow‐up (%pred)81.9 ± 8.177.4 ± 6.681.8 ± 6.9.04The percentage change during 12 months1.3 (9.7)0.0 (5.7)2.5 (8.1).15 Changes in lung function parameters after budesonide treatment in each cluster. Changes in lung function parameters between the budesonide and placebo groups across different clusters Comparison of lung function changes between budesonide and placebo groups in each cluster after 12 months of treatment. (A) Change in FEV1 percent predicted, (B) Change in FEV1/FVC ratio. Bars represent mean change, and error bars indicate standard deviation. Clusters 1 and 3 showed significant improvements in lung function with budesonide treatment compared to placebo, while Cluster 2 showed no significant difference between the two groups. When stratifying by PBE count in the budesonide group, distinct patterns emerged across clusters. In Cluster 1, ICS response did not differ significantly according to eosinophil status, although participants with PBE ≥300 cells/mm3 showed a numerically greater improvement in FEV1 compared to those with lower eosinophil counts. Notably, within Cluster 1, participants without peripheral blood eosinophilia had a significantly higher proportion of mild asthma (80.0% vs. 24.0%, p = .01), whereas those with PBE ≥300 cells/mm3 were predominantly classified as moderate asthma (76.0% vs. 20.0%). This eosinophil‐associated difference in asthma severity was not observed in Clusters 2 or 3 (Table 3). In Cluster 3, participants with PBE ≥300 cells/mm3 had greater baseline airway hyperresponsiveness, reflected by lower PC20 values compared to those without eosinophilia (0.7 vs. 2.0 mg/mL, p < .001), and lower baseline FEV1/FVC ratios (p = .01). Following budesonide treatment, those with higher eosinophil counts showed meaningful improvement in FEV1/FVC, with a change of +3.4 percentage points compared to +0.6 in the low‐eosinophil group (p = .002), reaching levels comparable to those in the low‐eosinophil group at 12 months in Cluster 3. In Cluster 2, no significant ICS response was observed regardless of eosinophil status (Table 3). TABLE 3Baseline characteristics and lung function changes according to peripheral blood eosinophil status within each molecular cluster.Cluster 1 (n = 35) p ValueCluster 2 p ValueCluster 3 p ValueWithout eosinophilia (n = 10)With eosinophilia (n = 25)Without eosinophilia (n = 9)With eosinophilia (n = 11)Without eosinophilia (n = 45)With eosinophilia (n = 84)Age (y)7.2 (1.3)8.1 (2.7).199.5 ± 2.19.0 ± 2.4.669.0 (3.4)9.2 (3.7).70Age at first asthma symptoms (y)3.1 ± 2.22.9 ± 1.9.853.1 ± 3.03.9 ± 2.3.532.0 (3.0)2.7 (2.5).44GenderMale6 (60.0%)14 (56.0%)1.007 (77.8%)5 (45.5%).2022 (48.9%)54 (64.3%).13Female4 (40.0%)11 (44.0%)2 (22.2%)6 (54.5%)23 (51.1%)30 (35.7%)EthnicityWhite9 (90.0%)22 (88.0%)1.006 (66.7%)5 (45.5%).8128 (62.2%)54 (64.3%).63Black1 (10.0%)3 (12.0%)3 (33.3%)5 (45.5%)9 (20.0%)20 (23.8%)Hispanic0 (0.0%)0 (0.0%)0 (0.0%)1 (9.1%)8 (17.8%)10 (11.9%)Body mass index (kg/m2)16.7 (1.3)16.9 (4.0).9117.9 ± 2.719.6 ± 3.1.2417.1 (3.1)17.5 (4.1).42Exposure to home tobacco smoke3 (30.0%)6 (25.0%)1.004 (44.4%)2 (18.2%).3423 (51.1%)32 (38.6%).24Baseline asthma severityMild8 (80.0%)6 (24.0%).014 (44.4%)4 (36.4%)1.0023 (51.1%)34 (40.5%).33Moderate2 (20.0%)19 (76.0%)5 (55.6%)7 (63.6%)22 (48.9%)50 (59.5%)Hay fever4 (40.0%)13 (52.0%).715 (55.6%)4 (36.4%).6524 (53.3%)50 (60.2%).57Atopic dermatitis1 (10.0%)5 (20.0%).652 (22.2%)2 (18.2%)1.0010 (22.2%)34 (40.5%).06Aeroallergen sensitization on skin testing5 (50.0%)25 (100.0%)<.0019 (100%)11 (100%)1.0036 (80.0%)81 (96.4%).004Baseline PC20 (mg/mL)0.8 (2.2)1.2 (1.4)10.6 (1.4)1.1 (1.1).912.0 (2.9)0.7 (1.1)<.001Baseline FVC (%pred)98.5 ± 15.1105.1 ± 13.6.22104.0 ± 15.2102.8 ± 10.0.84104.0 (9.0)104.5 (15.5).38FVC at 12 months (%pred)101.2 ± 14.8109.0 ± 12.8.14108.0 (26.0)103.5 (11.5).68107.3 ± 11.5108.1 ± 12.6.73FVC change from baseline at 12 months (%points)3.0 ± 8.24.1 ± 8.5.751.6 ± 6.8−2.3 ± 5.9.213.6 (11.2)1.1 (10.9).57Baseline FEV1 (%pred)91.0 ± 15.593.1 ± 14.4.7191.6 ± 13.389.4 ± 11.7.7199.0 (18.0)92.5 (19.5).21FEV1 at 12 months (%pred)94.3 ± 15.8101.3 ± 12.4.1893.1 ± 14.290.1 ± 9.2.59100.3 ± 13.2100.6 ± 13.3.89FEV1 change from baseline at 12 months (%points)4.4 (12.6)6.2 (13.3).362.1 ± 11.5−0.2 ± 10.3.665.8 (12.2)7.9 (17.7).17Baseline FEV1/FVC ratio (%)82.5 ± 8.678.4 ± 10.8.2979.0 (12.0)76.0 (6.0).4982.0 (8.0)79.0 (11.0).01FEV1/FVC at 12 months (% pred)82.2 ± 7.581.8 ± 8.5.8977.8 ± 7.877.1 ± 5.7.8383.5 (6.0)81.0 (7.5).29FEV1/FVC change from baseline at 12 months (% points)−0.6 (5.0)4.1 (11.2).090.2 ± 6.30.3 ± 6.2.980.6 (6.2)3.4 (8.3).002Baseline bronchodilator response (%)8.5 (6.2)11.0 (11.0).4011.0 (5.0)8.0 (6.5).427.0 (8.0)10.0 (14.0).04Peripheral blood neutrophil count (cells/mm3)1953.0 (2071.0)2736.0 (1864.0).643840.2 ± 1859.03310.0 ± 1810.0.533192.0 (1582.0)3207.0 (1529.2).84Peripheral blood lymphocyte count (cells/mm3)3033.0 (630.0)3283.0 (926.0).603039.2 ± 1166.83305.4 ± 876.0.572530.0 (888.0)3080.5 (1268.8).11Peripheral blood monocyte count (cells/mm3)560.2 ± 364.3450.0 ± 278.9.34436.9 ± 202.5469.8 ± 294.4.78354.0 (296.0)450.0 (308.2).05Peripheral blood eosinophil count (cells/mm3)36.0 (141.0)630.0 (381.0)<.001184.0 (132.0)558.0 (115.0)<.001160.0 (105.0)583.5 (308.5)<.001Peripheral blood basophil count (cells/mm3)0.0 (42.8)0.0 (72.0).320.0 (0.0)0.0 (66.0).350.0 (60.0)44.0 (78.2).16Total IgE56.5 (662.8)950.0 (1007.0).01522.0 (683.0)1051.0 (1091.0).21236.0 (533.0)660.0 (1309.5)<.001 Baseline characteristics and lung function changes according to peripheral blood eosinophil status within each molecular cluster. DISCUSSION In this study, we used k‐means clustering of miRNA expression profiles to classify participants into three possible molecular clusters. The differential miRNA expression patterns across these clusters highlight the molecular heterogeneity of asthma independent of clinical input variables. A total of 246 miRNAs were significantly differentially expressed, and each cluster showed distinct miRNA expression signatures with corresponding target genes and enriched pathways, suggesting biologically different mechanisms across the clusters. Although baseline clinical characteristics were broadly similar, the clusters differed in clinically relevant features, including lung function, eosinophil‐associated characteristics, and treatment‐related changes after budesonide exposure. In particular, Clusters 1 and 3 showed more favorable overall clinical patterns than Cluster 2 after budesonide treatment. In eosinophil‐stratified analyses, the clinical relevance of peripheral blood eosinophilia was not consistent across clusters: in Cluster 1, it appeared to be associated mainly with asthma severity, whereas in Cluster 3, it was associated with greater baseline airway hyperresponsiveness and greater improvement in FEV1/FVC after treatment. These findings suggest that the implications of eosinophilia may vary according to the underlying molecular context. We observed significant enrichment of the “interleukin‐4 and interleukin‐13 signaling” and “PIP3 activates AKT signaling” pathways across all asthma clusters. IL‐4 and IL‐13 are central cytokines in type 2 inflammation and play key roles in asthma pathogenesis. 21 The Phosphoinositide 3‐kinase (PI3K)/phosphatidylinositol‐3,4,5‐trisphosphate (PIP3) pathway is also involved in multiple aspects of asthma biology, including inflammatory cell survival, proliferation, migration, and activation, as well as airway remodeling. 22 The enrichment of these pathways across all clusters suggests that they may represent shared biologic features of pediatric asthma, regardless of atopy or peripheral blood eosinophilia. Cluster 1 showed clinically distinct patterns according to PBE status and was characterized by enrichment of several TLR signaling pathways. Participants without peripheral blood eosinophilia were more often classified as having mild asthma, whereas those with eosinophilia more often had moderate asthma, consistent with previous evidence linking eosinophilic inflammation to greater disease burden in pediatric asthma. 23 Notably, despite this difference in clinical severity, eosinophil‐stratified differences in ICS responsiveness were not prominent in this cluster. One possible explanation is that the enrichment of pathways such as the MyD88:MAL (TIRAP) cascade, TLR6:TLR2, TLR1:TLR2, TLR2, and TLR4 signaling reflects an innate immune milieu that may attenuate a more straightforward relationship between eosinophilia and corticosteroid responsiveness. Because these pathways have been implicated in neutrophil recruitment and Th17‐skewed inflammation, they may contribute to reduced steroid sensitivity in this molecular background. 24 , 25 , 26 However, this interpretation remains speculative and requires further mechanistic investigation. Clusters 2 and 3 exhibited enrichment patterns distinct from Cluster 1. Whereas Cluster 1 was characterized mainly by TLR signaling, Clusters 2 and 3 showed enrichment of VEGF and NTRK related pathways, which may indicate greater involvement of structural remodeling processes. Peripheral blood lymphocyte counts were also significantly higher in Cluster 1 than in Clusters 2 and 3. However, no clear structural changes were evident in Clusters 2 and 3, possibly because of the relatively young age of the cohort. Notably, despite partial overlap in enriched pathways, Clusters 2 and 3 were not clinically identical after budesonide treatment. Although changes in lung function over 12 months did not differ significantly, Cluster 3 showed more favorable FEV1 and FEV1/FVC values at 12 months than Cluster 2. This raises the possibility that a limited number of pathway differences may influence treatment‐related outcomes between molecularly related clusters. In Cluster 2, participants were unresponsive to ICS regardless of PBE status. The “FOXO‐mediated transcription” pathway was uniquely enriched in this cluster, raising the possibility that it may be associated with steroid resistance. The FOXO (Forkhead box O) transcription factors regulate inflammation and immune responses, and increased FOXO1 expression in airway macrophages has been reported in asthma. FOXO1 can promote inflammatory cytokines and receptors such as IL‐1β, TNF‐α, and TLR4, and may also enhance M2 macrophage polarization, thereby aggravating allergic airway inflammation and potentially reducing steroid responsiveness. 27 , 28 Although a direct link between FOXO‐mediated transcription and steroid resistance in asthma has not been established, our findings suggest a possible association that warrants further study. Cluster 3, the largest group, was characterized by enrichment of SCF‐KIT signaling, suggesting that mast cell‐related biology may be an important feature of this molecular cluster. Within this cluster, participants with peripheral blood eosinophilia had lower baseline PC20 values and lower baseline FEV1/FVC ratios than those without eosinophilia, indicating greater airway hyperresponsiveness and airflow limitation at study entry. At the same time, the eosinophilic subgroup showed a greater improvement in FEV1/FVC after budesonide treatment, so that the difference between the two groups became less apparent at 12 months. These findings suggest that, in Cluster 3, eosinophilia may identify a subgroup with more severe baseline airway dysfunction but preserved corticosteroid responsiveness. Given the enrichment of SCF‐KIT signaling, one possible interpretation is that eosinophilic inflammation and mast cell‐related pathways interact to amplify airway hyperresponsiveness in this cluster while remaining responsive to ICS treatment. 29 , 30 This study used the CAMP cohort, a clinically well‐characterized pediatric asthma cohort with standardized methodologies and provides new insight into mechanistic differences among possible asthma subtypes that are not evident from phenotype alone. Although the three clusters had broadly similar clinical characteristics, molecular profiling revealed meaningful differences in airway physiology and treatment‐related patterns. These findings suggest that miRNA‐based profiling may offer additional insight into asthma subtypes and their differential clinical behavior. MiRNAs may also serve as potential biomarkers in future clinical applications, and the enriched pathways identified here may help inform new therapeutic strategies. Several limitations should be noted. First, the identified clusters were not validated in an independent cohort. External replication will be important to determine the robustness and generalizability of these findings, especially because comparable pediatric asthma cohorts with serum miRNA data and longitudinal treatment response data are limited. Second, because the same miRNA features were used both to derive the clusters and to summarize differential expression across them, these differential‐expression results should be interpreted as descriptive of cluster composition rather than as an independent validation of cluster separation. Third, although the overall cohort size was substantial, stratification into molecular clusters and eosinophil‐defined subgroups resulted in relatively small subgroup sizes, which may have limited statistical power for some comparisons. Fourth, the miRNA profiles analyzed in this study were derived from serum, which reflects a composite signal from multiple circulating cell types rather than the airway compartment itself. Because cell‐type deconvolution was not performed, the cellular origin of the differentially expressed miRNAs remains uncertain. In addition, serum miRNA patterns may not directly reflect local airway biology, and the molecular clusters identified here may not fully correspond to airway‐level endotypes. Finally, because clustering was based solely on circulating miRNA profiles, other relevant biologic processes may not have been captured. Future studies incorporating external validation cohorts and airway‐based or multi‐omic data will be important to refine the biologic interpretation of these clusters. In conclusion, our findings emphasize the importance of molecular phenotyping in asthma. The distinct miRNA expression profiles and associated pathways observed across the three clusters suggest different pathogenic mechanisms that are not apparent from conventional clinical assessment alone. These results may help guide more targeted therapeutic approaches, improving outcomes for patients with asthma. Future studies should validate these clusters in larger populations and explore the therapeutic implications of the pathways identified in each cluster. AUTHOR CONTRIBUTIONS MY contributed to the study concept and design, data analysis and interpretation, and drafting of the manuscript. US, AT, AW, and SP contributed to microRNA data processing, bioinformatic analyses, and interpretation of molecular findings. AK and MM contributed to computational analysis, statistical methodology, and data curation. SW contributed to study design, cohort resources, and critical revision of the manuscript for important intellectual content. KT supervised the study, contributed to study conception and design, and critically revised the manuscript. CONFLICT OF INTEREST STATEMENT All authors declared that none have a conflict of interest related to this article. Supporting information Table S1. Figure S1. ACKNOWLEDGMENTS This work was supported by NIH R01 HL162570, R01 HL161632, R01 HL139634, R01 HL155742, and R01 HL127332. Yang M, Srivastava U, Kho A, et al. MicroRNA clustering in pediatric asthma reveals differential corticosteroid responsiveness. Pediatr Allergy Immunol. 2026;37:e70374. doi: 10.1111/pai.70374 Editor: Ayobami Akenroye REFERENCES Associated Data Supplementary Materials Table S1. Figure S1.