Age-related pharyngeal microbiome and host transcriptomic signatures underlying fever responses in RSV bronchiolitis

Abstract

Respiratory syncytial virus (RSV) bronchiolitis is the leading cause of hospitalization in infancy and exhibits pronounced age-dependent clinical heterogeneity. Fever becomes increasingly prevalent with age, yet whether febrile representation reflects a uniform inflammatory and immune phenotype across infancy remains unclear. In this prospective cohort of infants hospitalized with RSV bronchiolitis, we performed an integrated analysis of clinical features, pharyngeal microbiome composition, host transcriptomic profiles, and host-microbe interaction networks, with particular attention to age-related variation in fever-associated patterns. Clinically, fever prevalence exhibited a strong age-dependent increase across infancy. Correspondingly, canonical correspondence analysis identified age and fever as dominant gradients related to variation in both pharyngeal microbiome composition and host gene expression. Although no significant age-dependent correlations were observed at the global microbial and host transcriptomic levels in the fever-age interaction model, distinct patterns of microbial and host responses related to fever were observed across different age groups. Specifically, ranked gene set enrichment analysis indicated that febrile infants in early infancy showed relative attenuation of host defense-related programs, whereas older infants showed stronger enrichment of antiviral and inflammatory effector pathways, with more selective regulatory and signaling-associated patterns in late infancy. Integrated host-microbe network analysis further delineated a coherent developmental trajectory of fever-associated interaction architectures, evolving from densely interconnected regulatory networks in early infancy to modular, selectively coupled, host-centered configurations with advancing age. Together, febrile responses in RSV bronchiolitis should not be interpreted as a uniform biological phenotype across infancy and support age-aware interpretation of fever in pediatric RSV infection.

Highlights

• •Fever reflects age-related immune and host-microbe heterogeneity in RSV bronchiolitis.
• •Host transcriptomic responses to fever diverge markedly across infancy.
• •Early-life fever associates with immune suppression, later with effector activation.
• •Host-microbe networks show age-dependent immune maturation trajectories.

Introduction

Respiratory syncytial virus (RSV) is the most common cause of acute lower respiratory tract infection in young children worldwide, accounting for a substantial proportion of hospitalizations during early life (Li et al., 2022). Almost all children experience their first RSV infection by two years of age (Hall et al., 2009; Mazur et al., 2015; Hartmann et al., 2022). Despite advances in preventive and therapeutic strategies, including long-acting monoclonal antibodies and ongoing clinical development of antiviral therapies, RSV continues to impose a significant burden on pediatric health systems and remains a major cause of morbidity (Muller et al., 2023; Zhao et al., 2024; Feng et al., 2025; Yu et al., 2025).

Bronchiolitis caused by RSV frequently requires emergency evaluation and respiratory support (Ralston et al., 2014), and accumulating evidence indicates that severe RSV bronchiolitis in early life is associated with an increased risk of recurrent wheezing and the subsequent development of childhood asthma (Rosas-Salazar et al., 2023). Importantly, RSV bronchiolitis is not a uniform clinical entity; rather, it displays marked heterogeneity in clinical presentation, inflammatory manifestations, and disease trajectory. Clinically, younger infants are more likely to present with respiratory distress and hypoxemia, whereas fever becomes increasingly common with advancing age (Schroeder et al., 2013; Kawakami et al., 2018; Colosia et al., 2023). These observations suggest that fever during RSV infection may reflect age-dependent immune competence rather than a uniform marker of disease severity.

Fever is widely regarded as a hallmark of systemic inflammatory activation during infection (Blomqvist and Engblom, 2018). However, in early life, the immune system undergoes rapid maturation (Kollmann et al., 2012), raising the question of whether febrile responses in infants reflect a consistent biological process or differ across developmental stages. In RSV bronchiolitis, it remains unclear whether fever represents a uniform inflammatory phenotype or instead reflects age-dependent immune responses with distinct underlying biological features.

Recent studies have highlighted the importance of both airway microbiota (Mejias and Ramilo, 2020; Wei et al., 2026) and host transcriptional responses (de Steenhuijsen Piters et al., 2016; Fujiogi et al., 2022; Zivanovic et al., 2023) in shaping disease severity and clinical phenotypes in RSV infection. However, most analyses have focused either on individual omics layers or on pooled populations, with limited attention to how fever-associated biological signatures vary across age. In particular, it remains unknown whether fever-associated microbiome composition, host transcriptional programs, and host-microbe interactions are consistent across infancy or exhibit developmental heterogeneity.

To address these gaps, we conducted an integrated analysis of clinical features, pharyngeal microbiome profiles, host transcriptomes, and host-microbe interaction networks in infants hospitalized with RSV bronchiolitis. Using unified adjusted models and age-stratified analyses, we aimed to determine whether fever represents a consistent biological phenotype or a developmentally heterogeneous response across infancy.

Results

Clinical characteristics and age-dependent fever prevalence in RSV bronchiolitis

A total of 83 infants hospitalized with RSV bronchiolitis were included in the analysis (Table 1). The median age was 7.12 months (interquartile range [IQR], 3.73–13.29), and 32.5% were female. More than half of the infants were delivered by cesarean section, and 8.4% were born preterm (<37 weeks’ gestation). A documented history of eczema and wheezing was present in 28.9% and 13.3% of patients, respectively. At hospital presentation, fever was observed in 53.0% of infants, dyspnea in 48.2%, and 41.3% required supplemental oxygen. The median length of hospital stay was 4.76 days (IQR, 3.80–4.99). To examine age-related differences in clinical manifestations, patients were stratified into four developmental age groups (Supplementary Table S1). The prevalence of fever increased significantly with age, rising from 17.6% among infants aged 1–3 months to 76.0% among those aged 13–24 months (P = 0.001, Fig. 1A, Supplementary Table S1). In the adjusted logistic regression model that included variables such as oxygen therapy, antibiotic use, and disease duration, age remained independently associated with the occurrence of fever as both a categorical variable and a continuous variable (Supplementary Table S2). The results were consistent in the sensitivity analysis excluding premature infants, indicating that there is a clear developmental gradient in fever manifestation during early life.

Table 1.

Patient characteristics of infants hospitalized for RSV-bronchiolitis.

Variable	Overall (n = 83)
Characteristics
Age (months), median (IQR)	7.12 (3.73, 13.29)
Female sex	27 (32.5)
C-section delivery	44 (53.0)
Prematurity (<37 weeks)	7 (8.4)
History of eczema	24 (28.9)
History of wheezing	11 (13.3)
Prior antibiotic use	57 (68.7)
BMI for age
Overweight	6 (7.2)
Obesity	3 (3.6)
Thinness	3 (3.6)
Severe thinness	0 (0.0)
Clinical presentation
Fever	44 (53.0)
Dyspnea	40 (48.2)
Duration of symptoms at admission≤3	17 (20.5)
Duration of symptoms at admission>3	66 (79.5)
Therapies
Oxygen use therapy	27 (32.5)
Length of stay (days), median (IQR)	4.76 (3.80, 4.99)

RSV: respiratory syncytial virus. IQR: interquartile range. Data are n (%).

Fig. 1.

Clinical age gradient of fever and unified modeling of microbiome and host transcriptomic variation. A Prevalence of fever across age groups (1–3, 4–6, 7–12, and 13–24 months), showing an increasing trend with age. Numbers above bars indicate the proportion of febrile cases within each group. B, C CCA of (B) microbiome genus-level composition and (C) host transcriptomic profiles. Arrows indicate the direction and strength of association for clinical variables, including age, fever, antibiotic exposure, oxygen supplementation, disease duration, and preterm status. The proportion of variance explained by each axis is indicated. D Predicted probability of fever across age (modeled as a continuous variable) from logistic regression, with shaded areas indicating 95% confidence intervals. E, F Variance explained (r²) by clinical variables in PERMANOVA models for (E) microbiome and (F) host transcriptomic data. Statistical significance is indicated as follows: ∗P < 0.05, ∗∗P < 0.01, ns, not significant.

Age and fever as dominant gradients explaining pharyngeal microbiome and host transcriptomic variation but no robust global fever-by-age interaction effects

Canonical correspondence analysis (CCA) of the pharyngeal microbiome showed that age was the strongest explanatory variable (r² = 0.53, P = 0.001), followed by febrile status (r² = 0.17, P = 0.001) and disease duration (r² = 0.12, P = 0.004) (Fig. 1B, Supplementary Table S3). In the host transcriptome, age (r² = 0.18, P = 0.002) again accounted for significant component of variation (Fig. 1C, Supplementary Table S3). It is worth noting that antibiotic exposure is also a significant factor in explaining the microbiome and host transcriptomic variation.

Due to the strong clinical effect of age on fever prediction (Fig. 1D), and the observed age and fever have certain explanatory power for the overall differences in the microbiome and host transcriptome, we next applied unified adjusted models to formally test whether fever-associated variation in the microbiome and host transcriptome was modified by age. At the microbiome level, main-effects PERMANOVA showed that age group was independently associated with overall Bray-Curtis community structure (r² = 0.108, P = 0.001, Fig. 1E), whereas febrile status was not (r² = 0.016, P = 0.187). In the corresponding interaction model, the fever-by-age term was not significant (r² = 0.033, P = 0.428). For the host transcriptome, global structure was influenced primarily by febrile status (r² = 0.05, P = 0.003) and antibiotic exposure (r² = 0.034, P = 0.017, Fig. 1F), while no robust global fever-by-age interaction was detected in distance-based or transcriptome-wide interaction analyses after multiple-testing correction. Together, these results indicate that age represents the dominant shared structural factor across omics layers, while strong global age-modified fever effects were not supported in the unified adjusted models.

Modest and age-specific fever-associated alterations in the pharyngeal microbiome

To characterize the overall organization of the pharyngeal microbiome and its variation across age and fever status, we examined microbial composition and hierarchical clustering patterns (Fig. 2). At the genus level, the pharyngeal microbiome of infants with RSV bronchiolitis was highly heterogeneous across individuals but largely dominated by Streptococcus, followed by varying contributions from Rothia, Veillonella, Toxoplasma, Staphylococcus, Prevotella, Gemella, and Neisseria. Hierarchical clustering based on genus-level composition did not reveal discrete microbial clusters strictly segregated by age group or fever status. When stratified by age, differences in microbial α-diversity between febrile and afebrile infants were minimal across all groups (Supplementary Fig. S1). In contrast, fever-associated differences in β-diversity were observed in the 1–3 months group (P = 0.007) and 7–12 months group (P = 0.011), whereas no significant separation in other age groups, indicating that fever-related alterations in overall community structure were largely restricted to early infancy (Supplementary Fig. S1). Age-stratified, feature-level analyses further revealed pronounced age specificity in fever-associated microbial signatures (Supplementary Fig. S2). In the 1–3 months and 4–6 months groups, none of the top-ranked genera exhibited statistically significant associations with fever, and most estimated odds ratios were close to 1 or with broad confidence intervals (Supplementary Fig. S2A and B). In contrast, infants aged 7–12 months displayed multiple genera with statistically significant associations with fever, characterized by odds ratios consistently deviating from unity, such as febrile-associated genus Akanthomyces, Ustilaginoidea, and afebrile-associated genus Phascolarctobacterium (Supplementary Fig. S2C). In the 13–24 months group, only a small number of genera showed nominally significant associations like Allomeiothermus, while most features were associated with modest effect sizes (Supplementary Fig. S2D). Collectively, fever-associated microbial differences were generally subtle and context-dependent, without evidence of a conserved fever-specific microbiome signature across age groups.

Fig. 2.

Pharyngeal microbiome composition and hierarchical clustering across age and fever status. Heatmap and hierarchical clustering of genus-level relative abundance profiles of the pharyngeal microbiome in infants hospitalized with RSV bronchiolitis. Each row represents an individual infant, and each column corresponds to a bacterial genus. Relative abundances are shown as proportional stacked bars. Samples are hierarchically clustered based on genus-level composition using an unsupervised clustering approach. Annotation bars indicate age group (1–3, 4–6, 7–12, and 13–24 months) and fever status (febrile vs. afebrile).

Age-stratified host transcriptional programs associated with fever

Because transcriptome-wide gene-level significance was limited after multiple-testing correction, we next used ranked gene set enrichment analysis (GSEA) to characterize pathway-level differences associated with febrile status across developmental stages (Fig. 3).

Fig. 3.

Age-stratified GSEA of fever-associated host transcriptional programs. Heatmaps showing GO biological processes (A) and KEGG pathways (B) associated with febrile status across age groups (1–3, 4–6, 7–12, and 13–24 months), based on ranked gene set enrichment analysis. Color intensity represents normalized enrichment scores (NES), with red indicating positive enrichment in febrile infants and blue indicating negative enrichment. Asterisks indicate statistical significance (∗adjusted P < 0.05, ∗∗adjusted P < 0.01, ∗∗∗adjusted P < 0.001).

In the 1–3 months age group, febrile status was not associated with coordinated enrichment of canonical antiviral or inflammatory pathways. Positively enriched Gene Ontology (GO) terms were mainly dominated by extracellular structure organization, external encapsulating structure organization, cell adhesion, glutamatergic signaling, and sensory/neuroactive signaling-related processes (Fig. 3A). Consistently, positively enriched KEGG analysis included protein digestion and absorption, ECM-receptor interaction, cytoskeleton-related pathways, and neuroactive ligand signaling (Fig. 3B). By contrast, negatively enriched terms included processes related to viral entry, vesicular transport, and host cellular processing supporting the view that febrile responses in early infancy were not accompanied by a strong canonical inflammatory transcriptional program.

In infants aged 4–6 months, GSEA revealed a more heterogeneous transitional pattern. Positively enriched GO terms were dominated by protein quality control and cellular stress-response programs, including response to unfolded protein, response to topologically incorrect protein, regulation of RNA processing, and ubiquitin-dependent protein catabolism (Fig. 3A). Correspondingly, KEGG pathways enriched in the febrile direction included antigen processing and presentation, endocytosis, autophagy, and protein processing in the endoplasmic reticulum, whereas negatively enriched pathways involved drug/xenobiotic metabolism and selected signaling pathways (Fig. 3B). Together, these findings suggest that febrile responses in this intermediate age window were characterized by mixed antiviral, stress-adaptive, and metabolic features rather than a uniformly polarized inflammatory profile.

By 7–12 months, fever-associated transcriptional programs shifted toward broad and coordinated immune activation. Positively enriched GO pathways included macroautophagy, endosomal transport, immune response-activating signaling pathway, and regulation of innate immune response (Fig. 3A). KEGG analysis similarly demonstrated strong positive enrichment of endocytosis, autophagy, lysosome biogenesis, and multiple infection-annotated immune pathways, including Epstein-Barr virus infection, herpes simplex virus 1 infection, Salmonella infection, and measles, reflecting activation of shared host defense modules (Fig. 3B). In contrast, negatively enriched pathways were mainly related to olfactory and neuroactive signaling, indicating that the dominant febrile signature in this age group was a robust immune effector program.

By contrast, in infants aged 13–24 months, the febrile transcriptomic profile remained immune-related but appeared comparatively more selective and regulatory. Positively enriched GO terms included antigen processing and presentation, regulation of innate immune response, regulation of response to cytokine stimulus, and T cell activation involved in immune response (Fig. 3A). KEGG pathways enriched in the febrile direction included influenza A, coronavirus disease–COVID-19, antigen processing and presentation, neutrophil extracellular trap formation, and lipid and atherosclerosis, consistent with continued activation of host defense and inflammatory pathways (Fig. 3B). At the same time, several negatively enriched pathways involved olfactory/neuroactive signaling, lipid metabolic regulation, and selected hormone- or PI3K-Akt-related pathways, suggesting that febrile responses in later infancy were accompanied by a more refined balance between immune activation and broader regulatory signaling processes.

Sensitivity analyses combining infants aged 1–6 months yielded broadly similar pathway-level conclusions, with the merged early-infancy group still lacking the coordinated inflammatory effector enrichment seen in older infants (Supplementary Fig. S3). While some pathway-level features in the 4–6 months group overlapped with those observed in older infants, the overall transcriptional profile remained distinct, supporting a gradual developmental transition rather than an abrupt shift. Overall, GSEA indicated that fever-associated host responses were not uniform across infancy. Instead, they progressed from relative attenuation of classical immune-defense programs in early infancy, to a mixed stress-adaptive and antiviral profile in mid-infancy, followed by strong innate and inflammatory effector activation in 7–12 months, and a more selective immune-regulatory configuration in 13–24 months. These pathway-level results provide a functional overview of coordinated transcriptional trends but do not imply uniform activation at the individual gene level.

Transcriptional features of immune- and inflammation-related genes across age strata associated with fever

Given the limited number of transcriptome-wide differentially expressed genes after multiple-testing correction, we performed a targeted analysis focusing on a curated set of immune- and inflammation-related genes to illustrate gene-level expression patterns associated with febrile status across age groups. In age-stratified comparisons, nominally significant differences between febrile and afebrile infants exhibited substantial heterogeneity across developmental stages (Fig. 4, Supplementary Fig. S4). These gene-level findings are presented for descriptive and hypothesis-generating purposes and should not be interpreted as evidence of robust transcriptome-wide differential expression. Consistent with the pathway-level patterns identified by GSEA, both the direction and magnitude of gene-level expression changes varied across age groups.

Fig. 4.

Age-stratified immune- and inflammation-related host gene expression patterns associated with fever. Dot plots showing expression patterns of selected immune- and inflammation-related genes in febrile and afebrile infants with RSV bronchiolitis across four age groups: (A) 1–3 months, (B) 4–6 months, (C) 7–12 months, and (D) 13–24 months. Dot color represents the average gene expression level (log-transformed), and dot size indicates the proportion of samples expressing each gene within the corresponding group.

In the 1–3 months group, febrile infants showed broadly reduced expression of multiple genes involved in innate immune defense and myeloid-associated inflammatory responses, including FCGR3B, CXCR2, CXCL1, and several members of the S100 family, compared with afebrile counterparts (Fig. 4A, Supplementary Fig. S4A). Within the 4–6 months group, fever-associated transcriptional responses showed substantial gene-level heterogeneity. Genes belonging to the same functional modules, such as chemokine signaling, cytokine responses, and antimicrobial immunity, did not display concordant directional regulation at the pathway level, but instead exhibited gene-level divergence within shared functional modules (Fig. 4B, Supplementary Fig. S4B). By contrast, in the 7–12 months group, febrile infants demonstrated broadly elevated expression of genes involved in innate immune activation, chemokine signaling, and inflammatory amplification, including IL1B, TNF, CXCL16, and interferon-responsive genes (Fig. 4C, Supplementary Fig. S4C). In the 13–24 months group, febrile expression patterns were more selective and involved genes related to immune regulation, epithelial responses, and signal transduction rather than widespread activation of inflammatory or chemokine-driven gene programs (Fig. 4D, Supplementary Fig. S4D). In sensitivity analyses combining infants aged 1–6 months, gene-level differences associated with fever were modest and variable, without clear evidence of a coordinated transcriptional response (Supplementary Fig. S5).

Developmental progression of fever-associated host-microbe immune networks

Integrated network analysis revealed a progressive, age-dependent reorganization of fever-associated host-microbe interactions architectures across infancy (Fig. 5). In early infancy (1–3 months), fever-associated networks exhibited a densely interconnected but weakly specialized interaction topology, characterized by extensive coupling between multiple microbial taxa and host regulatory genes across modules (Fig. 5A). Network-defined module 1 integrated opportunistic or environmentally associated microbial taxa (e.g., Moellerella, Hydrogenibacillus, and Borreliella) with host genes involved in immune regulation (e.g., RFX1, SLC19A1, RPS6KA4), transcriptional control, and cell cycle (Supplementary Table S4). During mid-infancy (4–6 months), host-microbe networks underwent marked structural reorganization, characterized by pronounced modular segregation between host and microbial components (Fig. 5B). A distinct interaction module (module 2) was anchored by a phage-related microbial signal (Moineauvirus) together with host genes involved in innate immune signaling regulation and cellular signaling organization including components of NF-κB and MAPK signaling pathways (e.g., IKBKG and DUSP22), as well as genes related to signal–metabolic coupling, membrane phospholipid remodeling, and vesicle-mediated signal trafficking (Supplementary Table S5). This stage was distinguished by reduced direct host-microbe coupling and increased compartmentalization of network structure. By mid-to-late infancy (7–12 months), fever-associated networks displayed a streamlined and selectively connected architecture with fewer nodes (Fig. 5C). Network-defined module 3 exhibited a host-centered immune effector architecture, with multiple host genes forming network hubs, including SF3A1, ETAA1, ST13, MLX, and NIPSNAP2, which are involved in transcriptional regulation, replication stress responses, cellular stress handling, and immunometabolic adaptation. In contrast, microbial nodes were diverse and exhibited low connectivity, comprising a heterogeneous set of environmental-associated and opportunistic taxa without a dominant microbial hub (Supplementary Table S6). Both positive and negative host-microbe correlations were observed, indicating selective and directional coupling within an overall host-centered network organization. In late infancy (13–24 months), network topology further evolved toward a modular yet integrative configuration (Fig. 5D). Network-defined module 4 was centered on genes involved in signal integration, cytoskeletal organization, membrane trafficking, protein processing, metabolic adaptation, and transcriptional or non-coding RNA regulation, including IQGAP2, CCM2, TOM1L1, DCXR, PHF14, and STX18-AS1. In parallel, microbial signals, primarily represented by Moellerella and Geobacter, exhibited moderate connectivity and were embedded within defined host modules rather than dominating overall network structure (Supplementary Table S7). It should be noted that these network structures are based on correlation patterns and are therefore descriptive, without implying direct biological interactions or causal relationships.

Fig. 5.

Integrated host-microbe interaction networks associated with fever across age groups. Network visualizations depicting age-stratified host-microbe interaction architectures associated with febrile RSV bronchiolitis: (A) 1–3 months, (B) 4–6 months, (C) 7–12 months, and (D) 13–24 months. Nodes represent host genes or microbial taxa, and edges represent significant correlations between host and microbial features. Node shape indicates feature type (host gene or microbe), node color denotes network module membership, and node size is proportional to node degree. Edge color and width reflect the direction (positive- or negative-related) and absolute magnitude of correlations, respectively.

Discussion

This study provides an integrative, age-resolved view of how fever relates to host immune transcriptional responses and airway microbial interactions during RSV bronchiolitis in early life. Across clinical and multi-omic analyses, fever-associated signatures were strongly structured by age and were not consistent with a uniform inflammatory phenotype. While fever prevalence increased with age, global models did not support robust fever-by-age interaction effects, indicating that the observed differences are better interpreted as developmental heterogeneity rather than formal interaction effects. These findings should therefore be interpreted as descriptive age-stratified patterns rather than evidence of a unified or causal framework.

Host transcriptional responses to RSV bronchiolitis exhibit pronounced developmental heterogeneity in both magnitude and direction. In this study, fever-associated host responses ranged from lack of immune activation in early life to stronger immune-effector pathway enrichment in older infants, underscoring fever as a developmentally contingent immune phenotype rather than a uniform response to infection. Extensive transcriptomic profiling across in vitro systems, animal models, and clinical cohorts has established a conserved core RSV-induced immune signature, characterized by activation of innate and inflammatory pathways, including type I interferon signaling, NF-κB activation, chemokine production, and antigen processing and presentation, accompanied by viral strategies that antagonize host immunity and suppress adaptive immune responses (Brand et al., 2015; de Steenhuijsen Piters et al., 2016; Do et al., 2017; Mariani et al., 2017; Hu et al., 2020; Fujiogi et al., 2022). Furthermore, compared with blood-based analyses, airway transcriptomic profiling reveals a broader and more complex mucosal immune response during RSV infection (Do et al., 2017). These canonical transcriptional features were most evident in our cohort among infants aged 7–12 months, in whom febrile RSV bronchiolitis was marked by activation of antiviral, inflammatory, and immune effector programs. By contrast, early infancy did not show a coordinated canonical inflammatory transcriptional program, and several host defense-related pathways were relatively enriched toward afebrile infants. Patient-based transcriptomic studies consistently demonstrate age-dependent attenuation of host immune responses in early infancy. Whole-blood gene expression profiling of infants with RSV lower respiratory tract infection has shown that younger infants (<6 months) exhibit a predominance of down-regulated genes, globally reduced activation of innate and inflammatory pathways, and further suppression of adaptive immune responses compared with older children (Mejias et al., 2013). Because fever depends on coordinated activation of innate immune sensing and downstream pyrogenic cytokine production, its manifestation is inherently constrained by the developmental capacity of the host immune system. Collectively, fever during RSV bronchiolitis reflects age-dependent immune maturation rather than a uniform inflammatory response.

In contrast to the host transcriptome, fever-associated changes in the airway microbiome were modest and highly context-dependent. Although previous studies of RSV bronchiolitis have linked nasopharyngeal microbial communities enriched for potentially pathogenic genera, particularly Streptococcus and Haemophilus, to greater disease severity, underscoring the potential contribution of specific bacterial taxa to inflammatory responses during severe RSV infection (de Steenhuijsen Piters et al., 2016; Hasegawa et al., 2016; Feng et al., 2023; Kristensen et al., 2024). We did not observe a consistent fever-associated microbial signature across age groups. Instead, microbiome differences were subtle, varied across developmental stages, and did not form discrete community patterns. Importantly, the lack of segregation by age or fever status suggests that these factors did not give rise to distinct microbiome-defined community structures in this cohort. These findings suggest that microbial composition alone is unlikely to serve as a primary determinant of febrile responses, but may instead be embedded within age-specific host immune contexts.

Based on integrated host-microbe network analyses, we propose a fever-associated immune maturation framework to contextualize age-dependent differences in febrile responses during RSV bronchiolitis. In early infancy, febrile responses are relatively uncommon and are not accompanied by coordinated canonical inflammatory activation, consistent with developmental constraints of the immature immune system. As immune signaling capacity increases during mid-infancy, fever emerges alongside the priming and organization of innate immune signaling pathways, representing a transitional stage of immune readiness. By later infancy, febrile responses appeared more selective, in parallel with higher fever prevalence and greater inflammatory amplification observed clinically. Subsequently, immune responses appear to enter a remodeling phase characterized by regulatory feedback and stabilization, potentially enabling effective pathogen control while limiting excessive inflammation. This framework links age-related immune development to the epidemiology of fever in RSV bronchiolitis and suggests that fever reflects an age-dependent immune functional state rather than a uniform indicator of disease severity. Accordingly, age-aware interpretation of fever may improve clinical contextualization of RSV bronchiolitis, while future studies are needed to determine whether developmental stage-specific modulation of immune priming or inflammatory amplification can inform prognostic assessment or therapeutic strategies.

The present study has several limitations. First, its cross-sectional design precludes direct inference of within-individual temporal dynamics in airway microbiome composition and host immune responses across early infancy; longitudinal studies will be required to define true developmental trajectories. Second, although the overall cohort size supported integrated multi-omic analyses, age- and fever-stratified subgrouping reduced sample sizes, potentially limiting power to detect subtle microbiome differences, particularly at finer taxonomic resolution. In line with this, differential expression analysis identified limited numbers of genes after multiple testing correction, and most gene-level findings were based on nominal significance and should therefore be considered exploratory. Meanwhile, because of the observational design and limited subgroup sample size, it is difficult to rule out the influence of antibiotic exposure, which is associated with differences in the microbiome and the host transcriptome, and may to some extent interfere with the observed associations. In addition, other potentially relevant clinical and environmental factors, including feeding mode and RSV viral load, were not systematically available in the present cohort and therefore could not be adjusted for in the analyses. These unmeasured variables may also contribute to variation in both microbiome composition and host immune responses. Third, interaction effects between age and febrile status were not statistically significant in global models, suggesting that the observed age-dependent differences are better interpreted as stratified heterogeneity rather than formal interaction effects. Fourth, airway specimens were obtained from the pharyngeal compartment, which is ethically and practically preferable in young infants and has been shown to reflect key aspects of airway immune and microbial states (Poole et al., 2014; Marsh et al., 2016), but may not fully capture immune-microbial interactions in the lower airways. Finally, the integrated host-microbe network analyses are inherently associative and hypothesis-generating, and do not establish causal relationships, warranting further validation through longitudinal and mechanistic studies.

Conclusions

In conclusion, febrile responses in RSV bronchiolitis showed a robust clinical age gradient, but its associated microbiome and host transcriptional features were not uniform across infancy. While unified adjusted omics models did not support strong global fever-by-age interaction effects, age-stratified analyses identified developmentally heterogeneous pathway-level, gene-level, and host-microbe network patterns associated with febrile status. These findings support a more age-aware interpretation of fever in pediatric RSV infection and suggest that febrile presentation in RSV bronchiolitis reflects heterogeneous biological contexts across early life rather than a single shared inflammatory state.

Materials and methods

Study design and population

This was a prospective observational study conducted at Wuhan Children’s Hospital. Pediatric patients diagnosed with bronchiolitis and admitted to the hospital between October 1, 2020, and September 30, 2021 were consecutively enrolled, covering a full one-year period. Bronchiolitis was diagnosed according to established clinical criteria, based on the presence of acute lower respiratory tract infection symptoms, including cough, wheezing, tachypnea, and signs of increased work of breathing, with or without fever, in young children. Only patients with laboratory-confirmed respiratory syncytial virus (RSV) infection were included in the final analysis.

Inclusion criteria were: (1) hospitalization with a clinical diagnosis of bronchiolitis; (2) laboratory confirmation of RSV infection; (3) age ≤24 months at admission; and (4) availability of adequate pharyngeal airway specimens for microbiome and host transcriptomic analyses.

Exclusion criteria included: (1) pre-existing chronic lung disease (e.g., bronchopulmonary dysplasia), congenital airway or lung malformations, or significant congenital heart disease; (2) known primary or secondary immunodeficiency; (3) underlying genetic or metabolic disorders; (4) evidence of mixed respiratory infections, including co-detection of other respiratory viruses or bacterial pathogens in addition to RSV; and (5) insufficient clinical data or inadequate biological samples for downstream analyses.

Nucleic acid extraction and metatranscriptomic sequencing

Throat swabs were collected within 24 h of hospitalization and stored in −80 °C before usage. Total nucleic acids were extracted using the BeaverBeads ™ Viral DNA/RNA Kit 70406-32FU (Beaverbio, Suzhou, China) on TGuide S32 Automated Extraction (Tiangen, Beijing, China) following the manufacturer’s instructions. The RNA library was prepared using TruePrep® RNA Library Prep Kit for Illumina TR503 (Vazyme, Nanjing, China) according to the manufacturer’s instructions with several adjustment as in our previous study (Lu et al., 2021). For each library, 8 μL of total nucleic acids extracted from each sample was used without rRNA removal. Sequencing was carried out on the MGISEQ-T7 platform with 150 bp of paired-end raw reads generated, after converting sequencing library style to T7 library.

Raw data analysis

Quality control and preprocessing of sequencing data are vital for obtaining high-quality and high-confidence variants in downstream data analysis. Fastp (v0.23.4) (Chen et al., 2018) was used to filter out low-quality bases (quality score < 20) from sequencing reads. Bowtie2 (v2.5.4) (Langmead and Salzberg, 2012) was used to align sequencing reads to a reference genome to remove host ribosomal RNA (rRNA). STAR (Spliced Transcripts Alignment to a Reference) aligner (v2.7.11b) (Dobin et al., 2013) was then employed to distinguish host reads from non-host reads, aiding in the analysis of RNA-seq data. For taxonomic classification, kraken2 (v2.1.2) (Wood et al., 2019) was used to classify paired-end reads against a custom reference database comprising complete bacterial genomes, human viral genomes, and fungal genome assemblies, all acquired from NCBI’s RefSeq. Lastly, Bracken (v2.9) (Lu et al., 2017) was utilized to estimate species abundance based on k-mer profiles generated from sequence reads.

Microbiome analysis

Microbiome composition was analyzed at the genus level in infants with RSV bronchiolitis. Genus-level count data were used for diversity analyses, while reads-per-million (RPM)-normalized abundance tables were used for compositional visualization and feature association analyses. All microbiome analyses were conducted in R (version 4.5.1) using the ggplot2, the vegan (https://CRAN.R-project.org/package=vegan), and related packages.

Overall community structure was assessed using Bray-Curtis dissimilarity calculated from genus-level abundance data. Hierarchical clustering was performed using agglomerative clustering (hclust), and the resulting sample order was used for visualization of microbial composition. The top 20 genera ranked by total abundance were displayed as relative abundance bar plots, with samples ordered by hierarchical clustering or by clinical variables (age and febrile status). Sample-level annotations, including age group and fever status, were visualized alongside community composition.

Alpha diversity at the genus level was quantified using Chao1 richness, Shannon diversity, and Simpson indices. For age-stratified analyses, infants were grouped into four developmental age categories (1–3, 4–6, 7–12, and 13–24 months), and diversity indices were compared between febrile and afebrile infants within each group using the Wilcoxon rank-sum test.

Beta diversity was evaluated using Bray-Curtis dissimilarity and visualized by principal coordinate analysis (PCoA). Within each age group, differences in community composition between febrile and afebrile infants were assessed using permutational multivariate analysis of variance (PERMANOVA) implemented in adonis2 with 999 permutations.

To identify microbial genera associated with febrile status in an age-specific manner, a two-step feature analysis was applied. Random forest models were first used to prioritize genera contributing to discrimination between febrile and afebrile infants within each age group. Selected genera were subsequently evaluated using univariable logistic regression to estimate odds ratios and 95% confidence intervals for associations with fever. Results were visualized using forest plots on a logarithmic scale.

Differential gene expression analysis

Following the alignment of RNA-seq reads to the human reference genome by STAR, the featureCounts (Liao et al., 2014) tool was used to translate the data into a quantitative measure of gene expression. To generate a non-redundant gene expression matrix, duplicated gene symbols were collapsed by retaining the entry with the highest median expression across samples. Genes with zero counts in at least 50% of samples were removed. Samples with zero expression in more than 50% of retained genes were further excluded.

Differential gene expression analysis between febrile and afebrile infants was conducted using DESeq2 (Love et al., 2014). Given the limited sample size within age-stratified subgroups, transcriptome-wide multiple-testing correction resulted in a limited number of significant genes. Therefore, genes with a nominal P value < 0.05 and an absolute log2 fold change exceeding a data-driven threshold (mean absolute log2 fold change plus two standard deviations) were used primarily for descriptive and visualization purposes rather than as transcriptome-wide discovery claims.

For functional interpretation, ranked GSEA was performed using the R package clusterProfiler (Wu et al., 2021) based on ranked gene lists derived from differential expression. Genes were ranked according to the signed statistic from the linear model, and enrichment of GO biological processes and KEGG pathways was assessed using the gseGO and gseKEGG functions, respectively. Enriched terms with adjusted P values < 0.05 were retained.

Statistical modeling of fever-associated effects and age interaction

To assess whether fever-associated biological features were consistent across developmental stages, we constructed unified regression models incorporating both main effects and interaction terms between fever and age.

For clinical outcomes, logistic regression models were fitted with fever status as the dependent variable and age (modeled as either categorical age groups or a continuous variable) as the primary predictor, adjusting for relevant clinical covariates including oxygen therapy, antibiotic use, and disease duration. Model comparisons were performed using likelihood ratio tests.

For microbiome and host transcriptome data, we evaluated global compositional differences using distance-based multivariate analyses. Specifically, permutational multivariate analysis of variance (PERMANOVA) was performed on Bray-Curtis dissimilarity matrices for microbiome profiles and Euclidean distance matrices for host transcriptomic data. Models included fever status, age (categorical), their interaction term (fever × age), and the same clinical covariates (oxygen therapy, antibiotic use, and disease duration). Both marginal effects (by = “margin”) and sequential effects (by = “terms”) were examined. In addition, principal component analysis (PCA) was applied to host transcriptomic data, and linear models were used to assess associations between PCs and fever, age, and their interaction, adjusting for the same covariates.

Integrated host-microbe network analysis

To characterize age-specific fever-associated host-microbe interaction architectures, integrative multi-omics network analyses were performed separately within each age group using the DIABLO framework implemented in the mixOmics package (Rohart et al., 2017). For each age stratum, pharyngeal microbial abundance data (genus-level relative abundance), host gene expression data (normalized RPM values), and corresponding clinical metadata were first synchronized by retaining only samples shared across all datasets. Fever status was used as the response variable. Features with zero variance across samples were removed independently from microbial and host gene datasets to eliminate non-informative variables.

Microbial and host transcriptomic datasets were combined into a multi-block data structure and analyzed using block sparse partial least squares discriminant analysis (block sPLS-DA). The design matrix was specified to maximize correlations between microbial and host gene blocks while preventing within-block self-correlation. Model tuning was performed using repeated M-fold cross-validation to determine the optimal number of retained features (keepX) for each data block. The final DIABLO model was then trained using the selected parameters.

Features selected on the first latent component of the optimized DIABLO model were retained for downstream network construction. Pairwise associations between selected microbial taxa and host genes were quantified using Spearman rank correlation. Correlations exceeding a predefined absolute correlation threshold (0.9) and nominal significance level were retained as edges in the host-microbe interaction network.

Undirected host-microbe networks were constructed based on significant correlations, with nodes representing microbial taxa or host genes and edges representing positive or negative associations. Node degree was calculated to quantify connectivity, and highly connected nodes were defined as hubs based on upper quantile thresholds. Network modularity was assessed using the Louvain community detection algorithm, incorporating correlation-derived edge weights to identify densely connected host-microbe interaction modules.

All network analyses were conducted independently within each age group to enable direct comparison of age-dependent host-microbe interaction architectures associated with fever.

Data availablity

The sequencing data have been deposited in the Science Data Bank (https://doi.org/10.57760/sciencedb.36220).

Ethics statement

This study was conducted in accordance with the ethical standards of the institutional and national research committees and with the Declaration of Helsinki (1975), as revised in 2024. The research was approved by the Institutional Review Board and the Ethics Committee of Wuhan Children’s Hospital (Approval No. 2025R086-E01). All clinical data were extracted from electronic medical records and anonymized to ensure patient confidentiality.

Author contributions

Ying Li: formal analysis, investigation, methodology, visualization, writing-original draft; Jingyuan Zhang: data curation, investigation, validation, writing-original draft; Haizhou Liu: methodology; Muhammad Noman: validation, resources; Muhammad Arif Rizwan: resources; Ismaila Damilare Isiaka: resources; Hebin Chen: data curation, resources; Wenqing Li: data curation; Nan Zhao: data curation; Xiaoxia Lu: conceptualization, funding acquisition, project administration, writing-review & editing; Di Liu: conceptualization, funding acquisition, project administration, writing-review & editing; Yi Yan: conceptualization, funding acquisition, project administration, writing-review & editing.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. All authors have reviewed and approved this statement.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary Figure S1.

Age-stratified α- and β-diversity comparisons of the pharyngeal microbiome by fever status. Comparisons of within-sample (α-diversity) and between-sample (β-diversity) diversity between febrile and afebrile infants across four age groups. Panels show results for infants aged (A) 1–3 months, (B) 4–6 months, (C) 7–12 months, and (D) 13–24 months. α-diversity metrics include Chao1 richness, Shannon diversity, and Simpson diversity indices. β-diversity was assessed using PCoA based on Bray–Curtis dissimilarity. Each point represents an individual infant, with colors indicating fever status. P values indicate statistical significance of group differences within each age stratum.

Supplementary Figure S2.

Age-stratified feature-level associations with fever in the pharyngeal microbiome. Forest plots showing feature-level associations with fever status within each age group: (A) 1–3 months, (B) 4–6 months, (C) 7–12 months, and (D) 13–24 months. Odds ratios (ORs) and 95% confidence intervals were estimated for febrile versus afebrile infants using univariable logistic regression. The x-axis is displayed on a logarithmic scale, with the dashed vertical line indicating OR = 1. Points represent estimated effect sizes, and error bars denote 95% confidence intervals. Genera with statistically significant associations are highlighted in red, whereas non-significant associations are shown in blue.

Supplementary Figure S3.

GSEA of host transcriptional responses in combined age groups. (A) Heatmap of the top positively and negatively enriched GO biological processes across combined age groups (1–6, 7–12, and 13–24 months), based on ranked gene set enrichment analysis. (B) Heatmap of the top positively and negatively enriched KEGG pathways across the same age groups. Color intensity represents normalized enrichment scores (NES), with red indicating positive enrichment in febrile infants and blue indicating negative enrichment. Statistical significance is indicated by asterisks (∗adjusted P < 0.05, ∗∗adjusted P < 0.01, ∗∗∗adjusted P < 0.001).

Supplementary Figure S4.

Differential expression patterns of selected immune- and inflammation-related genes across age groups. Heatmaps showing the expression patterns of a curated set of immune- and inflammation-related genes in febrile and afebrile infants across age groups: (A) 1–3 months, (B) 4–6 months, (C) 7–12 months, and (D) 13–24 months. Gene expression values are scaled (z-score) within each gene. Samples are grouped by febrile status within each age group. The selected genes are derived from nominally significant differential expression analysis (unadjusted P < 0.05) and include key genes involved in innate immunity, chemokine signaling, and inflammatory responses.

Supplementary Figure S5.

Sensitivity analysis of host transcriptional differences in early infancy (1–6 months). Heatmap showing the expression patterns of selected immune- and inflammation-related genes comparing febrile and afebrile infants within the combined 1–6 months age group. Gene expression values are scaled (z-score) within each gene. The accompanying dot plot summarizes average gene expression (color scale) and the proportion of expressing cells/samples (dot size).