Mosquito repellency mediated by skin commensal Bacillus species via volatile metabolite reprogramming Authors: Zhenyu Gu, Heting Gao, Hongxia Guo, Zhihua Fan, Weixian Zhang, Dan Xing, Xinyu Zhou, Haotian Yu, Teng Zhao, Jiahong Wu, Chunxiao Li Published: 2026 ABSTRACT AbstractIndividual variation in human attractiveness to mosquitoes critically influences vector-borne disease transmission, and this phenomenon arises from multiple underlying factors. This study integrated human skin microbiome profiling, non-human primate in situ intervention, and volatilomics to uncover a microbial-mediated mechanism underlying host-specific mosquito repellency. Bacillus species were identified as a microbial signature associated with reduced mosquito attraction in humans. Targeted Intervention of B. licheniformis and B. paralicheniformis on primate skin was sufficient to significantly reduce mosquito attraction, thereby establishing a causal microbial contribution beyond descriptive correlations. Mechanistic analyses revealed that Bacillus enrichment reprogrammed the skin volatilome, specifically increasing the abundance of aromatic and fatty acid-derived effectors, such as p-Cresol and decanoic acid, which are associated with repellency phenotypes. Together, these findings revealed a potential role of skin commensals in shaping mosquito host-seeking behavior, providing a molecular basis for microbiome-informed and bio-inspired strategies for mosquito disease control.Subject terms: Applied microbiology, High-throughput screening, Metagenomics Introduction Mosquitoes, as medically important insects with global distribution, have triggered a public health crisis that has drawn international attention. Beyond direct discomfort from bites, such as skin redness, swelling, and itching, mosquitoes serve as the primary transmission vectors for various life-threatening mosquito-borne diseases, including dengue fever, Zika virus disease, and malaria1–4. Notably, malaria is currently the strongest selective force in the human genome and has been linked to the prevalence of sickle cell trait and G6PD deficiency5,6. In the absence of widely available vaccines and targeted therapeutics for most highly pathogenic mosquito-borne viruses, repellents serve as a practical auxiliary measure to limit pathogen transmission7,8. However, traditional plant-based products suffer from excessively short protection periods, while chemically synthesized repellents, despite strong efficacy, can exhibit mild toxicity and ecological pollution9,10. Improving the performance and safety of existing mosquito repellents, as well as exploring alternative volatile active substances, is therefore of great research significance for optimizing daily anti-mosquito strategies. Mosquito attraction to human hosts exhibits striking interindividual variation, with some people consistently attracting fewer mosquitoes than others under comparable environmental conditions. This heterogeneity has been repeatedly observed in both daily life and laboratory data. Extensive research has confirmed that human body temperature, exhaled carbon dioxide, and skin-surface volatile chemicals are key signals for mosquitoes to locate hosts11,12. Among these, individual differences in skin volatile metabolites are the core factor determining mosquito biting preferences13. Some populations form a natural defense barrier against mosquitoes due to skin odor composition, exhibiting significantly reduced attractiveness to mosquitoes7. As the largest human organ, the skin is colonized by a diverse and complex microbial community14,15. Volatile organic compounds (VOCs) produced by these microorganisms during metabolism can directly reshape the skin odor profile, thereby regulating mosquitoes’ host-seeking behavior16,17. This discovery highlights the potential of skin microbiota in repelling mosquitoes and provides an important entry point for exploring natural repellent resources. In recent years, with advances in high-throughput sequencing technologies, including metagenomics and metabolomics, the association between skin microbiota and mosquito attraction has attracted increasing research attention. However, a critical scientific gap persists in this field: although human body odor differences drive mosquito preferences and skin microbiota metabolism produces volatile compounds, the specific microbial taxa and underlying molecular mechanisms shaping interindividual mosquito attractiveness remain poorly defined and incompletely characterized. Previous studies have mostly been confined to correlation analyses between microbial community structure and mosquito attractiveness18,19. Accumulating evidence has confirmed the existence of the “microbiota–metabolite-mosquito behavior” regulatory chain, while the specific influential microbial species and their pivotal functional metabolites still require further consolidation and supplementation20–22. In addition, most existing studies use human volunteers or mice as models7,16,23. Although interindividual heterogeneity in host genetic, environmental, and behavioral factors do not shape the core human skin microbiome24,25, such individual variations can alter the overall microbial community composition and increase the complexity of experimental variable control26. While mouse models are well applicable for the validation of compound bioactivity, species-specific disparities in skin microecology restrict the translational potential of research on skin microbial colonization27. However, although relevant correlations have been documented in previous research, the causal and stable effects of specific skin-resident microbes on mosquito repellency via distinct metabolic products still lack comprehensive and in-depth mechanistic verification18. To address these limitations, this study integrates metagenomics, metabolomics, and a targeted dual-model design and employs a non-human primate (NHP) model and a mouse model in different experiments. The NHP model was used to conduct Bacillus spp. smearing experiments, which align with the translational requirements of skin microbial colonization research, while the mouse model was employed to complete the validation of repellent compounds, ensuring the accuracy of compound activity evaluation. This dual-model design effectively compensates for the shortcomings of a single model, thereby enhancing the reliability and reference value of the research conclusions. This study further complements previous efforts by exploring tentative causal links among the microbiota–metabolites–mosquito behavior, beyond conventional correlation-based analyses. Specifically, the present work characterized associations between skin microbial structure and mosquito attractiveness, screened and validated candidate microbial species with significant repellent activity, identified key potential metabolites, and dissected the underlying regulatory mechanisms. Experimental evidence suggests that skin-residing Bacillus spp. may serve as key “repellent probiotics”, which might exert mosquito-defensive effects by remodeling the skin volatile metabolome, without causing any skin health issues. This work provides a new perspective for understanding mosquito host selection, identifies promising candidates of natural repellent microorganisms and metabolic targets, and lays a preliminary theoretical foundation for developing microbiota-based natural repellents, offering potential leads for microbiome-informed mosquito bite reduction strategies. Results Skin microbiota composition is associated with differential mosquito attraction in humans Mosquito attraction assays were initially conducted on 37 healthy volunteers (Fig. 1A). Individuals were categorized into high- (n = 22) and low-attractiveness (n = 15) groups using Chi-square tests based on significant intra-individual differences between experimental and control arms (Fig. 1B). Validation of this grouping strategy confirmed that the high-attractiveness group not only exhibited significant within-group differences but also attracted significantly higher mosquito counts compared to the low-attractiveness group (Fig. 1B). Subsequent analysis of baseline characteristics, including age, gender, and BMI, revealed no significant differences between the two groups (P > 0.05), thereby excluding these factors as confounding variables (Fig. 1C). Distinct skin microbiome signatures characterize individuals with innate resistance to mosquitoes in human volunteers. A Using a single-channel olfactometer to measure the mosquito attraction ability of volunteers. B Comparison of mosquito attraction rates between the high-attraction and low-attraction groups (n = 30 for both the treatment group and control group). Grouping was determined by single-channel olfactometer assays (Chi-square test, one-sided, P < 0.05). C Demographic profiles of volunteers, stratified by mosquito attraction groups. D High-abundance microbial composition at the phylum levels across all volunteers. E High-abundance microbial composition at the genus level across all volunteers. F α-diversity indices (Shannon index and Simpson index) of the volunteer skin microbiota in the high-attraction group (n = 21) and low-attraction group (n = 16) (Wilcoxon rank-sum test, two-sided). (G LEfSe analysis identifying significant biomarkers differentiating the high-attraction group (n = 21) and low-attraction group (n = 16). Kruskal–Wallis H test (two-sided, P < 0.05) was used for intergroup differential screening followed by linear discriminant analysis (LDA score >3, P < 0.05). No additional multiple comparison adjustment was performed. H Comparison of absolute abundances for key differential species identified by metagenomic sequencing between the high-attraction group (n = 21) and low-attraction group (n = 16). All boxplots in this figure were constructed following the standard definition: horizontal lines inside each box represent the first quartile (Q1), median, and third quartile (Q3) of the dataset; whiskers stretch to the farthest data points within a distance of 1.5 times the interquartile range (IQR) relative to the box boundaries. Source data are provided as a Source Data file. Metagenomic sequencing identified the high-abundance skin microbiota of volunteers. At the phylum level, the top nine abundant phyla, including Proteobacteria, Firmicutes, Actinobacteria, and other taxa accounted for 99.92% of total relative abundance (Fig. 1D). At the genus level, the top 9 abundant genera, including Bacillus, Cutibacterium, and other taxa constituted 90.03% of relative abundance and were stably detected in all samples (Fig. 1E). Significant differences in skin microbiota structure existed between the two groups. At the phylum level, Proteobacteria were more abundant in the high-attractiveness group (65.49% vs. 27.65% in the low-attractiveness group), while Firmicutes were less abundant (10.93% vs. 65.57%). At the genus level, Bacillus abundance was significantly lower in the high-attractiveness group (0.57% vs. 62.72%). α-diversity analysis showed significantly higher Shannon and Simpson indices in the high-attractiveness group (P < 0.05), indicating greater species richness and evenness (Fig. 1F), suggesting a close association between microbial diversity and mosquito attraction phenotype. LEfSe analysis (LDA > 3, P < 0.05) identified 10 differential species. Staphylococcus epidermidis and other taxa were enriched in the high-attractiveness group, while Bacillus (including B. licheniformis, B. paralicheniformis), Bacillaceae, and Nitratireductor sp. OM-1 were enriched in the low-attractiveness group (Fig. 1G). Normalized absolute abundances of these key bacteria showed consistent variation trends with relative abundances (P < 0.05) (Fig. 1H). Collectively, microbiota differences between groups involve both overall diversity and high-abundance taxa of Proteobacteria and Firmicutes. These results indicate that variations in skin microbial composition and metabolic characteristics mediated by specific functional bacteria drive differential mosquito attraction, providing key targets for subsequent mechanistic exploration. For subsequent microbial colonization experiments and metabolomic analyses, Nitratireductor sp. OM-1 was excluded due to being unclassified. Among the enriched taxa, Bacillus species were prioritized for the following validation based on their consistent enrichment in low-attractiveness individuals, known colonization capacity, and reported metabolic versatility. Importantly, because commercially purchased standard strains were adopted in this study––and there are potentially genetic or metabolic differences between commercial and native strains––there may be subtle differences in functional phenotypes between commercial strains being tested and the native strains identified on subjects. Topical application of Bacillus reshapes the skin microenvironment and confers mosquito repellency in a NHP model To test whether Bacillus spp. enrichment is sufficient to causally reduce mosquito attraction, an in situ skin colonization model was established in cynomolgus monkeys. Three cynomolgus monkeys were subjected to baseline microbiota, metabolome, and attraction assessments. B. licheniformis and B. paralicheniformis were topically applied to the dorsal skin, followed by mosquito attraction assays and multi-omics analysis (Fig. 2A, B). Behavioral tests showed significantly reduced female Ae. aegypti attraction to Bacillus-treated skin compared to the control group, with RPI values < 1 at all time points. Significant reductions in attraction were sustained on days 2, 5, and 8 without attenuation (Fig. 2D). Dynamic monitoring of the microenvironment revealed that colonization by the two Bacillus strains significantly decreased microbial α-diversity (Fig. 2C). Prior to intervention, sequencing analysis of the baseline state identified 949 bacterial species across 19 phyla. Consistent dominant bacteria composition among monkeys indicated a stable baseline microbiota, where the top 9 abundant phyla (including Firmicutes and Actinobacteria) accounted for 99.96% of abundance, with Firmicutes (> 50% per site) as the high-abundance phylum (Fig. 2E). PCoA (Bray–Curtis distance) based on baseline data showed no site-specific clustering of microbial communities, confirming inter-site similarity (Fig. 2F). In situ inoculation of Bacillus reshapes the NHP skin microbiome and confers mosquito resistance. A Experimental design for skin bacteria application, microbiome, and metabolome detection in NHP. B Experimental design for mosquito attractiveness of NHP skin. C Changes in α-diversity (Shannon index) following the application of B. spp. (Wilcoxon rank-sum test, two-sided, P < 0.05, n = 6 for both the treatment and control groups). D Dynamic changes of mosquito attraction and RPI in NHPs across detection time points (n = 5 biological replicates per treatment at each time point). Data are presented as mean values ± SD. E Baseline microbial phylum composition across five distinct body sites of NHPs. F Pcoa analysis revealing baseline microbiome high similarity across various skin regions of NHPs. G Alterations in native dominant skin microbiota of NHP following B. licheniformis and B. paralicheniformis treatment (Wilcoxon rank-sum test, two-sided, P < 0.05, n = 6 for both treatment and control groups). H Spearman correlation analysis of dominant skin bacterial abundance and mosquito attraction (Spearman correlation analysis, two-sided test). All boxplots in this figure were constructed following the standard definition: horizontal lines inside each box represent the Q1, median, and Q3 of the dataset; whiskers stretch to the farthest data points within a distance of 1.5 times the IQR relative to the box boundaries. Source data are provided as a Source Data file. Furthermore, colonization significantly altered the relative abundances of original dominant species, specifically S. epidermidis, Acinetobacter chengduensis, and Comamonas testosteroni (Fig. 2G). Spearman correlation analysis showed significant negative correlations between Bacillus abundances and mosquito attraction rate, while S. epidermidis abundance was positively correlated with attraction rate (Fig. 2H). These results suggest that both B. licheniformis and B. paralicheniformis are functional effectors that reshape skin microbiota structure, modulate host skin microenvironment and odor profiles, and ultimately reduce mosquito attraction. Multi-omics integration reveals Bacillus-driven metabolic reprogramming PLS-DA of monkey skin metabolites showed no significant inter-site variability, with samples closely clustered in score plots (Fig. 3A), confirming dorsal skin metabolites reliably represent systemic profiles. UPLC-MS/MS and GC-MS analysis showed Bacillus colonization reshaped skin metabolic profiles. B. licheniformis treatment upregulated 67 and downregulated 12 metabolites; B. paralicheniformis treatment upregulated 22 and downregulated 4 metabolites, with 13 co-upregulated and 2 co-downregulated metabolites (Fig. 3B). Metabolomic reprogramming of skin and induced VOCs profiles following B. spp. colonization. A PLS-DA score plot verifying the consistency of baseline metabolomic profiles among different skin sites in NHP. B Differences in skin metabolites of NHPs following treatment with B. licheniformis and B. paralicheniformis. C Functional category enrichment analysis of skin compounds with differential expression before and after B. spp. application in NHPs. D Specific enrichment of functional pathways in B. licheniformis and B. paralicheniformis relative to native dominant skin bacteria. E Mantel test revealing associations between dominant skin bacterial genera and metabolite categories (Mantel test for positive correlation, one-sided). Source data are provided as a Source Data file. Pathway enrichment analysis revealed that KEGG pathway enrichment of the differential metabolites screened after Bacillus colonization showed significant enrichment in the “Phenylalanine metabolism” and “Fatty acid biosynthesis” pathways (Fig. 3C). Further verification combined with KEGG functional annotation of the Bacillus genome showed that a total of 54 metabolic pathways were enriched in its genome, including “Phenylalanine, tyrosine and tryptophan biosynthesis”, “Fatty acid metabolism”, “Fatty acid biosynthesis”, etc. Among them, functional genes related to the above three pathways consistent with differential metabolite enrichment were highly enriched, reflecting the consistency between genomic functions and metabolomic phenotypes (Supplementary Fig. S1). We then conducted functional enrichment analysis on bacteria depleted after Bacillus colonization (including S. epidermidis, A. chengduensis, and C. testosteroni) in order to elucidate the potential functional significance of Bacillus-mediated community remodeling on skin microecological functions. Functional enrichment results showed that the genes of these three bacteria were significantly enriched in 66 metabolic pathways, 45 of which were the same as those enriched in the Bacillus genome. There were also 22 specific metabolism-related pathways, mainly including functions related to aromatic compound degradation, and environmental substance transformation-related functions such as amino acid and vitamin metabolism, fatty acid degradation, and sulfur metabolism (Supplementary Fig. S2). In contrast, the Bacillus genome also had 11 specifically enriched pathways, whose major functions included carbohydrate metabolism, amino acid and vitamin metabolism, lipid metabolism, and antibiotic resistance-related functions (Fig. 3D). Based on the “pathway-product” association logic and combined with the pathway enrichment results of differential metabolites, this study identified 4 key VOCs downstream of the major pathways and clarified their potential metabolic pathway origins: p-Cresol is derived from the “Aromatic amino acid metabolism” pathway, Decanoic acid from the “Fatty acid biosynthesis” pathway, while 2-Chloroethyl benzoate and N,N-Dibutylformamide may be associated with fatty acid and aromatic compound-related metabolic pathways. Mantel test confirmed Bacillus abundance was significantly associated with lipids, organic acids, and phenylpropanoids (r>0.4, P<0.05) (Fig. 3E), supporting the bacteria-metabolite link. Functional verification of specific repellent metabolites in mice A murine model was used to verify the actual mosquito attraction-reducing activity of the four candidate compounds, which were screened and identified via metabolomic analysis of skin samples from bacteria-treated cynomolgus monkeys. (Fig. 4A). GC-MS quantitative detection results confirmed that the abundances of these four compounds in the skin samples of the two Bacillus treatment groups were significantly higher than those in the control group (Fig. 4B). Specifically, in the B. licheniformis treatment group: p-Cresol was upregulated by 2.15 fold, Decanoic acid by 2.16 fold, N,N-Dibutylformamide by 2.25-fold, and 2-Chloroethyl benzoate by 29.6 fold. In the B. paralicheniformis treatment group: p-Cresol was upregulated by 2.93 fold, Decanoic acid by 4.08 fold, N,N-Dibutylformamide by 2.30 fold, and 2-Chloroethyl benzoate by 5.55 fold. These results further verified the causal relationship between major pathway activation and target metabolite accumulation. Bacillus-derived metabolites serve as potent olfactory deterrents for host-seeking mosquitoes in mice. A Schematic of the murine model assay and validation of specific compounds’ mosquito repellent duration. B Abundance profiling of the metabolomics-identified specific compound. C Dose-dependent distribution of effective protection time for the four compounds. D Temporal dynamics of protection failure and time-to-event analysis of breakthrough bites and protected individuals (Log-rank test, two-sided, n = 4 for each concentration gradient, P < 0.05). All boxplots in this figure were constructed following the standard definition: horizontal lines inside each box represent the Q1, median and Q3 of the dataset; whiskers stretch to the farthest data points within a distance of 1.5 times the IQR relative to the box boundaries. Source data are provided as a Source Data file. The mosquito repellent time of each compound was determined to quantify its biological activity (Fig. 4C). The results showed that the repellent time of the four compounds increased with increasing concentration, showing a significant concentration-dependent effect. Even at low concentration (1 mg/ml), the repellent time of each compound was significantly higher than that of the solvent control group. These findings directly confirmed that the four compounds can induce a significant mosquito repellent response in mice, consistent with their upregulated abundances in Bacillus treatment groups, and further validated their role as repellent metabolites. Survival analysis was further performed to accurately calculate the effective protection time of the four compounds (Fig. 4D). The results showed that the survival curves of the four compounds were significantly separated from that of the solvent control group, with repellent proportion decreasing gradually as protection time extended. Among them, Decanoic acid exhibited the slowest decline rate in repellent proportion, maintaining a high repellent effect for a longer duration. In contrast, p-Cresol showed a relatively faster decline in repellent proportion, corresponding to its lower repellent activity compared with the other three compounds. Notably, with the standard requirement for mosquito repellents being 4 h of effective protection time, the other three compounds (Decanoic acid, N,N-Dibutylformamide, and 2-Chloroethyl benzoate) met this criterion, while p-Cresol did not. This analysis thus provided a precise characterization of the duration and efficacy of repellent activity for each compound. Discussion As the human body’s outermost protective organ, skin is a key interface through which vectors and vector-borne pathogens interact with hosts. Because skin surface odor is a key signal for mosquitoes to locate hosts. Skin microbiota are primary contributors to skin odor profiles22,28, and their compositional homeostasis directly shapes odor spectrum characteristics, which correlate with host attractiveness to mosquitoes29. Thus, maintaining skin microecological homeostasis is pivotal for reducing attraction to mosquitoes. This study integrated human mosquito attractiveness phenotyping, primate in situ skin colonization experiments, and multi-omics analyses to explore how two commensal Bacillus spp. (B. licheniformis and B. paralicheniformis) regulate host–mosquito attractiveness. Our results demonstrate that colonization by commercial Bacillus (matched with human strains) is associated with skin microbiota restructuring in the macaque model and volatile metabolite alterations in the mouse model, accompanied by reduced mosquito attraction. These findings provide a framework for understanding how skin microecology may contribute to host–mosquito interactions, and a working model is proposed to illustrate the potential underlying metabolic regulatory mechanism. Notably, multiple amplification was applied prior to high-throughput sequencing in this study, and inherent amplification bias inevitably distorted the compositional profile of human skin microbiota. This technical bias led to an overrepresentation of Bacillus taxa, accounting for more than 60% of the overall microbial abundance, which fails to reflect the actual proportion of Bacillus species in native human skin microbial communities. Nevertheless, such systematic deviation uniformly affected all enrolled samples from both high and low mosquito-attractiveness groups. Accordingly, the significant intergroup differences in the relative abundance of Bacillus licheniformis and Bacillus paralicheniformis remain credible and comparable, supporting the rationality of their correlation with mosquito attraction phenotypes. B. licheniformis and B. paralicheniformis belong to the B. subtilis group30,31. Studies confirm this group, including B. subtilis and B. licheniformis, exerts probiotic effects (e.g., inhibiting Salmonella infection, maintaining intestinal microbiota homeostasis, alleviating oxidative stress32–35), with efficacy and safety validated in clinical trials36–38. Consistent with their reported probiotic properties in other host-associated contexts, our results suggest that these Bacillus species can also modulate skin microbial communities. Bacillus strains secrete broad-spectrum antimicrobial substances such as gramicidin and polymyxin39–42, which counteract Staphylococcus spp. by disrupting quorum sensing and biofilm formation. Here, Bacillus colonization induced an overall shift in skin microbiota structure, reducing abundances of dominant commensals like S. epidermidis—a bacterium that enhances mosquito attractiveness23,43—and A. chengduensis. Inhibiting S. epidermidis likely represents one contributing pathway for these Bacillus spp. to reduce mosquito attractiveness to the host. Meanwhile, significant enrichment of Streptococcus and other bacterial taxa was detected, which are potentially associated with human skin infections and scarlet fever. Metabolically, Bacillus colonization induced significant changes in skin volatile metabolites. Four key VOCs were upregulated by both strains: 2-Chloroethyl benzoate, Decanoic acid, N,N-dibutylformamide, and p-Cresol, with functional validation confirming their direct mosquito repellent activity in mice. p-Cresol derives from aromatic amino acid metabolism44,45, and Decanoic acid is a fatty acid biosynthesis product46–48; both are associated with Culex spp. attractiveness49–51, and our study extends this association to Aedes spp.. In contrast, biosynthetic pathways of non-classical metabolites (2-chloroethyl benzoate, N,N-Dibutylformamide) lack clear annotations in skin microbe-host co-metabolism research52–54. Elucidating whether these compounds arise from indirect microbial transformation, host–microbe interactions, or environmental derivatization will require further investigation. Together, these observations underscore the need for future global health–oriented studies to explore how geography, diet, age, and the gut–skin axis shape Bacillus prevalence and skin repellent metabolomes across human populations, with the ultimate goal of identifying modifiable strategies to reduce mosquito-borne disease risk. Functional annotation clarified the competitive advantages of Bacillus spp. and their metabolic regulatory mechanisms. Compared with dominant bacteria (S. epidermidis, A. chengduensis), B. licheniformis and B. paralicheniformis are enriched in carbohydrate, lipid, and amino acid metabolic pathways. Skin surface carbohydrates and lipids (from sebum secretion) are core carbon sources for microbiota55–57; These Bacillus strains are hypothesized to consolidate their ecological niche by boosting the metabolism of such substrates, thereby sustaining stable colonization and driving microbiota restructuring. Specifically, their enriched “Phenylalanine, tyrosine and tryptophan biosynthesis” pathway provides substrates for p-Cresol synthesis, while the “Fatty acid biosynthesis” pathway consistents with Decanoic acid production. Meanwhile, suppressed bacteria (S. epidermidis, A. chengduensis) show impaired fatty acid degradation, further promoting Decanoic acid accumulation. Several limitations of this study should be acknowledged. Interspecies variation in the primate skin microbiome may limit direct extrapolation of our NHP findings to humans58. Although Bacillus colonization was sufficient to reduce mosquito attraction in a primate model, the natural persistence and ecological competitiveness of these strains on human skin remain to be determined. The present study did not conduct validation assays on Anopheles or Culex mosquitoes, nor did it verify the relevant findings in human cohorts. In subsequent studies, we will isolate and culture Bacillus strains directly from the skin of low‑attractiveness individuals and conduct further functional verification of additional native strains. In addition, while multiple repellent-active VOCs were identified, the precise microbial or host–microbial biosynthetic routes for some compounds require further clarification. Future research will include DEET and other conventional mosquito repellents as reference controls for comparative efficacy evaluation and broaden the range of tested mosquito species to encompass Culex and Anopheles species. These limitations highlight important directions for future investigation. Based on these findings, a working model of a skin microbiota–mediated repellent framework, herein referred to as the “Skin Probiotic Defense”, provides new mechanistic insights into host–mosquito interaction regulation. This framework posits that skin commensals like Bacillus spp. may contribute to reduced mosquito attraction by reshaping microbial community structure and associated volatile profiles. This perspective complements the conventional focus on single repellent molecules, emphasizing the skin microenvironment as a microbial community-modulated chemical ecosystem. In summary, this study identifies specific skin commensal bacteria as active biological sources of mosquito-reduced mosquito attraction. Through combined analyses of human cohorts, causal microbial application in a NHP model, and integrative multi-omics profiling, this study demonstrates that Bacillus spp. are consistently associated with reduced mosquito attraction and are sufficient to induce a robust reduced attractive phenotype on the skin. Simultaneously, the association between Bacillus abundance, altered metabolic pathways, and repellent-associated volatile compounds supports a plausible microbial–metabolite–behavior framework underlying this effect (Fig. 5). Unlike previous studies focusing primarily on individual VOCs, our findings highlight a microbiota-centered regulatory paradigm, where Bacillus colonization reshapes the skin metabolome and thereby modulates host attractiveness to mosquitoes. While the specific effector metabolites and metabolic pathways remain to be fully resolved, our data support an exploratory microbiota–metabolite–behavior framework linking Bacillus colonization to repellent-associated volatile profiles. Collectively, these results highlight the potential of a skin microbiota–derived attraction-reducing agent, expanding mosquito control strategies. Proposed working model illustrating how Bacillus spp. may reshape skin microbiota and volatile profiles to reduce mosquito attraction. On the one hand, it not only provides aromatic amino acid substrates for the repellent substance p-cresol but also directly participates in the synthesis of the repellent substance decanoic acid. On the other hand, the activity of inhibited skin microorganisms (such as S. epidermidis and A. chengduensis) decreases, leading to a reduction in attractive substances such as small-molecule fatty acids. Ultimately, this achieves the host’s protection against mosquitoes. Methods Volunteers and experimental materials Thirty-seven healthy volunteers were recruited via institutional advertisement notices and strictly screened according to predefined inclusion and exclusion criteria before enrollment, including 23 females and 14 males. Inclusion criteria: aged 22–32 years, no history of skin diseases, no use of antibiotics, hormonal drugs, or mosquito repellent products within the past month, no history of allergies to mosquito bites, and regular living habits. Three healthy male cynomolgus monkeys (Macaca fascicularis) aged 7–8 years were selected as experimental animals, reared under controlled indoor conditions (temperature: 22–25 °C, humidity: 50–60%) with free access to food and water. Experiments commenced after 1 month of acclimatization. Female Kunming mice aged 6–8 weeks were also utilized in the present study and raised under consistent indoor rearing environments (temperature: 26 ± 1 °C, relative humidity: 55% ± 5%, and the photoperiod was 14 h of light/10 h of dark with ad libitum feed and water supply. Laboratory-reared Aedes aegypti mosquitoes of the Xishuangbanna strain (Yunnan Province, China) were used. Female mosquitoes were maintained on 8% sugar water and fasted for 8 hours prior to experiments. Unfed female mosquitoes at 5–10 days post-eclosion were selected for attractiveness tests. The bacterial strains used were Bacillus licheniformis (BNCC336463, BeNa Culture Collection, BNCC, China) and Bacillus paralicheniformis (CICC10181, CICC, China Center of Industrial Culture Collection, China). For functional verification, four key VOCs with chromatographic purity (≥ 98%) were used, with CAS numbers provided as unique identifiers assigned by the Chemical Abstracts Service for each compound: p-Cresol (CAS: 106-44-5, Cat. No. C85751-100G, Sigma-Aldrich), Decanoic acid (CAS: 334-48-5, Cat. No. C1875-100G, Beijing Woke Biology Technology Co., Ltd.), 2-Chloroethyl benzoate (CAS: 939-55-9, Cat. No. YS121681-25g, Solarbio Science & Technology Co., Ltd.) and N,N-Dibutylformamide (CAS: 761-65-9, Cat. No. A14533-25ml, Beijing Woke Biology Technology Co., Ltd.). All compounds were dissolved in absolute ethanol and serially diluted to prepare working solutions at low (1 mg/mL), medium (10 mg/mL), and high (100 mg/mL) concentrations. Ethics All volunteers signed informed consent forms. The study protocol was approved by the Ethics Committee of Guizhou Medical University (Approval No.: 20230072). All procedures involving human participants were conducted in compliance with relevant ethical guidelines and regulations. The animal experimental protocol was approved by the Ethics Committee of the State Key Laboratory of Pathogen and Biosecurity, with the approval number for Macaca fascicularis: IACUC-IME-2023-012, and the approval number for Kunming mice: IACUC-DWZX-2025-A060. All procedures involving laboratory animals were conducted in accordance with the Guidelines for the Care and Use of Laboratory Animals. All ethical reviews and approvals were conducted in compliance with the national welfare and ethics guidelines of the People’s Republic of China (for human volunteers: GB/T 38736-2020; for laboratory animals: GB/T 35892-2018). Mosquito attractiveness test A single-channel olfactometer was used. Prior to sampling, all volunteers uniformly and thoroughly cleaned their forearms with unscented neutral soap and rinsed with deionized water. In addition, all participants were required to avoid food with irritant odors and cosmetics for at least 24 h before the experiment to eliminate exogenous odor interference. For the control, no odor source was placed. Thirty unfed female Ae. aegypti were released into the device. After stabilization, mosquitoes were allowed to choose freely for 1 hour, after which the number attracted to each capture area was counted. The relative attractiveness rate for each volunteer was calculated (the number of mosquitoes attracted by the experimental group—the number attracted by the control group). Volunteers were stratified into high-attractiveness (HA) and low-attractiveness (LA) groups based on the statistical significance of their attraction rates relative to the control (Chi-square test; P < 0.05 for HA, P > 0.05 for LA). Collection of volunteer skin microbiota samples and high-throughput sequencing Skin microbiota samples were collected from the forearms and palms of volunteers immediately following the attractiveness assay. To minimize contamination, sampling was conducted near an alcohol lamp. The skin surface was wiped thoroughly for 2 min using a sterile medical swab moistened with ST solution (0.9% NaCl, 0.1% Tween-20), applied with consistent pressure. Each swab was then transferred to a tube containing 1 mL of sterile PBS, stored at 4 °C, and processed within 2 h. All collections were performed by a single operator. A negative control (a swab immersed directly in sterile PBS) was included. Total genomic DNA was extracted using the E.Z.N.A.® Soil DNA Kit. DNA purity and concentration were measured with a NanoDrop2000, and integrity was checked via 1% agarose gel electrophoresis. Due to insufficient DNA yield and quality for direct library preparation, whole-genome amplification was performed using Multiple Displacement Amplification (MDA). Sequencing libraries were constructed from the amplified DNA, which involved fragmentation (ultrasonic shearing to 300–500 bp), end repair, A-tailing, adapter ligation, and PCR amplification. Library quality was assessed on an Agilent 2100 Bioanalyzer, followed by paired-end sequencing (PE150) on an Illumina NovaSeq 6000 platform to generate raw metagenomic data. Establishment of in situ colonization model on NHP skin To establish baseline skin microecology and metabolomic profiles, samples were collected from multiple sites on cynomolgus monkeys (back, armpit, upper limb, lower limb, and chest) using the same sterile swab method as for human volunteers, with three technical replicates per site. Total DNA was extracted and subjected to full-length 16S rRNA gene sequencing. Metabolomic profiling was performed using GC-MS and LC-MS platforms. The resulting microbial community structure and metabolite profiles were analyzed to characterize baseline conditions across sites. β-diversity analysis based on Bray–Curtis distance was further employed to assess the similarity between back skin and other body sites. Results revealed that the skin microbial community on the back was highly similar to those at other anatomical regions and thus representative of the overall skin microbiota in macaques. These baseline data supported the selection of the back as the target area for subsequent in situ colonization experiments with Bacillus strains. Commensal bacteria intervention and NHP skin microbiome sequencing Newly acquired B. licheniformis and B. paralicheniformis strains were cultured in nutrient broth at 30 °C. Glycerol stocks (25%) were prepared and stored at −80 °C. For experiments, stocks were revived and grown to log phase in nutrient broth; bacterial cells were harvested and resuspended at 1 × 10⁸ CFU/mL in fresh medium. These strains were commercially obtained instead of being isolated from the skin of low-attractiveness individuals, which may limit the simulation of the actual physiological and metabolic characteristics of native skin-derived bacteria. The back skin of cynomolgus monkeys was treated daily for three consecutive days by topical application. A 10 mL volume of secondarily activated bacterial suspension cultured to the logarithmic growth phase (1 × 108 CFU/mL) was evenly spread onto the shaved back skin with an area of ~7 × 20 cm using a sterile cotton swab; control animals received an equal volume of nutrient broth only. Attractiveness to Ae. aegypti was assessed before treatment (day 1) and on days 2, 5, and 8 post-treatments. Each monkey was placed inside a metal cage (40 × 45 × 70 cm) within an acrylic chamber (50 × 55 × 80 cm). Fifty female mosquitoes were introduced into a single port olfactometer connected to the chamber. To ensure data reliability, mosquito landing events were independently counted by two researchers who were blind to the treatment allocation, and the number attracted to the treatment side was recorded after 1 h (five replicates per time point). Mosquito response was quantified using the Relative Population Index (RPI). where Ta and Ca represent the average mosquito counts in the treatment group and the control group after treatment, respectively, and Tb and Cb represent the average mosquito counts in the treatment group and the control group before treatment, respectively. Skin swabs were collected from the treatment area at the same time point using the standardized sterile swab protocol. Total DNA was extracted by the CTAB method, and PCR amplification was performed with universal primers (27 F: 5’-GTGCCAGCMGCCGCGGTAA-3’, 1492 R: 5’-TACGGYTACCTTGTTACGACTT-3’). After verification of amplified products by agarose gel electrophoresis, single-molecule real-time sequencing was performed on the Pacbio Sequel II platform to analyze dynamic changes in skin microbial community structure before and after treatment. NHP skin metabolomics detection Skin metabolites were collected from the test area of cynomolgus monkeys before and after behavioral assays using a magnetic stir bar (Twister, Gerstel GmbH & Co. KG, Germany). The bar was rolled 100 times across the skin, then transferred to a vial. Samples were stored at −20 °C and analyzed within 72 h. Prior to analysis, Twisters were conditioned at 260 °C for 2 h. For GC-MS, bars were rinsed with warm pure water and placed in desorption tubes. For LC-MS, samples were extracted with 300 μL of 80% methanol, sonicated for 20 min, and centrifuged; the supernatant was transferred to an injection vial. GC-MS was performed using a Gerstel TDU2/Trace 1310/Q-Exactive system with helium carrier gas (1 mL/min). The oven program was: 40 °C (2 min), ramp to 230 °C at 4 °C/min, hold 5 min. EI ionization at 70 eV was used, with source/transfer line temperatures of 280/250 °C. Data were processed with Xcalibur and TraceFinder; compounds were identified using NIST17 and IAS Center Lab databases. LC-MS analysis employed an Agilent 1290 HPLC coupled to an AB SCIEX 6600 Q-TOF with ESI. Separation was on an HSS T3 column (2.1 × 100 mm, 1.8 μm) with water (A) and acetonitrile (B), both containing 0.1% formic acid. Gradient: 0–2 min (2% B), 2–14 min (2–98% B), 14–17 min (98% B), 17–17.1 min (98–2% B), 17.1–20 min (2% B). Flow rate: 350 μL/min; injection volume: 2 μL. Data were processed with MExplorer Ultimate; peaks with MS1 match and QC RSD > 30% were excluded. Determination of the protection time of metabolomic compounds Female Aedes aegypti (n = 100) were temporarily sedated with CO₂ only for sorting and transfer, then placed into a 25 cm³ rearing cage and allowed to fully recover for an adequate period before subsequent behavioral assays. No anesthesia was applied during the formal behavioral tests. Selected metabolite standards (identified from differential monkey skin metabolomics) were dissolved in absolute ethanol and diluted to three concentrations: 100, 10, and 1 mg/mL. The concentration gradient was set with reference to reported insect olfactory response studies and the physiological concentration range of skin volatile metabolites, to screen for dose-dependent behavioral effects7. SPF female mice were shaved on the back (2 × 2 cm) and fixed with a dedicated mouse holder for stable restraint during the experiment. The shaved area was treated with 200 µL absolute ethanol for 5 min. Mice were then introduced into the mosquito cage; pairs were considered suitable if ≥2 landing/biting attempts occurred within 2 min. For testing, 200 µL of compound solution was applied to the shaved skin. After 5 min, mice were exposed to mosquitoes, and the time to the first landing/biting attempt was recorded as the protection time (observation limit: 6 h). Tests were performed from low to high concentration with four replicates per concentration. Gloves were changed between solutions; mosquitoes and mice were replaced with different compounds. Bioinformatics analysis Raw data quality was assessed using FastQC (v0.11.9)59. Low-quality reads and adapters were trimmed with Trimmomatic (v0.39)60. Clean reads were aligned to the human reference genome (GRCh38.p13) with Bowtie2 (v2.4.5)61, and host sequences were removed using Samtools (v1.15.1)62. Taxonomic classification was performed with Kraken2 (v2.1.2)63, and species abundance was estimated with Bracken to generate taxonomic abundance profiles. LefSe was applied to identify differentially abundant taxa between attractiveness groups: the Kruskal–Wallis H test (α = 0.05) was used for screening, followed by linear discriminant analysis (LDA score >3.0). Quality control was conducted in QIIME264. After filtering low-quality sequences, chimeras were removed using UCHIME with the Gold database. ASVs were clustered at 97% similarity using DADA265 and annotated against the RDP database66. α- and β-diversity were calculated in QIIME2; group differences were tested with ANOSIM. Differentially abundant ASVs were identified using DESeq267. Peak detection, alignment, and quantification were performed in Progenesis QI to generate a metabolite × sample matrix. Following missing-value imputation and UV scaling, a PLS-DA model was built with the R package ropls, and model quality was assessed using Q² and R². Differential metabolites were selected via OPLS-DA (VIP > 1, P < 0.05). Functional annotation of protein-coding genes was performed using KOFAMSCAN (version 1.3.0) against the KEGG database to obtain KO (KEGG Orthology) numbers. KEGG pathway enrichment analysis was then carried out using the R package clusterProfiler based on the obtained KO assignments. Differential microbes and metabolites were integrated. Spearman correlation was used to compute associations ( | R | > 0.4, P < 0.05). The overall relationship between microbial communities and metabolomes was examined with the Mantel test. A correlation heatmap was generated in R (v4.4.1), and interactions were interpreted in the context of taxonomic and functional annotations. Data statistics and analysis Data was analyzed using R 4.4.1. Continuous variables are presented as mean values ± standard deviation. Normality was assessed with the Shapiro–Wilk test. Normally distributed data were compared between two groups using the independent samples t test, and among multiple groups using one-way ANOVA followed by Tukey’s post-hoc test. Non-normal data were analyzed with the Mann–Whitney U test (two groups) or the Kruskal–Wallis H test with Dunn’s post-hoc test (multiple groups). For mosquito behavioral assays (volunteer and NHP attractiveness tests), differences in mosquito counts were evaluated with the Chi-square test. Changes in attractiveness before and after treatment were assessed using the RPI analyzed by paired t test. Spearman correlation was used to examine associations between differential microbes, metabolites, and attractiveness parameters ( | r | > 0.4, P < 0.05). The overall relationship between microbial community structure and metabolomic profiles was tested with the Mantel test. All tests were two-tailed, with P < 0.05 considered statistically significant. Figures were generated using R 4.4.1 (ggplot2), with consistent formatting. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements We thank and acknowledge volunteers who participated in the experiment for their generous contributions to this research. During the preparation of this work, the authors used the assistance of artificial intelligence tools to refine language, enhance clarity and coherence, and improve overall readability of the manuscript. All schematic illustrations were created by the authors using Adobe Illustrator. Author contributions Conceptualization: C.L., T.Z., and J.W.; data curation: H. Gao and T.Z; Formal analysis: H. Gao and T.Z; funding acquisition: C.L., T.Z., and J.W.; investigation: Z.G., H. Guo, Z.F., and W.Z.; methodology: C.L. and T.Z.; project administration: C.L., T.Z., and D.X.; resources: X.Z., H.Y., and D.X.; software: H. Gao; supervision: C.L., T.Z., and J.W.; validation: Z.G., H. Guo, Z.F., and W.Z.; visualization: X.Z., H.Y., H. Gao, and Z.G.; writing—original draft: Z.G., H. Gao, and T.Z.; writing—review and editing: C.L. All authors read and approved the final version of the manuscript. Peer review Peer review information Nature Communications thanks Omar Akbari, Alicia Showering and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding The authors declare no relevant funding. Data availability The raw sequencing data generated in this study have been deposited at the National Genomics Data Center Genome Sequence Archive (GSA) under accession code PRJCA055301. The human-skin metagenomic sequencing data generated in this study have been deposited in the National Genomics Data Center Genome Sequence Archive under accession code CRA038283. The monkey-skin microbiome sequencing data generated in this study have been deposited in the GSA under accession code CRA038290. The metabolomics data generated in this study have been deposited in the OMIX under accession code OMIX019048. Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Zhenyu Gu, Heting Gao, Hongxia Guo, Zhihua Fan, Weixian Zhang. Contributor Information Teng Zhao, Email: zhaoteng2013@163.com. Jiahong Wu, Email: jiahongwu2008@hotmail.com. Chunxiao Li, Email: lichunxiao@bmi.ac.cn. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-78356-5. References Associated Data Supplementary Materials Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Reporting Summary Transparent Peer Review file Source Data Data Availability Statement The raw sequencing data generated in this study have been deposited at the National Genomics Data Center Genome Sequence Archive (GSA) under accession code PRJCA055301. The human-skin metagenomic sequencing data generated in this study have been deposited in the National Genomics Data Center Genome Sequence Archive under accession code CRA038283. The monkey-skin microbiome sequencing data generated in this study have been deposited in the GSA under accession code CRA038290. The metabolomics data generated in this study have been deposited in the OMIX under accession code OMIX019048. Source data are provided with this paper.