Classroom Microbiome Signatures of Pest Management Associate with Reduced Asthma Symptoms

Abstract

Background:

Integrated pest management (IPM) is thought to improve asthma symptoms through reduced mouse allergen exposure. Whether IPM acts through changes in mouse-associated microbes remains unknown.

Objectives:

To examine the effects of school-based IPM on the classroom microbiome, and to determine the association between intervention microbiome signatures and student asthma morbidity.

Methods:

In this ancillary study based on a randomized placebo-controlled clinical trial of school IPM and classroom high efficiency air purifiers (ClinicalTrials.gov NCT02291302), we performed deep metagenomics sequencing of longitudinally collected dust samples from 208 classrooms in 41 schools of 236 children with active, physician-diagnosed asthma with prospective follow-up of asthma severity during the school year. We assessed the effect of the interventions on classroom microbial communities in intention-to-treat analyses. Sparse Partial Least Squares models were used to identify microbial signatures of the interventions and the association between these microbial signatures and asthma morbidity was assessed using mixed effects models, controlling for covariates including mouse allergen exposure.

Results:

IPM significantly altered classroom bacterial and phage community structure and increased bacterial, archaeal, and fungal diversity. A classroom microbiome signature of IPM was identified (AUC=0.84) and was associated with lower odds of any asthma symptom days in the past two weeks (OR 0.47, 95% CI [0.22, 0.97], p=0.043) and lower Composite Asthma Severity Index (β −0.92, 95% CI [−1.51, −0.33], p=0.002), adjusting for mouse allergen levels and student characteristics.

Conclusion:

IPM-associated changes in the classroom microbiome are associated with lower asthma morbidity independent of mouse allergen exposure.

Capsule summary

The classroom microbiome was modified by integrated pest management (IPM). A classroom microbiome signature of IPM was associated with lower asthma burden independent of mouse allergen exposure.

INTRODUCTION

The environmental microbiome has emerged as a key determinant of both asthma inception and morbidity and represents a potentially modifiable target within the built environment, defined as human-made indoor spaces.¹ However, randomized controlled trials directly manipulating the environmental microbiome to improve asthma morbidity are lacking. Observational birth cohort studies have consistently demonstrated that features of early-life environmental microbial exposures, such as increased richness or the presence of specific microbial clades, are associated with the subsequent development of asthma.^2–4 These studies have largely relied on amplicon sequencing of the 16S rRNA gene, which only characterizes the bacteriome. In contrast, emerging studies suggest that in patients with established asthma, exposure to high environmental bacteriome diversity is a risk factor rather than protective for type 2 asthma endotypes, with the indoor mycobiome being a risk factor independent from the indoor bacteriome.⁵ Thus, environmental microbiome studies focused on asthma prevention in early life do not address the key issue of how these same microbial exposures affect morbidity in individuals with asthma, have largely focused on the bacteriome, and do not address whether the built environment microbiome can be successfully manipulated to improve the health of occupants.

We have previously shown in a pilot intervention trial of school-based integrated pest management (IPM) and classroom-based portable high efficiency particulate air (HEPA) purifiers involving 25 school-age children attending 3 schools that higher classroom microbial diversity is associated with increased asthma symptoms, and further that the IPM intervention changed overall classroom microbial community structure.⁶ Subsequently, we conducted a cluster-randomized, placebo-controlled trial of school IPM and classroom HEPA purification from 2015–2020, enrolling 236 children from 41 elementary schools with a primary outcome focused on improving asthma morbidity by reducing school allergen and particulate matter exposure (ClinicalTrials.gov identifier: NCT02291302). The primary analysis showed that clinical effects were not fully explained by changes in measured allergens,⁷ raising the possibility that non-allergen pathways may contribute to clinical outcomes. Historically, IPM is thought to improve asthma control through mouse allergen reduction in the environment, however, an alternate possibility is through changes in the mouse-associated environmental microbiome through pest control.^4,8

Here we tested the hypothesis that the school built environment microbiome is modifiable. We further hypothesized that a classroom microbiome signature of reduced mouse exposure predicts asthma symptoms independent of mouse allergen exposure. This represents the first randomized controlled trial to directly assess whether the classroom microbiome can be modified and whether those changes are associated with asthma morbidity.

METHODS

Study design

This was an ancillary study to the School Inner-City Asthma Intervention Study (SICAS-2: NCT02291302).^7,9 Caregiver surveys were distributed at participating schools during the spring¹⁰ to identify students in grades K-8 with active, physician-diagnosed asthma attending enrolled schools. Participants were enrolled in the summer and randomized to four different intervention groups; IPM and HEPA; IPM and sham HEPA; no IPM and HEPA; and no IPM & sham HEPA intervention. At the initial visit, informed consent was obtained, questionnaires (demographics, asthma characteristics, healthcare utilization, and home characteristics) were completed, and allergy testing, spirometry, and exhaled nitric oxide were performed. After the baseline environmental assessment during the academic school year, schools were randomized to receive IPM intervention or not. The IPM intervention consisted of pest management by trained professionals and enhanced cleaning by research staff.¹¹ HEPA intervention was conducted at the classroom-level where classrooms within each school were randomized to sham vs. functioning portable HEPA filter air purifiers (AP-1013-A, Woongjin Coway Co., Ltd). The details of both interventions are listed in Table E1.

Health outcome surveys and follow-up visit

Surveys were conducted by phone every two months during the school year to collect prospective health outcomes. Three times during the academic school year, students were briefly excused from class so that research staff could perform spirometry. From these, any asthma symptom days in the preceding two weeks (binary) and the Composite Asthma Severity Index (CASI)¹² were calculated as primary and secondary outcomes, respectively.

School environmental measurements

Dust samples were collected longitudinally throughout the school year, once in the fall prior to the study intervention and twice in the winter and spring/summer seasons as previously described.^9,13 Dust samples were weighed and used for allergen analysis using a multiplex assay (MARIA, Indoor Biotechnologies) as well as for nucleic acid extraction¹⁴ and shotgun metagenomics sequencing on the NovaSeq platform (Illumina), using a 2×150 bp paired-end protocol targeting 10 Gb per sample. Taxonomic profiling was performed using MetaPhlAn3¹⁵ and Marker-MAGu.¹⁶

Statistical analyses

The effects of the two interventions on the classroom microbiome and phageome were tested in an intention-to-treat analysis using linear mixed-effects models. For beta diversity indices, Permutational Multivariate Analysis of Variance was performed. To identify the most important features associated with the IPM intervention, separate sparse Partial Least Squares (sPLS) models were trained on microbiome and phageome features. Microbial features were used to train the sPLS models using an 80:20 train-test split and 10-fold cross-validation via ‘mixOmics’ and ‘mixOmicsCare’ packages.¹⁷

The distribution of asthma symptom days was highly zero-inflated (424 (59.9%) were zeros), thus the primary outcome was modeled as a binary variable indicating presence vs absence of asthma symptoms (Figure E1). The effect of the microbiome and phageome signature scores on the primary outcome of any asthma symptom days was examined using mixed-effects logistic models and on the secondary health outcome (CASI) using linear mixed-effects models. The health outcomes were linked to the environmental dust samples by the closest sampling date.

All analyses were performed using R version 4.4.2. Two-sided p-values <0.05 and q-values <0.1 after multiple testing correction were considered statistically significant.

Additional methods are in the Online Repository. Sequencing data are available at NCBI SRA (PRJNA1252025).

RESULTS

School, classroom, and participant characteristics

236 children with active, physician-diagnosed asthma attending 208 classrooms in 41 schools were enrolled. (Figure 1). Baseline school and classroom characteristics are listed in Table 1. More than 66% of participating classrooms were located in buildings constructed before 1950, the average class size was 18.6 ± 3.1 students, and most classrooms were passively ventilated, with only 23.6% having centralized heating, ventilation and air conditioning (HVAC) systems. Indoor air quality metrics were similar across intervention groups. Measured allergen captures exposure burden that is not fully represented by visual rodent indicators such as live mouse sightings or mouse dropping presence alone.^18,19 Though rodent indicators were observed in only 19 (46.3%) of 41 schools, mouse allergen was detected in all schools and 189 (90.9%) of 208 classrooms at the baseline visit, while rat allergen was detected in 14 (6.7%) classrooms. No classroom-associated variables were statistically different across intervention groups.

Figure 1. CONSORT flow diagram.

Participant flow from screening through randomization into four intervention groups (IPM & HEPA, IPM and sham HEPA, no IPM and HEPA, no IPM and sham HEPA) and corresponding classrooms, schools, and classroom dust sample counts.

Table 1. Classroom characteristics by intervention group.

Environmental and structural features of classrooms at baseline stratified by four randomized groups. Variables include air exchange rate, humidity, temperature, class size, ventilation type, built year, dust and allergen levels, indoor air pollutants, and visual rodent presence indicators. Continuous values presented as mean ± standard deviation, categorical variables presented as number, percentage.

Variables	Overall	Control	IPM	HEPA	IPM + HEPA	p-value8
N	208	48	46	57	57
Students per classroom	18.6 ± 3.1	18.5 ± 2.4	18.9 ± 3.8	18.2 ± 2.8	19.0 ± 3.4	0.480
Classroom dimensions (m3)1	259.4 ± 59.9	259.5 ± 57.6	255.1 ± 35.8	275.0 ± 86.1	248.4 ± 44.6	0.128
Ventilation
Air exchange rate (exchanges per hour)2	3.3 ± 0.6	3.3 ± 0.6	3.3 ± 0.6	3.3 ± 0.5	3.2 ± 0.6	0.915
Central HVAC	49 (23.6%)	10 (20.8%)	10 (21.7%)	16 (28.1%)	13 (22.8%)	0.815
Building built year						0.266
Before 1900	34 (16.3%)	10 (20.8%)	5 (10.9%)	9 (15.8%)	10 (17.5%)
1900 – 1949	104 (50.0%)	25 (52.1%)	21 (45.7%)	29 (50.9%)	29 (50.9%)
1950 – 1969	31 (14.9%)	5 (10.4%)	11 (23.9%)	4 (7.0%)	11 (19.3%)
1970 or later	39 (18.8%)	8 (16.7%)	9 (19.6%)	15 (26.3%)	7 (12.3%)
Indoor exposures
Mouse allergen (ng/mg dust)3	2.5 ± 11.4	1.3 ± 3.6	3.5 ± 15.1	2.7 ± 16.0	2.4 ± 6.4	0.829
Mouse allergen detected3	189 (90.9%)	41 (85.4%)	42 (91.3%)	51 (89.5%)	55 (96.5%)	0.258
Endotoxin (EU/mg dust)4	100.5 ± 130.9	95.5 ± 141.4	91.2 ± 85.8	84.6 ± 105.6	128.3 ± 168.7	0.313
Indoor air quality
CO2 (ppm)2	793.7 ± 168.8	791.9 ± 154.0	798.4 ± 154.3	783.7 ± 184.0	802.6 ± 182.8	0.957
Relative humidity (%)2	48.7 ± 7.9	49.6 ± 6.9	47.1 ± 8.2	50.5 ± 7.2	47.2 ± 8.8	0.099
Temperature (°C)2	20.8 ± 1.3	20.8 ± 1.3	21.1 ± 1.2	20.4 ± 1.4	20.8 ± 1.3	0.149
Coarse PM (μg/m3)5	7.5 ± 3.1	6.9 ± 3.0	7.7 ± 3.2	7.3 ± 2.8	8.1 ± 3.4	0.309
PM2.5 (μg/m3)6	5.7 ± 2.2	6.1 ± 2.8	5.7 ± 2.2	5.5 ± 2.0	5.4 ± 1.9	0.521
Rodent indicators7
Droppings seen	40 (19.2%)	7 (14.6%)	14 (30.4%)	9 (15.8%)	10 (17.5%)	0.177
Mouse seen	3 (1.4%)	2 (4.2%)	0 (0%)	1 (1.8%)	0 (0%)	0.256

IPM, Integrated pest management. HEPA, High efficiency particulate air. HVAC, Heating, ventilation and air conditioning. PM, particulate matter.

¹There were 10 missing values in classroom dimensions.

²There were 37 missing values in air exchange rate, CO₂, relative humidity, and temperature.

³There were 11 missing values in Mus m 1 (mouse allergen) levels.

⁴There were 6 missing values in endotoxin levels.

⁵There were 33 missing values in coarse PM.

⁶There were 32 missing values in PM_2.5.

⁷There were 5 missing values in both rodent indicators. Positive rodent indicators were observed in 19 (46.3%) of 41 schools, and mouse allergen (Mus m 1) was detected in all schools and 189 (90.9%) of 208 classrooms (with 11 missing) at baseline. Baseline rodent indicators were assessed during a single afternoon visit by the research team at the beginning of the school year and likely underestimate true rodent infestation as mice are nocturnal and mobile.

⁸Statistical comparisons were performed using one-way analysis of variance for continuous variables and the chi-square test for categorical variables.

Baseline characteristics of study participants are summarized and listed in Table 2. The average age of the students was 8.7 years, with 52% male, and 17.4% belonging to non-Hispanic white race and ethnicity. Household income was relatively low, with caregivers of only 42.4% of students reporting annual household incomes over $35,000, which is approximately 150% of the U.S. federal poverty level for a household of four. On average, enrolled students had mild to moderate asthma severity. There were no significant differences in baseline demographics or asthma severity among intervention groups.

Table 2. Participant characteristics including demographic, clinical, and treatment data.

Baseline characteristics of students stratified by four randomized groups. Variables include age, sex, race/ethnicity, household income, allergen sensitization, asthma symptom burden, and asthma treatment type. Continuous values presented as mean ± standard deviation, categorical variables presented as number, percentage.

Variables	Overall	Control	IPM	HEPA	IPM + HEPA	p-value10
N	236	59	59	59	59
Age, years	8.7 ± 2.0	8.4 ± 2.0	8.7 ± 2.0	8.8 ± 2.1	8.7 ± 1.8	0.742
Male	123 (52.1%)	38 (64.4%)	27 (45.8%)	26 (44.1%)	32 (54.2%)	0.104
Non-Hispanic White1	41 (17.4%)	13 (22.0%)	14 (23.7%)	7 (11.9%)	7 (11.9%)	0.161
Household income over $35,0002	100 (42.4%)	25 (42.4%)	31 (52.5%)	23 (39.0%)	21 (35.6%)	0.088
Grade3						0.733
K	18 (7.6%)	4 (6.8%)	7 (11.9%)	4 (6.8%)	3 (5.1%)
1	32 (13.6%)	10 (16.9%)	6 (10.2%)	8 (13.6%)	8 (13.6%)
2	47 (19.9%)	16 (27.1%)	10 (16.9%)	9 (15.3%)	12 (20.3%)
3	43 (18.2%)	9 (15.3%)	9 (15.3%)	14 (23.7%)	11 (18.6%)
4	36 (15.3%)	6 (10.2%)	10 (16.9%)	7 (11.9%)	13 (22.0%)
5	35 (14.8%)	9 (15.3%)	12 (20.3%)	8 (13.6%)	6 (10.2%)
6	10 (4.2%)	0 (0%)	2 (3.4%)	5 (8.5%)	3 (5.1%)
7	8 (3.4%)	3 (5.1%)	2 (3.4%)	1 (1.7%)	2 (3.4%)
8	6 (2.5%)	2 (3.4%)	1 (1.7%)	2 (3.4%)	1 (1.7%)
Characteristics of skin test sensitivity4
Any sensitization	136 (57.6%)	34 (57.6%)	38 (64.4%)	30 (50.8%)	34 (57.6%)	0.513
Mouse allergen	54 (22.9%)	11 (18.6%)	16 (27.1%)	12 (20.3%)	15 (25.4%)	0.674
Rat allergen	35 (14.8%)	12 (20.3%)	9 (15.3%)	6 (10.2%)	8 (13.6%)	0.595
German cockroach allergen	55 (23.3%)	14 (23.7%)	17 (28.8%)	16 (27.1%)	8 (13.6%)	0.349
Dust mite allergen	69 (29.2%)	13 (22.0%)	26 (44.1%)	15 (25.4%)	15 (25.4%)	0.288
Cat allergen	45 (19.1%)	8 (13.6%)	16 (27.1%)	8 (13.6%)	13 (22.0%)	0.349
Dog allergen	18 (7.6%)	3 (5.1%)	7 (11.9%)	5 (8.5%)	3 (5.1%)	0.595
Mold allergen	27 (11.4%)	6 (10.2%)	7 (11.9%)	3 (5.1%)	11 (18.6%)	0.349
Asthma symptom days in the past two weeks5	2.0 ± 3.5	2.7 ± 4.4	2.1 ± 3.6	1.2 ± 2.0	2.2 ± 3.5	0.127
CASI6	3.2 ± 2.3	3.1 ± 2.1	3.1 ± 2.2	3.2 ± 2.3	3.6 ± 2.6	0.546
ACT score7	22.4 ± 2.9	22.1 ± 2.9	22.8 ± 2.7	22.1 ± 3.2	22.6 ± 2.8	0.431
ACT score over 197	189 (80.1%)	48 (81.4%)	48 (81.4%)	46 (78.0%)	47 (79.7%)	0.496
Steroid use within 12 months	2 (0.8%)	2 (3.4%)	0 (0%	0 (0%)	0 (0%)	0.109
Antibiotic use within 12 months8	9 (3.8%)	3 (5.1%)	3 (5.1%)	1 (1.7%)	2 (3.4%)	0.736
Daily asthma controller use	129 (54.7%)	27 (45.8%)	31 (52.5%)	34 (57.6%)	37 (62.7%)	0.290
% predicted FEV19	98.5 ± 17.5	99.1 ± 15.1	98.7 ± 16.0	96.5 ± 16.3	99.6 ± 21.9	0.790
% predicted FVC9	98.4 ± 15.4	98.6 ± 12.5	98.3 ± 14.8	98.5 ± 15.6	98.2 ± 18.5	0.999
FEV1/FVC ratio9	85.0 ± 7.8	84.4 ± 7.1	85.9 ± 7.9	84.4 ± 6.9	85.1 ± 9.1	0.680
FeNO	21.1 ± 20.6	18.2 ± 13.7	21.2 ± 19.9	18.5 ± 21.1	27.5 ± 25.9	0.112

IPM, Integrated pest management. HEPA, High efficiency particulate air. CASI, Composite asthma severity index. ACT, Asthma control test. FEV, Forced expiratory volume. FVC, Forced vital capacity. FeNO, Fractional exhaled nitric oxide.

¹There were 5 missing values in race/ethnicity.

²There were 77 missing values in household income, of which 56 caretakers responded they did not know their income, and 20 declined to answer.

³There was 1 missing value in grade.

⁴There were 5 missing values in any sensitization and most individual allergen measures, 26 in rat allergen sensitization, and 6 in dust mite sensitization. Any allergen sensitization was defined as a positive skin prick test response with mean wheal diameter ≥ 3 mm above saline control to at least one of the following allergens: mouse (Mus m 1), rat (Rat n 1), German cockroach (Bla g 2), cat (Fel d 1), dust mite (Der f 1, Der p 1), dog, Aspergillus, Alternaria tenuis, birch pollen (Bet v 1), Cladosporium, Penicillium, oak tree pollen, timothy grass, and ragweed.

⁵Asthma symptoms include wheezing, cough, chest tightness, having to stop activity due to asthma symptoms, and nocturnal awakening due to asthma.

⁶There were 9 missing values in composite asthma severity index (CASI). The score was imputed using day symptoms and albuterol use, night symptoms and albuterol use, controller treatment, lung function measures, and exacerbations.

⁷There were 8 missing values in asthma control test (ACT) score.

⁸Antibiotic use within 12 months refers to use for any reason.

⁹There was 1 missing value in % predicted FEV1, FVC, and FEV1/FVC ratio. Percent predicted values were calculated as the ratio of the pre-bronchodilator measured value to the predicted normal value based on age, sex, height, and race/ethnicity.

¹⁰Statistical comparisons were performed using one-way analysis of variance for continuous variables and chi-square tests for categorical variables, with missing observations excluded on a variable-specific basis. The Benjamini-Hochberg false discovery rate method was applied to the p-values for the seven individual allergen sensitization variables; the p-value for any sensitization and all remaining p-values were unadjusted.

Multikingdom characterization of the classroom microbiome

A total of 570 classroom dust samples, 153, 129, 155, and 133 from each experimental group, were collected and underwent deep metagenomics sequencing (Figure 1, E2). All reagent-only negative controls resulted in non-measurable DNA concentrations after library preparation and were considered to have failed sequencing. The median sequencing depth was 90.7 million reads per sample, with 37.7 million reads per sample after removing human reads. While bacteria were present in all 570 samples, archaea (a separate microbial kingdom composed of single-celled prokaryotes), fungi, and phages were detected in 242 (42.5%), 496 (87.0%), and 568 (99.6%) of samples, respectively. A total of 2,957 distinct bacteria, 35 archaea, 25 fungi, and 4,440 phages were identified at the species level. Figure 2A–D illustrates the top 10 most prevalent species stratified by microbial kingdom.

Figure 2.

Species-level composition of the classroom microbiome, stratified by microbial kingdom and correlation between classroom exposures. Each stacked bar along the x-axis represents a classroom microbiome sample, and the y-axis shows the relative abundance of detected microbes. Within each kingdom, species are displayed in order of relative abundance (ranging from 0% to 100%) for that classroom, with microbial taxa displayed in the legend in order of greatest to least abundance. Each panel shows taxa in (A) bacteria, (B) archaea, (C) fungi, (D) phages, which are labeled by viral species genome bin (vSGB) identifier and annotated in parentheses with their predicted microbial host at the genus level to provide biological context. Empty spaces in panel B and C represent samples that have not resulted in reads in that kingdom but had identified species from other kingdoms. Panel (E) represents the Spearman’s correlation matrix between classroom microbial and allergen exposures. Microbial kingdom-specific measures include total microbial load and measures of alpha diversity including species richness and Shannon indices.

The most abundant bacteria observed universally in all classroom dust samples belonged largely to the Actinobacteria and Proteobacteria phyla, with three being human-associated commensals (Collinsella massiliensis, Cutibacterium acnes, and Sutterella parvirubra) and the remainder previously described in environmental sources such as lakes, soil, or wastewater. For archaea, Methanobrevibacter smithii, a dominant human gut-associated methanogen was prevalent, as well Methanosarcina thermophila, an environmental methanogen, and Nitrosoccosmicus oleophilus, an environmental chemolithoautotroph that oxidizes ammonia to nitrite. Analysis of fungi showed that two human skin associated microbes were prevalent (Malassezia restricta and Malassezia globosa), as well as Saccharomyces cerevisiae, a facultative fermenter of sugars ubiquitous in plants and transient colonizer of human mucosa. The most prevalent phage vSGB5853, with predicted bacterial host Cutibacterium (a common human skin commensal), was present in 56.6% of samples. A total of 26 potential contaminants were identified (Table E2), though all were removed after applying a 5% prevalence filter.

To better understand whether other environmental pollutant measures such as allergen levels or microbial load indicators serve as a surrogate for microbial diversity, Figure 2E shows a heatmap with Spearman’s correlation between classroom measures of microbial load (endotoxin, (1,3)-beta-D-glucan, total microbial cells), allergen levels, and alpha diversity based on sequencing (species richness and Shannon index); the alpha diversity of each microbial kingdom (bacteria, archaea, fungi, phage) were evaluated separately. No measure of classroom microbial load nor allergen measure was significantly correlated with alpha diversity, indicating that allergen exposure, microbial load and microbial diversity represent independent exposures.

Effect of IPM and HEPA intervention on classroom microbiome across microbial kingdoms

The IPM but not the HEPA intervention changed the classroom microbiome with an increase in alpha diversity across several kingdoms. In intention-to-treat analyses, IPM increased species richness for bacteria (37.30, 95% CI [4.21, 70.39]) and fungi (0.45, 95% CI [0.16, 0.75]), increased the Shannon index for bacteria (0.19, 95% CI [0.01, 0.36]), archaea (0.05, 95% CI [0.01, 0.10]), and fungi (0.15, 95% CI [0.04, 0.26]), Figure 3A and Table E3. After IPM, total microbial load (microbes per mg dust) decreased only for archaea (−0.36, 95% CI [−0.71, −0.01]) in the unadjusted analyses. In sensitivity analyses adjusting for season, all associations with alpha diversity remained unchanged; however, bacterial and archaeal loads decreased after IPM. Similar intention-to-treat analyses were conducted to estimate the effect of the HEPA intervention (Table E4 and Figure E3); none of the alpha diversity indices were affected by HEPA intervention, for any kingdom.

Figure 3.

IPM intervention effects on (A) microbial biomass, species richness, and diversity and (B) differentially abundant species tested by linear mixed-effects models accounting for nested repeated measurements within classrooms and schools. For panel A, values reflect estimates from unadjusted and season-adjusted models with corresponding confidence intervals of species richness, Shannon diversity index and microbial load (log₁₀[cells/mg dust]) after IPM. Total phage load was not estimated. For panel B, colors represent the phylum for bacterial taxonomy and the predicted microbial host taxonomy of phages. Multiple hypothesis testing was controlled using false-discovery rate (q < 0.1). Analyses were stratified by kingdom; no archaeal and fungal species showed statistically significant changes. All models were fit on centered log-ratio transformed relative abundance data. Overall, 10 bacterial species were more abundant and 8 were less abundant post-IPM intervention; among phages, 7 increased, and 2 decreased.

IPM but not the HEPA intervention changed microbial community structure. Unadjusted and adjusted effects of IPM and HEPA interventions on beta diversity based on Jaccard index are summarized and listed in Table E5. Unadjusted analysis showed that IPM affected bacteria and phage beta diversity, a measure of differences in community composition, with 0.2% of variance explained for both kingdoms (p = 0.039 and p < 0.001, respectively), while HEPA intervention showed no variance explained for any kingdom. Season-adjusted analyses showed similar pattern as unadjusted analyses.

Similarly, IPM but not the HEPA intervention changed the differential abundance of bacteria and phage species (Figure 3B and Figure E4). In intention-to-treat analyses, the relative abundance of 10 bacteria and 7 phages were increased, while 8 bacteria and 2 phages were decreased, after IPM. Specifically, the relative abundance of Klebsiella michiganensis, Escherichia sp. ESNIH1, Agreia sp., Delftia acidovorans, Arthrobacter agilis, Belnapia rosea, Leclercia adecarboxylata, Hephaestia caeni, Curtobacterium sp., and Sphingobium yanoikuyae were increased after IPM while Prevotella intermedia, Xanthomonas oryzae, Pseudochrobactrum asaccharolyticum, Stenotrophomonas acidaminiphila, Alcaligenes faecalis, Pseudomonas sp., Dermabacter hominis, and Hymenobacter sp. were decreased. Three phages, vSGB10767, vSGB11614 and vSGB57960, predicted to infect Stenotrophomonas, were increased in relative abundance. Meanwhile, in intention-to-treat analyses, the HEPA intervention was not associated with any differentially abundant microbial species. After adjusting for season in a sensitivity analysis, three bacterial species, Acinetobacter variabilis, Hymenobacter actinosclerus, and Roseateles terrae, were significantly decreased after IPM intervention.

Microbiome signature of IPM intervention

Given the differences noted after the IPM intervention, we sought to identify a core classroom microbiome signature representing the IPM intervention. This was achieved using a sparse Partial Least Squares (sPLS) model; separate microbiome (bacteria, archaea, fungi) and phageome signatures were identified given potential phage-host interactions.²⁰ Using an 80:20 train-test split, the final sPLS model was constructed with 32 microbial species from latent component 1 and 32 from latent component 2; influential microbes are shown in the loading plots in Figure 4. The colors indicate whether the species had higher mean relative abundance after (+ IPM) or before (- IPM) the intervention across all post-intervention time points combined. The model area under the curve (AUC) showed 0.85 in the training set and 0.84 in the validation set, suggesting generalizability. Three influential microbes have previously been identified as potential rodent gut commensals (Sutterella parvirubra,^21,22 Kosakonia cowanii,²³ and Lactobacillus murinus^24,25). Other influential microbes were first identified as environmental extremophiles, organisms adapted to extreme environments, such as Cryobacterium arcticum²⁶ and Hymenobacter amundsenii²⁷ which were first isolated from Arctic and Antarctic soil, or previously described as UV-tolerant microbiota,²⁸ including Deinococcus actinosclerus, Deinococcus gobiensis. Other extremophiles included Thermoleophilum album, Dichotomicrobium thermohalophilum, and Hymenobacter amundsenii. Many of these extremophiles have previously been identified as constituents of the urban subway microbiome data as part of the MetaSUB project,^29,30 while the remaining two thermophiles (Thermoleophilum album³¹ and Dichotomicrobium thermohalophilum³²) and one acidophile (Acidisphaera rubrifaciens³³) were not present in the MetaSUB dataset but have been reported in environmental microbiome studies from a polluted canal in Brooklyn, NY;³⁴ intertidal microbial mats in Shark Bay, Australia;³⁵ and subways in Bangkok, Thailand,³⁶ respectively.

Figure 4.

Classroom microbiome signature of integrated pest management (IPM) intervention and corresponding feature loadings identified by sparse Partial Least Squares (sPLS). Colors indicate whether the species had higher mean relative abundance after (+ IPM) or before (- IPM) the intervention. The sPLS model was trained to classify pre- vs. post-IPM profiles; seasonal variation was examined separately in sensitivity analyses. Model performance using Area Under the Curve (AUC) was evaluated in both train and test datasets. Component 1 and 2 explained 100% and 72.7% of outcome variance along each dimension, respectively.

The phageome IPM signature showed smaller AUC (train: 0.79, test: 0.82) (Figure E5). Five signature phages had predicted hosts that overlapped with the microbiome signature taxa: vSGB37240 (Comamonas), vSGB16463 (Leptotrichia), vSGB13148 (Pseudomonas), and vSGB82073/vSGB82074 (Ligilactobacillus, basonym Lactobacillus).

Association between classroom microbiome and phageome signatures of IPM and asthma morbidity

Microbiome and phageome signature scores of IPM were generated for each classroom microbiome sample on a continuous scale from 0 to 1, with a higher score more indicative of a post-IPM classroom microbial profile. Post-intervention, the classrooms randomized to IPM had a higher IPM microbiome score than the control group, though the effect waned over time (Figure E6). Using generalized linear mixed-effects models, we evaluated the effect of these signatures on any asthma symptom days in both univariate and multivariate models, adjusting for student characteristics (age, sex, race and ethnicity, household income, mouse allergen sensitization), classroom mouse allergen levels, and accounting for repeated measures within students and clustering of classrooms within schools (Table 3). In both unadjusted and adjusted models, the classroom microbiome signature of IPM was protective against asthma symptoms, reducing odds of any asthma symptom days by 53% (OR 0.47, 95% CI [0.22, 0.97], p = 0.043 in adjusted models). The IPM microbiome signature was also protective for our secondary outcome, the Composite Asthma Severity Index (CASI), which accounts for asthma symptoms as well as controller medication use and lung function (Table 3). In both unadjusted and adjusted analyses, higher IPM microbiome signature scores were associated with a lower CASI (β −0.92, 95% CI [−1.51, −0.33], p = 0.002), indicating reduced severity. These associations remained statistically significant when the models were adjusted for any allergen sensitization rather than mouse-specific sensitization (Table E6). In post-hoc exploratory analyses evaluating potential effect modification by mouse sensitization status, interaction p-values between mouse sensitization and the IPM microbiome score were 0.090 and 0.544 for asthma symptom days and CASI, respectively. The association remained significant in both mouse and non-mouse sensitized subgroups for CASI and in the mouse-sensitized subgroup only for any asthma symptom days (Table 4). In contrast, classroom randomization to the IPM intervention was associated with lower odds of any asthma symptom days but was not associated with CASI (Table E7).

Table 3. Effect of classroom microbiome signature of integrated pest management (IPM) intervention on asthma outcomes.

Associations were assessed with binomial models and linear mixed effect models for primary and secondary outcomes, respectively. The classroom microbiome signature and their loadings were selected and computed with sparse Partial Least Squares (sPLS) method.

		Unadjusted		Adjusted
Outcome	Variable	OR [95% CI]	p-value	OR [95% CI]	p-value
Any asthma symptom days in the past two weeks1	Microbiome score (IPM)	0.45 [0.22, 0.90]	0.025	0.47 [0.22, 0.97]	0.043
	Age	0.94 [0.85, 1.05]	0.282	0.96 [0.86, 1.07]	0.461
	Sex (Male)	0.88 [0.57, 1.33]	0.538	0.87 [0.56, 1.36]	0.548
	Non-Hispanic White	0.94 [0.56, 1.58]	0.820	1.17 [0.67, 2.07]	0.582
	Mouse allergen sensitization	0.98 [0.59, 1.62]	0.929	1.05 [0.63, 1.74]	0.860
	Income > 35k	0.73 [0.48, 1.12]	0.152	0.74 [0.46, 1.17]	0.197
	log10(mouse allergen ng/mg)	1.01 [0.85, 1.21]	0.874	1.02 [0.85, 1.23]	0.814
		β [95% CI]	p-value	β [95% CI]	p-value
Composite Asthma Severity Index2	Microbiome score (IPM)	−0.77 [−1.34, −0.21]	0.008	−0.92 [−1.51, −0.33]	0.002
	Age	−0.08 [−0.21, 0.04]	0.184	−0.05 [−0.19, 0.08]	0.425
	Sex (Male)	0.12 [−0.39, 0.62]	0.650	0.11 [−0.42, 0.63]	0.685
	Non-Hispanic White	−0.69 [−1.31, −0.06]	0.032	−0.52 [−1.21, 0.17]	0.137
	Mouse allergen sensitization	−0.05 [−0.65, 0.54]	0.860	0.06 [−0.55, 0.67]	0.841
	Income > 35k	−0.61 [−1.12, −0.11]	0.018	−0.48 [−1.03, 0.07]	0.085
	log10(mouse allergen ng/mg)	0.15 [−0.02, 0.31]	0.075	0.16 [−0.01, 0.33]	0.060

¹Wheezing, cough, chest tightness, having to stop activity due to asthma symptoms, and nocturnal awakening due to asthma.

²Imputed using day symptoms and albuterol use, night symptoms and albuterol use, controller treatment, lung function measures, and exacerbations.

Table 4. Post-hoc subgroup analysis stratified by mouse sensitization status.

Effect of IPM microbiome score on any asthma symptom days and Composite Asthma Severity Index (CASI) stratified by mouse sensitization status in exploratory subgroup analyses. p-values for the mouse sensitization × IPM microbiome score interaction were 0.090 and 0.544 for asthma symptom days and CASI, respectively.

		Mouse Sensitized Subgroup		Mouse Not Sensitized Subgroup
Outcome	Variable	OR [95% CI]	p-value	OR [95% CI]	p-value
Any asthma symptom days1	Microbiome score (IPM)	0.13 [0.03, 0.60]	0.009	0.63 [0.27, 1.48]	0.294
	Age	0.91 [0.75, 1.11]	0.361	0.97 [0.85, 1.11]	0.642
	Sex (Male)	1.07 [0.48, 2.36]	0.875	0.84 [0.50, 1.41]	0.512
	Non-Hispanic White	0.72 [0.27, 1.94]	0.514	1.38 [0.70, 2.71]	0.358
	Income > 35k	1.22 [0.54, 2.77]	0.639	0.62 [0.35, 1.08]	0.091
	Mouse allergen ng/mg, log10	1.11 [0.80, 1.56]	0.527	0.98 [0.79, 1.22]	0.860
		β [95% CI]	p-value	β [95% CI]	p-value
Composite Asthma Severity Index2	Microbiome score (IPM)	−1.07 [−2.09, −0.06]	0.038	−0.86 [−1.57, −0.16]	0.016
	Age	−0.04 [−0.28, 0.20]	0.747	−0.05 [−0.20, 0.11]	0.558
	Sex (Male)	1.04 [0.04, 2.03]	0.042	−0.18 [−0.79, 0.43]	0.561
	Non-Hispanic White	−0.76 [−1.99, 0.47]	0.219	−0.48 [−1.28, 0.33]	0.245
	Income > 35k	−1.19 [−2.23, −0.15]	0.026	−0.24 [−0.88, 0.39]	0.453
	Mouse allergen ng/mg, log10	0.17 [−0.10, 0.44]	0.216	0.17 [−0.03, 0.38]	0.096

¹Wheezing, cough, chest tightness, having to stop activity due to asthma symptoms, and nocturnal awakening due to asthma.

²Calculated using day symptoms and albuterol use, night symptoms and albuterol use, controller treatment, lung function measures, and exacerbations.

The classroom phage signature of IPM was not associated with either outcome in both unadjusted and adjusted analyses (Table E8); any asthma symptom days (OR 0.74, 95% CI [0.33, 1.64], p = 0.455) or CASI (β −0.45, 95% CI [−1.18, 0.28], p = 0.222).

The associations between other measures of the classroom microbiome (such as alpha diversity and total microbial load) and asthma outcomes were also investigated and the results are listed in Table E9. Species richness, Shannon index, and total microbial cells/mg dust from all kingdoms were not associated with any asthma symptom days nor CASI.

DISCUSSION

In this randomized placebo-controlled trial of a school-level IPM and classroom-level HEPA purifier intervention, we show that multiple microbial kingdoms are altered by IPM but not HEPA. An IPM-derived classroom microbiome signature was associated with lower odds of any asthma symptom days and lower CASI, independent of mouse allergen exposure and student covariates. Further, neither bulk microbial load (endotoxin, β-glucan, total microbial cells) nor alpha diversity correlated with morbidity. This underscores that community composition, rather than “how much,” is the salient exposure. Together, our results support the concept that the built environment microbiome may be a modifiable exposure to improve asthma morbidity.

Prior intervention trials for asthma targeted allergen and indoor air pollutants such as tobacco smoke or particulate matter.^11,37,38 Published intervention trials focused on the indoor built environment microbiome have largely employed antimicrobial approaches to pathogen eradication such as with ultraviolet germicidal irradiation³⁹ and probiotic-based cleaning systems,⁴⁰ and largely take place in office buildings or hospitals and did not address asthma outcomes. Many such studies rely on pre-post comparisons rather than randomized designs,^41,42 which may be challenging to interpret when microbial composition and clinical outcomes such as asthma exacerbations are known to have seasonal variation.⁴³

The lack of a detectable HEPA effect on the classroom dust microbiome is not surprising. HEPA filtration primarily reduces airborne particulates but has a more limited impact on settled dust,⁴⁴ which is commonly used as an integrated reservoir of indoor microbial DNA for sequencing-based exposure assessment. The airborne microbiome is transient, low biomass, and technically challenging for metagenomics.⁴⁵ Prior studies show indoor microbial communities are driven mainly by persistent sources such as occupancy (both human and animal), resuspension from surfaces, and long-term accumulation within dust reservoirs rather than short-term airborne variation.⁴⁶ Accordingly, air filtration interventions may reduce bioaerosols without measurably altering dust microbial composition.

In post-hoc exploratory analyses, the primary clinical trial showed that there was a significant improvement in asthma symptom days for the IPM group in the first two post-intervention visits that waned over time;⁷ IPM was performed only once in intervention schools after the baseline visit. Here we focused on a mechanistic understanding of how IPM might influence the classroom environment by measuring a classroom microbiome signature of IPM to allow for the efficacy of IPM associated environmental changes to vary across classrooms and over time. We show that this microbiome signature of IPM was associated with both improved asthma symptoms and improved CASI, independent of mouse allergen levels and mouse sensitization status. It is possible that an ‘omics approach to measuring an environmental exposure would better capture the time-varying nature of the IPM intervention particularly as it relates to health, similar to recent precision nutrition studies which show that ‘omic based approaches to measuring diet show superiority over surveys for assessing health effects.⁴⁷ Another recent trial of IPM focused on the home environment and enrolled only mouse-sensitized children with asthma, and found no effect of IPM.¹¹ We have previously shown that home and school dust microbiomes are compositionally distinct; the school but not home microbiome was associated with asthma symptoms.⁶ Therefore, our findings should not be interpreted as directly contradicting prior home-based IPM trials, but rather as suggesting that school-based IPM may alter a distinct classroom microbial exposure environment which is different from the home environment.

This finding of a disconnect between sensitization, allergen exposure, and clinical outcomes has been observed for other allergens.^48,49 We have reported in an earlier observational study of elementary school children with asthma conducted from 2008 to 2013 that the association between higher school mouse allergen exposure and increased asthma symptom days in students is independent of allergic sensitization to mouse allergen, and that allergic sensitization to mouse allergen did not serve as an effect modifier between mouse allergen levels and asthma symptom days¹³. There may be a direct irritant effect of the allergen or direct allergen activation of the innate immune system,⁵⁰ although the current study further supports an additional alternate mechanism - through rodent associated changes in the classroom microbiome. Recent work suggests that microbial proteases, which are not assessed in clinical allergy testing, may serve as either allergens or can initiate and regulate allergic inflammation.^51,52 This may be a potential mechanism by which rodent-associated changes in the classroom microbiome may contribute to the relationship between mouse exposure and asthma outcomes in students regardless of sensitization status.

There has been longstanding interest in how environmental microbial exposures influence asthma, though many of these studies focus on early life microbial exposures and risk of developing asthma with fewer focused on exacerbations in those already diagnosed with asthma. Early studies on microbial exposure and asthma development relied on measuring endotoxin,⁵³ a component of the gram-negative microbial cell wall, with conflicting epidemiologic evidence.^54,55 With the recent recognition that endotoxin from different bacterial species can be either pro- or anti-inflammatory⁵⁶ studies have transitioned to using microbial sequencing to characterize the built environment microbiome.³ However, studies on asthma exacerbations in established asthmatics are still largely limited to characterizing endotoxin exposure,^57–60 with the few using sequencing^61–63 to characterize microbial exposures largely bacteriome focused. To our knowledge, there are no other randomized placebo controlled clinical trials of environmental interventions to manipulate the built environment microbiome and improve asthma severity. Here we show that the IPM intervention changes the classroom microbiome, and that a classroom microbiome signature of IPM was protective against asthma morbidity independent of mouse allergen exposure. We characterize not just the bacteriome, but also the archaeome, mycobiome, and phageome of classrooms, and show that microbial diversity is not correlated with measures of microbial load such as endotoxin. Taken together, our findings offer evidence for the health-modulating functions of the built environment microbiome and offer a broader paradigm, beyond allergen and pollutant avoidance, for environmental interventions to improve asthma control.

Mechanistically, IPM’s source-control levers such as pest suppression, sanitation and food control, and sealing likely act by several mechanisms. First, IPM alters potential rodent-linked fecal inputs, as reflected in our IPM signature taxa Lactobacillus murinus, Kosakonia cowanii, and Sutterella parvirubra. Second, IPM selects for oligotrophs/extremophiles that tolerate drier, cleaner surfaces (e.g., Deinococcus actinosclerus, Deinococcus gobiensis, Cryobacterium arcticum, Hymenobacter amundsenii, Thermoleophilum album, Dichotomicrobium thermohalophilum). Several IPM-increased lineages (e.g., sphingomonads) possess atypical outer-membrane lipids; instead of lipopolysaccharides they contain glycosphingolipids,⁶⁴ with lower TLR4 reactogenicity than canonical Enterobacterales lipopolysaccharides,⁶⁵ suggesting a less pro-inflammatory dust mixture as a plausible pathway. Third, IPM may alter the microbiome of occupants. Our prior study of academic research mouse facilities show that they have distinct microbiomes that at least transiently influence colonization of the upper airways of occupants.⁸ Finally, microbial proteases may drive or initiate allergic inflammation.^51,52 However, more work is needed to elucidate the exact mechanisms at play.

Our study has several strengths. To our knowledge, this is the first randomized placebo- controlled trial evaluating the effect of an environmental intervention on the built environment microbiome and asthma morbidity; this study design allows us to infer the causal effect of IPM on the classroom microbiome. Environmental samples were collected longitudinally throughout the academic school year, representing different seasons. We performed deep metagenomics sequencing, allowing us to characterize the effect of the intervention on different microbial kingdoms; prior studies in this area are largely focused on the bacteriome and typically at the genus or higher taxonomic level rather than the species level. Beyond the built environment microbiome, we accounted for the effect of other environmental exposures, including allergens, that are known to influence asthma outcomes. Unlike many other built environment studies of the microbiome, we have simultaneous interrogation of occupant health.

Nevertheless, there are a few limitations. Our study was conducted in the northeastern United States and may not be representative of regions with different climates. We cannot exclude the possibility that some of the effects of IPM on the classroom microbiome can be attributed to non-rodent factors, such as the durable effects of enhanced cleaning (even with bleach alternatives) or changes in food-related classroom behaviors given the multi-faceted nature of the IPM intervention, though some of the species in the IPM microbiome signature are compatible with rodent-associated inputs. The post-intervention period was limited to the winter, spring, and summer seasons as the baseline visit occurred in the fall for all intervention groups; the fall is when asthma symptoms peak for many school age children.^66–68 We are not currently able to validate our findings in an external cohort, as to our knowledge our study is the only randomized controlled trial of integrated pest management in a school setting with collection of classrooms settled dust samples throughout the school year. The associations we observed correspond to small-to-moderate standardized effect sizes, which may be due to our study population which was composed of children with mild asthma, near-normal lung function, and low prior corticosteroid. However, school IPM is a relatively inexpensive intervention thus even modest effects may be cost-effective.⁶⁹

In this randomized controlled school trial of school-based environmental interventions, an IPM intervention restructured the multi-kingdom classroom microbiome. The IPM-driven classroom microbiome signature was associated with lower odds of any asthma symptom days and lower asthma severity, independent of mouse allergen exposure. Bulk microbial load and alpha diversity did not track with morbidity, highlighting the primacy of community composition over quantity and diversity. Targeting the built environment microbiome, independent of other indoor pollutants, offers a practical path to improve asthma control in schools and supports incorporating microbiome-based biomarkers into environmental interventions to improve respiratory health.

Key messages.

• Integrated pest management can modify the classroom microbiome

• The classroom microbiome is associated with asthma symptom burden in school-age children

Abbreviations

OR

Odds ratio

IPM

Integrated pest management

HEPA

High efficiency particulate air

sPLS

sparse Partial Least Squares

AUC

Area under the curve

SICAS

School Inner-City Asthma Intervention Study

HVAC

heating, ventilation and air conditioning