Analysing indoor mycobiomes through a large‐scale citizen science study in Norway

In the built environment, fungi can cause important deterioration of building materials and have adverse health effects on occupants. Increased knowledge about indoor mycobiomes from different regions of the world, and their main environmental determinants, will enable improved indoor air quality management and identification of health risks. This is the first citizen science study of indoor mycobiomes at a large geographical scale in Europe, including 271 houses from Norway and 807 dust samples from three house compartments: outside of the building, living room and bathroom. The fungal community composition determined by DNA metabarcoding was clearly different between indoor and outdoor samples, but there were no significant differences between the two indoor compartments. The 32 selected variables, related to the outdoor environment, building features and occupant characteristics, accounted for 15% of the overall variation in community composition, with the house compartment as the key factor (7.6%). Next, climate was the main driver of the dust mycobiomes (4.2%), while building and occupant variables had significant but minor influences (1.4% and 1.1%, respectively). The house‐dust mycobiomes were dominated by ascomycetes (⁓70%) with Capnodiales and Eurotiales as the most abundant orders. Compared to the outdoor samples, the indoor mycobiomes showed higher species richness, which is probably due to the mixture of fungi from outdoor and indoor sources. The main indoor indicator fungi belonged to two ecological groups with allergenic potential: xerophilic moulds and skin‐associated yeasts. Our results suggest that citizen science is a successful approach for unravelling the built microbiome at large geographical scales.

Fungi, one of the most diverse kingdoms of life, with essential ecosystem functions (Willis, 2018), are also present in the built environment, where the extreme environmental conditions (dry and generally warm) favour certain species. The overall assembly of fungi in buildings can be termed the "indoor mycobiome" and is largely composed of saprotrophs that degrade available organic substrates and stress-tolerant ascomycetes, including ubiquitous airborne mould genera (e.g., Cladosporium, Penicillium, Aspergillus and Alternaria).
Wherever enough moisture is present, fungi grow and subsequently emit spores, fragments of hyphae, volatile organic compounds and mycotoxins that act as sources of indoor pollutants (Flannigan & Miller, 2011;Nevalainen et al., 2015;Rintala et al., 2012). Dampnessand mould-related indoor air quality problems are a public health concern due to their association with adverse health effects, such as allergies, asthma and other respiratory symptoms (Fisk et al., 2007;Mendell et al., 2011).
Microbiological assessments in the built environment focus mainly on air and dust samples indicative of human exposure indoors. The fungal content of these samples can be analysed using different approaches: microscopy, culturing, chemical analyses and DNA-based methods (Nevalainen et al., 2015). Considering the wellknown limitations of culture-based methods (Amann et al., 1995), a shift toward DNA-based methods has taken place in recent decades.
High-throughput sequencing (HTS) of amplified markers (DNA metabarcoding) has recently become a key tool for surveying fungal communities in environmental samples (Lindahl et al., 2013;Nilsson et al., 2019). In the last decade, many studies have used DNA metabarcoding to reveal the microbiome of residential buildings in different parts of the world (Gilbert & Stephens, 2018), mainly focusing on bacteria (Adams et al., 2015;Lax et al., 2014), but also on fungi (Adams et al., 2013a(Adams et al., , 2013bAmend et al., 2010;Barberán et al., 2015a;Tong et al., 2017).
The indoor mycobiome is determined primarily by large-scale environmental gradients such as climate, but local environmental variation within individual buildings, including differences in construction features and building functions, can also contribute to shaping the fungal diversity and composition (Adams et al., 2016;Gilbert & Stephens, 2018;Stephens, 2016). A first global survey analysing 72 settled-dust samples from buildings in six continents revealed that the indoor fungal diversity is significantly higher in temperate zones than in the tropics, with latitude being the best predictor of the indoor mycobiome composition, while neither building design nor function had any significant effect (Amend et al., 2010).
Both culture-and DNA-based studies have demonstrated that outdoor air is the main source of indoor fungi. Adams et al. (2013aAdams et al. ( , 2013b observed that indoor fungi are dominated by those spreading from outdoor air, and the mycobiome of indoor surfaces displayed similar patterns to outdoor air in the same locality. Barberán et al. (2015aBarberán et al. ( , 2015b analysed dust microbiomes collected inside and outside 1200 houses across the USA and confirmed that most indoor fungi were derived from outdoor sources.
They further identified geographical patterns in the indoor mycobiomes that could be explained by climate, soil and vegetation variables.
DNA-based dust studies have indicated that various building features and occupant characteristics are also key determinants of the indoor mycobiome Kettleson et al., 2015). In this regard, Yamamoto et al. (2015) claimed that indoor emissions associated with occupant activities were the primary sources of airborne allergenic fungal particles. However, taken together, it is well accepted that the indoor mycobiome is determined largely by the outdoor environment, while bacteria are more strongly influenced by occupants and their activities (Adams et al., 2016;Barberán et al., 2015a;Gilbert & Stephens, 2018;Lax et al., 2014;Stephens, 2016).
Except for the pioneering global study by Amend et al. (2010), and the continental-scale study across the USA by Barberán et al. (2015aBarberán et al. ( , 2015b, the majority of existing DNA-based mycobiome studies have focused on specific building units at a local scale. A few regional studies have also targeted some large cities, like Munich (Weikl et al., 2016) and Hong Kong (Tong et al., 2017). Given that the indoor mycobiome is highly influenced by the outdoor air, we can expect significant differences between houses inherent to their outdoor regional climate and environment. Revealing the indoor mycobiome and characterizing the variations across houses from different geographical regions of the world will provide basic knowledge for improved indoor air quality management and the identification of health risks.
Our study area, Norway, possesses marked climatic and environmental gradients, enabling us to assess to what degree the outdoor environment, vs. building features and occupant characteristics, influence the indoor mycobiomes. To represent a broad sample of buildings, we organized a citizen science dust sampling campaign in houses throughout Norway coupled with subsequent DNA metabarcoding analyses of the mycobiomes. Previous studies have demonstrated that citizen science, coupled with HTS approaches, is a promising avenue for conducting large-scale microbiome studies, including the built and human microbiomes (Barberán et al., 2015a;McDonald et al., 2018).
More specifically, we addressed and tested the following research questions and hypotheses: (i) which factors shape the indoor mycobiomes? In this regard, we investigate whether regional-scale variation in climate (and other regional-scale variables), building features or occupant characteristics are the main determinants. Here we hypothesize (H1) that all three categories influence the indoor mycobiomes, but regional-scale climate is the most important driver.
Next, we ask (ii) which fungi dominate the house-dust mycobiomes in Norway. We hypothesize (H2) that ascomycetes, and especially stress-tolerant ascomycetes, are the dominant groups in this environment. We also ask (iii) how much of the indoor mycobiome overlaps with the outdoor mycobiome. In relation to this question, we hypothesize (H3) that a major fraction of the indoor fungi derives from outdoor sources, while a relatively minor fraction originates from indoor sources.

| Citizen science dust sampling campaign
To increase the number of study houses and cover a broad geographical area, citizen scientists were recruited through scientific F I G U R E 1 Overview of the citizen science dust sampling campaign in Norway. (a) Schematic overview of the metadata for each house: outdoor metadata that mainly include climatic variables (green), building features (violet) and occupant characteristics (blue). The sampling points (house compartments) are indicated with red dots. The building variable "Dust coverage" corresponds to the percentage of dust covering the study surface at the living room, as measured on the adhesive tape (Mycotape2 To minimize the influence of seasonality effects, all samples were collected in a short time span during spring 2018, mainly in May (from April 27 to June 5). In total, 269 houses were sampled from mainland Norway, covering its major climatic gradients ( Figure 1b; Figure S2). Two houses from Longyearbyen, in the Arctic Archipelago of Svalbard, were also included.

| Environmental data
Metadata about the study houses and their occupants were provided by the volunteers through an online questionnaire at the UiO website. In addition to the location of houses including their addresses and geographical coordinates (latitude and longitude), the follow- and annual precipitation BIO12) were extracted at 30-seconds resolution (~1 km 2 ) using the r package dismo (Fick & Hijmans, 2017).
Moreover, data for 116 environmental variables related to geology, topography, climate and hydrology were also explored. They were kindly provided by the authors of a recent study modelling the vegetation types in Norway (Horvath et al., 2019). The contribution of all continuous variables, 46 of 116 from this data set plus the six previously extracted from WorldClim, were evaluated by principal component analysis (PCA) using the r package ade4 (Dray & Dufour, 2007) ( Figure S3). Based on this PCA, 10 continuous variables were selected for the statistical analyses: the six detailed WorldClim bioclimatic variables plus growing season length, snow-covered area in February, snow water equivalent in April and potential incoming solar radiation. Two additional categorical variables from the vegetation study (Horvath et al., 2019): land cover AR50 (developed area/ agricultural area/forest/barren land/bog and fen/fresh water) and bedrock nutrient (poor/average/rich), as well as the dust coverage measured on the adhesive tapes, were included in the final selection (32 variables; Figure 1a). The internal transcribed spacer 2 (ITS2) region of the nuclear rDNA was amplified using the primers gITS7 5′-GTGARTCATCGARTCTTTG-3′ (Ihrmark et al., 2012) and ITS4

|
5′-TCCTCCGCTTATTGATATGC-3′ (White et al., 1990). Both forward and reverse primers were designed with 96 unique tags (barcodes) of 7-9 bases at the 5′-end, which differed in at least three positions from each other. To avoid tag switching errors (Carlsen et al., 2012), samples were combined in pools of 96 samples, each with a unique tag combination (Table S1). Nine pools (96 samples each) were analysed in this study, and each of them included an extraction blank, a PCR (polymerase chain reaction) negative and a mock community that was used as a positive control (details in the Supporting Information

| Bioinformatics pipeline
After an initial quality checking of sequencing results using fastqc version 0.11.2 (Babraham Bioinformatics Team), samples were demultiplexed independently (R1 and R2) with cutadapt version 1.8 (Martin, 2011) allowing zero mismatches in tags and primers; these were simultaneously removed along with sequences shorter than 100 bases. The demultiplexed R1 and R2 reads were kept separate for the next analyses using dada2version 1.12 (Callahan et al., 2016): (i) quality filtering and trimming, (ii) dereplication, (iii) generating error models and denoising, (iv) merging in contigs, (v) creating the table of amplicon sequence variants (ASVs) and (vi) removal of chimeras. Additional clustering of ASVs in operational taxonomic units (OTUs), as recommended in previous studies , was done using vsearch version 2.11.1 (Rognes et al., 2016) at 98% similarity. This clustering level is similar to the 98.5% level used to define the species hypotheses (SHs) in the UNITE database . OTUs containing only one read (singletons) were removed after clustering. To correct for potential over-splitting of OTUs due to remaining sequencing errors, the OTU table was curated using lulu with default settings (Frøslev et al., 2017).
Taxonomic assignment of the OTUs was carried out using vsearch against the eukaryotic ITS data set from unite version 8.0 (UNITE Community, 2019a). Two filters were subsequently applied on the resulting OTU table to select those OTUs that contained at least 10 reads and showed at least 70% identity in the taxonomic assignment. Finally, we selected the OTUs assigned to the kingdom Fungi on the quality-filtered table. To refine the taxonomic annotation of the top-100 most abundant fungi, a double-checking was done on those OTUs that initially failed at the species level. This was performed using blast+ version 2.8 against both UNITE and NCBI databases (UNITE Community, 2019b). Ecological trophic modes and guilds for the identified taxa were annotated using the funguild tool (Nguyen et al., 2016). More details on the bioinformatics analyses and the assessment of control and replicates samples are provided in the Supporting Information (Table S2, Figure S4 and supplementary methods).   Figure S7). In contrast to richness and diversity, the compositional dissimilarity (beta diversity) was higher among the outdoor samples (Figure 2b).

| Determinants of the community composition
We observed a marked compositional difference between indoor and outdoor mycobiomes, as revealed by NMDS ordination of all dust samples (Figure 3a). House compartment (outside, living F I G U R E 2 Box plots visualizing diversity patterns in the three studied house compartments. A total of 269 houses were assessed, including dust samples from the outside (n = 266), living room (n = 270) and bathroom (n = 271).
(a) Alpha diversity (richness) and Shannon index; (b) beta diversity. All statistics were calculated from the rarefied matrix (6632 OTUs). All differences between outdoor and indoor compartments (outside vs. living room and outside vs. bathroom) were highly significant according to Tukey HSD test (p < 1e-05), while no significant difference was found between living room and bathroom (p > .05) Anova, p = 9.9e-09 Anova, p < 2.2e-16 Anova, p < 2.2e-16  (Table 1), altogether accounting for about 15% of the variation (Figure 4). Climatic variables were also important for the fungal community composition in the dust samples, as seen in the ordination plot. Various climatic variables correlated with the second ordination axis (Figure 3b; Table S3). Together, climatic variables accounted for 4.18% of the variation among all dust samples, which increased to 6.79% for the outdoor samples when analysed separately (Figure 4). The four most important climatic variables were annual temperature variation (temperature seasonality BIO4), mean temperature of the warmest (BIO10) and the driest (BIO9) quarter, as well as annual precipitation (BIO12) ( Table 1). There was a clear geographical signal in the fungal community composition. This was especially the case for the outdoor samples, but also, to a lesser extent, for the indoor samples ( Figure   S8a,c, Table S3), which again relate to the regional climate variability in the study area (Figure 1). Building features and occupant  for the indoor data set (Table 1). The more occupants there were in houses, the more similar the indoor samples were to outdoor samples in fungal community composition (Figure 3b).

| Dominant fungi in house dust
The taxonomic assignment for the most abundant fungi is shown in Figure 5 and Figure S9 (FUNGuild annotation) and ( Figure 5a). The third most abundant phylum was Mucoromycota, showing higher percentages of sequences in the indoor samples (2.1% living room and 1.5% bathroom) compared to outside (0.3%).
As seen from the OTU ordination plot in Figure 3c, there was a broad-scale structuring of the major taxonomic groups. Both for the indoor and the outdoor samples, ascomycetes were in general more associated with areas with higher precipitation, lower mean temperature of the warmest quarter and low degree of seasonality in temperature, while the basidiomycetes showed the opposite pattern, being associated with more continental climates (Figure 3b,c).
At the order level, there were also marked differences between indoor and outdoor samples (Figure 5b). Eurotiales, the most common order, including 20.8% of the total sequence count, was far more abundant in the indoor compartments (30.5% in living rooms and 25.2% in bathrooms) than outside (6.5%). The same trend appeared for Saccharomycetales, Agaricales, Helotiales, Malasseziales and

Mucorales. In contrast, Capnodiales, Pucciniales, Lecanorales and
Chaetothyriales were clearly more abundant in the outdoor samples.
Like for the order level, there were also clear trends for the most common genera (Figure 5c) 12.6% of the total sequences; Figure 3d). In addition, the yeast genera Malassezia and Aureobasidium were particularly abundant in bathrooms.

| Indoor vs. outdoor mycobiomes
A large proportion of the fungi (36.3% of the OTUs) were present in all three house compartments and 50.6% of the OTUs were shared between indoor and outdoor compartments (Figure 6a left). However, after excluding low-abundance OTUs (with <10 reads per sample), only 27.4% of the OTUs were shared between outdoor and indoor samples (Figure 6a right), indicating that the relatively high overlap was largely driven by rare fungi. In addition, comparing the overlap on a houseby-house basis revealed that only 15% of the OTUs on average were F I G U R E 4 Venn diagram summarizing the variation partitioning analysis (VPA). The three groups of variables are indicated in colours ("Building," "Occupants" and "Climate") and compared to the factor "House compartment"; see Table 1    indicator OTUs) and Agaricales (15.5%), and numerous outdoor indicator OTUs in Lecanorales (16.5%), Chaetothyriales (16.5%) and Capnodiales (13.4%) ( Figure S10). OTUs with the highest indicator values (IndVal > 50%) for indoor and outdoor environments are detailed in Table 2. Overall, indoor indicator fungi were mostly characterized by their allergenic potential and association with human skin and material colonization, while outdoor indicator fungi were associated with rock-inhabiting fungal taxa.  The fungal community composition in house dust was clearly different between indoor and outdoor samples. After accounting for

| Determinants for the indoor dust mycobiome
the key effect of the house compartment (7.66% of the variation), our results corroborated the first hypothesis (H1), namely that regional-scale climate is the most important driver of the mycobi-
TA B L E 2 (Continued) (Adams et al., 2013a(Adams et al., , 2014Sylvain et al., 2019). This trend was reported for the fungal diversity and biomass in settled dust from water-damaged units of a housing complex in San Francisco, with the lowest diversity inside units with visible moulds (Sylvain et al., 2019).
However, that finding was associated with the influence of a few dominant taxa, which were probably growing and spreading from mould colonies indoors. In this regard, Adams, Amend, Taylor, and Bruns (2013) demonstrated that local sources of abundantly sporulating fungi might distort the perception of species richness and community composition assessed by PCR-based HTS approaches, where a few abundant species can mask the presence of rarer fungi during the PCR.
In addition, several studies have reported a global trend for fungal diversity and richness that increase with latitude (Amend et al., 2010;Větrovský et al., 2019). Our study also supports this trend, as slightly higher alpha diversities were obtained for houses in northern Norway.
In agreement with previous studies in the built environment, which mainly described air-and dust-borne communities, the mycobiomes in studied houses were clearly dominated by ascomycetes (⁓70%) with Capnodiales and Eurotiales as major orders in abundance, corroborating our hypothesis H2. These orders are well known for their stress tolerance; Capnodiales (with Cladosporioum as the dominant genus in our data set) is particularly rich in extremotolerant species, including saprobes, plant pathogens, endophytes, epiphytes and rock-inhabiting fungi (Ametrano et al., 2019;Crous et al., 2009), while Eurotiales contains many xerophilic fungi (especially Aspergillus and Penicillium species) that are able to grow on substrates with low water activity (aw ≤ 0.85) like household dust (Flannigan & Miller, 2011;Pettersson & Leong, 2011).
Interestingly, we observed a distinct difference in the overall distribution of Ascomycota and Basidiomycota; the former was to a higher extent connected to areas with high annual precipitation and longer growing season, while basidiomycetes were more prevalent in continental areas with high degree of seasonality and high snow cover during winter. More than reflecting the actual biogeography of the two phyla, we speculate that this pattern may partly be due to temporal differences in the vegetation period across the study

| Overlap between indoor and outdoor mycobiomes
In light of previous studies (Adams et al., 2013a; 2015a), we expected that a major part of the indoor fungi originated from outdoor sources (H3). Barberán et al. (2015a) (Barberán et al., 2015b;Flannigan, 2011;Nevalainen et al., 2015;Rintala et al., 2012;Shelton et al., 2002). They are especially abundant indoors, as part of household dust or colonizing building materials and foodstuffs, which become relevant sources for further conidial dispersion (Andersen et al., 2011;Flannigan & Miller, 2011). Wallemia is an extreme xerophilic basidiomycete, commonly found in dust due to its ability to grow at low water potential, aw < 0.75 (Flannigan & Miller, 2011; showing a prevalence in indoor environments (Dannemiller et al., 2014;Findley et al., 2013;Flannigan, 2011;Maestre et al., 2018;Rintala et al., 2012;Tong et al., 2017). The fourth mostabundant species (OTU3, 5.4% of the total reads, present in 96% of houses), with the highest indoor IndVal (91.2%), was identified as Saccharomyces sp., a relevant genus in food production that includes S. cerevisiae (baker's and brewer's yeast) and has previously been reported in indoor environments (Barberán et al., 2015a;Flannigan, 2011). The majority of these indoor fungi have been described as potential allergenic taxa (Esch et al., 2001;Yamamoto et al., 2012). Lastly, there was a significantly higher occurrence (mean = 21% of study houses) of indoor indicator species compared to outdoor indicators (9%), supporting that there is a consistent indoor core mycobiome.
Outdoor dust mycobiomes, collected at the doorframe of the main entrance outside the buildings, also showed striking differences compared to the indoor mycobiomes. Besides the prevalence of Cladosporium and Thekopsora (18% and 16% mean relative abundance per samples, respectively), the indicator species analysis revealed that outdoor samples were distinctly enriched in so-called rock-inhabiting fungi, including lichen-forming fungi of the order Lecanorales (16% of outdoor indicator OTUs), as well as fungi affiliated to Chaetothyriales (16%) and Capnodiales (13%). They are well known for their multistress tolerance and prevalence in diverse outdoor environments including rocks and buildings, where they are exposed to solar radiation, desiccation and rehydration, temperature fluctuations, osmotic stress, pollutants and lack of nutrients (Ametrano et al., 2019;Gorbushina, 2007).

| Concluding remarks
In summary, we have shown that numerous factors are related to the composition of the indoor mycobiomes, but together only explain a small fraction of the community composition. This seems to be a general feature of fungal communities. Further observational or experimental studies should be addressed to assess the causal effect(s) of one or a few factors using a balanced and cross-factorial design.
For example, regional environmental variation can be removed by focusing on a smaller geographical area where two factors, such as number of occupants and building types, can be systematically evaluated.
Our main findings are in line with previous indoor mycobiome studies, identifying climatic variables as the key determinants of the indoor mycobiome. Building features and occupant characteristics had a significant but smaller influence. The indoor dust mycobiome represents a mixture of fungi from outdoor and indoor sources, which could also be the reason why a higher fungal richness was observed indoors. The indoor core mycobiome is characterized by two ecological groups with allergenic potential, xerophilic moulds and skin-associated yeasts. In contrast, rock-inhabiting fungi, well known for their multistress tolerance and ability to form biofilms on buildings, were the main outdoor indicator fungi.
Despite methodological limitations related to the citizen science sampling (e.g., nonuniform means of collection, small amount of dust collected with subsequently low DNA yields and low number of samples per house), this approach turned out to be highly effective and we were able to obtain a large number of samples covering Norway in a relatively short time. The DNA analyses revealed that most samples could be used in statistical analyses, with no divergent outlier samples. Moreover, most indoor and outdoor samples fall into two separate clusters, supporting that the samples were collected according to our instructions. We believe the citizen science approach holds large opportunities for further broad-scale sampling within countries and continents, but also at a global scale. Not only can indoor environments be sampled this way, but also various outdoor environments such as soil and plants. In addition to democratization of science, citizen science is a way to reduce unnecessary travelling and related carbon emissions.