healthmedicinet

Passive samplers in “real life” and experimental approaches

We employed both observational and experimental approaches to compare bacterial and
fungal quantity as well as bacterial composition across sampler types. To compare
the passive samplers in situ, multiple materials were used side by side in occupied
buildings for 1 month across two geographic locations, the United States and Finland
(Table 1). In addition, we situated different sampler types in an experimental chamber in
which known and homogenous dust, collected from the vacuum bags of local homes, was
aerosolized (Additional file 1). Within these different approaches, in total, five different materials were considered
as passive samplers. The most basic was an empty (growth-medium-free) polystyrene
petri dish 11], 12], 17], the use of which was inspired by the “pizza box” dustfall collector developed by
Würtz et al. 7]. The second was a polytetrafluoroethylene fiber sampling cloth, known as TefTex,
used as a surface wipe 18] in the Canadian Healthy Infant Longitudinal Development (CHILD) Study (http://www.canadianchildstudy.ca). The remaining three materials were different brands of dry sweeping cloths typically
used in household cleaning: Lysol and Swiffer for the USA-based sampling and Zeeman
for the Finnish-based sampling, referred to as EDC1, EDC2, and EDC3, respectively.
The use of dry sweeping cloths as so-called “electrostatic dustfall collectors” (EDCs)
was first reported by Noss et al. 6] and subsequently applied to study a variety of (micro)organisms and their products
in settled dust 5], 10], 13], 14], 19].

Table 1. Summary of the different observational and experimental settings in which different
passive samplers were compared

Bacterial composition across samplers

Several lines of evidence indicate that, within each experimental setting, bacterial
composition was similar within a sampling environment regardless of the sampler type
used to characterize that environment. That is, bacterial composition of the passively
collected dust correlated most strongly with the particular environment in which the
sample was collected rather than with the particular method of dust collection, and
this was true both for in situ building samples (Fig. 1a, b) and for experimental conditions (Fig. 1c). Statistical analysis confirmed that the sampling environment was the single largest
predictor of microbial community composition within a study and that sampler type
was found to have much less predictive power, even if differences between sampler
types reached statistical significance (Table 2). Moreover, we utilized supervised learning to determine if unlabeled communities
could be classified as belonging to a particular sampler type based on a set of labeled
training communities 20]. The interpretation of the technique is based on a ratio of classification error
to that of baseline error. For each of the USA homes, Finland buildings, and experimental
chamber, this ratio was ~1, indicating that the classifier performed no better than
random guessing at which sampler types from which experimentally unlabeled microbial
communities were derived (Table 2). On the other hand, the ratio of classification error to baseline error for classifying
sampling environment was ?2.3, indicating that the classifier performs at least twice
as well as random guessing for determining the particular dust environment. Lastly,
we examined the diversity of taxa detected in the different sampler types within a
given study component (USA homes, Finland buildings, and chamber), as this study was
not focused on how diversity compared across the environments. Using a mixed effect
model, Shannon diversity was not found to vary across the sampler types (ANOVA p??0.05), and observed richness significantly varied only in the chamber component
(ANOVA p??0.05), where it was lower in the EDCs compared to other sampling approaches.

Fig. 1. Bacterial community composition across experimental localities. Panels are a USA homes, b Finland buildings, and c experimental chambers, and community distances are visualized based on the Bray-Curtis
community distance. Different sampling localities or rounds appear as different colors, and different sample types are marked with different symbol shapes. Except in the chamber study, samplers were tested in duplicate, so symbols will
repeat

Table 2. Factors influencing bacterial community composition in settled dust samples. Permanova
analyzes the statistical variance in biological Bray-Curtis dissimilarity among bacterial
communities explained by different measured variables, where R²
represents the variance explained be each factor and the corresponding p value. The ratio in supervised learning refers to the ratio of the error in classifying microbial communities into categories
of factors to the baseline error of random assignment, where a ratio of ~1 indicates
no better classification than random

In addition, our data speak to two aspects of sampling repeatability. In the USA homes,
samplers were placed at two heights, and in the Finland buildings, duplicate samplers
were placed side by side at the same location. In each of these trials, duplicate
samples were statistically indistinguishable with regard to bacterial composition
(Table 2).

The taxonomic composition observed was largely consistent with other recent studies
of indoor bacterial microbiomes (e.g., 21], 22]). Ten groups—the Staphylococcaceae, Micrococcaceae, Moraxellaceae, Corynebacteriaceae,
Streptococcaceae, Sphingomonadaceae, Bartonellaceae, Enterobacteriaceae, Rhodobacteraceae,
and Streptophyta—combined to ~50 % of sequence reads (Additional file 2). Within the chamber trials, for which the microbial community composition of the
input dust is known through direct sequencing, there are modest differences in the
compositional proportions between the vacuum dust and passive samplers. However, the
passive samplers are all skewed in the same direction, such that Pseudomonadales,
Enterobacteriales, and Streptophyta are underrepresented in the passive collectors,
relative to their abundance in the vacuum dust that was aerosolized into the chamber
(Fig. 2). Figure 2 highlights the top-most abundant taxa by sequence reads, and the full dataset is
available as Additional file 2.

Fig. 2. The top-most 16 bacterial orders detected in the experimental chamber. Left column is the input vacuumed dust, and the four right columns are the passively settled dust in the different sampler types

Within the building-based observations, taxa tended to vary in their relative abundances
rather than in their detection. For example, within Finland buildings, 21 of the 25
most abundant taxa found in the petri dishes were common to the top taxa detected
in the EDC and 15 were common to the top taxa in the TefTex. It was only the more
rare taxa that were detected in one sampler and missed entirely in others. For instance,
a bacterial operational taxonomic unit (OTU) belonging to the family Dermatophilaceae
represented 0.08 % of the sequences in the Petri dish sequences and 0.004 % of the
sequences in the EDC but was not detected in the TefTex samples. Within USA homes,
Streptophyta (likely chloroplasts) comprised a much larger percentage of the reads
in petri dishes than the other sampler types.

Fungal data were available for only one component of the study, that from USA homes.
Using an approach similar to that used for bacteria, the sampling environment of the
USA homes explained over half the variation in fungal composition while sampler type
was not a significant predictor (see further details in Additional file 3).

Microbial quantity across samplers

Quantitative PCR was used to estimate the microbial quantity collected in each of
the samplers. Tables 3 and 4 report the bacterial and fungal counts, respectively, and additional quantitative
PCR (qPCR) markers and more detailed information on analyses of the Finland building
samples are included (Additional file 4). Because experimental protocols were different in the USA and Finland (see the “Methods”
section), absolute values of microbial quantities across study components are difficult
to compare. This was particularly the case for the extraction protocol of EDC and
TefTex samplers, where the Finnish protocol included a rigorous and more efficient
dust extraction procedure. In the USA homes, the highest yields of microbial biomass
were found in the petri dish, followed by TefTex and the two EDCs, which had similar
yields. For bacteria, the mean ratios of biomass detected relative to the highest
yield in the petri dish—normalized for sampling surface area—were 0.3 for TefTex,
0.2 for EDC1, and 0.4 for EDC2; for fungi, the mean ratios were 0.2 for TefTex, 0.1
for EDC1, and 0.1 for EDC2. In the Finland buildings, the highest yields for microbial
groups were generally ranked as the petri dish, EDC, and then TefTex samplers, although
house 3 was an exception. For bacteria, the mean ratios of biomass detected relative
to the highest yield in the petri dish were 0.4 for TefTex and 0.6 for EDC3; for fungi,
the mean ratios relative to petri dishes were 0.4 for TefTex and 0.8 for EDC3. The
relative differences across locations matched predictions based on occupancy, although
we acknowledge low sample numbers. For example, within the USA, quantities were lowest
for house 1, which was occupied by a single occupant, and highest for house 3 occupied
by a family of five with three dogs. In Finland, houses showed higher microbial biomass
than work settings (one labspace, two offices). In contrast to the home settings,
yields from the chamber did not show such clear trends. In the chamber, which had
much higher particle loading onto the samplers compared to the buildings, TefTex samplers
most often showed the highest yields, followed by the petri dish samplers. For bacteria,
the mean ratios of biomass detected relative the highest yield in TefTex were 0.7
for petri dish, 0.5 for EDC1, and 0.2 for EDC2; for fungi, the mean ratios were 0.7
for petri dish, 0.5 for EDC1, and 0.2 for EDC2.

Table 3. Bacterial quantity across sampler types and experimental conditions. Values reported
are mean and standard deviations of cell equivalents per 100 cm
²
of sampler per time of exposure (day for USA and Finland, hour for chamber). Note
that bacterial determinations relied on different qPCR protocols in the USA/chamber
studies and the study part in Finland, and thus, absolute values are not well comparable
between study parts but are comparable between sampler types within environment

Table 4. Fungal quantity across sampler types and experimental conditions. Values reported
are mean and standard deviations of cell equivalents per 100 cm
²
of sampler per time of exposure (day for USA and Finland, hour for chamber). Note
that fungal determinations relied on different qPCR protocols in the USA/chamber studies
and the study part in Finland, and thus, absolute values are not well comparable between
study parts but are comparable between sampler types within localities

Side-by-side samplers in the Finland component of the study allows for examination
of the correlation between duplicate samplers. Table 5 summarizes Pearson’s correlations of duplicate sampler qPCR determinations. Overall,
strong and highly significant correlations were observed for the duplicate determinations
in most cases, except in some cases for the TefTex material. The highest correlations
were found for EDC3, followed by petri dish, and then TefTex. Although limited by
a small number of different sampling environments and duplicate samples, analyses
of the intraclass correlation (ICC, comparing the within-location variance to between-location
variance) and coefficient of variation (CoV) of duplicates showed similar trends,
with highest correlation/lowest variation observed for EDC3, followed by petri dish
sampling, then the TefTex material. Lastly, correlations of biomass determinations
between different sampler types were strong (Pearson correlation 0.85 for each sampler pairwise
correlation). Further information is detailed in Additional file 4.

Table 5. Pearson correlation coefficients of naturally log-transformed qPCR data for duplicate
determinations from sample pairs in Finland locations