patrickbdevaney/mia9 · Datasets at Hugging Face

text stringlengths 0 6.44k
Diversity 2020, 12, 0324 4 of 17
2.2. Soil Collection
In June 2018, we collected soil microbiomes from 96 sites. We colocated our fungal community
sampling with 96 understory plant community plots established in 2009. These plots were evenly
distributed among islands (12 per island) and among experimental treatments, i.e., island core type
treatments (48 plots on peat core versus 48 plots on limestone core island) and tree planting density
treatments (24 plots per density). On each island, the three plots were randomly assigned to locations
within each tree density quadrant (e.g., Figure S2). For each sampling point, we used a soil corer to
aseptically collect approximately 50 mL of soil at 15–25 cm below the soil surface and approximately
0.25 m from the center of the 1 m2 plot (indicated by permanent PVC site marker). Soil samples were
placed in sterile conical tubes and transported on ice to University of Miami’s campus where they were
kept at −80 ◦C.
2.3. DNA Sequencing and Data Processing
We extracted DNA from 0.25 g of each soil sample using the E.Z.N.A. Soil DNA kit
(Omega, D5625-01, Norcross, GA, USA). We removed PCR inhibitors from DNA samples with
agarose gel electrophoresis and extracted genomic bands (i.e., >15 kilobases; E.Z.N.A Gel Extraction kit,
Omega, D2500-01, Norcross, GA, USA). We performed quality control by endpoint PCR of the fungal
ribosomal ITS1 region to confirm that enough PCR inhibitors were removed to obtain DNA of sufficient
quality for amplification. The ITS1 region was amplified via PCR using two-step dual-indexing [45],
and amplicon libraries were sequenced at the University of Minnesota Genomics Center (UMGC)
using the Illumina MiSeq platform (v3, 300 bp paired-end, San Diego, CA, USA). We also sequenced
three negative controls in which ultrapure water (G-Biosciences, St. Louis, MO, USA; 786–293) was
used in place of soil during extractions to confirm that samples were not contaminated during the
DNA extraction process. UMGC demultiplexed reads using bcl2fastq. We denoised reads, joined paired
ends, and grouped reads into exact sequence variants (ESVs; i.e., operational taxonomic units at
100% sequence similarity) with the QIIME2 pipeline (v2019.1, [46]). We rarefied sequences to 4000 reads
per sample to account for unequal sequencing depths and excluded samples with less than 4000 reads
(12 of 96 samples). We classified ESVs into fungal “species” if they shared at least 97% sequence
similarity to reference sequences in the UNITE database (version v7_01.12.2017, [47]). For the fungal
community at each site, richness was determined based on the number of rarefied ESVs, and Shannon
diversity was calculated using sequence reads of ESVs as estimates of abundance. Note that using
ESVs, compared to OTUs (Operational Taxonomic Units) with 97% sequence similarity, may result
in higher absolute fungal richness, but ESVs and 97% OTUs are often highly correlated, such that
comparisons among treatments are robust to sequence binning choices [48].
2.4. Data Analyses
To understand which abiotic and biotic factors explain fungal community diversity and richness,
we performed model selection with the abiotic and biotic factors in Table 1 as explanatory variables and
identified the best model based on AIC using dredge (R package MuMin; [49]). We also included tree
island identity as a random effect in the model selection to account for understory plant community
variation between islands. We confirmed that model variables were not highly correlated with each
other before performing model selection (all correlation coefficient <0.7; Figure 2). To investigate
the possible causes of variation in fungal community composition, we performed a distance-based
redundancy analysis (dbRDA, Bray-Curtis) with the same abiotic and biotic factors as explanatory
variables using capscale (R package vegan; [50]). Indicator taxa analysis was used to identify which
of the 500 most common fungal taxa were associated with differences in community composition
between treatments for significant categorical variables (e.g., limestone versus peat core). We calculated
‘indicator values’ of the fungal taxa based on the relative abundance and consistency in each treatment
and determined the significance of these values with permutation tests (R package indicspecies [51,52]),
Diversity 2020, 12, 0324 5 of 17
which were FDR (False Discovery Rate) corrected for multiple comparisons. We also conducted a
Mantel test between fungal and plant community distance matrices using mantel (R package vegan, [50])
to examine whether the variation in fungal community composition can be explained by variation in
understory plant community composition.
Table 1. Best models for the response of fungal diversity and richness to abiotic and biotic tree
island factors.
Island Factors Fungal Diversity Fungal Richness
Relative Water Level F1,68 = 8.63, p = 0.0045 F1,65 = 2.40, p = 0.1255
Island Core – F1,6 = 1.18, p = 0.3188
Canopy Openness – F1,65 = 9.41, p = 0.0031
Tree Density – F1,65 = 1.99, p = 0.1626
Understory Evenness F1,68 = 8.83, p = 0.0041 F1,65 = 7.66, p = 0.0073
Understory Richness – F1,65 = 1.78, p = 0.1864
Estimated Understory Biomass – –
NOTE: All environmental factors with statistical results listed in the table were part of the model selection’s best
fit model, and significant terms are bolded. Double dashes indicate those environmental variables that were not
included in the best model. ‘Understory’ metrics refer to characteristics of the understory plant community.
To investigate the distribution of functional guilds of fungi, we used the FUNGuild database
(version v1.1; [53]), which assigned each taxa identified at the species level via UNITE to a functional
guild (e.g., leaf saprotroph or fungal parasite) and a trophic mode (e.g., symbiotrophic, pathotrophic)
(Figure S3A,B). ESVs that were not identified to the species level were not included in this analysis.
We also filtered to include only taxa that had guild and trophic mode assignments with the confidence
rankings “Highly Probable” and identified the three most common guilds based on their normalized
ESV counts. For our analyses, we focused on functional guild assignments using model selection to
determine which biotic and abiotic variables explained variation in the number of different guilds
present at a site (guild richness) and the relative abundances of the top three most common functional
guilds. As above, our model selection used linear mixed effects models with abiotic and biotic factors
as explanatory variables and tree island identity as a random effect. All statistical analyses were
performed in R ver 3.6.0.
3. Results
3.1. Sequencing and ESV Taxonomic Assignments Statistics
After quality control filtering, 84 of the 96 samples had sufficient read counts to reach our
saturation cutoff (≥4000 reads) required to be included in analyses. Across these samples, there were
over 9000 ESVs, with an average of 243 ± 11 ESVs per sample. Using the UNITE database, we identified
approximately 35% of all ESVs and an average of 100 ± 5 taxa per sample (mean ± s.e) to at least
the order level. Approximately 20% of all ESVs were identified in UNITE to the species level,
allowing them to be included in our functional analyses using FUNGuild. Fusarium keratoplasticum was
most common across sites, occurring in 71 of the 84 sites, followed by Penicillium sp. (69 of 84 sites) and
Retroconis fusiformis (69 of 84 sites). An unidentified fungal taxa had the greatest number of reads on
average (1149) followed in abundance by an ESV from the order Tremellales (614) and Inocybe curvipes
(597). Demultiplexed sequence data, the ESV table, and the taxonomy table have been submitted to
NCBI (BioProject: PRJNA639837).
3.2. Fungal Community Diversity and Richness
The relative water level and evenness of the understory plant community were identified
by model selection as key environmental characteristics explaining variation in fungal diversity
(Table 1). Relative water level and understory plant community evenness as well as canopy openness,
Diversity 2020, 12, 0324 6 of 17
understory plant richness, core type, and tree density were also included in the best model for fungal
richness (Table 1). Further investigation of the significant explanatory variables in the best fit models
revealed a positive relationship with several environmental factors for both diversity and richness.
For example, fungal diversity increased with increasing relative water level (F1,68 = 8.63, p = 0.004;

Diversity 2020, 12, 0324 4 of 17

2.2. Soil Collection

In June 2018, we collected soil microbiomes from 96 sites. We colocated our fungal community

sampling with 96 understory plant community plots established in 2009. These plots were evenly

distributed among islands (12 per island) and among experimental treatments, i.e., island core type

treatments (48 plots on peat core versus 48 plots on limestone core island) and tree planting density

treatments (24 plots per density). On each island, the three plots were randomly assigned to locations

within each tree density quadrant (e.g., Figure S2). For each sampling point, we used a soil corer to

aseptically collect approximately 50 mL of soil at 15–25 cm below the soil surface and approximately

0.25 m from the center of the 1 m2 plot (indicated by permanent PVC site marker). Soil samples were

placed in sterile conical tubes and transported on ice to University of Miami’s campus where they were

kept at −80 ◦C.

2.3. DNA Sequencing and Data Processing

We extracted DNA from 0.25 g of each soil sample using the E.Z.N.A. Soil DNA kit

(Omega, D5625-01, Norcross, GA, USA). We removed PCR inhibitors from DNA samples with

agarose gel electrophoresis and extracted genomic bands (i.e., >15 kilobases; E.Z.N.A Gel Extraction kit,

Omega, D2500-01, Norcross, GA, USA). We performed quality control by endpoint PCR of the fungal

ribosomal ITS1 region to confirm that enough PCR inhibitors were removed to obtain DNA of sufficient

quality for amplification. The ITS1 region was amplified via PCR using two-step dual-indexing [45],

and amplicon libraries were sequenced at the University of Minnesota Genomics Center (UMGC)

using the Illumina MiSeq platform (v3, 300 bp paired-end, San Diego, CA, USA). We also sequenced

three negative controls in which ultrapure water (G-Biosciences, St. Louis, MO, USA; 786–293) was

used in place of soil during extractions to confirm that samples were not contaminated during the

DNA extraction process. UMGC demultiplexed reads using bcl2fastq. We denoised reads, joined paired

ends, and grouped reads into exact sequence variants (ESVs; i.e., operational taxonomic units at

100% sequence similarity) with the QIIME2 pipeline (v2019.1, [46]). We rarefied sequences to 4000 reads

per sample to account for unequal sequencing depths and excluded samples with less than 4000 reads

(12 of 96 samples). We classified ESVs into fungal “species” if they shared at least 97% sequence

similarity to reference sequences in the UNITE database (version v7_01.12.2017, [47]). For the fungal

community at each site, richness was determined based on the number of rarefied ESVs, and Shannon

diversity was calculated using sequence reads of ESVs as estimates of abundance. Note that using

ESVs, compared to OTUs (Operational Taxonomic Units) with 97% sequence similarity, may result

in higher absolute fungal richness, but ESVs and 97% OTUs are often highly correlated, such that

comparisons among treatments are robust to sequence binning choices [48].

2.4. Data Analyses

To understand which abiotic and biotic factors explain fungal community diversity and richness,

we performed model selection with the abiotic and biotic factors in Table 1 as explanatory variables and

identified the best model based on AIC using dredge (R package MuMin; [49]). We also included tree

island identity as a random effect in the model selection to account for understory plant community

variation between islands. We confirmed that model variables were not highly correlated with each

other before performing model selection (all correlation coefficient <0.7; Figure 2). To investigate

the possible causes of variation in fungal community composition, we performed a distance-based

redundancy analysis (dbRDA, Bray-Curtis) with the same abiotic and biotic factors as explanatory

variables using capscale (R package vegan; [50]). Indicator taxa analysis was used to identify which

of the 500 most common fungal taxa were associated with differences in community composition

between treatments for significant categorical variables (e.g., limestone versus peat core). We calculated

‘indicator values’ of the fungal taxa based on the relative abundance and consistency in each treatment

and determined the significance of these values with permutation tests (R package indicspecies [51,52]),

Diversity 2020, 12, 0324 5 of 17

which were FDR (False Discovery Rate) corrected for multiple comparisons. We also conducted a

Mantel test between fungal and plant community distance matrices using mantel (R package vegan, [50])

to examine whether the variation in fungal community composition can be explained by variation in

understory plant community composition.

Table 1. Best models for the response of fungal diversity and richness to abiotic and biotic tree

island factors.

Island Factors Fungal Diversity Fungal Richness

Relative Water Level F1,68 = 8.63, p = 0.0045 F1,65 = 2.40, p = 0.1255

Island Core – F1,6 = 1.18, p = 0.3188

Canopy Openness – F1,65 = 9.41, p = 0.0031

Tree Density – F1,65 = 1.99, p = 0.1626

Understory Evenness F1,68 = 8.83, p = 0.0041 F1,65 = 7.66, p = 0.0073

Understory Richness – F1,65 = 1.78, p = 0.1864

Estimated Understory Biomass – –

NOTE: All environmental factors with statistical results listed in the table were part of the model selection’s best

fit model, and significant terms are bolded. Double dashes indicate those environmental variables that were not

included in the best model. ‘Understory’ metrics refer to characteristics of the understory plant community.

To investigate the distribution of functional guilds of fungi, we used the FUNGuild database

(version v1.1; [53]), which assigned each taxa identified at the species level via UNITE to a functional

guild (e.g., leaf saprotroph or fungal parasite) and a trophic mode (e.g., symbiotrophic, pathotrophic)

(Figure S3A,B). ESVs that were not identified to the species level were not included in this analysis.

We also filtered to include only taxa that had guild and trophic mode assignments with the confidence

rankings “Highly Probable” and identified the three most common guilds based on their normalized

ESV counts. For our analyses, we focused on functional guild assignments using model selection to

determine which biotic and abiotic variables explained variation in the number of different guilds

present at a site (guild richness) and the relative abundances of the top three most common functional

guilds. As above, our model selection used linear mixed effects models with abiotic and biotic factors

as explanatory variables and tree island identity as a random effect. All statistical analyses were

performed in R ver 3.6.0.

3. Results

3.1. Sequencing and ESV Taxonomic Assignments Statistics

After quality control filtering, 84 of the 96 samples had sufficient read counts to reach our

saturation cutoff (≥4000 reads) required to be included in analyses. Across these samples, there were

over 9000 ESVs, with an average of 243 ± 11 ESVs per sample. Using the UNITE database, we identified

approximately 35% of all ESVs and an average of 100 ± 5 taxa per sample (mean ± s.e) to at least

the order level. Approximately 20% of all ESVs were identified in UNITE to the species level,

allowing them to be included in our functional analyses using FUNGuild. Fusarium keratoplasticum was

most common across sites, occurring in 71 of the 84 sites, followed by Penicillium sp. (69 of 84 sites) and

Retroconis fusiformis (69 of 84 sites). An unidentified fungal taxa had the greatest number of reads on

average (1149) followed in abundance by an ESV from the order Tremellales (614) and Inocybe curvipes

(597). Demultiplexed sequence data, the ESV table, and the taxonomy table have been submitted to

NCBI (BioProject: PRJNA639837).

3.2. Fungal Community Diversity and Richness

The relative water level and evenness of the understory plant community were identified

by model selection as key environmental characteristics explaining variation in fungal diversity

(Table 1). Relative water level and understory plant community evenness as well as canopy openness,

Diversity 2020, 12, 0324 6 of 17

understory plant richness, core type, and tree density were also included in the best model for fungal

richness (Table 1). Further investigation of the significant explanatory variables in the best fit models

revealed a positive relationship with several environmental factors for both diversity and richness.

For example, fungal diversity increased with increasing relative water level (F1,68 = 8.63, p = 0.004;