Site description and sample collection

Soil samples were collected from two long-term warming soil experiments located at the Harvard Forest in Petersham, MA (42°30′30″N, 72°12′28″W), as described in [27]. The Prospect Hill Soil Warming Study was established in 1991 [34], and the Soil Warming x Nitrogen Addition (SWaN) Study was established in 2006 [24]. Warmed plots are heated continuously +5°C above ambient using buried resistance cables placed 10 cm below the soil surface and spaced 20 cm apart. Soil samples were collected in 2019 on July 15^th and October 19^th, from Prospect Hill and SWaN, which had been warmed for 28 and 13 years, respectively, at the time of sampling. The experiments are located adjacent to one another and have the same dominant plant overstory (Acer rubrum, Betula lenta, Betula papyrifera, Fagus grandifolia, Quercus velutina, Quercus rubra) and soil type (coarse-loamy incepitsols). The temperature ranges from a mean low of 3.3°C to mean high of 13.1°C [35]. The mean annual precipitation, including snow, is 1107 mm distributed evenly across seasons [35]. Both Prospect Hill and SWaN organic and mineral soils are acidic, with the organic horizon having pHs between 3.8 and 4.2 and the mineral soils having pHs between 3.9 and 4.4 [14, 36, 37]. The organic horizon in the control, unheated plots has a depth of approximately 5 cm, and the organic horizon in the heated +5°C plots has a depth of approximately 2–3 cm [14]. Duplicate cores were taken from each plot to 10 cm depth using a 5.7 cm diameter tulip bulb corer. Cores were separated into the organic and mineral horizons, duplicate soil cores were pooled by depth increment, roots and rocks were removed, and then the soil was sieved <2 mm. After sieving, samples were kept at ambient temperature, and within 4 hours of collection, samples were taken back to the lab for further analyses. The full sampling design was two sites (Prospect Hill, SWaN) x two treatments (control, heated) x two seasons (summer, fall) x two soil layers (organic horizon, mineral soil) x five replicate plots for a total of 78 samples (one sample from the SWaN experiment had a mineral horizon that was beyond the reach of the corer for the July sampling and this same plot was subsequently not sampled in October).

Ecosystem functionality measurements

To assess ecosystem functionality, we utilized five soil properties or functions that were measured as a part of Domeignoz-Horta et al. (2023), including microbial biomass carbon, respiration, and the potential activities of four enzymes: phenol oxidase and peroxidase, β-glucosidase (BG), and N-acetyl-glucosaminidase (NAG). Additionally, we measured total soil organic carbon and total nitrogen. Altogether, there were seven different soil functions or properties measured for this study.

To measure total soil organic C and N and soil water content, soils were weighed into pre-weighed aluminum tins and dried in a 65°C oven until they reached a constant mass the same day the samples were brought back to the lab. Constant mass was verified by weighing the soils multiple times throughout the drying process. The drying temperature was selected so soils could be utilized for a separate analysis that was performed in Domeignoz-Horta et al. (2023) but not included in this study. After the soils achieved a constant mass, the tins were weighed again. The dry soil weight was subtracted from the wet soil weight, then divided by the dry soil weight to calculate water content. For total C and N, the dried soils were ground to a fine powder using a mortar and pestle. The soil was then weighed and packaged in duplicate into tins, which were run on a Perkin Elmer 2400 Series II CHN Elemental Analyzer with acetanilide as a standard at the University of New Hampshire Water Quality Analysis Lab. Total C and total N or duplicates were averaged.

Within three days of soil sampling, microbial biomass carbon was measured. Soils were stored at 15°C in the intervening time between sample collection and MBC measurement. The three day storage was to allow for similar treatment as samples used in carbon use efficiency calculations for Domeignoz-Horta et al. (2023). Four replicate soil samples were each split into two subsamples, with one group serving as a control and one group which was fumigated with chloroform vapors under vacuum pressure for 24 hours. Dissolved organic carbon (DOC) was then extracted from both the unfumigated and fumigated samples using 15 mls of 0.05 M K₂SO₄ and quantified on a Shimadzu TOC analyzer. Microbial biomass carbon was determined by subtracting the DOC concentration in the unfumigated subsample from the fumigated subsample.

Soils were stored at room temperature (20°C) overnight before being aliquoted for soil respiration measurements. Soil respiration was measured on triplicate subsamples (0.15 or 0.3 g for the organic horizon or mineral soil, respectively) that were placed into Hungate tubes. Tubes were then sealed and incubated at 15°C for 24 hours. A 30 ml headspace sample was then taken and injected into an infrared gas analyzer (Quantek 906) to measure CO₂ concentrations.

Prior to extracellular enzyme activity assays, soils were stored for no more than 4 days at 15°C after soil sample collection. Potential extracellular enzyme activity was measured using fluorescent substrates. Soil slurries were prepared with 1.25 g wet weight soil and 175 mls of 50 mM pH 4.7 sodium acetate in a Waring blender. For the BG and NAG assays, 200 μls of soil slurry was pipetted into black 96 well plates, and for the oxidative enzyme assay, 500 μls of soil slurry was pipetted into deep well plates. Plates were then placed in a 15°C incubator for 25 minutes to allow for temperature acclimation. This temperature reflects the average air temperature between the summer sampling and the fall sampling. After temperature acclimation, either 50 μls of 4000 μM 4-methylumbelliferyl β-D-glucopyranoside, or 50 μls of 2000 μM 4-methylumbelliferyl N-acetyl-glucosaminidase were added to each well. For assessing phenol oxidase and peroxidase activity, 500 μls of 25 μM L-DOPA + 0.03% H₂O₂ were added to each well. Each plate contained a standard curve as well as a slurry-only control. All plates were read on a SpectraMax M2 plate reader. BG and NAG plates were measured at 360/450 nm excitation/emission wavelengths after substrate addition. Oxidative enzyme plates were incubated for 4 hours after substrate addition, and then 100 μls were removed, transferred to a clear 96 well plate, and read at 460 nm. Since phenol oxidase and peroxidase both act on L-DOPA and 0.03% H₂O₂, any enzyme activity measured using these substrates was called oxidative enzyme activity. All enzyme activity was normalized for each sample by the sample’s microbial biomass carbon.

Ecosystem multifunctionality calculation

Multifunctionality was calculated for the organic horizons and mineral soil in R [38] using the multifunc package [39]. We included microbial biomass carbon, soil respiration, total carbon, total nitrogen, N-acetyl glucosaminidase activity, β-glucosidase activity, and oxidative enzyme activity in the ecosystem multifunctionality index. These functions represent both process rates (respiration, potential enzyme activity) and nutrient pools (total C, total N, microbial biomass) [40]. The organic horizon and mineral soil layers were analyzed separately due to documented differences in soil parameters [14] and microbiome properties [11, 41]. A multifunctionality index was calculated for each sample by taking the average of the z-score transformed ecosystem functions. While there are different ways of assessing diversity-multifunctionality relationships, we selected the averaging approach for ease of interpretation. Individual ecosystem functions’ relationship with diversity was also investigated. For this approach, the raw untransformed function measurements were used.

DNA extraction and library preparation

Bacterial diversity was measured using 16S ribosomal RNA (rRNA) gene amplicon sequencing. After sampling, soils were stored for three days at 15°C to keep treatments consistent for an analysis in Domeignoz-Horta et al. (2023) that was not included in this study. After this three day storage period, soils were stored at -80°C until extraction. DNA was extracted from soils using the Qiagen Powersoil kit following the manufacturer’s protocol, and DNA concentration was measured using the Picogreen dsDNA kit (ThermoFisher Scientific).

We sequenced the 16S rRNA gene V4 region using the primers 515F (5’ - GTG YCA GCM GCC GCG GTA A- 3’) and 806R (5’ - GGA CTA CNV GGG TWT CTA AT - 3’) following the Earth Microbiome Project protocol [42], with minimal modifications. Forward primers contained the 5’ Illumina adapter, 12 basepair unique Golay barcode, forward primer pad and linker, and 515F. Reverse primers contained the 3’ reverse complement Illumina adapter, reverse primer pad and linker, and 806R. Templates were amplified in duplicate prior to sequencing using a 25 μl reaction containing 10 μM 2X Invitrogen Platinum Hot Start master mix, 0.5 μl 10 μM forward primer, 0.5 μl 10 μM reverse primer, 13 μl molecular grade H₂O, and 1 μl template. Samples were amplified in an Eppendorf Mastercycler Pro thermocycler with the program parameters of 94°C for 3 min, 35 cycles of 94°C for 45 s, 50°C for 60 s, 72°C for 90 s, and a final extension at 72°C for 10 min. Technical replicate amplicons were combined and visualized on a 1% agarose gel. We quantified the amplicons with the PicoGreen Assay for dsDNA (ThermoScientific) and pooled 300 ng of each sample. We used AMPure XP magnetic beads (Beckman Coulter) to clean the pooled library. Prior to sequencing, the quality of the amplicon library was checked using Qubit Fluorometer (Thermo Fisher Scientific Inc) and 2100 Bioanalyzer DNA 7500 assay (Agilent Technologies, Inc). The library quantification was done using NEBNext Library Quant Kit for Illumina (New England Biolabs). The amplicon library was spiked with Illumina PhiX and sequenced on Illumina MiSeq platform using the v2–300 cycle kit with 156 bp paired end chemistry (Illumina Inc). Custom Read1, Read2, and Index Sequencing primers were used for sequencing. Negative controls with no template were PCR amplified and sequenced as well. Amplicon library quality assessment and sequencing was performed at the Genomics Resource Laboratory (RRID:SCR_017907), Institute for Applied Life Sciences, University of Massachusetts Amherst, MA, USA.

We were also interested in measuring fungal community diversity using ITS amplicon sequencing. However, we were unable to successfully PCR amplify enough samples to have triplicates for each combination of season, soil layer, warming treatment, and warming duration. Thus, the focus of this study considered only on the bacterial community.

Sequencing processing and analysis

Raw FastQ files were demultiplexed and primers trimmed using cutadapt [43]. Bacterial 16S rRNA gene sequences were processed with the DADA2 pipeline (v.1.18.0) [44], which generates amplicon sequence variants (ASV), in R. Forward reads were trimmed to 150 bp, and reverse reads were trimmed to 140 bp. Any reads with expected error higher than 2 or with any ambiguous nucleotides were discarded, and all PhiX sequences were also removed. Error rates for the forward and reverse reads were estimated separately using the function ‘learnErrors’. ASVs were called for the forward and reverse reads using their respective error models. Denoised forward and reverse ASVs were merged using the function ‘mergePairs’ with default parameters. Prior to assigning taxonomy, any chimera sequences were identified de novo and removed. Taxonomy was assigned using SILVA (v 138.1) [45]. All sequences that were classified at the family level as “mitochondria” or at the order level as “chloroplast” were filtered out from the final ASV table. To account for differences in sequencing depth between samples, raw ASV counts were normalized with library size estimation factors calculated in DESeq2 [46, 47]. Any ASVs that had 1 or 0 reads total across all samples after normalization were discarded.

Our final data set comprised 69 out of 79 possible samples due to difficulty with successfully PCR amplifying samples during library preparation. In the final data set, each combination of season, soil layer, warming treatment, and warming duration had at least three replicates out of the original five replicates, except the October organic horizon control treatment from the 13-year experiment, which had only two out of five replicates. At the end of sequence processing, the data set had 4,345,430 reads, with an average depth of 62,977 reads per sample, and 8,754 total ASVs. The no template negative controls had fewer than 500 reads. After correcting with a sample specific library size estimate calculated in DESeq2 [46], the total read count was 4,159,442, with an average read count of 60,281 per sample. Rarefaction curve analyses indicated that for the samples that were successfully sequenced, the sequencing efforts were adequate.

Since a variety of diversity metrics have been correlated with EMF, we calculated Simpson diversity, Shannon’s H, Chao1 estimated richness, community structure (PCoA axes), and evenness (Pielou’s J). Diversity metrics were calculated using the packages vegan [48] and phyloseq [49]. The distance matrix for the principal coordinate analysis was calculated using Bray-Curtis dissimilarity using the function ‘vegdist’ in the package vegan.

Statistical analysis

All statistical analyses were done in R [38]. Prior to any statistical analysis, we evaluated whether warming treatment and warming duration could be combined into one variable, where a single “control” treatment (or “0 year warmed”) would derive from combined Prospect Hill and SWaN plot control samples, Prospect Hill warmed samples would be “28 year warmed”, and SWaN warmed samples would be “13 year warmed”. Both EMF and bacterial diversity metrics were tested for normality using a Shapiro-Wilk test. After checking for normality, we tested for significant differences between the control plot samples from Prospect Hill and SWaN using either a Wilcoxon rank sum test if the data were not normally distributed (as was the case for both the organic horizon and mineral soils’ EMF), or a t-test if the data were normally distributed (as was the case for the organic horizon and mineral soils’ bacterial diversity metrics). We found no significant differences between the control plots’ EMF in Prospect Hill and SWaN. However, we did find significant differences in bacterial diversity between the control plots from the two experiments (p = 0.0260). As a result, warming treatment and warming duration were combined into a single variable when EMF was the response variable, but not when bacterial diversity was the response variable.

We also investigated whether percent water content correlated with ecosystem multifunctionality or any of the bacterial diversity metrics in the organic horizon or mineral soils. Prior to correlation testing, ecosystem multifunctionality, Shannon H, Chao1 estimated richness, and Pielou J were tested for normality using a Shapiro-Wilk test. Of these four variables, the organic horizon bacterial diversity metrics and mineral soil Shannon’s H were normally distributed. All other variables were not normally distributed. For variables that were normally distributed, we used a Pearson correlation test. For variables that were not normally distributed, we used a Spearman correlation test.

We built a set of general linear models using raw or log-transformed data to investigate the effects of warming, soil layer, season, and/or site on four response variables: individual ecosystem functions, nutrient pools, EMF, or bacterial diversity. When normality of model residuals was not satisfied or the response variable was not normally distributed, we constructed generalized linear models with non-Gaussian distributions. We considered both additive and interactive effects between warming, soil layer, season, and/or site.

A set of candidate generalized linear models were constructed with either EMF or bacterial diversity as the response variable. First, models were built using a Gaussian distribution and model residuals were tested for normality, except for models where Chao1 estimated richness was the response variable. When Chao1 estimated richness was the response variable, models utilized a Poisson or negative binomial distribution, which are both appropriate for count data. Poisson models for Chao1 estimated richness were checked for overdispersion by looking at the ratio between residual deviance and degrees of freedom, and if overdispersion was observed (ratio > 1.1), a negative binomial distribution was used with the R package MASS [50]. For the mineral soils, any model where EMF was the response variable had a log-transformation applied to EMF to make the residuals normally distributed. The results from the log-transformed models are reported without back-transformation. We found that the different diversity metrics all performed similarly based on their ΔAICc, so we elected to utilized Shannon’s H, Pielou’s J, and Chao1 estimated richness as predictors since taxon richness and community evenness have both been found to be drivers of ecosystem function [51, 52].

We also tested whether warming treatment, warming duration, soil layer, and/or season affected the relationship between the individual ecosystem functions or nutrient pools and bacterial diversity by constructing linear models. Model residuals were tested for normality using a Shapiro-Wilk test. If residuals were not normally distributed, a log-transformation was applied. If the model residuals were still not normally distributed after a log-transformation, we constructed a set of generalized linear models using a gamma or inverse Gaussian distribution. Residuals for these models were validated using the DHARMa package [53]. Results for models where a log-transformation was applied or for models that did not use a normal distribution are reported without back-transformation.

Finally, we then investigated how warming treatment, warming duration, soil layer, and/or season affected the relationship between EMF and bacterial diversity. For all of these analyses, we constructed a set of candidate linear models. In these sets of candidate models, we considered all predictors and possible interactions between warming treatment and diversity, warming treatment and horizon, warming treatment and season, or warming treatment and warming duration. All model residuals were checked for normality, and if necessary a log-transformation was applied to the response variable. Estimates and errors from the log-transformed models are reported without back-transformation.

For all of the candidate sets of models, model fit was assessed using AICc in the package [54]. If models were within 2 ΔAICc units, we considered the models equally supported, and used a likelihood ratio test for the nested models to select a model. If the models with similar ΔAICc were not nested, we selected the model that explained more variance in the data. Model residual deviance was also compared to model null deviance to examine model fit. When significant interactions were found between warming treatment, warming duration, and/or season, a Tukey’s HSD test was utilized to further investigate the interaction. Otherwise, p-values had a Benjamini-Hochberg correction applied to them to account for multiple comparisons. We utilized a p-value cutoff of 0.05 for statistical significance. If the p-values were between 0.05 and 0.1, we report it as a trend. All estimates are reported with a 95% confidence interval.

After analyzing the bacterial diversity and finding a significant difference in bacterial evenness between the warmed and control treatments in the summer organic horizon samples, we conducted a post-hoc analysis to see whether any of the dominant ASVs were differentially abundant between control and heated plots. We used three different methods: Analysis of Community Microbiomes with Bias Correction in the R package ANCOM-BC (v.2.0.3) [55], log2 fold change using the package DESeq2 [46], and a χ² test between ASV counts in the heated and control plots. Dominance was determined based on a relative abundance threshold of ≥ 0.001 within each sample. Of the 8,629 ASVs in the organic horizon summer dataset, 256 ASVs were identified as dominant, which accounted for 2.967% of the total ASVs. For all three differential abundance methods, p-values were adjusted for multiple comparisons using a Benjamini-Hochberg correction.

Bacterial diversity

The effect of warming on bacterial evenness varied by season in the organic horizon, where the best fitting model based on ΔAICc was the interaction model with warming and season (Table 1). Warming had a negative impact on Pielou’s J in the summer but no effect in the fall (Tukey HSD summer: -0.024 ± 0.011, p = 0.026; Tukey HSD fall: p = 0.801). This pattern was not observed in the mineral soil, where warming did not affect bacterial evenness. None of the bacterial diversity metrics in the organic horizon had a significant correlation with soil water content (S1 Table). There was also a season-dependent warming trend on Shannon diversity, although this effect was not significant (p = 0.072). Since in the organic horizon in the summer warming decreased bacterial evenness but did not affect Chao1 richness, we carried out a post-hoc differential abundance analysis on dominant ASVs to determine if changes in the abundance of dominant ASVs could explain the decrease in evenness. None of the dominant ASVs were differentially abundant between the warmed and control plots.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The effects of warming treatment, season, or site on diversity metrics in the organic horizon and mineral soil.

https://doi.org/10.1371/journal.pone.0311364.t001

Chao1 estimated richness significantly differed between the two different sites, Prospect Hill (28 year warmed) and SWaN (13 year warmed) in the organic horizon. The older experimental site had around 90% of the number of taxa compared to the younger site (Table 1). Soil water content did not have a significant correlation with any of the bacterial diversity metrics in the mineral soils (S1 Table). Unheated control plots from the 13-year warmed and 28-year warmed experiments had significantly different bacterial diversity metrics and so were not combined when bacterial diversity was the response variable.

Ecosystem multifunctionality

Ecosystem multifunctionality was affected only by season and not warming treatment or warming duration. In the organic horizon, EMF was significantly higher in fall compared to summer (0.089 ± 0.064, p = 0.022, Fig 1A). Similarly, in the mineral soil, EMF trended higher in fall compared to summer, although this difference was not statistically significant (p = 0.090, Fig 1B). Neither EMF in the organic horizon nor EMF in the mineral soils had a significant correlation with soil water content (S1 Table). Due to known differences between the organic horizon and mineral soil [14], ecosystem multifunctionality was calculated for the soils separately.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Ecosystem multifunctionality across warming treatment and season.

Ecosystem multifunctionality (EMF) was calculated by averaging z-score standardized ecosystem function measurements for each sample, and EMF was calculated separately for the organic horizon (A) and mineral soils (B). In the mineral soils (B), EMF was log-transformed for general linear model construction, but original non-log-transformed values are displayed.

https://doi.org/10.1371/journal.pone.0311364.g001

For both the organic horizon and mineral soils, neither 13 years nor 28 years of warming significantly altered EMF. In mineral soils, EMF tended to increase as warming duration increased, but these differences were not significant. In the organic horizon soils, we did not observe any trend in EMF with longer duration of heating.

Ecosystem multifunctionality-bacterial diversity relationship

In the organic horizon, season had a significant effect on the relationship between EMF and bacterial Chao1 estimated richness, where fall had significantly higher intercept for EMF than summer, irrespective of warming treatment or duration (Fig 2A, 0.108 ± 0.074, p = 0.020). We observed a similar, seasonal trend in the relationship between EMF and Shannon’s diversity (S2A Fig, 0.105 ± 0.075, p = 0.052) and Pielou’s evenness (S2C Fig, 0.104 ± 0.074, p = 0.055). In the organic horizon soils, ecosystem multifunctionality did not correlate with bacterial Chao1 estimated richness, Shannon diversity, or Pielou evenness. The relationship between EMF and Chao1 estimated richness had a progressively lower intercept as warming duration increased. However, neither the 13-year warmed nor the 28-year warmed plots showed a relationship between EMF and Chao1 estimated richness that was significantly different from the control plots.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The relationship between ecosystem multifunctionality and bacterial Chao1 estimated richness.

Ecosystem multifunctionality (EMF) was calculated by averaging z-score standardized function measurements for each sample. Shaded regions represent a 95% confidence interval. Mineral soil EMF was log-transformed and is displayed without back-transformation.

https://doi.org/10.1371/journal.pone.0311364.g002

In the mineral horizon, Shannon diversity (S2B Fig), Chao1 estimated richness (Fig 2B), or Pielou evenness (S2D Fig) were not correlated with EMF, and the intercept for the relationship between EMF and Chao1 estimated richness did not change as warming duration increased. There also was no effect of season on the intercept for EMF for all three of the diversity metrics in the mineral soil.

Individual ecosystem functions-bacterial diversity relationship

In the organic horizon, no individual ecosystem function or nutrient pool had a significant relationship with any diversity metric. MBC and oxidative enzyme activity both varied significantly by season, with MBC higher in fall compared to summer and oxidative enzyme activity higher in fall compared to summer for both the Shannon diversity and Chao1 estimated richness models (Table 2 and S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Single ecosystem function-diversity relationships in the organic horizon.

https://doi.org/10.1371/journal.pone.0311364.t002

In the mineral soil, β-glucosidase activity was the only ecosystem function that had a relationship with any bacterial diversity metric, where Shannon’s H had a negative relationship with β-glucosidase activity (Table 3 and S3 Table). Similar to the organic horizon, significant seasonal differences were observed for MBC, oxidative enzyme activity, and β-glucosidase activity. Both MBC and oxidative enzyme activity were higher in fall compared to summer, whereas the opposite was observed for β-glucosidase activity, irrespective of the diversity metric (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Single ecosystem function-diversity relationships in the mineral soil.

https://doi.org/10.1371/journal.pone.0311364.t003