High Arsenic Levels Increase Activity Rather than Diversity or Abundance of Arsenic Metabolism Genes in Paddy Soils

ABSTRACT

Arsenic (As) metabolism genes are generally present in soils, but their diversity, relative abundance, and transcriptional activity in response to different As concentrations remain unclear, limiting our understanding of the microbial activities that control the fate of an important environmental pollutant. To address this issue, we applied metagenomics and metatranscriptomics to paddy soils showing a gradient of As concentrations to investigate As resistance genes (ars) including arsR, acr3, arsB, arsC, arsM, arsI, arsP, and arsH as well as energy-generating As respiratory oxidation (aioA) and reduction (arrA) genes. Somewhat unexpectedly, the relative DNA abundances and diversities of ars, aioA, and arrA genes were not significantly different between low and high (∼10 versus ∼100 mg kg⁻¹) As soils. Compared to available metagenomes from other soils, geographic distance rather than As levels drove the different compositions of microbial communities. Arsenic significantly increased ars gene abundance only when its concentration was higher than 410 mg kg⁻¹. In contrast, metatranscriptomics revealed that relative to low-As soils, high-As soils showed a significant increase in transcription of ars and aioA genes, which are induced by arsenite, the dominant As species in paddy soils, but not arrA genes, which are induced by arsenate. These patterns appeared to be community wide as opposed to taxon specific. Collectively, our findings advance understanding of how microbes respond to high As levels and the diversity of As metabolism genes in paddy soils and indicated that future studies of As metabolism in soil or other environments should include the function (transcriptome) level.

IMPORTANCE Arsenic (As) is a toxic metalloid pervasively present in the environment. Microorganisms have evolved the capacity to metabolize As, and As metabolism genes are ubiquitously present in the environment even in the absence of high concentrations of As. However, these previous studies were carried out at the DNA level; thus, the activity of the As metabolism genes detected remains essentially speculative. Here, we show that the high As levels in paddy soils increased the transcriptional activity rather than the relative DNA abundance and diversity of As metabolism genes. These findings advance our understanding of how microbes respond to and cope with high As levels and have implications for better monitoring and managing an important toxic metalloid in agricultural soils and possibly other ecosystems.

INTRODUCTION

Arsenic (As) is a toxic metalloid widely distributed in Earth’s crust. Relatively large amounts of As are introduced into the environment from natural sources, such as volcanic eruptions, weathering of rocks, and geothermal activities, as well as anthropogenic activities such as mining, pigment production, and application of As-based pesticides in agriculture (1, 2). Because of its pervasive presence in the environment, microorganisms have evolved the capacity to metabolize As (3), and this capacity is hypothesized to be an ancient mechanism that emerged at least 2.72 billion years ago (4, 5).

Microbial-mediated As biotransformation includes As detoxification to mitigate toxicity and As respiration to generate energy. Several mechanisms have evolved to detoxify both inorganic and organic As. The inorganic arsenate [As(V)] compounds in cytoplasm can be reduced by a reductase (ArsC) (6), followed by arsenite [As(III)] efflux outside the cell, which is catalyzed by two evolutionarily unrelated As(III) efflux permeases, i.e., ArsB and Acr3 (7). Additionally, As(III) can be methylated by an As(III) S-adenosylmethionine (SAM) methyltransferase (ArsM) to the more toxic organic methylarsenite [MMAs(III)] (8), followed by the organoarsenical detoxification pathways, which recently have been shown to be present in various microbial genomes. The latter detoxification pathways include MMAs(III)’s oxidization to methylarsenate [MMAs(V)], which is catalyzed by the MMAs(III)-specific oxidase ArsH (9) and actively transported outside the cell by the MMAs(III) efflux permease ArsP (10), or MMAs(III)’s demethylation to the less toxic As(III) by the ArsI lyase, which cleaves the carbon-As bond in MMAs(III) (11). These As resistance genes are usually organized in an ars operon, which is controlled by one of three known ArsR transcriptional repressors (ArsR1, ArsR2, and ArsR3) regulated selectively by As(III) and one (ArsR4) by MMAs(III), depending on the exact version of the operon (12, 13). The As respiratory pathways include oxidation of As(III) either to detoxify it to less toxic As(V) or couple it to ATP production (14) and dissimilatory reduction that couples As(V) reduction to anaerobic heterotrophic growth (15). As(III) oxidation is catalyzed by As(III) oxidase (Aio), whose large subunit (AioA) has been well characterized in both bacteria and archaea (16). On the other hand, As(V) reduction is catalyzed by respiratory As(V) reductase (Arr), whose large catalytic subunit (ArrA) serves as a reliable marker for the process (17).

Due to the anaerobic conditions typically prevailing in paddy soils, As(III) is commonly the predominant form of As, which is more mobile and bioavailable to microbes for biotransformation (18–20). Indeed, microbial genes involved in As resistance, such as arsC and arsM, and genes involved in As respiration, such as aioA and arrA, have been detected to be widely present in the paddy soils with As contamination (21), and their activity to biotransform As has also been experimentally demonstrated (22–24). Recent studies of soils with low As levels, i.e., typically less than 15 mg kg⁻¹ (or ppm), also showed relative high abundance and diversity of microbial As metabolism genes, most notably arsB, acr3, arsH, arsR, arsC, arsM, aioA, and arrA genes in paddy soils (25) and arsC, arsM, aioA, and arrA genes in estuarine sediments across southeastern China (26). These results reveal high prevalence of As metabolism microbes in the environment even in the absence of high concentrations of As. However, these previous studies were carried out at the DNA level and, thus, the expression of the As metabolism genes detected remains speculative. Further, these DNA-based results showing prevalent As metabolism microbes in both high- and low-As-level environments are somewhat surprising because gene functions and species selected by a specific factor tend to be more abundant in ecosystems where the factor (selective pressure) is stronger, as this has been shown for various soil factors, such as pH, total organic carbon (TOC), total nitrogen (TN), phosphorus (TP) (27), and organic pollutants (28).

Here, we applied both metagenomics and metatranscriptomics to investigate the resident (at the DNA level) and active (at the RNA level) As-associated microbial genes in paddy soils of different As concentrations, aiming to address the following questions. (i) Do As metabolism genes in high-As-level paddy soils show higher relative abundances and/or transcriptional activities than their counterparts in low-As-level paddy soils to cope with the higher As level present? (ii) Which are the key As metabolism genes and pathways coping with high As levels? (iii) What factors other than As levels are driving microbial communities compositional and gene transcriptional differences in soils with As contamination?

RESULTS

Soil physicochemical characteristics.

A total of 18 samples, three biological replicates from the same field for each site, were collected from six paddy soils in Hunan, China, with different As concentrations (see Fig. S1 and Table S1 in the supplemental material). As concentrations in these paddy soils varied significantly (analysis of variance [ANOVA], P < 0.01) between high (67.3 to 104.0 mg kg⁻¹) and low (2.5 to 10.8 mg kg⁻¹) As levels. The high As concentrations in the paddy soils likely resulted from the historic mine activities occurring in the same area. The total concentrations of nitrogen (TN) in high As paddy soils were also significantly higher (ANOVA, P < 0.01) than those with low As (1.64 to 1.77 g kg⁻¹ versus 1.17 to 1.24 g kg⁻¹). No significant difference was found in the other soil physiochemical characteristics, including pH, oxidation-reduction potential (Eh), total concentration of carbon (TC), phosphorus (TP), sulfur (TS), and concentrations of heavy metals, including Pb, Cd, Cr, Sb, and Fe. Even though the Eh in sample CZ_AsHig and Pb and Cd concentrations in sample YCP_AsLow were substantially different from those for the other samples, these differences in soil properties did not reach statistical significance between high- and low-As paddy soils, presumably due to small sample size. Details of the soil properties are summarized in Table S2.

Broad microbial community diversity indices.

Six shotgun metagenomes and six shotgun metatranscriptomes were obtained after the triplicate DNA and RNA samples from each site were pooled. For metagenomes (Table S3), 12 to 15 Gbp (13 Gbp, on average) of shotgun metagenomic data and an average read length of ∼115 bp were acquired after trimming for each pooled DNA sample. On average, 6.6 Gbp (ranged from 3.9 to 8.3 Gbp) of shotgun metatranscriptomic data and average read length of ∼129 bp were obtained after trimming and removal of rRNA reads for each pooled RNA sample (Table S4). The coverage achieved by sequencing based on the read redundancy value estimated by Nonpareil was around 22 to 27%, except for sample SKS_AsHig, which had a lower alpha-diversity (measured by Nonpareil) than the others (Nonpareil diversity [Nd] 22.6 versus 23.6 to 23.8; note that Nd is in log scale) and, thus, higher coverage (42%). No significant difference (P > 0.05) was found in the alpha-diversity between high- and low-As paddy soils (average of 23.3 versus 23.7). The average genome size of microbial community was 5.7 Mbp and 5.6 Mbp in high- and low-As paddy soils, respectively. Considering the difference of average genome size of microbial community between these paddy soils was no more than 2-fold, estimation of As metabolism genes abundance was essentially the same between the RPKM (the number of reads mapping on reference sequences/gene length in kb per one million reads analyzed) and RPKG (RPKM/average genome size) metrics.

Paddy soil community taxonomic composition.

16S rRNA gene (16S) fragments extracted from the metagenomes were about 0.03 to 0.04% of the total reads (Table S3) and around 65 to 70% of them were classifiable at the phylum level. Bacteria were predominant in these paddy soils, accounting for 61 to 69%, while Archaea accounted for only 0.5 to 2.4%; the remaining reads were unclassified against the SILVA database (29). The dominant bacteria comprised of phyla Proteobacteria, Acidobacteria, Chloroflexi, Nitrospirae, Verrucomicrobia, and Bacteroidetes, representing 50 to 59% of the total 16S-carrying reads. The archaeal community was mostly represented by the Crenarchaeota phylum at 0.05 to 0.78% of the total, followed by Euryarchaeota at 0.36 to 0.77% (Fig. 1A). There were overall no significant relationships between the level of As and community composition, assessed by Bray-Curtis similarity in 16S rRNA gene-based operational taxonomic units (OTUs) (nonmetrical multidimensional scaling [NMDS] plot; Fig. S2). DESeq2 analysis (Fig. 1B) revealed statistically significantly lower abundance of members of Euryarchaeota, i.e., Methanomicrobia and Thermoplasmata (0.15% versus 0.85% and 0.02% versus 0.07%, respectively; P < 0.05) in high- versus low-As paddy soils. In contrast, the abundance of Chloroflexi, mostly comprised of Ellin6529 (1.44% versus 0.52%), S085 (0.28% versus 0.09%), Thermomicrobia (0.06% versus 0.01%), and Chlorobi OPB56 (0.16% versus 0.05%), GAL15 (0.21% versus 0.01%), and OP11-WCHB1-64 (0.07% versus 0.01%), was significantly (P < 0.05) higher in high- versus low-As paddy soils. We also attempted to recover metagenome-assembled genomes (MAGs; see details in the supplemental material), which yielded only five MAGs of high quality (completeness of >80% and contamination of <5%), presumably due to the high complexity of the data sets. These MAGs were assignable to the taxa Deltaproteobacteria, Thermomicrobia, Clostridia, Acidobacteriaceae, and Solibacteres, which is consistent with the dominant microbes in these sites, as revealed by the read-based findings reported above.

FIG 1.

Comparison of microbial community composition in high- versus low-As sites. (A) Phylum-level community composition and abundance of microbes in paddy soils based on taxonomic classification of identified 16S rRNA gene sequence fragments. The relative abundance of each taxon was normalized by the total number of 16S rRNA gene sequence fragments (both assigned and unassigned) obtained in each corresponding metagenome. Only the top 20 most abundant phyla are shown. (B) Heatmap of microbes at the class level showing statistically significant differences in abundance between high (red) and low (blue) As paddy soils (negative binomial test, adjusted P < 0.05). Log₂ normalized counts were used to calculate the abundances of different microbes at the class level.

Paddy soil community functional composition.

Protein-coding fragment sequences were predicted for about 90 to 92% and 44 to 59% of the trimmed reads in metagenomics and metatranscriptomics, respectively (Tables S3 and S4). The high-As microbial communities clustered separately from their low-As counterparts based on functional gene content (summarized as annotation counts from SEED subsystems), although they were also quite different from each other and, thus, did not cluster tightly together (Fig. 2A). It was not unexpected to observe clustering at the functional but not taxonomic composition of the sampled communities, considering that functional gene content has been shown to be more strongly and faster influenced by environmental conditions than phylogenetic beta-diversity (30). DESeq2 analysis revealed no statistically significant changes in the relative abundance of genes predicted from metagenomics of different As levels. In contrast, genes associated with As resistance were significantly (adjusted P < 0.05) more expressed based on metatranscriptomes by a log₂ fold change of 1 to 2.3 in high- versus low-As paddy soils (Fig. 2B and C). Other gene functions, including phage tail fiber, phage photosynthesis, phage capsid, phage cyanophage, and CBSS-316279, were also found to be differentially expressed between high and low As (Fig. 2B and C). This could be related to “hitchhiking” effects (31, 32), functions not assessed here, and/or experimental noise, and were not examined further here.

FIG 2.

Effect of high As levels on the functional composition of microbial communities. (A) NMDS plot of the Bray-Curtis metric based on the counts of SEED subsystems (level 3) annotated from metagenomes in high (red) and low (blue) As paddy soils. (B) Heatmap of SEED subsystems (level 1) annotated from metatranscriptomes showing statistically significant differences in abundance between high (red) and low (blue) As paddy soils (negative binomial test, adjusted P < 0.05). Log₂ normalized counts were used to calculate the abundances of different SEED subsystems (level 1) annotated from metatranscriptomes. (C) Log₂ fold change of SEED subsystems (level 1) annotated from metatranscriptomes that were significantly differentially abundant between high- and low-As paddy soils (negative binomial test, adjusted P < 0.05).

Relative DNA abundance and expression of As resistance and respiratory genes.

Overall, the relative abundances of ars genes, i.e., the cumulative abundance of arsR (arsR1, arsR2, arsR3, and arsR4), acr3, arsB, arsC (arsC1, arsC2, and acr2), and arsM, arsI, arsP, and arsH genes, were comparable between high- and low-As paddy soils as revealed by metagenomes (1.2 × 10⁻² ± 9.7 × 10⁻³ versus 6.1 × 10⁻³ ± 5.3 × 10⁻⁴ RPKM; Fig. 3A). An outlier to this result was sample SKS_AsHig, which had three times higher relative abundance of ars genes than other samples (Fig. S3). To robustly assess transcriptome activity while accounting for DNA abundance, the transcriptional level of ars genes were estimated by normalizing metatranscriptomic counts by the relative abundance of the same ars genes identified in metagenomes. More than 80% of the As metabolism genes detected in metatranscriptomes were identified in the metagenomes, representing the majority of the As metabolism genes in the samples. The normalized transcriptional expression levels of ars genes were significantly higher (P < 0.05) in high- versus low-As paddy soils (2,074 ± 701 versus 710 ± 337 RPKM; Fig. 3A). Overall, the 13 ars genes responsible for As transcriptional repressor (arsR), As(III) efflux (acr3 and arsB), As(V) reduction (arsC including arsC1, arsC2 and acr2), As(III) methylation (arsM), MMAs(III) demethylation (arsI), MMAs(III) oxidation (arsH), and MMAs(III) efflux (arsP) all consistently showed higher transcriptional activities in high- versus low-As paddy soils (Fig. 3B). However, for sample HP_AsLow, arsH, arsR, and arsP showed a slightly higher transcriptional activity than one of the samples from high-As-level sites (CZ_AsHig or SKS_AsHig), which could be attributed to high soil heterogeneity or biases during library preparation and/or sequencing.

FIG 3.

Relative abundance and transcription of ars, aioA, and arrA genes in high- and low-As paddy soils. (A) The relative abundance and transcription of ars genes is the cumulative of arsR (arsR1, arsR2, arsR3, arsR4), acr3, arsB, arsC (arsC1, arsC2, and acr2), arsM, arsI, arsP, and arsH genes. Relative abundance of As metabolism genes in metagenomes was normalized by the statistic RPKG (reads per kb per genome equivalent; see Materials and Methods for more details). The As metabolism gene transcriptional activity was evaluated by the relative abundance of As metabolism genes in metatranscriptomics (RPKM) normalized by the relative abundance of the same gene identified in metagenomics (RPKG). Horizontal lines represent the median value; the asterisk indicates statistically different means at P < 0.05 based on one-way ANOVA. (B) Transcriptional activity of the As metabolism genes and the cellular location of their encoded proteins. Transcriptional activities (the same values as for panel A) of ars (acr3, arsB, arsC, arsM, arsI, arsH, arsP, and arsR), aioA, and arrA genes in high (red) and low (blue) As paddy soils visualized in a heatmap that was produced using the pheatmap package in R 3.5.1. The color ranges from red to blue indicate transcriptional activity from high to low.

For aioA genes (Fig. 3A), which encode respiratory As(III) oxidase in the periplasm, a higher relative DNA abundance was revealed, although it was not statistically significant (P = 0.114), in high- versus low-As paddy soils (1.0 × 10⁻⁴ ± 6.2 × 10⁻⁵ versus 2.9 × 10⁻⁵ ± 3.2 × 10⁻⁷ PRKG, respectively). The transcriptional activity of aioA genes was strongly significantly (P < 0.05) increased in high- versus low-As paddy soils (117 ± 39 versus 34 ± 31 PRKM, respectively). Further, there was no significantly higher abundance (P = 0.598) or transcriptional activity of arrA genes (P = 0.33), which are responsible for respiratory As(V) reduction, between high- and low-As paddy soils (Fig. 3A). The relative abundances of arrA genes in metagenomes were comparable between high- and low-As paddy soils (1.0 × 10⁻⁴ ± 5.8 × 10⁻⁵ versus 7.4 × 10⁻⁵ ± 5.9 × 10⁻⁵ RPKG, respectively), while their transcriptional activity was slightly higher, but not significant (P = 0.33), in paddy soils with high versus low As levels (39 ± 27 versus 19 ± 16 RPKM, respectively).

Responses to high As level in paddy soils are community wide rather than taxon specific.

To evaluate which clades expressed higher As genes in high As soils and, thus, assess whether the transcriptomic shifts of As metabolism genes described above were due to systematic community-wide response as opposed to a few taxa, the Ars-, AioA-, and ArrA-carrying reads in metagenomes (DNA reads) and metatranscriptomes (RNA reads) were placed on the their representative phylogenetic clade (typically >95% nucleotide identity within versus <95% identity between clades), and the ratio of RNA/DNA reads assigned at the same reference clade (or gene-based OTU) was compared between samples (Fig. 4 and Fig. S4 to S12). Overall, in high-As soils, more than 60% (ranging between 63% and 98%; Table S6) of the total clades recruited reads from metatranscriptomes for each As metabolism gene, except for arsB, suggesting that the high As levels mostly induced a community-wide response. Moreover, around 54 to 78% of the total clades for Ars, AioA, and ArrA proteins showed a higher ratio of RNA/DNA reads in high- than low-As paddy soils (Table S6), revealing that the As metabolism genes were expressed more strongly in high- than low-As samples by many distinct clades (taxa). For arsB, due to relatively lower abundance of this gene compared to the others [e.g., 75 to 375 times lower than acr3, which is also responsible for As(III) efflux], only 17% of the total clades recruited arsB-carrying reads from high-As paddy soils and no clade recruited reads from the low-As paddy soils.

FIG 4.

Community-wide shifts as an effect of high As levels. Phylogenetic placement of arsC-carrying reads identified in metagenomes (DNA reads, in blue) and metatranscriptomes (RNA reads, in red) from high (A) and low As paddy soils (B). For this analysis, the three metagenomes and metatranscriptomes from the same As level (i.e., high or low) sites were combined. The pie size indicates the DNA abundance, and the ratio indicates transcriptional activity for each clade (i.e., fraction of RNA versus DNA reads assigned to the class with the same thresholds for a match; see Materials and Methods for further details). The clades are colored on the outside based on the ArsC type, i.e., ArsC1 (blue), ArsC2 (red), and ACR2 (green). Phylogenetic placement of other As metabolism genes carrying reads are provided in the supplemental material. Note that several clades in the high As sample recruited more RNA than DNA reads (red color dominates), whereas the opposite pattern was observed for most clades in the low-As data sets.

Environmental factors affecting the composition of microbial communities in metagenomes.

To investigate the differences between microbial communities in high- versus low-As soils, five previously characterized samples with high As (total As concentration ranged from 34 to 821 mg kg⁻¹) affected by a mine field in Hunan, China (33), and eight samples with low As (total As concentration ranged from 2 to 16 mg kg⁻¹) in Hunan, Zhejiang, Jiangxi, and Guangdong, China (25), were included in comparisons with the samples determined by our study. Because these metagenomes had more than 2-fold variation in sequencing effort applied, we subsampled the metagenomes to the same level (2 Gbp) in order to avoid false-positive signal in detecting features as differentially abundant due to differences in coverage alone (34). Principal coordinate analysis (PCoA)-based analysis of Mash distances of whole metagenomes showed that samples were mostly separated based on their geographic distances, e.g., the most distinct metagenomes were those that originated form the most distant sites (Fig. 5A). The Adonis dissimilarity test corroborated that geographic distribution patterns imposed significant (P < 0.01) changes in microbial community compositions. Notably, As level did not cause any significant changes (P > 0.05) of the whole microbial communities based on metagenomic (DNA) abundances. For the As-associated microbial communities (Fig. S13), PCoA-based analysis of Mash distances of all As gene-carrying reads recovered from metagenomes revealed a weaker effect of geographic distance in shaping diversity patterns compared to the whole microbial communities. In contrast, As level showed a relatively higher effect, especially on the As-associated microbial communities in the mine field, which is characterized by higher As level, ranging from 34.1 to 821.2 mg kg ⁻¹. However, when the analysis was restricted to the paddy soils characterized by our study, no clear significant effect of high- versus low-As level (∼10 versus ∼100 mg kg⁻¹) was observed, which was consistent with the NMDS plot of Bray-Curtis metric based on the counts of As genes annotated from metagenomes (Fig. S14).

FIG 5.

Physicochemical factors affecting whole microbial community composition. (A) PCoA plot of site-to-site similarity matrix based on Mash distances of whole metagenomes from this study and 13 selected metagenomes from previous studies. Symbols in different colors indicate samples from different provinces (green, Hunan; red, Guangdong; blue, Jiangxi; purple, Zhejiang). The shapes of symbols indicate different As levels of these samples (circle, high As levels; triangle, low As levels). Note that samples were clustered according to their geographic distance, not As levels (i.e., CL-CZ, 445 km; CL-JX, 588 km; CL-ZJ, 938 km; CL-GD, 949 km; most distinct metagenomes were clustered further apart). (B) NMDS plot of Mash distances of whole metagenomes determined by this study. The correlations of microbial community differences with environmental factors are shown (arrows). Fe and P concentrations were significantly correlated with the variations observed between the microbial communities based on the Adonis test (P < 0.05).

Based on the samples determined by our study, the concentrations of total phosphorus (TP) and iron (Fe), in contrast to As, were significant factors in driving the variations observed between the microbial communities (Fig. 5B), explaining about 30% of the differences among the sampled total microbial communities. A significant (P < 0.01) interaction was also found between these two factors. Interestingly, the TP concentration also significantly correlated with the relative abundance of different ars genes in the metagenomic data sets (R² = 0.83, P < 0.05). The As concentration was found to significantly correlate with the variance in ars gene transcriptional activities assessed by Bray-Curtis distance between paddy soils (Adonis test, R² = 0.69, P < 0.05) and the transcriptional activities of aioA genes (Pearson’s correlation = 0.85, P < 0.05). In addition to As, the concentration of total nitrogen (TN) (R² = 0.73, P < 0.01) and chromium (R² = 0.72, P < 0.05) also significantly correlated with differences in the transcriptional activities of ars genes, and the concentration of antimony (Sb) (Pearson’s correlation = 0.83, P < 0.05) significantly correlated with the transcriptional activities of aioA genes (Table S7). No significant interaction was found between these environmental factors as revealed by permutational ANOVA (PERMANOVA). For the transcriptional activities of arrA genes, no environmental factor was found to affect these activities among the different paddy soils examined.

DISCUSSION

Arsenic metabolism genes were pervasively present in the paddy soils either with high (67.3 to 105 mg kg⁻¹) or low (2.5 to 10.8 mg kg⁻¹) As levels, and their relative DNA abundance was mostly comparable between these soils (Fig. 3). Microbial capacity of As metabolism has been shown to be an ancient trait developed in response to the pervasive presence of As on early Earth (3). Therefore, it is likely that the presence and relative abundances of As metabolism genes in the paddy soils are a legacy effect of an early life characteristic, at least to some extent (4). Further, the As resistance mechanisms, including the most well-studied detoxification As(V) reduction system (ArsC) followed by the efflux of reduced As(III) (ACR3 and ArsB) (1), and the recently identified organoarsenical detoxification systems (ArsM, ArsI, ArsH, and ArsP) (10, 35) have been detected in various microbes (36). The respiratory reduction of As(V) or oxidation of As(III), which is thought to have developed in the Archaean era (16, 37), have also been described in a broad diversity of microbes (38). Consistent with these findings, the prevalent As metabolism genes have been previously reported in various wetlands with low As levels, such as paddy soils and estuarine sediments (<15 mg kg⁻¹) (21, 25, 26), corroborating the wide prevalence As metabolism genes in our paddy soils.

Notably, the As concentration in high-As paddy soils, up to ∼100 mg kg ⁻¹ in the paddy soils studied here, did not significantly correlated with a higher relative abundance of As metabolism genes in the resident As-associated microbial community. Previous study of As metabolism genes in mining-impacted fields showed a significant (P < 0.05) linear correlation of arsC and aioA gene abundance with the increasing of As concentration from 34 to 821 mg kg ⁻¹ (33). We studied the correlation of ars gene abundances with As concentrations using our own and the previously determined samples by Luo and colleagues and found, consistent with the previous study, a significant correlation (P < 0.05) between As metabolism gene abundance and high As concentration (especially in the 410 to 812 mg kg ⁻¹ range; see Fig. S15 in the supplemental material). These findings indicated that the 12-fold higher As concentration in high (average of 92 mg kg ⁻¹) versus low (average of 8 mg kg ⁻¹) paddy soils of our study was not strong enough to significantly differentiate As genes abundances and that higher difference in As level and/or high absolute As concentrations (e.g., >410 mg kg ⁻¹) is probably necessary in order to elicit more clear gene content and abundance differences. The possibilities that some of the As we measured in our high-As soils is not bioavailable (and, thus, not selecting for more As metabolism genes), especially considering that the water soluble or exchangeable (bioavailable) As only accounts for about 0.4 to 24% of the total As in paddy soils (39). Therefore, to further corroborate these findings, the selection pressure exercised by different As levels and As species should also be measured in future studies.

Nevertheless, the As present in the soil has apparently induced the ArsR-mediated upregulation of the ars operon and Aio and Arr and, therefore, resulted in the increased activities of As metabolic pathways, even though the difference in As concentration between high- versus low-As paddy soils was probably not great enough to induce difference in DNA abundance due to most bacteria being intrinsically tolerant to such As concentrations (40). The higher abundance of ars (arsR, acr3, arsB, arsC, arsM, arsI, arsP, and arsH) and aioA genes at the expression (but not DNA) level presumably reflected a truly stronger response of As genes to increased levels of As in the paddy soils at the RNA level than DNA level. Among the As metabolism proteins, both the ArsR and AioA proteins are induced by As(III), which is the dominant As species in the paddy soils during the flooded period (18, 19) and most likely contributed to the significantly increased transcriptional activities of ars and aioA genes in soils with higher As levels (Fig. 3). Overall transcriptional activity of the arrA gene, which is responsible for dissimilarity As(V) reduction (38), was not significantly different between high- versus low-As paddy soils (Fig. 3). Because ArrA is induced by As(V), it is possible that the dominant As species, i.e., As(III) in paddy soils (18, 19), did not significantly upregulate its transcriptional activity directly. Moreover, our study showed the increased transcriptional activity of As metabolism genes was due to several distinct clades (community wide) in the high-As paddy soils (Fig. 4 and Fig. S4 to S12), and this result again corroborated the pervasive presence of As metabolism mechanisms in microbes of the paddy ecosystem.

It is also noticeable that the relative abundances of aioA genes (DNA level) were about 3.5 times higher in high- than low-As paddy soils, although this was not statistically significant (P = 0.114). In contrast, the relative abundance of ars genes at the DNA level was comparable between high and low As levels (Fig. 3A). Considering that the As respiratory As(III) oxidation catalyzed by Aio in microbes can couple As(III) oxidation to ATP production, while the Ars is only responsible for As resistance (14, 36), our finding is consistent with the expectation that energy production genes (e.g., aioA) will be at higher demand at both the DNA and transcriptional (RNA) levels with increased availability of the substrate, while for the resistance genes (e.g., ars) transcriptional response might be adequate with increased substrate and the substrate might not be similarly toxic to all species in order to select for the corresponding genes.

The As-associated microbial populations in high As soils mostly clustered together and separate from their low As counterparts based on the similarities among the expression levels of the ars, aioA, and arrA genes (Fig. S14), except for sample SKS_AsHig, which could be attributed to high soil heterogeneity or the nonsignificant effect of As levels on arrA genes. In contrast, the As-associated populations did not show any obvious clustering in terms of high As versus low As soils based on metagenomic (DNA) abundances (Fig. S14). These results are overall consistent with the relative abundance and expression of As metabolism genes, indicating that high As selects more for transcriptome response than gene abundance at the DNA level.

In terms of whole microbial communities, geographic distances, rather than As levels, were found to shape microbial community differences among the paddy soil sites studied here (Fig. 5). The geographic patterns of soil microbial communities have been widely studied previously and are mostly attributed to varied physicochemical and carbon sources in soil environments (41–43). It is possible that the As levels in these paddy soils did not possess a strong enough selective pressure on the whole microbial communities. In contrast to As, the concentrations of TP (0.31 to 0.46 g kg⁻¹) and Fe (14.8 to 27.0 mg kg⁻¹), which were about 1.5 to 1.8 times different between high and low paddy soils, were significant in determining microbial community diversity among our samples (Fig. 5). Previous studies have shown that available organic soil P has an important influence on microbial community composition, shifting both the composition and function of soil microbes (44, 45). Especially for rhizosphere-associated microbes such as the dominant Proteobacteria in paddy soils (46), their community structure can be altered by soil P availability (47, 48). The Fe in paddy soils is likely to impact the microbially mediated redox processes (49), and its presence in the wetland environments has been demonstrated to correlate with microbial biogeographic patterns. It is also important to mention that the effect of Fe and TP on As microbes and genes may be interlinked due to the significant (P < 0.01) interaction of P and Fe, since P is often the limiting nutrient in many terrestrial ecosystems due to strong sorption on Fe(III) hydroxides (50). A significant positive correlation between P and Fe in porewaters as well as soil matrix was previously observed in flooded soils with rice growth (51). Moreover, their interaction could also affect As behavior in paddy soils (52) because iron oxyhydroxides could sequester As(V) (53) and reduce As bioavailability to microbes.

Overall, our study revealed that the high As level of ∼100 kg mg⁻¹ increased As metabolism gene activities rather than their abundance or diversity in paddy soils. These findings advance our understanding of how microbes respond to high As levels and the diversity of As metabolism genes in paddy soils. Even in cases where there may not be significant difference in the As-associated microbial communities at the DNA level, increased activity of the As-metabolic genes would indicate significant functional contributions to As biotransformation in the paddy ecosystem, which would be relevant for managing these ecosystems. Therefore, the potential functional contributions of the ubiquitous As metabolism genes in the paddy ecosystem, and likely other environments, should be examined not only at the DNA level but also at the transcript and possibly protein levels. The As-associated microbial populations in soils with As could have significant contributions to the biogeochemical cycling of As and the MAGs reported here could facilitate future studies in this area. The work presented here could also serve as a guide for the number of samples to obtain, amount of sequencing to apply, and what bioinformatics analyses to perform for studying the dynamics of the As or other elements metabolism pathways.

MATERIALS AND METHODS

Sites description and sample collection.

Soil samples were collected from six distinct paddy fields in Hunan province, China, including three sites with high As levels (67.3 to 104.0 mg kg⁻¹) in Cili (CL_AsHig), Chenzhou (CZ_AsHig), and Shuikoushan (SKS_AsHig), and three sites with low As levels (2.5 to 10.8 mg kg⁻¹) in Hanpu (HP_AsLow), Lianhua (LH_AsLow), and Yuchangping (YCP_AsLow). Eighteen soil samples, three replicates from the same field for each site, were collected from the soil surface (0 to 20 cm) in July 2016 during the rice-growing period (flooded). Paddy soil samples were placed in sterile plastic bags and transported to the laboratory on ice for DNA and RNA extraction and soil physiochemistry analysis. Soil properties, including pH, Eh, total concentration of carbon (TC), nitrogen (TN), phosphorus (TP), sulfur (TS), arsenic (As), lead (Pb), cadmium (Cd), chromium (Cr), antimony (Sb) and iron (Fe), were determined on a composite sample of the three replicates following standard methods (21). Details of the underlying methods are provided in the supplemental material.

Soil DNA and RNA isolation, library preparation, and sequencing.

Soil DNA and total RNA was extracted from 0.5 g and 2 g of soil using the FastDNA spin kit (MP Biomedicals) and E.Z.N.A. soil RNA Midi kit (Omega, Bio-Tek Inc., USA), respectively, according to the manufacturer’s protocol. For each site, DNA or total RNA extracted from the triplicates were pooled to obtain enough high-quality material for each site and to overcome high sample-to-sample heterogeneity, which is characteristic of soil microbial communities. The quality of DNA and total RNA was assessed by a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific, Inc., Wilmington, DE, USA), and the ratio of absorbance at 260 nm and 280 nm was around 1.8 to 2.0, indicating good quality for downstream analysis. Total RNA of the paddy soils was subjected to an rRNA removal procedure using the Ribo-Zero rRNA removal bacteria kit (Illumina Inc., San Diego, CA, USA) by following the manufacturer’s protocol.

Paired-end DNA and cDNA libraries (2 × 150 bp) were constructed using a TruSeq DNA sample prep kit (Illumina Inc., San Diego, CA, USA) and TruSeq RNA sample prep kit (Illumina Inc., San Diego, CA, USA), respectively, by following the manufacturer’s protocol. The six DNA and cDNA libraries were sequenced on an Illumina HiSeq 3000 platform at Majorbio Bio-pharm Technology Company in Shanghai, China, with HiSeq 3000/4000 PE cluster kit and HiSeq 3000/4000 SBS kit (Illumina Inc., San Diego, CA, USA), respectively, according to the manufacturer’s protocol.

Taxonomic classification, functional annotation, and population genome binning of soil metagenomes and metatranscriptomes.

The raw paired-end reads of metagenomic and metatranscriptomics data sets were trimmed using SolexaQA (54), removing reads by Phred quality score (Q < 20) and minimum fragment read length of 50 bp after trimming for downstream analyses. Parallel-META version 2.4 (55) was used with default settings to recover 16S rRNA gene fragments from metagenomes and to identify and remove residual rRNA sequences after rRNA subtraction from metatranscriptomes. 16S rRNA gene fragments from metagenomes were then processed for operational taxonomic unit (OTU) picking using close OTU picking, defined at the 97% nucleotide sequence identity threshold, and taxonomic identification with the SILVA database (29) as implemented in QIIME 1.9.1 (56) and classified at the phylum and class levels. The relative abundance of each taxon at each level was normalized by the total number of 16S rRNA gene sequence fragments recovered from the corresponding metagenome.

Protein-coding sequences from short-read metagenomes and metatranscriptomes were predicted using FragGeneScan (57) with default settings and searched against Swiss-Prot database (UniProt, downloaded in April 2019) using BLASTp (BLAST+ version 2.2.28) (58). Only the best match with amino acid identity of ≥50% and query sequence length coverage of ≥70% by the alignment was retained. Protein-coding sequences were also functionally annotated, using the same threshold, against the SEED database using the subsystem categories (59). Differentially abundant taxa based on 16S rRNA gene sequence fragments and SEED subsystems of metagenomics and metatranscriptomics between high and low As levels were determined with the DESeq2 package using the negative binomial model, adjusted for false discovery rate (60). Heatmap in pheatmap R package (R version 3.5.1) (61) was used to visualize the significantly differentially abundant taxonomic levels and SEED subsystems (adjusted P < 0.05). Metagenome-assembled genomes (MAGs) were recovered using both MaxBin version 2.1.1 (62) and MetaBAT version 2.12.1 (63) as described in more detail in the supplemental material.

Assessment of the abundance and transcriptional activity of arsenic metabolism genes. (i) Building an As metabolism protein reference database.

To identify and quantify As metabolism proteins containing reads, in-house databases of eight As resistance proteins (Ars, i.e., ArsR [ArsR1, ArsR2, ArsR3, and ArsR4], ACR3, ArsB, ArsC [ArsC1, ArsC2, and ACR2], ArsM, ArsI, ArsP, and ArsH) and As respiratory oxidation (AioA) and reduction (ArrA) proteins were constructed and manually curated as described before (5). Briefly, verified arsenic metabolism protein sequences were downloaded from UniProt or NCBI’s NR database to build HMM profiles and searched against 786 representative species that are sampled from an original data set used for reconstruction of the tree of life (species tree) (64) to identify homologs of As metabolism proteins. The resulting matching sequences were manually inspected for the presence of conserved domains based on the Pfam models (release 28.0) (65) and forming deep-branching clades in the phylogenetic tree of the (manually) verified reference sequences to further determine if the sequence in question was truly an As metabolism protein. The tree was built using maximum likelihood as implemented in RAxML v8.4.1 with the PROTGAMMAAUTO model (66). Sequences forming long branches and/or lacking one or more of the conserved domains were removed from further analysis.

To build a more comprehensive As metabolism protein database that better captures the diversity of As metabolism proteins, proteins on the contigs assembled using IDBA with default settings (67) from metagenomes and metatranscriptomes (for more details, see supplementary) were also searched against the curated As metabolism protein database using BLASTp (BLAST + version 2.2.28) with a cutoff for a match of amino acid identity of ≥40% and reference length coverage of ≥70% by the alignment. A total of 1,040 Ars proteins, including 87 ACR3, 281 ArsC, 1 ArsH, 69 ArsI, 173 ArsM, 21 ArsP, and 408 ArsR, were identified using this approach. However, for the AioA and ArrA proteins that are longer than the Ars proteins (800 to 900 versus 200 to 400 amino acids), no match was identified, and it was likely attributable to the limiting assembly, consistent with the relatively low sequencing effort. Therefore, in order to obtain a comprehensive database for AioA and ArrA proteins, the AioA and ArrA references in the curated As metabolism database were searched against the UniRef90 (TrEMBL, downloaded in October 2018) database for additional homologs. A more stringent cutoff compared to the Ars proteins mentioned above, i.e., identity of ≥40% and reference length coverage of ≥90% by the alignment, was used to filter the AioA and ArrA proteins in the UniRef90 database. Identified Ars protein sequences from metagenomes and metatranscriptomes and AioA and ArrA protein sequences from the UniRef90 database were searched the against Pfam database (release 32.0) (65) in June 2019 and inspected for the presence of known conserved domains. The resulting matching sequences were aligned using MAFFT version 7.407 (68) and visually inspected for forming deep-branching clades in a maximum likelihood phylogenetic tree as described above to determine if the sequences were truly As metabolism proteins. These phylogenetic trees were also used for placement of As metabolism proteins encoding reads as described below.

(ii) Estimation of As metabolism gene abundance and transcriptional activity.

Short sequences from metagenomes and metatranscriptomes were searched against the comprehensive Ars, AioA, and ArrA reference databases using BLASTx search (BLAST+ version 2.2.28) and filtered for best matches with amino acid identity of ≥90%, alignment length of ≥25 amino acids, and E value of ≤1E−5. MicrobeCensus was used to assess the average genome size of each sample community (69) and normalize the total base pairs assigned to each As gene-by-gene length, i.e., RPKM, and genome equivalent, i.e., RPKM/average genome size, providing an estimated RPKG [(reads mapped to gene)/(gene length in kb)/(genome equivalents)] of how many cells from those sampled encode the gene of interest. Relative abundance of As metabolism genes in metatranscriptomics was calculated by the statistic RPKM (reads per kb per millions of reads), i.e., RPKM = (the number of reads mapping on reference sequences)/(gene length in kb)/(the total number of reads after removal of rRNA reads). The As metabolism genes transcriptional activity was evaluated by the relative abundance of As metabolism genes in metatranscriptomes (RPKM) normalized by the relative abundance of the same gene variant or OTU (at >90% amino acid level) identified in metagenomes (RPKM; we used RPKM to normalize the relative abundance of the same gene identified in metagenomes here instead of RPKG in order to keep the normalization criteria consistent between metagenomes and metatranscriptomes in calculating the gene’s transcriptional activity) and visualized in heatmap using the pheatmap package in R 3.5.1 (61). One-way ANOVA, performed with the vegan package (70) in R 3.5.1, was used to compare the difference of As metabolism gene abundances and transcriptional activities in high- versus low-As-level paddy soils.

(iii) Phylogenetic placement of As metabolism gene carrying reads.

Arsenic metabolism protein-containing reads identified by BLASTx search were first translated to amino acid sequences by FragGeneScan (57) with default settings. The amino acid fragments were then added to the As metabolism protein reference alignments using MAFFT version 7.407 (68) with the addfragments option and were placed in the corresponding phylogenetic tree of different As metabolism proteins (described above) using RAxML v8.4.1 (66) with the -f v option. The placement of the target As metabolism protein-containing reads was visualized in iTOL (71) after processing the resulting visualization jplace file generated by RAxML with JPlace.to_iTOL.rb from the Enveomics Collection (72).

Alpha and beta community diversity estimation.

The abundance-weighted average coverage of the sampled microbial communities achieved by our sequencing effort was estimated by Nonpareil (73, 74) based on the level of redundancy of the sequence reads of metagenomes. Mash distances (75) were used to assess the sample-to-sample sequence composition similarity at the whole-metagenome level based on shared kmers. Because the coverage of these metagenomics varied among different studies, the data sets were subsampled to 2 Gbp to make them comparable. Principal coordinate analysis (PCoA) and/or nonmetrical multidimensional scaling (NMDS) plot were used to visualize the site-to-site similarity matrices of microbial community compositions based on Mash distances, OTU tables, and annotation counts from SEED subsystems based on Bray-Curtis distance. NMDS plot of annotation counts of ars (i.e., arsR [arsR1, arsR2, arsR3, and arsR4], acr3, arsB, arsC [arsC1, arsC2, and acr2], arsM, arsI, arsP, arsH), aioA, and arrA genes from metagenomes and metatranscriptomes based on Bray-Curtis distances were used to visualize the site-to-site similarity of As-associated microbial community compositions.

The Adonis test was used to determine correlations between physicochemical variables measured at each site and similarity in microbial community composition (Mash distance) or ars gene abundances/transcriptional activities (Bray-Curtis distances) and tested for significance based on a permutation test with 999 interactions. The interactions between different physicochemical variables were assessed by PERMANOVA, while the correlation of physicochemical variables with aioA and arrA gene abundances and transcriptional activities in different sites were tested by Pearson correlation. These analyses were performed in R 3.5.1 with the vegan (70) and ggplot2 (76) packages.

Data availability.

Metagenomics and metatranscriptomics data sets have been deposited in the National Center for Biotechnology Information (NCBI)’s Sequence Read Archive (SRA) database under the BioProject PRJNA616041. NCBI SRA numbers for all data sets are provided in Table S1 in the supplemental material.