Permafrost thaw ponds and lakes are widespread across the northern landscape and may play a central role in global biogeochemical cycles, yet knowledge about their microbial ecology is limited. We sampled a set of thaw ponds and lakes as well as shallow rock-basin lakes that are located in distinct valleys along a north–south permafrost degradation gradient. We applied high-throughput sequencing of the 16S rRNA gene to determine co-occurrence patterns among bacterial taxa (operational taxonomic units, OTUs), and then analyzed these results relative to environmental variables to identify variables controlling bacterial community structure. Network analysis was applied to identify possible ecological linkages among the bacterial taxa and with abiotic and biotic variables. The results showed an overall high level of shared taxa among bacterial communities within each valley; however, the bacterial co-occurrence patterns were non-random, with evidence of habitat preferences. There were taxonomic differences in bacterial assemblages among the different valleys that were statistically related to dissolved organic carbon concentration, conductivity and phytoplankton biomass. Co-occurrence networks revealed complex interdependencies within the bacterioplankton communities and showed contrasting linkages to environmental conditions among the main bacterial phyla. The thaw pond networks were composed of a limited number of highly connected taxa. This “small world network” property would render the communities more robust to environmental change but vulnerable to the loss of microbial “keystone species”. These highly connected nodes (OTUs) in the network were not merely the numerically dominant taxa, and their loss would alter the organization of microbial consortia and ultimately the food web structure and functioning of these aquatic ecosystems.
Permafrost is widespread in Arctic and boreal regions (Schuur et al.,
2008) and is estimated to contain ca. 1700 Pg of organic carbon
(McGuire et al., 2009; Tarnocai et al., 2009). Permafrost thawing
and erosion is evident by the northward retreat of the permafrost boundary
(Thibault and Payette, 2009). In some northern regions this has led to
the expansion of permafrost thaw ponds and lakes (thermokarst systems;
Grosse et al., 2013), whereas in other regions there
has been a contraction and loss of these waterbodies (e.g., Andresen
and Lougheed, 2015). These thermokarst systems are part of circumpolar and
global biogeochemical cycles (Abnizova et al., 2012; Walter et
al., 2007). Although some are carbon sinks
(Walter Anthony et al., 2014), others are net
sources of carbon dioxide (CO
Microbial communities are among the main drivers of key biogeochemical processes (Ducklow, 2008), and in thermokarst systems they are composed of functionally diverse taxa (Crevecoeur et al., 2015; Rossi et al., 2013). In particular, these systems are favorable for bacterial methanotrophs (Crevecoeur et al., 2015) as well as archaeal methanogens (Mondav et al., 2014), and the relative activity of these two groups will affect methane balance and the net emission of greenhouse gases. Identifying factors that shape microbial communities in these aquatic systems is therefore essential for understanding the functional significance of these permafrost thaw systems in the global carbon budget.
Aquatic bacterial communities are thought to be selected by a combination of bottom-up (resource availability) and top-down (viral lysis, grazing) controls. Less studied are bacteria–bacteria interactions (facilitation, competition), which may further contribute to nonrandom distributions observed among microbial taxa (e.g., Horner-Devine et al., 2007). Examining co-occurrence patterns has the potential to unveil ecological processes that structure bacterial communities. Specifically, patterns of co-occurrence may reveal to what extent groups of microbes share habitat preferences, to what extent there may be ecological linkages among bacterial taxa and with other planktonic organisms, and the extent of phylogenetic closeness of co-occurring bacterial taxa given that closely related taxa may share life strategies and ecological traits.
Across northern landscapes, both regional (e.g., climate and the degradation state of permafrost) and local (e.g., nutrients, dissolved organic carbon and oxygen) conditions are likely to influence the distribution of bacterial communities of thaw ponds and lakes. These thermokarst systems show a high degree of limnological (Deshpande et al., 2015) and bacterial heterogeneity (Crevecoeur et al., 2015), making them suitable models to investigate the co-occurrence patterns among bacterial taxa as well as their network relationships within microbial consortia. The main objectives of this study were to characterize the ecological linkages within microbial communities as a response to permafrost thawing. Our hypotheses were that (i) bacterial communities follow co-occurrence patterns along the permafrost degradation gradient, due to distinct habitat preferences among bacteria, and (ii) these habitat preferences relate to differences in the phylogenetic structure of bacterial communities.
To test the above hypotheses, we employed high-throughput sequencing of the 16S rRNA gene to determine the composition of bacterial communities in thaw ponds and lakes of Nunavik (Québec, Canada) along a north–south permafrost degradation gradient. In addition, we sampled rock-basin lakes that were under the same regional climate but whose formation was not related to climate change. We investigated the relationships among bacterial taxa and local environmental conditions by means of network analysis, which has been applied with success elsewhere to evaluate microbial distribution patterns (Barberan et al., 2012; Peura et al., 2015; Steele et al., 2011) and responses to environmental perturbation (Araújo et al., 2011). We then examined the potential linkages between the bacteria and phytoplankton, phototrophic picoplankton and zooplankton biomass in the ponds and lakes.
Surface water (0.2 m) from 29 thermokarst ponds was collected from 1 to 13
August 2012 in two types of permafrost landscapes. Thaw ponds were located
in the vicinity of Whapmagoostui-Kuujjuarapik (W-K: lat 55
At each site, temperature, conductivity, dissolved oxygen and pH were
measured using a 600R multiparametric probe (YSI, Yellow Springs, OH, USA).
Water for dissolved organic carbon (DOC) and chlorophyll
Samples for zooplankton were collected using a 35
DOC concentrations were analyzed on a Shimadzu TOC-5000A carbon analyzer and
nutrients were analyzed using standard methods (Stainton et al., 1977).
Colored dissolved organic matter (CDOM) was measured by spectrophotometric
analysis of absorbance at 254 nm by water filtered through 0.2
Phytoplankton biomass was estimated as Chl
Bacterial community composition (BCC) was determined by 454-pyrosequencing of
the V6–V8 regions of the 16S rRNA gene. In brief, water was sequentially
filtered through a 20
All sequence data processing was within the QIIME v1.8.0 pipeline (Caporaso et al., 2010b). Reads were first pre-processed by removing those with a length shorter than 300 nucleotides. The remaining reads were then processed through a QIIME denoiser. Denoised sequence reads were quality-controlled and chimeras were detected using UPARSE (Edgar, 2013). Operational taxonomic unit (OTU) sequence representatives were aligned using PyNAST (Caporaso et al., 2010a) with the pre-aligned Greengenes 16S core set (DeSantis et al., 2006) as a template and taxonomically classified using the Mothur Bayesian classifier (Schloss et al., 2009). The reference database was the SILVA reference database (Pruesse et al., 2007) modified to include sequences from our in-house, curated northern 16S rRNA gene sequence database (Lovejoy et al., 2015). Sequences classified as plastid or mitochondrial 16S were removed from the analyses.
All phylogenetic analyses were based on a phylogenetic tree constructed with an approximate maximum-likelihood (ML) approach using FastTree v.2.1 (Price et al., 2010) following the procedures described in Monier et al. (2015). UniFrac dw4000 (weighted) and duw4000 (unweighted) distances (Lozupone and Knight, 2005) among the different microbial communities were all computed based on the OTU approximate ML phylogenetic tree. Clustering of UniFrac distances was performed using the unweighted pair group method with arithmetic mean (UPGMA) algorithm, and cluster robustness was assessed using 1000 jackknife replicates (on 75 % subsets). Beta-diversity significance was assessed using UniFrac Monte Carlo significance test on dw4000 with 10 000 randomizations, as implemented in QIIME.
We investigated community phylogenetic diversity as defined by Faith (1992), along with other diversity metrics such as phylogenetic species richness and evenness (Helmus et al., 2007), using the R package “picante” v1.5 (Kembel et al., 2010). Community phylogenetic structure was investigated with the calculation of the net relatedness index (NRI) that measures the phylogenetic relatedness for each community. Specifically NRI determines whether OTUs are more closely related to co-occurring relatives than expected by chance (Webb et al., 2002).
All statistical analyses were carried out using R 3.0.3 (R Core Team, 2014). Abiotic and biotic environmental variables were log-transformed, with the exception of pH (already on a log scale). All analyses were performed on the subsampled data set (4000 sequences per sample) with a total number of 2166 OTUs.
Dissimilarities in community composition among the different valleys were visualized using cluster and principal coordinate analyses. A rank abundance plot was generated to identify the bacterial dominants.
The taxonomic uniqueness of sites as well as the taxa that contribute the most to these compositional differences was evaluated by means of local contribution to beta-diversity (LCBD; Legendre and De Cáceres, 2013). Differences in LCBD, phylogenetic diversity, species richness and structure across spatial scales were tested using ANOVA followed by Tukey's HSD test and regression models to identify links between site uniqueness and environmental variables.
Significant associations between the abundance of bacterial OTUs and the
five valleys were further assessed by correlation indices (as a measure of
habitat preferences), including the point biserial correlation statistic
Co-occurrence analyses were performed using the overall data set and each of the data sets for the five individual valleys. The data were filtered by using only those OTUs with a minimum of 20 reads and that were detected in at least three different ponds. This filtering step removed poorly represented OTUs and reduced the network complexity, resulting in a core community of 294 OTUs.
Randomness in co-occurrence of OTUs in the regional and individual valley
data sets was tested in a null model using the quasiswap algorithm
(Miklós and Podani, 2004) and
Network analyses were conducted on the filtered OTU data set. In addition, a
total of eight physicochemical variables (DOC; TP; TN; pH; SUVA
To examine associations between the bacterial OTUs and their environment, we
analyzed the correlations of the OTUs with each other and with biotic and
abiotic variables using the maximal information coefficient (MIC;
Reshef et al., 2011). The MIC value indicates
the strength of the relationship between two variables and is analogous to
The relationship between the connectivity of OTUs (as indicated by the degree value in the network) and their corresponding abundance was examined in generalized linear models in order to relax the normality assumptions. OTU abundance was first calculated per individual pond as the product of percent of total reads and total bacterial abundance. The total abundance of an OTU in the data set was then obtained by summing the abundance calculated for each pond. A heat map was produced to examine the variability in the ecological preference among the 30 most connected OTUs.
The phylogenetic composition of bacterial communities differed significantly
among valleys (dw4000, UniFrac weighted significance test;
Community phylogenetic analysis based on NRI indices showed that all site
clusters had significant phylogenetic structure (positive NRI values; one-sample
The local contribution to beta-diversity (LCBD) values indicated the
compositional uniqueness of local bacterial communities. One-way ANOVA
showed that pond location had a significant influence on compositional
uniqueness (
The thaw pond communities were dominated by OTUs that were assigned to
Betaproteobacteria, particularly the order Burkholderiales, which was well
represented in all communities (35.4 % of the total number of reads).
Actinobacteria (24.5 % of total reads) were mainly represented by OTUs
assigned to the family ACK-M1 (60.5 % of Actinobacteria reads). Among
Bacteroidetes, which accounted for up to 15.7 % of the total number of
reads, Sphingobacteriales were highly represented and were dominated by the
family Chitinophagaceae that contributed up to 4.7 % of total number of
reads. Other dominant OTUs were within the Verrucomicrobia (6.8 % of total
reads) (Table 1). Among the 30 most abundant taxa, some were highly
associated with a specific valley, whereas others were not detected in
certain valleys (Fig. 2a). This pattern remained when considering the
ensemble of the 2166 OTUs (Fig. S4). Specifically, 272 OTUs (11.3 % of
the 2166 detected in this data set) showed a significant association in the
indicator value analysis (the point biserial statistic r.g.) considering
habitat combinations. Among the 272 OTUs showing a significant habitat
preference, 246 were associated with a single valley: 13, 12, 31, 99 and 91
OTUs were associated with the BGR, NAS, KWK, SAS and RBL valleys,
respectively. Four OTUs were associated with the discontinuous permafrost
landscape and three with the sporadic permafrost landscape (Table 2). There
were distinctions between ponds located in the sporadic versus discontinuous
permafrost landscapes. In particular, OTUs closely related to methanotrophs
were prominent within the sporadic permafrost landscape type: OTUs closely
related to
Five most abundant (number of reads) OTUs across spatial scales. Finest taxonomy assignments are presented with a minimum confidence of 0.8.
Heat map representation of habitat preference of the 30 most
abundant
To test for differences in co-occurrence patterns between microbial
communities across the permafrost landscape, we first selected OTUs that had
at least 20 reads and were detected in at least three different ponds. The
bacterial OTUs were not randomly distributed among the different valleys
when considering the entire region (
Results of indicator species analysis. Valley refers to the valley
(or combination of valleys) for which the OTUs obtained the highest
correlation. We indicate the correlation value (r.g.) and its statistical
significance (
The OTU co-occurrence patterns as well as the relationships among both biotic and abiotic variables were investigated by network analysis. The most connected nodes (degree > 10) were related to three abiotic variables (DOC, conductivity and TP) and one biotic variable (phototrophic picoeukaryotes). The topology of the networks is presented in Table 4. For the whole regional network, a total of 248 nodes and 968 edges were detected, which was fragmented in 3 components including 2 small components composed of 2 and 3 nodes (Fig. S5). The observed characteristic path length of 3.06 and clustering coefficient of 0.25 were both greater than estimates originating from the random network of similar size. In addition, the observed : random network clustering coefficient ratio (log response ratio of 0.92) showed that the network had “small world” properties – i.e., the nodes were more connected than expected in a random network (Table 4). The frequency distribution of nodes followed a power law function, which indicated that the network was composed of a few highly connected nodes, as opposed to an even distribution of connectivity (Fig. S6).
Results of co-occurrence analyses for the dominant OTUs (20 reads, 3 sites). Significant results are presented in bold. SES refers to standardized effect size.
Four main bacterial phyla were well represented in the networks: Proteobacteria (83 nodes), Bacteroidetes (56 nodes), Actinobacteria (42 nodes), and Verrucomicrobia (24 nodes). Although edges between nodes that referred to bacterial OTUs dominated the network, connections between bacterial OTUs and both biotic and abiotic variables were detected (Fig. S5). For example, conductivity and DOC were amongst the most connected nodes, illustrating their importance in the network. The subnetwork built around DOC showed a diverse bacterial consortium with a slight dominance of Actinobacteria (Fig. 3a). Phototrophic picoeukaryotes were the most connected node among biotic variables. The subnetwork built around that variable showed strong co-occurrence between picoeukaryotes and Actinobacteria (Fig. 3b). The co-occurrence network around the group Chitinophagaceae showed that these OTUs were associated with different environmental variables including DOC, dissolved oxygen, conductivity, abundance of phototrophic picoeukaryotes, cladocerans and rotifers (Fig. 4a) and had recurrent, strong co-occurrences with Actinobacteria, especially with organisms closely related to ACK-M1 (Fig. 4b). The analysis of the linearity of the latter association indicated a positive co-occurrence between OTUs closely related to members affiliated with the ACK-M1 (aka AcI) group of Actinobacteria and Chitinophagaceae (Fig. 5c). Other examples of strong linkages between OTUs are given in Fig. 5, with illustrations of positive co-occurrence (Fig. 5a) and non-coexistence (Fig. 5b).
Subnetworks organized around DOC
Subnetworks organized around bacterial OTUs closely related to
Chitinophagaceae. Panel
Topology of the thermokarst systems co-occurrence networks. Regional corresponds to a network built around the selected 294 OTUs whereas Hubs refers to a network where the most connected 24 OTUs from the whole network (Fig. S5a) were removed prior to this analysis. Random refers to topology characteristics of Erdős–Rényi random networks of similar size.
In general, our results indicated that the most abundant OTUs were also the
most connected ones (
We further investigated the implications of the removal of the top 24 connected OTU nodes (hubs), which represented a removal of 10 % of nodes, and the results showed a high level of fragmentation of the network and a drop in node degree (Table 4, Fig. S8).
Analysis of the network hubs further showed that the top 24 were mainly
composed of Actinobacteria OTUs, in particular members of Actinomycetales and
Acidimicrobiales. In addition, OTUs assigned to Betaproteobacteria
represented a large fraction of these highly connected OTUs, including the
typical freshwater
The main goal of the present study was to identify co-occurrence patterns among bacterial communities in thaw ponds and lakes in the changing subarctic landscape. Consistent with our first hypothesis, there was a nonrandom distribution of bacterial taxa across the distinct valleys sampled in this study. The results showed that thaw ponds communities from the same valley, especially those located in the sporadic permafrost landscape, tended to be more similar in terms of bacterial community composition than communities originating from ponds located in other valleys. Furthermore, the thaw ponds differed taxonomically from the rock-basin reference lakes, with specific bacterial OTUs associated with a particular valley or permafrost landscape type. Contrary to our second hypothesis, that differences in habitat preferences among bacterial communities were related to distinct phylogenetic structure, we found no evidence for differences in the community phylogenetic relatedness between the different valleys. The same bacterial phyla occurred throughout the region, and variability among ponds in the same valley was greater than the differences among valleys.
Associations between bacterial OTUs in permafrost thaw
ponds and lakes.
Non-random distribution patterns among bacterial taxa were detected, indicating that bacterial taxa in our study region tended to co-occur more than expected by chance. Non-random assembly patterns indicate the dominance of deterministic processes such as environmental filtering in shaping community composition (Horner-Devine et al., 2007). The bacterial communities of freshwater ecosystems elsewhere (Eiler et al., 2011), as well as in certain terrestrial (Barberan et al., 2012) and marine (Steele et al., 2011) ecosystems, have also been reported to have distributional patterns that relate to the environment. Such patterns may depend on niche breadth and competitive abilities (Székely et al., 2013), grazing and viral lysis susceptibilities (Chow et al., 2014; Miki, 2008) and dispersal capabilities (Fahlgren et al., 2010; Hervàs and Casamayor, 2009). The patterns described here are for the free-living fraction of bacterial assemblages, which raises the question of whether such patterns remain for the attached fraction of the communities. The latter may represent a substantial part of the total communities given that these waterbodies can contain a large content of suspended solids. Previous studies comparing the compositional patterns in bacterial communities between the free-living and attached fractions showed that these two distinct lifestyle have a similar community composition (Crevecoeur et al., 2015), indicating that the patterns described here may reflect patterns for the entire community.
No significant relationship was found between distribution patterns and
environmental heterogeneity. This was unexpected, as previous studies have
shown that thermokarst systems are heterogeneous environments with marked
differences in community composition across the different valleys associated
with distinct environmental variables (Crevecoeur et al., 2015; Comte et al.,
2015). In agreement with Heino and Grönroos (2013), we
suggest that the relationship between distribution pattern and environmental
heterogeneity may be scale-dependent such that environmental heterogeneity
may have effects on the bacterial taxa distribution patterns at the overall
study region scale and not at the valley scale as tested here. The results
did show differences in the phylogenetic composition of bacterial
communities among the different valleys, which highlight distinct habitat
preferences among taxa (Figs. 2, S4). In particular, the combination
of LCBD and regression analyses indicated that the compositional uniqueness
of thaw ponds and lakes was positively related to DOC concentrations, a well-known determinant of bacterial communities and processes
(Kritzberg et al., 2006;
Ruiz-González et al., 2015). Along with the variations in permafrost
degradation state across the study region, there were also differences among
valleys in terms of availability and origin of carbon subsidies. The
northern sites are located within the discontinuous permafrost area where
most of the soil remains frozen and is thus not available for microbial
degradation, while in the southern sporadic area, permafrost is highly
degraded (Bouchard et al., 2014) and large amounts of ancient
permafrost carbon may be available for microbial processes. Consistent with
this pattern, elevated concentrations and high rates of CO
The mean NRI across all communities was significantly greater than zero. This provides evidence for a dominant role of environmental filtering on community composition (Kembel, 2009). The corollary is that a set of environmental variables constrained community composition, resulting in taxa that were closer phylogenetically and more ecologically similar than if stochastic processes (including dispersal) drove community assembly. In fact, there is no corridor such as streams that connects the ponds, and thus local dispersal processes are unlikely to explain the local phylogenetic structure of the thaw pond communities (Comte et al., 2015). Similar results were obtained for microbial community studies in the ocean (Monier et al., 2015) and on groundwater communities (Stegen et al., 2012).
No significant difference in NRI was found among the different valleys, but this result likely reflects the high variability within individual valleys. In particular, two ponds in the NAS valley had higher values of NRI in comparison to their neighboring ponds. These two ponds had specific environmental characteristics including high concentrations of suspended clay particles and low phytoplankton concentrations, which may have favored certain environmental specialists. The rock-basin waters had higher NRI values than the thaw ponds, indicating that their assemblages were more ecologically similar to each other than those originating from thaw ponds and lakes. This could relate to their respective histories in that the rock-basin lakes originate from deglaciation followed by retreat of the Tyrrell Sea ca. 8000 years ago and have thus been exposed to longer-term ecological processes.
The extent of permafrost erosion (permafrost landscape type) appeared to influence phylogenetic structure. When controlling for the two outliers mentioned above (NAS-A and NAS-B), the northern communities (BGR, NAS) had a greater phylogenetic distance among co-occurring taxa than expected by chance (lower NRIs) in comparison to communities from the thaw ponds located in valleys from sporadic permafrost (KWK, SAS). This suggests that taxa from SAS valley (and to a lesser extent KWK) tend to be more ecologically similar to each other than those from northern valleys, reflecting strong environmental filtering by variables such as DOC concentration, as previously documented in this valley (Comte et al., 2015). These findings are in line with studies elsewhere that showed that clustered communities are mainly retrieved from environments that have constrained environmental conditions (Monier et al., 2015).
The extent to which closely related bacterial taxa may coexist is still a subject of considerable discussion (Mayfield and Levine, 2010). Previous studies on aquatic microbial communities have shown that closely related taxa have coherent temporal dynamics and share similar ecological niches (Andersson et al., 2009; Eiler et al., 2011). Co-occurrence networks enable the depiction and visualization of co-occurrence patterns among OTUs, and they provide a way of identifying potential ecological niches within microbial consortia. Network analyses have recently been applied to a wide range of microbial communities and biomes, and specific associations among bacterial OTUs and with environmental variables have been reported (Barberan et al., 2012; Chow et al., 2014; Eiler et al., 2011; Steele et al., 2011).
Our results point toward the importance of environmental filtering for community assembly in thaw ponds and lakes. In co-occurrence networks, correlations between OTUs and environmental variables highlight the conditions that may favor particular assemblages. Specifically, our co-occurrence networks identified two abiotic variables (DOC and conductivity) to be among the most connected nodes (Fig. S5b), and these variables separated according to landscape type: the northern ponds located in the discontinuous permafrost landscape had high conductivity and low DOC, whereas southern sites within the sporadic permafrost landscape had high DOC and lower conductivity (Table S2; further details are given in Comte et al., 2015). The analysis of the DOC subnetwork showed that only a few OTUs were significantly and directly related to DOC; these included OTUs assigned to Actinobacteria as well as OTUs closely related to bacterial methanotrophs and taxa involved in the degradation of complex organic polymers (Fig. 3a). Among phylogenetically related microbes, unique combinations tended to co-occur (Fig. 4a). For example, some OTUs assigned to the Chitinophagaceae appeared to be significantly related to different abiotic and biotic variables, which in turn suggested niche separation.
In addition to the bottom-up factors that shape bacterial communities,
recent work on microbial networks has highlighted the role of top down
processes such as grazing and viral lysis in affecting prokaryotic community
structure and co-occurrence patterns
(Chow et al., 2014; Steele et al., 2011). In the present study, phototrophic
picoeukaryote abundance (degree
In general, relationships among microbes dominated the network, rather than
those between microbes and abiotic or biotic environmental parameters (Fig. S5). There was overlap in terms of community composition among the
different valleys (Fig. 1), with shared dominant taxa (Table 1, Fig. S2).
Although this may indicate that some OTUs may respond similarly to specific
environmental factors and outcompete others, some associations may be the
result of substrate interdependencies. One example is the relationship
between bacteria able to degrade chitin and others that take up the
resulting hydrolysis products (Beier and
Bertilsson, 2013). OTUs closely related to bacteria in the Chitinophagaceae,
a group known to be involved in the degradation of chitin and other complex
polymeric organic matter (del Rio et al., 2010), were well
represented in our study area, and have also been found in other cold
terrestrial environments
(Franzetti et al., 2013; Ganzert et al., 2011). The subnetwork built around this group
showed that these OTUs are linked to other phyla (Fig. 4a), notably certain
Actinobacteria (Fig. 4b). The dominants were closely related to clade Ac1,
which is known to include specialists that use hydrolysis products from
chitinolytic bacteria (Beier and Bertilsson, 2011). The analysis of
linearity of the associations between the corresponding OTUs showed a
positive co-occurrence (Fig. 5c), consistent with bacterial network
relationships. Although other examples of positive co-occurrence among
bacterial OTUs were identified in the data set (Fig. 5a), there was also
evidence of “non-coexistence” (sensu Reshef et al., 2011) among certain OTUs: in
the northern, less degraded permafrost valley (BGR), OTU 1242
(Betaproteobacteria
The microbial networks for the thermokarst systems had “small world”
properties, with only a few, highly connected nodes, which can be viewed as
“keystone species”. This property would render the networks more resistant
to environmental change, but vulnerable to the loss of these nodal species
(Montoya et al., 2006). The bacterial hubs were
identified as typical freshwater, terrestrial and marine taxa (Table S3),
and some of them were closely related to taxa that are involved in key
biogeochemical processes such as nitrogen fixation and degradation of
complex polymers, or that are known to be restricted in niche breadth, for
example to cold environments. In accordance with Peura et al. (2015),
the importance of a taxon in a microbial network may be less associated with
its abundance but instead determined by its connectivity, as represented by
node degree for example. Thus many of the hub taxa identified in this study
could be defined as keystone microbial species (Table S3). These
“keystone” OTUs identified as hubs were not merely the abundant OTUs (Fig. 2b),
but some were rare and potentially important actors for the functioning
of these ecosystems. For example, the nitrogen-fixing bacterium
The thaw ponds and lakes sampled in the present study showed large variability in their bacterial community structure, even among waterbodies in a single valley. This underscores the heterogeneous nature of permafrost aquatic environments, and is consistent with their known limnological variability. A small number of taxa occurred in high abundance and dominated many of the communities; these northern dominants included members of the betaproteobacterial order Burkholderiales and the actinobacterial family ACK-M1; other dominants included members of the Bacteroidetes family Chitinophagaceae and Verrucomicrobia. Despite this variability and the existence of common taxa, there were taxonomic differences among different valleys and between permafrost landscape types, implying some degree of habitat selection.
The bacterial networks further showed that DOC and conductivity played an important role in the co-occurrence patterns of bacterial OTUs, corresponding at least in part to differences in these two environmental variables among valleys (Table S2). Strong positive associations as well as non-coexistence among OTUs were detected, and the resultant networks were composed of a limited number of highly connected OTUs. This “small world network” property would render these communities more resistant to environmental change but sensitive to the loss of their hub OTUs, which themselves showed some degree of habitat specificity. With ongoing global warming, these waters are likely to experience the effects of increased permafrost erosion and associated changes in their chemical environment, including shifts in DOC and conductivity. If such changes eventually cause the loss of “keystone species” that form the hubs of the present microbial networks, there would be a major disruption of community structure, with potentially large biogeochemical consequences.
We are grateful to M. Bartosiewicz, B. Deshpande, A. Matveev, A. Przytulska-Bartosiewicz and C. Tremblay from Whapmagoostui-Kuujjuarapik CEN station and the pilots of Canadian Helicopter Ltd. for their assistance in the field. We are also grateful to Paschale N. Bégin for zooplankton enumeration, M.-J. Martineau for pigment analyses, and I. Laurion (INRS-ETE) for flow cytometry. Computing support from CLUMEQ/Compute Canada, aid from A. Monier for bioinformatics and phylogenetic analyses, advice from A. Eiler for network analyses, and insightful comments from the two anonymous reviewers and the editor were also greatly appreciated. We acknowledge the Natural Sciences and Engineering Council (NSERC) of Canada funding for Discovery grants to W. F. Vincent and C. Lovejoy and Discovery Frontier (ADAPT) grant to W. F. Vincent, the support from the Networks of Centres of Excellence program ArcticNet to W. F. Vincent and C. Lovejoy, and the Canadian Research Chairs Program to W. F. Vincent. Additional support from Fonds de Recherche du Québec Nature et Technologies (FRQNT) to CEN is acknowledged. J. Comte was partially supported by a FRQNT postdoctoral fellowship and the EnviroNorth CREATE program from NSERC. Edited by: I. Laurion