Protistan community analysis: key findings of a large-scale molecular sampling.

Protists are perhaps the most lineage-rich of microbial lifeforms, but remain largely unknown. High-throughput sequencing technologies provide opportunities to screen whole habitats in depth and enable detailed comparisons of different habitats to measure, compare and map protistan diversity. Such comparisons are often limited by low sample numbers within single studies and a lack of standardisation between studies. Here, we analysed 232 samples from 10 sampling campaigns using a standardised PCR protocol and bioinformatics pipeline. We show that protistan community patterns are highly consistent within habitat types and geographic regions, provided that sample processing is standardised. Community profiles are only weakly affected by fluctuations of the abundances of the most abundant taxa and, therefore, provide a sound basis for habitat comparison beyond random short-term fluctuations in the community composition. Further, we provide evidence that distribution patterns are not solely resulting from random processes. Distinct habitat types and distinct taxonomic groups are dominated by taxa with distinct distribution patterns that reflect their ecology with respect to dispersal and habitat colonisation. However, there is no systematic shift of the distribution pattern with taxon abundance.

Introduction

Protists are abundant and diverse in aquatic and terrestrial ecosystems and fulfil critical ecosystem functions (del Campo and Massana, 2011; Triadó-Margarit and Casamayor, 2012; Bates ; Geisen ). They not only strongly contribute to primary production (Field ) and bacterial grazing (Boenigk and Arndt, 2002; Glücksman ), but are also major players in diverse nutrient cycles (Finlay and Esteban, 1998; Coleman and Whitman, 2004) and as such, they form the basis of aquatic and terrestrial food webs along with bacteria.

The ecological key role of protists is sharply contrasted by an only vague understanding of their diversity. During the past decade, the assumption of unlimited dispersal of protists was progressively replaced by the ‘moderate endemicity model' (Foissner, 2006; Bass and Boenigk, 2011; but see Livermore and Jones, 2015). With the increasing acceptance of dispersal limitations in protists, the research focus turned to the underlying patterns of protist distribution and the generalisability of such patterns across protistan taxa and across different habitat types (Katz ; Bik ). Owing to methodical limitations, analyses of protistan distribution patterns up to recently were centred around a handful of more or less abundant taxa (for example, Lopéz-García ; Moreira and López-García, 2002). High-throughput sequencing has already revolutionised environmental surveys of microbial organisms and opened the door for large-scale analyses of molecular microbial community analyses. Generating many millions of reads, these methods allow for deep-sequencing of microbial communities (Caporaso ) and make possible a much more realistic understanding of protistan diversity and distribution than former, lower capacity technologies (Medinger ; Lecroq ; Degnan and Ochman, 2012). The current high-throughput sequencing technologies now allow for sufficiently deep community analysis. Nevertheless, most studies have been restricted to few sampling sites (Caron and Countway, 2009; Stoeck ; Nolte ; Caporaso ). Only recently, large-scale multisample sets, specifically from the marine biome (de Vargas ), allow for deeper insights into protist distribution patterns.

Here, we address the generalisability of protist distribution patterns with respect to habitat type, taxonomic group and taxon abundance based on a large data set comprising 232 samples from soil as well as fresh, brackish and marine waters. Besides judging the relative importance of habitat type, seasonality and geographic region in shaping protist distribution patterns, we evaluate the question on whether or not rare taxa show similar distribution patterns to the abundant taxa.

Materials and methods

Sampling and sample preparation

Sampling sites and sampling procedures of the assembled data set are summarised in Supplementary Table S1. The 10 different sampling campaigns representing eight different habitat types are referred to as Alpine spatial, Alpine seasonal, Lake District, Winterpico, Estuarine, Soil, Whale Fall, Borehole, heterotrophic nanoflagellates (HNF) and Biofilm within this paper. Briefly, they comprise an alpine transect of freshwater samples (Austria, 32 lakes at 450–2050 m elevation) and a seasonal sampling of three Austrian lakes (Lake Mondsee region, 9–28 samples per lake, range: April to December), a seasonal sampling in the English Lake District, partially size-fractionated (UK, 7 sites, 2–3 samples per site) and size-fractionated samples of the same sites in winter (Winterpico), an estuarine gradient (UK, River Colne, 11 sites, partially seasonal), a comparative soil sampling from fallow and re-cultivated fallow fields (UK, 15 sites across England), a sampling on the seabed of and from below a sunken whale (Whale fall) (North Sea, Sweden), a collection of aquifer samples from boreholes (UK, Berkshire), an experimental set-up of a size-filtered (5 μm) freshwater sample fed with different food bacteria and focusing on heterotrophic nanoflagellates (Czech Republic, Římov reservoir) and an experimental set-up of freshwater biofilms treated with the viricide TamiFlu (Southern UK). Owing to the different natures of the collected samples, slightly modified sampling protocols and DNA-isolation methods (compared in Supplementary Table S1) had to be used.

PCR and pyrosequencing

PCR amplifications targeting the SSU V9 region were conducted using primers 1391F (Lane, 1991; Stoeck ) and Euk B (Medlin ; Stoeck ) both carrying a 5′-tail for the 454 sequencing (adapter A: GCCTCCCTCGCGCCATCAG, adapter B: GCCTTGCCAGCCCGCTCAG) to amplify a broad spectrum of eukaryotes. The final concentrations in all of the PCR reactions were as follows: 1 μl of DNA template in 20 μl PCR reaction with 0.4 units of Phusion polymerase; primers at 0.25 μm final concentration; dNTPs at 0.2 mm final concentration; 4 μl Phusion buffer; and 12.2 μl water. The PCR conditions consisted of an initial denaturation at 94 °C for 4 min and 35 cycles of: 30 s at 95 °C, annealing for 60 s at 60 °C, elongation for 2 min at 72 °C, followed by a final extension step of 10 min at 72 °C. We received sequences of about 200bp from PCR. Pyrosequencing was carried out on a 454 Genome Sequencer FLX System using the Titanium chemistry (454 Life Science, Branford, CT, USA) with 50–100 amplified samples placed onto one 454 sequencing platform, respectively.

Bioinformatics

For the bioinformatics analyses, we used a standardised pipeline including (i) quality filtering, (ii) clustering and (iii) taxonomy annotation. Low-quality tails were removed from the reads and trimmed reads with an average Phred quality score less than 25 were discarded. Additionally, we removed all reads with at least one base with a quality of less than 15 and also all reads that contained errors in the primer regions. Multiplex Identifiers were used to separate the different samples. Chimeras were removed using UCHIME (USEARCH v.6) (Edgar ). Subsequently, the sequences that passed the quality filtering were clustered into Operational Taxonomic Units (OTUs) with UCLUST version 6 (Edgar, 2010) at 97% sequence identity for further analyses. Sequences were sorted by abundance beforehand clustering and clustering was performed for all sequences of all sampling campaigns at a time (that is, not individually by campaigns). For all OTUs, we used BLASTn version 2.2.25+ (Altschul ) with the nucleotide collection database nr/nt and the NCBI Taxonomy Database (http://www.ncbi.nlm.nih.gov/taxonomy; RefSeq Release 57 (14 January 2013)) to annotate the OTUs with taxonomic information. For reasons of comparison, we also processed our data with the CANGS pipeline (Pandey ) and received similar results.

Taxonomic analysis of the sequence data set

OTUs and their reads from bioinformatic filtering and assembling were used to analyse the community structure and distribution patterns of OTUs (abundant vs rare, habitat-wise, and taxon-wise). For analyses on higher taxonomic group level, the following taxa were selected: Ciliophora, Dinophyceae, Apicomplexa, Alveolata rest, Bicosoecida, Oomycetes, Chrysophyceae, Synurophyceae, Bacillariophyta, Stramenopiles rest, Cercozoa, Rhizaria rest, Heterolobosea, Euglenida, Kinetoplastida, Euglenozoa rest, Excavata rest, Choanoflagellida, Chytridiomycota, Microsporidia, Ascomycota, Basidiomycota, Glomeromycota, Amoebozoa, Apusozoa, Glaucocystophyceae, Rhodophyta, Viridiplantae without Embryophytes, Cryptophyta, Katablepharidophyta and Haptophyta. OTUs affiliated with metazoa and embryophytes (as multicellular organisms), as well as with Bacteria and Archaea, have been excluded before further analyses. Thus, only protists were analysed since they were the focus of our investigation. The ‘rest' categories within the analysis are meant to collect minor groups and OTUs without further affiliation within Alveolata, Stramenopiles, Rhizaria and Excavata in order to fully show the protistan diversity present in the samples. Eukaryotic OTUs that are not met by one of the given taxonomic groups or have not been deleted from the data set for the given reasons are shown as ‘others'. Furthermore, OTUs that could not be affiliated to entries in the NCBI database are labelled as ‘unknown'.

Statistical analysis of the sequence data set

The specific nature of the data matrix that results from deep-sequencing diversity counting is characterised by a multitude of zero and single counts of OTUs (zero: 98.35%, single counts: 0.83%, two counts: 0.27%, ten counts: 0.014%, etc.) not fitting traditional statistical methods of diversity comparison. Therefore, we used a two-step standardisation of site data; first, rarefaction of the sites was done using the ‘drarefy' function provided by the VEGAN R package (Oksanen ), followed by Hellinger transformation (Legendre and Gallagher, 2001). After application of these procedures, the resulting pretransformed data matrix of OTUs can be submitted to Ward cluster analysis and further multivariate analyses (see Figures 1a and b; Legendre and Gallagher, 2001; Borcard ). Cluster analysis could be calculated on both the OTU and the meta-group level (for chosen meta-groups see above). The principle component analysis based on Hellinger distances shows those taxonomic groups with most explanatory value for clustering and separation of samples (see Figure 1b).

Figure 1

Results of cluster analysis of sites. (a) Bars (left to right) are ordered by clusters from cluster analysis which almost entirely fit the actual sampling campaigns (see cluster names); black arrows at bottom indicate those sites within clusters not identified with the respective sampling campaign; within clusters, order (left to right) is by species richness of identified taxonomic groups, respectively; species richness is shown as height of bars out of 100; community composition is shown as colour of bars by identified taxonomic groups (bottom) and merged or unidentified reads (top) (procedure: drarefy, see Materials and methods). (b) PCA showing clusters from cluster analysis (encircled) and significant taxonomic groups for sorting of sites (arrows) (pretreatment of data: drarefy+Hellinger transformation, see Materials and methods); PCA was calculated on the basis of percentage of taxonomic groups within sites. Abbreviation: PCA, principle component analysis.

As reads per sample differ decisively between samples, rarefaction values at a read level of 100 reads (including extrapolation for a few samples, see Colwell ) were used to compare species richness and composition of OTUs among samples. For the shown composition within each sample, the ‘drarefy' function (R package VEGAN; Oksanen ) was used to minimise the bias in the rarefaction analysis. The more classical ‘rrarefy' is not suitable here, as recently pointed out by McMurdie and Holmes (2014). Thus, more exact relationships of meta-groups from higher read levels are kept intact (see Figure 1a); however, this could possibly result in OTU percentages per meta-group of less than 1 at a read level of 100. The community structures of the different samples were analysed using the mean relative abundance of meta-group reads per sampling, which showed significant differences between meta-groups among sampling campaigns (see Figure 2, also compare Supplementary Table S2).

Figure 2

Bars in individual graphs showing percentage of indicated taxonomic group in sampling campaigns (bars on the x-axis=campaigns, respectively, from left to right: 1—Soil, 2—Borehole, 3—Biofilm, 4—Whale fall, 5—Estuarine, 6—Lake District 1, 7—Lake District 2, 8—Alpine Lakes, 9—Winterpico, 10—HNF). Columns (a–f) sorting out taxonomic groups with highest percentage in (a) Soil, (b) Borehole, (c) Biofilm, (d) Whale fall and Estuarine (salt and brackish waters), (e) Lake District 1 and 2 and Alpine Lakes (freshwater) and (f) Winterpico and HNF (size-fractionated samplings). Abbreviation: HNF, heterotrophic nanoflagellates.

Weibull distribution in protistan OTUs

Following preliminary tests, the distribution patterns of individual OTUs did not follow a Poisson distribution (P<0.05) or a normal distribution (Ricci, 2005). Although not common, a two-parameter Weibull density distribution, which is used in survival and engineering analyses, fitted our data. Recently, a three-parameter Weibull empirical cumulative distribution function was used for fitting diversity curves (Livermore and Jones, 2015), showing the applicability of the Weibull distribution in ecology. For our data, the fitting of the Weibull distribution among sites (y-axis=no. of sites, x-axis=abundance of OTU in these sites) passed the KS-test (that is, the null hypothesis could not be rejected, P-value: >0.05) for almost all OTUs and was consequently used as the distribution for further analysis. All KS-tests were performed with R, referring to Birnbaum and Tingey (1951). All OTUs passed the KS-test (that is, the Weibull distribution could not be rejected with all P-values >0.9) when P-values were corrected for multiple testing by using p.adjust (method= ‘fdr', R package ‘stats'). When corrections for multiple testing were disregarded, the KS-test rejected the Weibull distribution for 10 out of 1249 OTUs. Even though these 10 OTUs passed the test when correctly applied (that is, corrected for multiple testing), we decided to mark these OTUs in the figures (Figures 3a and b, top). The shape parameter k of the Weibull distribution is suitable for comparing the distribution of abundant and less abundant taxa as the total abundance primarily affects the scale, but not the shape parameter. Therefore, we focus on the comparative analysis of the parameter k. For k=1, the two-parameter Weibull distribution switches to that of the exponential distribution, for k ~3.6 it switches to a Gaussian distribution. The k-values in this study all lie between 1 and 3.6 forming curves of positive skewness. The skewness of the curves as a function of k indicates the value of evenness of OTU reads over sites and is, therefore, suitable of showing distribution patterns of taxa. Shape values k of Weibull distributions were compared using Mann–Whitney U-tests. All U-tests were performed with Sigmaplot 12.5 (Systat Software Inc., Erkrath, Germany) referring to Mann and Whitney (1947). The control data set, tested against Bacillariophyta as the closest subgroup data set within the analysed data, was generated by the function ‘rnorm' with n=55 (normally distributed), average value and median=1 (Ricci, 2005), and with the same standard deviation value as calculated for the Bacillariophyta data set. We also provide a graph with exemplary Weibull curves of typical OTUs in the Supplementary Information on methods.

Figure 3

Distribution patterns as functions of the log abundance of OTUs in the samplings, OTUs are also sorted (left to right) by read abundance (logarithmic scale). The grey line shows the relation between read abundance and the number of sites where the distinct OTU occurs. (a) Weibull shape of individual OTUs (=distribution of reads over sites) as red and black dots for all Eukaryota (top), OTUs occurring in freshwater (middle) and OTUs occurring in soil (bottom); x indicate the 10 OTUs for which Weibull distribution was rejected when corrections for multiple testing were disregarded (P<0.05, see Materials and methods); for OTUs occurring in less than 15 sites, a calculation of Weibull shape is not possible (unreliable), these dots were set to 0 instead of filling the lower right part of the plots; OTUs are sorted (left to right) by read abundance (logarithmic scale); the grey line (for values see the right y-axis) showing the connection of read abundance and presence in sites; the red dotted line showing the median shape value of all Eukaryota; black dotted lines showing the median shape value of freshwater and soil, respectively. (b) Graphs showing OTUs as a function of log of occurrence in sites and log of abundance of reads (left half of the figure) and as their Weibull shape value (=distribution of reads over sites) (right half of the figure) for all Eukaryota (top—red dots), Amoebozoa (upper middle—black dots), Chrysophyceae (lower middle—black dots) and Bacillariphyta (bottom—black dots); the 10 OTUs for which Weibull distribution was rejected (P<0.05, see Materials and methods) are indicated by an x; the dotted black vertical line in the left-hand graphs indicates the read abundance threshold up to which Weibull shape value calculation (right-hand graphs) was carried out (also see the x-axis); additionally, for OTUs occurring in less than 15 sites, a calculation of Weibull shape was not possible (unreliable), these dots were set to 0 instead of filling the lower right part of the plots.

Results

The large-scale analysis undertaken herein—carried out as a multiple sample habitat comparison—showed that protistan communities differ decisively. They did so in terms of OTU presences in a distinct sample and habitat as well as in the ratio of higher taxonomic groups (meta-groups) (Figure 1a). Taking all samples into account, Ciliophora were dominating in terms of d-rarefied sequence abundance at 13% followed by Bacillariophyta at 9.7% and Chrysophyceae at 9%.

Community composition patterns

Despite many OTUs being distributed across different campaigns and sample types, many OTUs were either specific for distinct habitats/campaigns or their relative abundance deviated considerably from that in other habitats/campaigns. Cluster analysis based on protistan sequence analysis separated different habitat types. Specifically soil, biofilm and brackish sediments (estuarine), and freshwaters were distinguished and each showed a specific community composition (Figure 1b). Size-fractionated samples, such as the heterotrophic nanoflagellates data set, were also separated by cluster analysis.

This differentiation based on OTUs was also reflected by shifts in the community composition on the level of higher taxonomic groups (Figures 1b and 2). Substrate-bound taxa were, as expected, specifically abundant in sediments. However, while Amoebozoa and Cercozoa were specifically more abundant in the soil samples (see column A in Figure 2), Heterolobosea, Kinetoplastida, Apusozoa and to some extent Chrysophyceae were specifically abundant in the aquifers of the borehole samples (see column B in Figure 2). The Peridiosporomycota (Oomycetes) were found to be relatively abundant in both habitats. In the size-fractionated samples, small heterotrophic nanoflagellates taxa showed relatively high sequence abundances, specifically Chrysophyceae, Katablepharidiophyceae, Choanomonada and Bicosoecida (see column F in Figure 2). In our biofilm samples, we found an increased sequence abundance of Chlorophyta and streptophytic algae (labelled as ‘Viridiplantae' without Embryophytes), Cryptophyta, Bacillariophyta and Chytridiomycota. Substrate-bound taxa did not have particularly high sequence abundances in the biofilm samples (see column C in Figure 2). The marine samples showed comparatively low read counts of Chlorophyta, streptophytic algae, Chytridiomycota and Cryptophyta, but comparatively high reads of Bacillariophyta in the salinity gradient along the Colne estuary, and of Dinophyceae and Euglenida at the whale fall site (see column D in Figure 2).

These habitat-specific community profiles were similar for almost all samples of the respective habitat types. Cluster analysis revealed only very few samples that would be sorted into the wrong campaign based on sequence data (see arrows in Figure 1a).

Differential distribution patterns of protistan taxa

For the analysis of OTU distribution patterns among sites, we analysed the shape parameter k of the Weibull distribution (see the Materials and methods section). This analysis only revealed a minor modulation in the shape factor of the Weibull distribution between abundant and less abundant taxa. The mean Weibull shape value k is rather similar irrespective of the relative overall abundance of the OTUs (Figure 3a, top). Shape values of individual OTUs scatter around a median of 1.67, but there was no distinct change in the median k for less abundant OTUs. The median shape value of 1.67 indicates that taxon distribution is not due to unlimited random dispersal (which would favour a value of 1 or lower, that is, a colonisation from one site with high frequency), but on the contrary that some factors work against invasion and/or colonisation of habitats (significant U-test of shape value 1.67 vs 1, P-value: <0.001). One might argue that this effect is caused by the different habitat types included in the analysis. However, even when the analysis is restricted to one habitat type within a distinct geographic region, taxon distribution cannot exclusively be explained by unlimited random dispersal (for example, soil sites UK: shape parameter=2.099, significant U-test vs 1, P-value: <0.001; for example, Austrian lakes+Lake District 1: shape parameter=1.711, significant U-test vs 1, P-value <0.001; compare methods in this paper and Figure 3a—mid and bottom). In contrast, the shape parameter is even higher when the analysis is restricted to soils, indicating that random dispersal is even less important in soils compared with aquatic habitats. However, the scatter around the median k increases for rare taxa (Figure 3b, top right). Similarly, the scatter of sites of occurrence as a function of total read number increases (Figure 3b, top left). In contrast, the most abundant taxa do have very similar distribution patterns, indicated by both a very similar shape parameter in the Weibull distribution (Figure 3b, top right) and a narrow range of the ratio of number of sites vs total read abundance (Figure 3b, top left). Both observations indicate a stronger differentiation of distribution patterns in rare taxa, whereas abundant taxa show relatively uniform distribution patterns. Owing to this, within the rare taxa, a differentiation into more generalist taxa (occurring at a relatively high number of sites) and more specialised taxa (restricted to a few sites) is pronounced.

Similarly, distribution differs between taxonomic groups. Diatom OTUs, as one extreme, are characterised by a comparatively low k and occur at a relatively low number of sites for a given total read count. Amoebozoa, on the other hand, are characterised by comparatively high shape values and occur at a relatively high number of sites for a given total read count. The distribution pattern of Amoebozoa indicates a broader (more generalist) distribution, whereas the Diatom distribution pattern reflects a narrower (more specialised) distribution (significant U-test of shape values for Amoebozoa vs Bacillariophyceae, P-value: <0.001). The Chrysophyceae are somewhat in between the two strategies. Nevertheless, k is significantly larger as would be expected for random distribution for all taxa (Amoebozoa: shape parameter=2.139, significant U-test vs 1, P-value: <0.001; Bacillariophyceae: shape parameter=1.449, significant U-test vs 1, P-value: <0.001; Chrysophyceae: shape=1.595, significant U-test vs 1, P-value: 0.009) (Figure 3b, right and left).

Non-random distribution patterns are also supported by principle component analysis (Figure 1b). Based on the molecular diversity, the samples clustered according to habitat types. Nevertheless, within habitat types, regional effects were pronounced. Samples from the same sampling site clustered together irrespective of seasonal effects. Furthermore, samples from the same region (for example, from Austria) mostly clustered together.

Discussion

Here, we address the generalisability of protist distribution patterns. We demonstrate that protist taxon distribution varies depending on habitat types, taxonomic groups and taxon abundance.

With our analysis, we offer a large-scale and cross-biome investigation of protist diversity and distribution. The standardised comparison across diverse habitats is, to our knowledge, unique to this study and includes standardised nucleic acid processing, sequencing, bioinformatics, and statistical analyses of sediment, biofilm and water samples of diverse locations. Owing to the approach of comparing different habitat types, sampling methods had to be altered. Standardisation is otherwise complete and, therein, goes beyond comparative platform approaches (Caporaso ; Mahé ). Further, even though an unbiased view on protist diversity is difficult to achieve (compare Supplementary Information on methods), the standardisation over all 232 samples of the analysis allows for addressing overarching patterns of protist diversity in a comparative approach.

General patterns of dispersal limitation

Protist distribution and potential dispersal limitations have been in the focus of protist ecology and diversity research since the early observations. The scientific dispute can be traced back at least to de Candolle (1820) and Ehrenberg (1849, 1850). With the catchy phrase ‘Everything is everywhere, but the habitat selects' Baas-Becking formulated the central idea behind the ongoing dispute (Bass and Boenigk, 2011; Boenigk, 2014). In this form, the idea of global dispersal was revived in the 1980s, resulting in the recent debate on protist dispersal and endemism as summarised by O'Malley (2007), de Wit and Bouvier (2006) and Bass and Boenigk (2011). It is now generally believed that protist distributions cannot be explained exclusively by unlimited dispersal (Foissner, 2006; Bass and Boenigk, 2011), expressed, for instance, by the moderate endemicity model (Foissner, 1999, 2004). However, microbial organisms may, in general, show wider distribution patterns as compared with higher organisms (Livermore and Jones, 2015). We demonstrate that the distribution patterns of individual protistan OTUs, specifically as expressed by the shape parameter of the Weibull distribution, violate the assumption of unlimited dispersal. This is consistent with findings of Livermore and Jones (2015) who demonstrated for bacteria that microbial diversity patterns non-randomly vary between ecosystem types. Our analysis further shows that patterns of non-random dispersal are stronger in soil habitats than aquatic habitats.

Consistency of distribution patterns within habitat types

Molecular data produce strong protistan community signatures, that is, habitats being characterised by distinct protistan communities (Bates ; Tedersoo ; de Vargas ). Consistent with earlier findings from microscopy and Sanger sequencing studies (Fenchel, 1994; Anderson, 2012; Tikhonenkov ), soil and aquifer protist communities are dominated by substrate-bound taxa (Foissner, 1991; Novarino et al., 1997; Adl, 2003; Adl and Coleman, 2005; Risse-Buhl ). Likewise, the marine samples showed a high proportion of sequences affiliated with diatoms, dinoflagellates and euglenids (Ingmanson and Wallace, 1995; Graham ), whereas Synurales, Cryptophytes, Chlorophyta and Chytridiomycota were particularly prominent in freshwater samples (Dokulil ; Graham ).

The consistency of distribution patterns with respect to habitat type was also reflected at the level of individual OTUs. Principle component analysis, as well as cluster analysis, revealed similar community profiles for samples originating from similar habitats. These community profiles were highly specific for distinct habitat types. Even seasonal variation did not blur the distinctiveness of community structure. Therefore, community profiles based on high-throughput sequencing data (in this case the 18S rDNA V9 region) appear highly promising for overarching habitat comparisons as well as for long-term monitoring studies.

The relatively uniform distribution pattern of abundant taxa is contrasted by a high variation in rare taxa

Molecular studies typically report a high fraction of rare taxa. The high number of rare taxa led to the assumption that rarity might be an evolutionarily advantageous trait in these organisms (Logares ) and favoured speculations on potentially deviating distribution patterns between rare and abundant taxa (for example, Nolte ). In contrast to these earlier studies, we show that rare and abundant taxa have generally similar distribution patterns. However, whereas the distribution patterns of individual taxa are relatively uniform for abundant OTUs, the deviation among individual rare OTUs is high. Further, the proportion of rare and abundant taxa differs distinctively by habitat type: the soil samples are characterised by a high evenness, that is, a multitude of OTUs appear at higher abundances in many of the samples. On the contrary, aquatic sites have a comparatively lower evenness with usually only a small number of dominant species reaching higher abundances, accompanied by many rare taxa, at a distinct point in time. Thus, in addition to taxonomic profile/community structure differences between habitats, abundance and diversity patterns appear to be characteristic for different habitat types.

Differences in distribution patterns indicate differential niche widths between taxa (Boenigk ; Bass ). In general, broad niche widths and high ecophysiological tolerances should be reflected by broad distributions of taxa across different samples. In contrast, taxa with more restricted niche widths and/or ecophysiological tolerances should be restricted to fewer samples. The largely uniform distribution patterns for abundant OTUs indicate that dispersal and success of colonisation is largely comparable between taxa, resulting in similar distribution patterns. Within the rare biosphere, however, distribution patterns vary strongly between individual taxa, presumably indicating a more selective niche adaptation. Thus, we suspect that the rare biosphere offers an even stronger potential use for comparative habitat analyses, including biological monitoring, as compared to the more abundant taxa, which are usually the focus of such studies.

Differential patterns between taxonomic groups

With respect to taxonomic groups, amoebozoan taxa (that is, amoebozoan OTUs in our analysis) are often characterised by broader distribution patterns, indicating a more generalist strategy, whereas the diatom taxa (that is, OTUs) are often characterised by narrower distribution patterns, indicating a more specialised strategy. The Chrysophytes, for instance, show intermediate distribution patterns. These differences in distribution presumably indicate a differential dispersal and proliferation among taxa even on higher taxonomic levels, including not all, but most OTUs of a higher taxonomic group. With regard to taxonomy, it seems plausible that distributions are markedly different between soil-inhabiting and endobenthic taxa, such as Amoebozoa, which are presumably less prone to random dispersal, and planktonic or epibenthic taxa.

Conclusions

With our analyses, we show that limited dispersal and distribution in protists differ by habitat type as well as by taxonomic group. Further, rare and abundant taxa do not show generally different patterns of distribution. The variation of distribution patterns in rare taxa is, however, much more pronounced making the rare biosphere as revealed by molecular investigations a promising candidate for comparative habitat analyses. For analyses as undertaken herein, standardisation of a large number of samples is essential and can be provided by large-scale, methodologically unified data sets and parallel investigations made possible by high-throughput methods. Such investigations are more directly comparable than data sets generated by different ‘wet' laboratory conditions (PCR protocols, cycling conditions, reagents, etc.) and different bioinformatic methods. This unified/standardised approach is recommended for revealing robust and fine-scale differences between community structures and distributions of taxa that might be missed or skewed by non-standardised comparisons. This is important for further illuminating the biodiversity and biology of protists on Earth and thus broaden our knowledge on these and other under-researched micro-organisms.