Study site and soil sampling

Projected climate change for Denmark in the 2100th century indicates drier summers, which are experimentally simulated on the CLIMAITE study site (Larsen et al., 2011). The CLIMAITE experimental site is situated in a dry heath-/grassland 50 km NW of Copenhagen, Denmark (55° 53′ N, 11° 58′ E). The mineral fraction of the soil consists of 92% sand, 5.8% silt and 2.2% clay (Nielsen et al., 2009). The site is well drained with an organic top layer (O-horizon). The pH CaCl2 in the O-horizon is 3.3 increasing to 4.5 in the lower B-horizon. The dominant vegetation consists of the dwarf shrub Calluna vulgaris (c. 30% cover) and the perennial grass Deschampsia flexuosa (c. 70% cover). The annual mean temperature is 8 °C with a mean precipitation of 613 mm (Danish Meteorological Institute, www.dmi.dk). Since 2005, a complete three-factorial treatment with increased CO 2 , temperature and summer drought has been maintained in 12 four-chamber octagons (7 mm in diameter), where the fourth octagon is a control plot with no treatments. Each treatment is replicated seven times. These treatments are intended to mimic the projected climate change for the region in 2075. Drought is induced once or twice a year by automatic rain shelters, which exclude the precipitation continuously for 2–5 weeks until the water content plunges below 5% by volume in the upper 20 cm of the soil. Larsen et al. (2011) provide a detailed description of the experiment.

In November 2010, we sampled topsoil (0–8 cm, O-horizon) from two of the six drought plots 1–2 (Dry1) and 3–1 (Dry2) and from two of the control plots 9–1 (Control1) and 11–4 (Control2). To minimise the spatial variation, we pooled and mixed three subsamples from each plot for the subsequent analyses.

Comparisons of hypervariable regions

The entire 18S region typically spans a length of ~2000 bp, whereas HTS methods have not been able to amplify >500–700 bp at most. Therefore, it is necessary to identify good representative parts of the 18S region for HTS analyses. Prime candidates are the eight hypervariable regions labelled V1–V5 and V7–V9 (the V6 hypervariable region found in bacterial 16S is absent from eukaryotic 18S (Howe et al., 2011). As the V4 region is the longest of the eight hypervariable regions, we found it the most attractive for taxonomic annotation. However, we wished to make sure that the diversity in V4 correlated reasonably well with the total 18S diversity as compared with the other hypervariable regions.

An appropriate clustering level-threshold constitutes another special problem in HTS analyses. Thus, it is necessary to choose a percentage-wise separation threshold when clustering the obtained sequences into OTUs. To obtain a robust theoretical foundation for HTS analysis, we first obtained two data sets of named cercozoan 18S Sanger-generated sequences. In June 2015, we obtained one set of 63 species from GenBank. This set contained the whole 18S region including all eight hypervariable regions in their entirety. We used the 63 sequence set to identify the V4 region as the best representative of the entire 18S diversity. The other, larger set consisted of 193 partial sequences that were at least 1500 bp long and all contained the V4 region. We included only named species documented in recent papers (Ekelund et al., 2004; Hoppenrath and Leander, 2006; Lara et al., 2007; Wylezich et al., 2007; Bass et al., 2009a, b; Burki et al., 2010; Chantangsi and Leander, 2010a, b; Heger et al., 2010; Heger et al., 2011; Howe et al., 2011; Yabuki and Ishida, 2011). This sequence set represented all nine cercozoan classes sensu Cavalier-Smith and Chao (2003), and contained no duplicate names or synonyms. We used this set of 193 sequences to evaluate the interspecific distances in the 18S region most suitable for separation of OTUs in the Cercozoa, and to test the effects of different OTU separation thresholds. Names and accessions numbers of the 63+193 sequences are listed in Supplementary data (tables 1 and 2).

In order to precisely identify the position of the hypervariable regions in the Cercozoa, we use the E-ins-i algorithm in MAFFT (Katoh and Standley, 2013) to align the 63 sequences along with the complete sequence of the fungus AF258606 Scytalidium hyalinum, which had the start and end of each V1–V9 regions fully annotated in its documentation on GenBank. Using the command dist.seqs in MOTHUR (Schloss et al., 2009), counting all indels as one event without penalising end gaps, we then calculated all possible uncorrected P-distances between the sequences for the whole 18S and for each of the eight hypervariable regions V1–V9, and correlated these P-distances for each region with the complete 18S. In this manner, we tested how well the genetic diversity of the respective regions correlated with that of the complete 18S region. All graphics and statistics were done in R (Ihaka and Gentleman, 1996).

Sequence variation between known species

To examine the congruence between already described Cercozoa and the genetic distances in V4, we aligned the data set of 193 sequences with MAFFT using the E-ins-i algorithm and calculated the uncorrected P-distances for all 18 527 pairs in MOTHUR. We then clustered the V4 region of these 193 sequences with the furthest neighbour-algorithm (implemented in MOTHUR) for all thresholds between 0 and 10%.

DNA extraction, primer design and initial cloning check

We extracted DNA from 0.5 g of fresh soil within 24 h of soil sampling. We used a genomic mini spin kit for universal DNA isolation (A&A biotechnology, Gdynia, Poland) with a standard protocol (Yu and Mohn, 1999). Based on the 193-sequence alignment, we designed primers that would amplify the majority of named key soil cercozoan genera within Granofilosea, Imbricatea, Cryomonadida, Cercomonadida, Glissomonadida and Euglyphida with no—or in some cases one—mismatch in the primer sequence. We accepted a slight bias against some members within these taxa, and against some genera, for example, Cyphoderia, Platyreta, Arachnula and Filoreta (two mismatches) and some bias against Chlorarachniophyta, Phytomyxea and Ascetosporea (notably Haplosporida and Mikrocytida) and other endomyxan lineages, and the genera Helkesimastix, Sainouron, Cholamonas (Cavalier-Smith et al., 2009), Reticulamoeba (Bass et al., 2012), and Rosculus and Guttulinopsis (Bass et al., submitted) with three or more mismatches in each primer.

From the alignment of all these genera, we used the representative sequence AF411270 Cercomonas longicauda as template in Primer3 (Rozen and Skaletsky, 2000) to find two compatible primers: the forward primer Cerc479F (5’TGTTGCAGTTAAAAAGCTCGT-3’, Tm=57.8 °C) and the reverse primer Cerc750R (5’TGAATACTAGCACCCCCAAC-3’, Tm=57.5 °C). To check the specificity of the primers, we performed an initial PCR and cloning of 50 sequences. The PCR master mix (25 μl) consisted of 1 × High Fidelity buffer (Invitrogen, Carlsbad, CA) with MgCl 2 , 0.25 mM deoxynucleotides mixture, 1 μl 100 × bovine serum albumin, 0.5 IU Phusion Hot Start DNA polymerase (5 units μl–1, Invitrogen 0.4 μM) of each primer, 1 μl DNA template. The PCR incubation conditions consisted of an initial denaturation step of 94 °C for 5 min; 30 cycles of denaturation at 94 °C for 60 s, annealing at 55 °C for 60 s and elongation at 68 °C for 60 s; and finally, an extension step of 72 °C for 7 min. We chose to lower the annealing temperature from the theoretical optimum of the primers to compensate for the mismatches. Cloning was performed using TOPO TA Cloning Kit from Invitrogen, and sequencing of 50 supposedly positive clones from this PCR was done by MACROGEN in Seoul, South Korea.

DNA amplification and GS-FLX Pyrosequencing

The samples were prepared for GS-FLX pyrosequencing in a two-step PCR. We used a Platinum Taq DNA High Fidelity polymerase (5 units μl−1, Invitrogen); otherwise the master-mix and the PCR incubation conditions were as above. To eliminate as many primer-dimers as possible, the products were incubated at 70 °C for 4 min and then stored immediately on ice before electrophoresis. We loaded the PCR products on a 1% agarose gel with ethidium bromide, which confirmed the presence of a single band in the desired length of ~250–300 bp with ultraviolet illumination. The bands of PCR products were excised from the agarose gel and purified by the Montage DNA Gel Extraction kit (Millipore, Bedford, MA).

The second PCR amplification was performed with fusion primers consisting of the raw primers above with the B-adaptors and four MID-tag barcodes of 10 bp added upon the forward primer and was amplified using only 15 PCR cycles. Otherwise, PCR incubation conditions, electrophoresis, gel excision and purification were as above. The amplified DNA from the second PCR was quantified with the Qubit dsDNA HS Assay Kit and the Qubit fluorometer (Invitrogen, Life technologies, Carlsbad, CA, USA) and mixed in approximately equal molar concentration (5 × 106 copies μl−1) to ensure an approximately equal representation of sequences on each sample. A GS-FLX Titanium sequencing run was then performed on a 70_75 GS PicoTiterPlate (PTP) using a GS-FLX Titanium pyrosequencing system according to manufacturer instructions (Roche Diagnostics, Basel, Switzerland) at the National High-throughput DNA Sequencing Centre (Copenhagen, Denmark).

Bioinformatic analyses

In several taxonomic groups, including Rhizaria, error rates primarily linked to singletons and homopolymers may cause a considerable overestimation of diversity; especially in the V4 compared with the V9 region. GS-FLX Titanium was particularly susceptible to such errors compared with the GS-FLX standard kit, even when reads are clustered up to a 3% level (Behnke et al., 2011). Hence, to eliminate such errors, we applied a strict quality sorting approach in our analysis; we eliminated singletons and long homopolymers, and chose a conservative 5% clustering threshold.

The titanium run produced 689 988 reads. We analysed it through the Qiime pipeline (Caporaso et al., 2010) and discarded all reads that had a quality score below 25 or had any mismatches in the primer or MID-tag sequences. We also discarded reads with a length outside 200–1000 bps, as the shortest cercozoan among the 193 named species had a V4 length of 218 bp. This left 494 963 reads, which were run through ACACIA (Bragg et al., 2012) to discard homopolymers >6 bp. Chimeras were removed with UCHIME (Edgar et al., 2011). This removed further 22 254 and 12 310 reads, respectively. The rest were clustered at 5% with UCLUST (Edgar, 2010), and 838 post-clustering singletons were subsequently discarded. Representative sequences from the resulting 1745 OTUs were blasted using nblast (Altschul et al., 1990). We removed another 160 OTUs that either had no BLAST hit (two OTUs), were presumed chimera with different BLAST hits of the 5’ and 3’ end (2), had top hits to non-target organisms (17 fungi, 2 ciliates, 1 heterokont), or had a query coverage of 60% or below (136 OTUs); and thus perhaps were chimeras. All the rest blasted to Cercozoa with a similarity of 80% or more to the most similar hit in GenBank. The final data set consisted of 443 350 sequences, distributed on the plots with 85 305 from Control1, 94 975 from Control2, 116 388 from Dry1 and 146 682 from Dry2. To obtain comparable data for rarefaction curves and statistical tests, we further resampled down to 85 305 sequences per plot. The data (the sff file) and barcode information has been archived on GenBank in the Sequence read Archive under the experiment number SRX1054896.

Taxonomic affiliation of OTUs

In some groups of organisms, an argument for choosing a particular clustering threshold can be made by identifying a ‘barcoding gap’ (or barcoding window); that is, a gap between non overlapping distributions of taxa. This approach has, for example, been used in several fungal groups (Frøslev et al., 2007; Jargeat et al., 2010; Harder et al., 2013). Unfortunately, our analysis of the 193 sequences assigned with a name shows that no such barcoding gap exists in Cercozoa (see also results and discussion), as the distribution of the interspecific diversity extends continuously all the way down to 0%. Hence, to eliminate as much artefactual diversity as possible without over-compromising phylogenetic resolution, we chose a conservative 5% level for OTU separation. We first blastn-searched the remaining 1143 OTUs locally against a custom cercozoan database (David Bass, in preperation). This enabled us to group them roughly into higher level groups approximating to Order/Class. We then used GenBank to blastn-search for representatives for these higher level groups. The top blastn hits, including as many named or otherwise characterized database sequences as possible, were retrieved and aligned with the OTUs generated in this study using the E-ins-i algorithm in MAFFT (Katoh and Standley, 2013) and phylogenetically analysed using RaxML BlackBox (Stamatakis et al., 2008). We constructed an ML tree in RAxML BlackBox (using the GTRGAMMA) of the HTS sequences within a set of longer 18S reference sequences using the approach described in Dunthorn et al. (2014). An approach using only V4 for the whole analysis gave a similar result but with less backbone resolution because this approach removes informative data from the analysis.

We used the resulting trees to assign OTUs as far as possible to named genera, higher level groups or environmental clades. We used a similar taxonomic framework as the one used in several recent studies of cercomonads, glissomonads, Granofilosea and other rhizopodial forms, as well as Cercozoa in general (Bass et al., 2009a, b; Howe et al., 2009, 2011). We included the % sequence identity in the OTU name to indicate the degree of similarity of the OTU to the best-matched database sequence. We made no attempt to assign any OTUs to species level. This may be possible for 100% complete 18S ribosomal DNA reads, however, in the vast majority of cases it would be misleading to imply such a high resolution. We considered OTUs assigned to the orders Euglyphida, Cryomonadida, and Thecofilosea, and the genera Trinema, Rhogostoma, Corythion, Cyphoderia, Ovulinata, Euglypha, Trachelocorythion, Assulina, Pseudodifflugia, Tracheleuglypha and Ebria as testate.