Evaluating High-Throughput Sequencing Data of Microalgae Living in Melting Snow: Improvements and Limitations

Abstract

Melting snow fields are an extremophilic habitat dominated by closely related Chlamydomonadaceae (Chlorophyta). Microscopy-based classification of these cryophilic microalgae is challenging and may not reveal the true diversity. High-throughput sequencing (HTS) allows for a more comprehensive evaluation of the community. However, HTS approaches have been rarely used in such ecosystems and the output of their application has not been evaluated. Furthermore, there is no consensus on the choice for a suitable DNA marker or data processing workflow. We found that the correct placement of taxonomic strings onto OTUs strongly depends on the quality of the reference databases. We improved the assignments of the HST data by generating additional reference sequences of the locally abundant taxa, guided by light microscopy. Furthermore, a manual inspection of all automated OTU assignments, oligotyping of the most abundant 18S OTUs, as well as ITS2 secondary structure analyses were necessary for accurate species assignments. Moreover, the sole use of one marker can cause misleading results, either because of insufficient variability within the locus (18S) or the scarcity of reference sequences (ITS2). Our evaluation reveals that HTS output needs to be thoroughly checked when the studied habitats or organisms are poorly represented in publicly available databases. We recommend an optimized workflow for an improved biodiversity evaluation of not only snow algal communities, but generally ‘exotic’ ecosystems where similar problems arise. A consistent sampling strategy, two- molecular marker approach, light microscopy-based guidance, generation of appropriate reference sequences and final manual verification of all taxonomic assignments are highly recommended.

Introduction

Extremophilic microalgae cause a distinct colouration of the melting snow from green to different shades of yellow, orange and red in alpine and polar regions (Kol 1968, Hoham and Duval 2001, Komárek and Nedbalová 2007, Remias 2012, Leya 2013, Anesio et al. 2017). Snow algae have evolved a range of adaptive strategies to overcome a multitude of environmental stresses including low temperatures, freezing, desiccation, nutrient scarcity and extreme irradiation. Thus, they are of general interest to studies of a wide range of fundamental cellular processes. These eukaryotic photoautotrophs mostly belong to the Chlamydomonadaceae (Chlorophyceae) and their carotenoid-rich immotile stages (cysts), which are adapted to these harsh conditions, are predominately found throughout the melt season (Remias et al. 2010).

Traditionally, the red snow phenomenon has been associated with Chlamydomonas nivalis, which has to be regarded as a collective taxon (Kol 1968). Yet, a plethora of further species can be found in melting snow including Chlainomonas sp. (Remias et al. 2016), Chloromonas nivalis (Remias et al. 2010, Procházková et al. 2018b) or Chloromonas brevispina (Matsuzaki et al. 2015). Nonetheless, we are likely only scratching the surface of snow algal diversity and the above mentioned species constitute a small proportion of the true diversity. The establishment of new strains is very time-intensive. Attempts to induce germination of cysts collected in the field has not been successful in many cases and culturing may also induce pleomorphism. Moreover, cells of one species transgress through a variety of morphological and physiologic changes during the melt season. This poses another challenge for the microscopy-based identification and classification.

In contrast, high-throughput sequencing (HTS) of field samples allows for a comprehensive assessment of the microbial community composition of a natural ecosystem. With its broad application in many other environments (Grossmann et al. 2016), it is striking how rarely such approaches have been used for microalgae in general (Bradley et al. 2016), and psychrophilic algae in particular. So far only a few studies have targeted snow algal communities in the Arctic (Lutz et al. 2015a, 2015b, 2016, 2017, Segawa et al. 2018), in the US (Brown et al. 2016), in Japan (Terashima et al. 2017) and in Antarctica (Segawa et al. 2018). However, HTS data on European Alpine communities is completely absent in the literature.

The nature of HTS to produce large datasets in a mostly automated way also come at a cost; that is some of the data processing steps are to a certain degree a ‘black box’. In addition, several technical biases have to be taken in account. These may include effects due to extreme GC/AT ratios (Oyola et al. 2012), the choice of primers and library preparation methods (Schirmer et al. 2015), annealing temperature (Schmidt et al. 2013), DNA polymerase (Brandariz-Fontes et al. 2015), amplicon size variability (Schirmer et al. 2015) or from the sequencing technology itself (Schloss et al. 2011). Currently, there is no clear consensus on the choice for a suitable DNA marker for phototrophic eukaryotes. Several markers, which are being employed for algal species delimitation, have been summarized in Leliaert et al. (2014). Comprehensive reference data bases exist for the 18S rDNA marker (Quast et al. 2012), which confers a higher universal applicability. However, this marker is not sufficiently variable to distinguish among closely related taxa (Hall et al. 2010). In contrast, the internal nuclear rDNA transcribed spacer 2 (ITS2 rDNA) inherits high taxonomic resolution, but also in some cases high intragenomic variation (Thornhill et al. 2007, Simon and Weiß 2008, Alanagreh et al. 2017), which may affect OTU clustering and taxonomic identification. Since the rRNA cassette can vary in copy numbers per organism and the ITS regions are free to independently drift within the same organism, a potential overestimation of OTUs may occur, especially in some fungal species (i.e., one species can split into several OTUs (Lindner and Banik 2011, Lindner et al. 2013). Nevertheless, intergenic variation of ITS2 in algae is generally considered lower compared to fungi. Moreover, there is a dearth of appropriate ITS2 rDNA references sequences (Yao et al. 2010, Buchheim et al. 2011). Hence, at least a two-marker approach is advisable (Chase and Fay 2009).

Here for the first time we evaluated the application of HTS for the characterization of snow algal communities in the extreme habitat of melting European Alpine snow fields. Samples from the Tyrolean Alps were collected during two consecutive summer seasons and were used as a case study. We combined the power of the two markers 18S and ITS2 rDNA for HTS sequencing. These data were complemented by traditional Sanger sequencing to gain more and longer reference sequences, as well as microscopic observations. In addition, all automated taxonomic assignments were manually verified by using the NCBI BLAST server, as well as oligotyping of the most abundant 18S rDNA OTUs and ITS2 rDNA transcript secondary structure analyses.

We hypothesize that the application of HTS needs to be optimized for ‘exotic’ environments where generally a limited number of reference sequences is available. This will help us to better evaluate true snow algal diversity in the future and also improve studies of other ecosystems, where similar problems arise.

Material and Methods

Field work and sample preparation

For Sanger sequencing, cells of virtually monospecific patches of snow were identified by light microscopy. Three spots were each caused by one of the locally abundant taxa: Chlamydomonas nivalis (Procházková et al. 2018a), Scotiella cryophila (Remias et al. 2018) and Chloromonas brevispina were harvested in the Kühtai region (Table 1). One Sanger sample (Chloromonas brevispina) was collected in 2016 in course of this study, and two data sets derived from previous Sanger sequencing works of the authors were used for comparisons (Table 1). For HTS, two field samples (WP79, WP99) containing mixed communities of several snow algae were harvested in the summers of 2015 and 2016. They were collected from a non-permanent, flat snow field in the Tyrolean Alps in Austria (Table 1). The site was dominated by an alpine meadow covered by a melting snow pack with characteristic reddish surface coloration. The 2016 Sanger and HTS samples (Table 1) were collected about 2 weeks earlier in the melting season than the 2015 (WP79) sample. Cell harvest was performed as previously described by Procházková et al. (2018a) using a sterilized stainless steel shovel, putting the snow into sterile sampling bags and keeping it cold/frozen until returning to the laboratories for microscopic analyses and DNA extractions. The presence of algae and the species composition were evaluated using an Evolution field microscope (Pyser SGI, USA). For HTS, we intentionally used two different sampling approaches: In 2015 (sample WP79) sampling included the complete snow column from the snow surface to the soil layer (approximately 30 cm; Fig. S1), whereas in 2016 (sample WP99) surface and ground snow were not harvested to avoid allochthonous organisms (airborne and soil alga can occur in snow, Stibal and Elster 2005).

Table 1.

Overview of snow algae samples from Kühtai in the Austrian Alps with three locally abundant taxa (causing monospecies bloom) harvested for Sanger sequencing (Sanger) and two samples of mixed cryflora communities collected for high throughput sequencing. Species name/sample codes, collection date, sampling altitude (m a.s.l.), geographic position (GPS) are shown.

Species/sample	Date	Altitude	GPS	Sequencing	Reference
Chlamydomonas nivalis DL07	28.05.2016	2380	N47° 13.709 E11° 00.949	Sanger	(Procházková et al. 2018a)
Chloromonas brevispina K-2	28.05.2016	2430	N47 13.753 E11 00.737	Sanger	this study
Scotiella cryophila K-1	05.06.2009	2432	N47°13.748 E11°00.704	Sanger	(Remias et al. 2018)
WP79	11.06.2015	2300	N47°13.422, E11°01.310	HTS	this study
WP99	31.05.2016	2299	N47°13.416, E11°01.260	HTS	this study

Light Microscopy

Cells were analyzed in the laboratory with a Nikon Eclipse 80i light microscope equipped with a Plan Fluor 1.3 100x objective and a Nikon DS-5M digital camera.

Sanger sequencing of locally abundant taxa

Total genomic DNA was isolated from monospecific field blooms of Chlamydomonas nivalis and Scotiella cryophila with the DNeasy Plant Mini kit (Quiagen) as previously described (Procházková et al. 2018b). The field sample containing Chloromonas brevispina was lower in biomass (<20 mg wet weight). In that case DNA was extracted using the Instagene Matrix (Bio-Rad Laboratories, USA) according to Remias et al. (2016). DNA isolates were diluted to concentration of 5 ng µl^–1. The 18S and ITS2 rDNA regions were amplified using existing primers (Table S1). Polymerase chain reactions were performed according to Procházková et al. (Procházková et al. 2018b): For 18S rDNA, initial denaturation at 95 °C for 10 min was followed by 25 cycles of denaturation at 95 °C for 1 min, annealing at gradient temperature 53, 56, 59 and 61 °C for 1 min and elongation at 72 °C for 2 min. Final elongation was at 72 °C for 10 min. For ITS2 rDNA, initial denaturation at 97 °C for 5 min was followed by 37 cycles of denaturation at 95 °C for 1.25 min, annealing at gradient temperature 56, 59, 61 and 64 °C for 2 min and elongation at 72 °C for 4 min. Final elongation was at 72 °C for 7 min. Each 20 µl of PCR reaction for 18S amplification contained 5 µl of DNA isolates, 0.8 µl of each 10 µM primer, 1.6 µl of 25 mM MgCl₂, 1.5 µl of 2 mM dNTPs, 2 µl of 10× Taq buffer + KCl–MgCl₂, 7.8 µl sterile Milli–Q water, and 0.5 µl of 1U.µl^–1 Taq DNA polymerase (Fermentas, USA). Each 35 µl PCR reaction for ITS2 rDNA contained 1 µl of DNA isolates, 1.4 µl of each 10 µM primer, 2.8 µl of 25 mM MgCl₂, 2.6 µl of 2 mM dNTPs, 3.5 µl of 10× buffer Taq buffer + KCl–MgCl₂, 21.8 µl sterile Milli–Q water, and 0.5 µl of 1U.µl^–1 Taq DNA polymerase (Fermentas). The PCR products were stained with bromophenol loading dye, quantified on a 1.5% agarose gel and stained with GelRed (Biotium). The amplification products were purified and sequenced on the Applied Biosystems automated sequencer (ABI 3730×l) at Macrogen (Netherlands). Chromatogram data of forward and reverse sequences of 18S and ITS2 rDNA markers were visually inspected and edited in the program FinchTV 1.4.0 (Geospiza, USA). The contig of each molecular marker was assembled in SeqMan 5.06 (DNASTAR Inc., USA).

High-throughput sequencing

DNA was extracted from field samples WP79 and WP99 using the PowerSoil^® DNA Isolation kit (MoBio Laboratories). The 18S and ITS2 rDNA amplicons were prepared according to the Illumina “16S Metagenomic Sequencing Library Preparation” guide (Illumina n.d.). In brief, 18S rDNA genes were amplified using the eukaryotic primers 528F (5’ GCGGTAATTCCAGCTCCAA) and 706R (5’ AATCCRAGAATTTCACCTCT; Cheung et al. 2010) spanning the V4-V5 hypervariable regions. ITS2 rDNA genes were amplified using the primers 5.8SbF (5’ GATGAAGAACGCAGCG; Mikhailyuk et al. 2008) and ITS4R (5’ TCCTCCGCTTATTGATATGC; White et al. 1990). All primers were tagged with the Illumina adapter sequences. Polymerase chain reactions (PCR) were performed using KAPA HiFi HotStart ReadyMix. Initial denaturation at 95 °C for 3 min was followed by 25 cycles of denaturation at 95 °C for 30 s, annealing at 55 °C for 30 s and elongation at 72 °C for 30 s. Final elongation was at 72 °C for 5 min. All PCRs were carried out in reaction volumes of 25 µl containing 12.5 µl of ReadyMix, each 5 µl of the forward and reverse primer and 12.5 ng of DNA template in 2.5 µl. All pre-amplification steps were done in a laminar flow hood with DNA-free certified plastic ware and filter tips. Amplicons were barcoded using the Nextera XT Index kit. The pooled library was sequenced on the Illumina MiSeq using paired 300-bp reads at the University of Bristol Genomics Facility. 18S and ITS2 rDNA raw sequences have been deposited to the European Nucleotide Archive (ENA) under accession number PRJEB24479.

Processing of HTS sequences

The sequencing quality of each de-multiplexed fastq file was analyzed using the FastQC software (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The low quality 3’ ends of all reads were trimmed. All forward reads were trimmed by 20 bp and all reverse reads by 100 bp. All other processing steps were performed in Qiime (Caporaso et al. 2010). The trimmed paired end reads were joined before further processing and additionally filtered only allowing a minimum Phred quality score of Q20. Reads that could not be joined or were below the quality cut-off were excluded from the analysis. Chimeric sequences were removed using USEARCH 6.1.

The software ITSx (Bengtsson-Palme et al. 2013) was used to extract the ITS2 rDNA regions from all sequences to avoid the inclusion of the highly conserved neighbouring genes (i.e., 5.8S and 28S). Inclusion of these regions in the identification process would otherwise lead to misleading results. HMMER (Eddy 1996) was used to predict the origin of the sequences (e.g., Chlorophyta, Fungi) based on Hidden Markov Models.

OTUs were picked de novo and clustered at 99.5% and 94.0% similarity for the conserved 18S and more variable ITS2 rDNA markers, respectively. Singletons (OTUs containing only 1 sequence, likely derived from sequencing errors) were removed from both the 18S and ITS2 rDNA data sets prior to further analysis.

Taxonomic identities were assigned for representative sequences of each OTU using three steps. (1) The BLAST (Kent 2002) assignment method implemented in Qiime was used with the default minimum percent similarity of 90% to consider a database match a hit (unless a customized script is being used to overwrite the default setting and to increase the similarity threshold). The publicly available and Qiime-compatible Silva (Quast et al. 2012) database was used for the assignment of the 18S rDNA data set and extended with 223 additional sequences of cryophilic algae kindly provided by Dr. Thomas Leya from the CCCryo - Culture Collection of Cryophilic Algae (Fraunhofer IZI-BB). Sequences assigned to Opisthokonta, Amoebozoa, Alveolata and Rhizaria were removed from the OTU table. For the taxonomic assignment of the ITS2 rDNA sequences, a custom database with the limited number of available reference sequences for (cryophilic) green algae was downloaded from NCBI (Table S2). (2) The reference sequences of the locally abundant taxa derived from Sanger sequencing (see above) were included into the classification schema. (3) Manual comparisons were carried out for the representative sequences of the OTUs with their respective reference sequences including verification of sequence identities for 18S and ITS2 and their secondary structures for ITS2 (see below).

Manual post-processing of HTS data

The representative sequences (i.e., the cluster seeds in the OTU picking process) of the 50 most abundant 18S and ITS2 rDNA OTUs comprising >85% and >98% of the total community, respectively, were manually submitted to the BLAST (Kent 2002) web server to search NCBI for close hits to algal taxa. The used BLAST nucleotides parameters were the following: megablast (highly similar sequences), ´others´ as database search set, uncultured/environmental sequences were included, other algorithm parameters were kept with default values.

Fragments of 18S rDNA sequences of phytoplankton assemblages and prokaryotic and eukaryotic alpine permafrost communities are usually clustered into OTUs at 97% identity in HTS studies (Frey et al. 2016, Tragin et al. 2017). A far stricter threshold is required for this molecular marker when snow algal communities dominated by Chlamydomonadales are investigated since several species in this group are very closely related and in some cases differ by only 1 bp over the length of the amplicon (e.g., Chloromonas fukushimae GsCl-11 (AB906342) and Chloromonas tughillensis UTEX SNO91 (AB906348)). Thus, an identity threshold of 2 bp nucleotide difference representing ~99.4% identity on a 342 bp sequence of the 18S rDNA marker (and to allow for sequencing errors (Bradley et al. 2016), had to be passed in order to be considered as a database match. Sequences below this threshold were recorded as “no blast hit”.

The three most abundant OTUs of the 18S rDNA data were further subjected to oligotyping (Eren et al. 2013, Lutz et al. 2018). All sequences contained in one OTU were extracted individually and trimmed to the same length of 340 bp using Fastx Trimmer (http://hannonlab.cshl.edu/fastx_toolkit/). The number of components (i.e., nucleotide position with the highest entropy) to be used was chosen based on the entropy analysis of the sequence alignment. Noise filtering was carried out using a minimum substantive abundance of 50.

For ITS2, a minimum similarity of 94.0% had to be passed in order to be considered a database match. The chosen threshold is in par with several findings on the level of identity of algal ITS2 rDNA. In Gonium pectorale less than 5% of the nucleotide positions differ in pairwise comparisons and less than 7% vary between all clones (Coleman et al. 1993). Similarly, only few nucleotide differences have been reported among strains of Chloromonas reticulata (3.4–4.1%), and of Chlamydomonas reinhardtii NIES-2463 and SAG 11-32a (3.3%), with the latter one being able to cross and produce zygotes (Matsuzaki et al. 2015). A similar low identity threshold for OTU picking (95%) was also successfully applied during Illumina barcoding of soil fungal communities (Schmidt et al. 2013). If an OTU passed the identity threshold of ≥94.0%, it was retained for ITS2 rDNA transcript secondary structure analysis. A schematic overview of the workflow can be found in Fig. S2. The ITS2 sequences were folded using the Mfold server (http://mfold.rna.albany.edu/?q5mfold; Zuker 2003; note: HTS delivers DNA based data, but during RNA folding, thymin [‘T’] is converted to uracil [‘U’]). The model of the secondary structure with the minimum free energy that was consistent with the specific features of nuclear rDNA ITS2 and that contained four helixes and U–U mismatch in helix II (Coleman 2007) was selected. The ITS2 sequences and secondary structures were automatically and synchronously aligned (Schultz and Wolf 2009) using 4SALE (Seibel et al. 2006, 2008), and subsequently manually validated and corrected. First, structure based information (i.e., consensus of all secondary structures and all secondary structures displayed separately) was visually inspected to detect misaligned sequences. This was followed by the manual editing of the secondary structure in order to provide accurate sequence-structure alignments in the context sensitive editing mode (i.e., sequences and secondary structure information are used to validate whether a binding in the context is possible or not). The alignment consisted of the reference species and all OTUs assigned to this species based on preliminary pairwise comparisons in BLAST. Species delimitation was performed in 4SALE and was based on the detection of CBCs, both nucleotides of a paired site mutate while the pairing itself stays stable (e.g. paired sites A - U mutated into G - C). A search for CBCs can only be performed in homologous positions of the ITS2 molecule, which can be unambiguously aligned. For Chlorophyceae, the consensus secondary structure model of ITS2 was identified and the conservation level of individual ITS2 sequence positions (i.e., a position is conserved above 70%) in the alignment was specified (Caisová et al. 2013). Comparisons of the ITS2 secondary structure prediction of the reference sequences gained from Sanger sequencing and those from the HTS data set that were preliminarily assigned to those reference sequences were performed. Based on these comparisons the number of haplotypes of each reference species was assessed. Even a single CBC in helices II and III of the ITS2 secondary structure may indicate sexual incompatibility as has been shown in crossing experiments (Coleman 2000, 2009). Two CBCs in the conserved helix III region of the ITS2 suggested separation between two sister species which differ in their cell morphology, i.e., Chloromonas reticulata and Chloromonas chlorococcoides (Matsuzaki et al. 2012). It has been shown that the probability of a CBC representing two distinct species is 93% (Wolf et al. 2013). Thus, the absence of a CBC in the homology positions of the ITS2 of two OTUs was required in order to be recorded as a match. The secondary structure of nuclear rDNA ITS2 was drawn using VARNA version 3.9 (Darty et al. 2009).

Results

Community composition based on light microscopy (LM)

The most abudant taxa in the two HTS samples (WP79, WP99) according to LM were Chloromonas brevispina, Chlamydomonas nivalis and Scotiella cryophila. All cells of these species were immotile cysts containing a secondary red carotenoid pigmentation, more or less masking the chlorophylls (Fig. 1, Fig. S3). The macroscopical appearance of the harvested snow was red at the surface, turning greenish deeper in the snow for sample WP79 (Fig. S1) and yellowish for WP99. Sample WP79 additionally contained a number of unidentified unicellular green algae.

Fig. 1.

Light micrograph of cells in field sample WP79. The three locally abundant snow algae identified using morphological features were Cr. brevispina (Cb), Scotiella cryophila (Sc) and Cd. nivalis (Cdn). Other cells observed in this sample were Cr. nivalis (Cn), unknown unicellular green alga (U), fungus (F) or bacteria (B). Scale bar: 10 µm.

Output of Sanger sequencing of the locally abundant taxa

Long sequences of multiple DNA regions containing 18S rDNA (about 1700 bp) and ITS2 rDNA (~ 200-1550 bp, Fig. 2) were obtained from the samples with monospecific algal blooms of the three locally abundant taxa. The used primers (Table S1) amplify a larger fragment than ITS2, the actual length of the ITS2 rDNA for Chlorophyta (most common photosynthetic members of snow communities) varies between taxa in a range between 180 and 480 bp (Buchheim et al. 2011). For instance, the AL1500af and LR3 primers (Table S1) are complementary to the end of 18S rDNA and 26S rDNA, and therefore resulting in the amplification of an approximately 1550 bp region. Chloromonas brevispina K-2 sequences of 18S rDNA and ITS2 rDNA were gained in the course of this study, whereas Sanger sequences from two other species have been recently published - Chlamydomonas nivalis DL07 (Procházková et al. 2018a) and Scotiella cryophila K-1 (Remias et al. 2018). Sequences of these three reference species are available in NCBI under the accession numbers listed in Table S3. These sequences were combined and used for the generation of a custom reference sequence database.

Fig. 2.

Schematic workflow for the optimised molecular evaluation of snow algal community structures and diversity, which heavily relies on the combined power of light microscopy, Sanger sequencing and high-throughput sequencing. All details on the markers 18S rDNA and ITS2 are highlighted in red and green, respectively.

Output, evaluation and optimization of the 18S rDNA HTS data

A total of 433,190 18S rDNA sequences passed the quality control and 208,517 sequences could be assigned to green algal taxa (Table 2). The remainder of the sequences was assigned to mostly fungi, as well as Alveolata and Rhizaria (data not shown). 50 OTUs made up >87% of the total community composition (Table S4) and they were selected for the data evaluation and workflow optimization (Fig. 2). An overview of the 10 most abundant OTUs (>78% of the total community) can be found in Table 3.

Table 2.

Overview of HTS data. Number of 18S rDNA and ITS2 sequences before and after quality filtering, as well as the number and percentage of sequences assigned to green algal taxa (the remaining sequences were mostly assigned to Fungi, Alveolata and Rhizaria; data not shown).

Marker	Samples	No. of sequences before quality filtering	No. of sequences after quality filtering	No. and percentage of sequences assigned to green algae
18S	WP79	221837	184921	88795 (48.0%)
	WP99	294484	248269	119722 (48.2%)
ITS2	WP79	187543	116446	81022 (69.6%)
	WP99	204224	156958	108889 (69.4%)

Table 3.

Algal community composition based on the 18S rDNA data set and comprising the 10 most abundant OTUs that made up >78% of the community. Table shows the discrepancies between OTU assignments using (1) Qiime and the publicly available Silva database, (2) Qiime and additional reference sequences of the locally abundant taxa (underlined), and (3) the manual verification of taxa assignments in NCBI, only allowing up to 2 bp nucleotide difference with the respective reference sequence in order to record it as a match. Ambiguous hits are sequences that share the same level of similarity with an inspected OTU, and thus cannot be unambiguously assigned. Our results reveal that the manual verification in NCBI is essential and, thus, highly recommended. A more comprehensive list comprising the 50 most abundant OTUs with their corresponding OTU identification numbers can be found in the Table S4.

OTU ID	WP79 [%]	WP99 [%]	(1) Qiime + Silva	(2) Qiime + Silva + local references	(3) Qiime + Silva + local references + manual verification
denovo 14334	33.0	66.8	Chloromonas sp. Gassan-A LC012753.1	Chloromonas brevispina K-2	Ambiguous hits: Chloromonas brevispina K-2, Scotiella cryophila K-1, Chloromonas sp. TA AB902996, Chloromonas sp. Gassan-B LC012714.1
denovo 45654	18.7	0.1	Mesotaenium sp. AG-2009-1 FM992335.1	Ancylonema nordenskioeldii AF514397.2	Ancylonema nordenskioeldii AF514397.2
denovo 36485	0.9	13.6	Uncultured Chlamydo-monadaceae AB902971.1	Chlamydomonas nivalis DL07	Chlamydomonas nivalis DL07
denovo 40226	8.2	0	Botrydiopsis constricta AJ579339.1	Botrydiopsis constricta AJ579339.1	Botrydiopsis constricta AJ579339.1
denovo 15070	4.6	1.0	Uncultured Chloromonas AB903008.1	Chloromonas cf. alpina CCCryo 033-99 HQ404865.1	Chloromonas platystigma strain CCCryo 020-99
denovo 20542	4.6	0	Uncultured Dunaliellaceae EF023287.1	Uncultured Dunaliellaceae EF023287.1	Chloroidium saccharophilum isolate HST10K KX024691.1
denovo 101	3.2	1.1	Chloromonas sp. D-CU581C AF517086.1	Chloromonas cf. rostafinskii CCCryo 025-99 AF514402.1	Ambiguous hits: Chloromonas sp. NIES-2379 AB906350.1, Chloromonas rostafinskii strain CCCryo 025-99 AF514402.1
denovo 23251	3.0	<0.1	Chloromonas sp. Gassan-A LC012753.1	Chloromonas brevispina K-2	Ambiguous hits: Chloromonas brevispina K-2, Chloromonas sp. Hakkoda-1 LC012710.1, Chloromonas sp. Gassan-A LC012709.1
denovo 30051	0.4	1.9	Uncultured Chloromonas AB902984.1	Uncultured Chloromonas AB902984.1	No blast hit
denovo 36086	2.1	0	Prasiola furfuracea AF189073.1	Prasiola furfuracea AF189073.1	No blast hit
	21.3	15.5	Other	Other	Other

The largest proportion of the sequences (WP79: 33.0%, WP99: 66.8%) was clustered in one OTU ‘denovo14334’ (99.4% similarity). The initial species assignment solely using the Qiime-compatible Silva database resulted in Chloromonas sp. Gassan-A LC012753.1 (Table 3 – column (1)). Other abundant OTUs were assigned to Mesotaenium sp. and several “uncultured Chloromonas and Chlamydomonadaceae” without a species affiliation (Table 3). The addition of the generated reference sequences of the locally abundant taxa resulted in new assignments (26 out of the 50 most abundant OTUs) and in some cases in the clarification of the species assignment (Table S4 – column (2)). For instance, the “uncultured Chlamydomonadaceae” could be identified as Chlamydomonas nivalis (WP79: 0.9%, WP99: 13.6%).

The representative sequences of the 50 most abundant OTUs in 18S rDNA were subjected to a manual BLAST search against NCBI GenBank since the taxa assignments in Qiime (Caporaso et al. 2010) uses a low default value of 90% minimum percent similarity to place taxonomic strings onto OTUs. The aim was to verify the actual identity percentage and whether there are closer hits that are not present in the Silva database. Indeed, the manual verification step improved the taxonomic assignment of another eight taxa. However, it also revealed that 15 OTUs shared the same identity with several species. For instance, a vast number of Chloromonas species including Chloromonas brevispina K-2, Chloromonas sp. TA 8 (AB902996.1), Chloromonas sp. Gassan-A (LC012714.1), Chloromonas sp. Gassan-B (LC012714.1), Chloromonas polyptera (JQ790556) and Chloromonas sp. Hakkoda-1 (LC012710.1), as well as Scotiella cryophila K-1 (MG253843; considering two ambiguous positions in the reference which can code for the same nucleotides) shared more than 99% identity in the hypervariable V3-V4 region of the 18S rDNA sequences (Table 3 and Table S4). The same applied to Raphidonema sempervirens (AF514410.2), Raphidonema nivale (AB488604.1) and Stichococcus sp. (KP081395.1), which also shared more than 99% identity. Thus, those OTUs could not unambiguously be assigned on the species level (see ambiguous hits in Tables 3 and Table S4). The most abundant OTU, denovo14334, showed a difference of 1 bp to Chloromonas brevispina K-2, Chloromonas sp. TA 8 (AB902996.1), Chloromonas sp. B (LC012714.1) and Scotiella cryophila K-1 (MG253843), and thus several species are likely unresolvable combined in this OTU. Furthermore, several OTUs showed differences of more than 2 bp to the closest reference and their assignment was therefore discarded in this step and recorded as “no blast hit” (2 OTUs in Table 3 – column (3), and 15 OTUs in Table S4 – column(3)). The percentage of sequence cover in the pairwise comparison with the reference sequence was also considered. For example, the OTU denovo30674 was initially assigned to Prototheca cutis (AB470468.1; < 2 bp difference). Yet, manual post-processing revealed that the sequence cover was only 18%. Thus, this assignment was discarded.

Furthermore, based on microscopic observations and on known algal distribution patterns for this European Alpine location, some taxa assignments had to be scrutinized despite a very high similarity of the amplicons sequences with the suggested affiliations. These included Ancylonema nordenskioeldii, which mainly thrives on glacial surfaces of Polar regions (Remias et al. 2012, Lutz et al. 2018), and Chloromonas polyptera, which can be found in Maritime Antarctica in the vicinity of penguin rockeries (Remias et al. 2013).

Since several OTUs seemed to contain multiple taxa, oligotyping was carried out to further refine the taxonomic assignments. OTU ‘denovo14334’ consisted of four oligotypes of which the most abundant shared 100% similarity with Chloromonas brevispina K-2 in WP79 and several uncultured snow algae species in WP99 (Table 4, Table S5). Oligotyping of the OTUs ‘denovo45654’ and ‘denovo36485’ also resulted in the resolution of another two oligotypes each.

Table 4.

Oligotyping of the three most abundant 18S rDNA OTUs. Table shows the individual oligotypes that were conflated in the three most abundant 18S rDNA OTUs (Table 3) and could not be detected by conventional OTU clustering. The refined taxonomic assignments and their respective relative abundances were then used for the final description of the snow algal community composition in Figure 5. For instance, OTU ‘denovo14334’ was assigned to Chloromonas brevipsiona K-2 (WP79: 33%). However, oligotyping revealed that only one oligotype within this OTU corresponded to this species, which decreased its relative abundance from 33.0% to 23.4% in WP79.

OTU Oligotype	Taxa assignment	Similarity [%]	WP79 [%]	WP99 [%]
denovo14334	Ambiguous hits: Chloromonas brevispina K-2, Scotiella cryophila K-1, Chloromonas sp. TA AB902996, Chloromonas sp. Gassan-B LC012714.1	99.4	33.0	66.8
TTT	Uncultured snow algae LC371427.1, LC371425.1, LC371423.1, LC371419.1, LC371414.1	100	1.0	60.7
TCT	Chloromonas brevispina K-2, Chloromonas sp. Gassan-A LC012753.1, Chloromonas sp. Hakkoda-1 LC012710.1,	100	23.4	<0.1
CTT	Chloromonas sp. Gassan-B LC012714, uncultured Chloromonas sp. ANT1 AB903007.1 and Chloromonas sp. TA8 AB902996.1, Chloromonas polyptera JQ790556.1, uncultured Viridiplantae HQ188979.1	100	8.5	0.2
TTC	28 hits	>99	0.1	5.9
denovo45654	Ancylonema nordenskioeldii AF514397.2	100	18.7	0.1
A	Ancylonema nordenskioeldii AF514397.2	100	17.2	0.1
C	Mesotaenium berggrenii var. alaskana JF430424.1, Mesotaenium sp. AG-2009-1 FM992335.1	99.4	1.5	<0.1
denovo36485	Chlamydomonas nivalis DL07	100	0.9	13.6
T	Chlamydomonas nivalis DL07, 14 hits including several Chlamydomonas nivalis and uncultured snow algae strains	100	0.8	12.3
C	14 hits including several Chlamydomonas nivalis and uncultured snow algae strains	100	0.1	1.3

Output, evaluation and optimization of the ITS2 rDNA HTS data

A total of 273,404 ITS2 rDNA sequences passed the quality control and 189,922 sequences could be assigned to Chlorophyta (Table 2). The remainder of the sequences was assigned to mostly Fungi and Alveolata (Table S6). 38 taxa made up >98% of the total community composition (Table S7) and were selected for the data evaluation and workflow optimization (Fig. 2, Fig. S2). An overview of the 10 most abundant OTUs (>88% of the total community) can be found in Table 5.

Table 5.

Algal community composition based on the ITS2 data set and comprising the 10 most abundant OTUs that made up >88% of the community. Table shows the discrepancies between OTU assignments using (1) Qiime and a custom database downloaded from NCBI, (2) Qiime and additional reference sequences of the locally abundant taxa (underlined), and (3) the manual verification of taxa assignments. The latter one comprised the comparison of the representative sequences of the OTUs with their respective reference sequences in terms of sequence identities (≥94.0% similarity required) and their secondary structures (absence of compensatory base changes (CBC) in homology positions required. A more comprehensive list comprising the 38 most abundant taxa with their corresponding OTU identification numbers can be found in the Table S7. Our results reveal that the manual verification including secondary structure prediction and CBC search is essential, and thus, highly recommended

OTU ID	WP79 [%]	WP99 [%]	(1) Qiime + NCBI database	(2) Qiime + NCBI database + local references	(3) Qiime + NCBI database + local references + manual verification (sequence similarity [%], sequence cover [%])
denovo 99	1.8	59.1	Chloromonas sp. CCCryo289-06 HQ404893.1	Chloromonas brevispina K-2	No blast hit (88%, 83%, 6 CBC when compared denovo99 and Chloromonas brevispina K-2 – one CBC out of it is located in the most conserved part of structure, i.e. in top close to the 5´end of III helix, see Fig. 4)
denovo 20	5.5	28.2	Chlamydomonas nivalis GU117577.1	Chlamydomonas nivalis DL07	Chlamydomonas nivalis DL07 (100%, 100%)
denovo 107	39.8	<0.1	Chloromonas sp. CCCryo289-06 HQ404893.1	Chloromonas brevispina K-2	Chloromonas brevispina K-2 (100% identical except for one nucleotide - instead of ʿRʿ in reference sequence, there was ’G’, 100%)
denovo 100	13.7	<0.1	Chloromonas pichinchae HQ404889.1	Chloromonas pichinchae HQ404889.1	no blast hit
denovo 63	10.9	0.2	No blast hit	Scotte Ila cryophila K-1	Scotiella cryophila K-1 (100%, 100%)
denovo 142	8.2	1.7	Chloromonas rostafinskii HQ404863.1	Chloromonas rostafinskii HQ404863.1	No blast hit
denovo 130	1.4	3.5	No blast hit	No blast hit	No blast hit
denovo 181	4.1	0	No blast hit	No blast hit	No blast hit
denovo 254	3.1	0.1	Chloromonas rostafinskii HQ404863.1	Chloromonas rostafinskii HQ404863.1	No blast hit
denovo 23	0.1	2.2	No blast hit	No blast hit	No blast hit
	11.4	5.0	Other	Other	Other

The initial assignments based on the limited number of sequences available at NCBI resulted in species assignments for 6 out of 10 OTUs (Table 5- column (1)) and for 17 out of the 38 OTUs (Table S7 – column (1)). After including the Sanger derived reference sequences of the three locally abundant taxa into the classification schema, eight assignments for denovo OTUs were improved (Table 5 – column (2), Table S7 – column (2)) and one OTU previously without blast hit could be newly assigned (i.e. denovo63 to Scotiella cryophila K-1). Chloromonas brevispina K-2 and Scotiella cryophila K-1 contributed significantly to the pool of detected sequences.

After conducting ITS2 rDNA transcript secondary structure analyses and CBC detection (Fig. S4), the haplotype diversity of three locally abundant species revealed to be lower than expected from the HTS output (Fig. 3 – column ´before´). In the analysis of the 38 most abundant OTUs, which comprised >98% of the community (Table S7), only one haplotype for each species was recovered (and was shown to be 100% identical with the reference sequence of a locally abundant taxon) (Fig. 3 – column ´after´, Table 5). When the analysis was extended to the whole data set (160 OTUs), other haplotypes were recovered (Fig. S5), which were less abundant in terms of their read numbers. For instance, the dominant haplotype of Scotiella cryophila K-1 in WP79 accounted for 8,853 reads, whereas two other haplotypes comprised only two reads each. The major haplotype of Chlamydomonas nivalis DL07 in WP99 included 30,734 reads and the second haplotype only three reads. A dramatic drop of OTUs assigned to Chloromonas brevispina prior and after the secondary structure analysis (and associated CBC counts) was revealed (Fig. 3, Fig. S5). The major haplotype of Chloromonas brevispina K-2 in WP79 was represented by 32,258 reads and the second haplotype by five reads only. The most abundant OTU ‘denovo99’ (59% of sequences in WP99), which had previously been assigned to Chloromonas brevispina K-2, had six CBCs in the conservative regions (including one CBC in the most conserved apex of helix III close to the 5´end) of the ITS2 structure in comparison with the reference sequence (Fig. 4). Taking into account that the presence of at least one CBC between two organisms in the conserved regions of ITS2 is predicting a failure to sexually cross (Coleman 2009), we infer that OTU ‘denovo99’ might represent an independent though undescribed species. Furthermore, several OTU assignments, meaning the correct placement of taxonomic string onto OTUs, were below the chosen similarity threshold of 94.0% with respect to the reference sequences. These included the three OTUs (denovo142, denovo254 and denovo127) initially assigned to Chloromonas cf. rostafinskii (HQ404863.1) and the OTU ‘denovo100’ assigned to Chloromonas pichinchae CCCryo 261-06 (HQ404889.1). Consequently, these unknown species contributed to 69% of the unidentified ITS2 diversity in sample WP79. The suitability of the used ITS2 identity threshold (≥94.0%) was verified by checking the presence of CBCs in a few representative OTUs with a lower similarity threshold. For instance, the OTU ‘denovo100’ shared 92% identity with Chloromonas pichinchae CCCryo 261-06 and one CBC was detected when the secondary structures of these sequences were compared. In contrast, denovo85 was 95% identical with the reference species and no CBC was found. Thus, denovo85 was assigned to Chloromonas pichinchae.

Fig. 3.

Haplotype diversity of the three locally abundant species (Chloromonas brevispina K- 2, Chlamydomonas nivalis DL07, Scotiella cryophila K-1; identified by light microscopy in Fig. 1 and Fig. S3) before and after comparison of the ITS2 rDNA transcript secondary structures (Table S7) of the 38 most abundant OTUs in the two samples (WP79 and WP99). A sequence identity of ≥94.0% and an absence of CBCs in homologic positions in comparison to a reference species were required for an assignement to each taxon.

Fig. 4.

Comparison of the secondary structure of ITS2 rDNA transcripts between Chloromonas brevispina K-2 (accession number MG791868) and the closely related OTU ‘denovo99’. Differences characteristic for the latter are shown by nucleotides outside the structure and are linked by dotted lines; middle and top part of helix I (encircled) represent an expansion segment, whose length is not conserved and in which positions are <70% conserved according to consensual secondary structure model of Chlorophyceae (Caisová et al. 2013). Therefore, this part of helix I characteristic for denovo99 is shown outside the structure and is linked by dotted lines. ITS2 displays four helices, the U–U mismatch in helix II (arrows) and the YGGY motif on the 5’ side near the apex of helix III (bold letters), all of which are hallmarks of eukaryotic nuclear rDNA ITS2 secondary structures (Coleman 2003). Compensatory base changes between the two species are indicated by rectangles in helixes I, II and (mainly) III. In addition, Chloromonas brevispina was identical with OTU ‘denovo107’, except for one ambiguous base (the nucleotide position marked by an asterisk in helix VI). (Note: HTS delivers DNA based data, but during RNA folding, thymin [‘T’] is converted to uracil [‘U’].)

Community composition based on the 18S and ITS2 rDNA HTS data sets

The presence of Chloromonas brevispina K-2 and Chlamydomonas nivalis DL07 were confirmed by both the 18S and the ITS2 rDNA data. Whereas Scotiella cryophila K-1 was one of the more abundant species in the ITS2 data in WP79 (10.9%), it could not be unambiguously assigned in the 18S data (Table 3 and Table S4, denovo14334). Several other, less frequent species, (e.g, Chloromonas sp. Hakkoda-1 (LC012710.1), Chloromonas platystigma (AF514401.1), Chloroidium saccharophilum (KX024691.1), Chloromonas cf. rostafinskii (AF514402.1) could be detected in the 18S rDNA data, yet, were absent in the ITS2 data.

In contrast, the 18S rDNA data of WP99 was dominated by a taxon sharing 100% smiliarity with several uncultured snow algal species, and Chlamydomonas nivalis (18S: 13.6%, ITS2: 28.2%; Tables 3 and 5, Figure 5). A considerably higher abundance of sequences with no species assignment was present in the ITS2 data sets (WP79: 37.3%, WP99: 68.9%) in comparison to the 18S rDNA data sets (WP79: 3.6%, WP99: 3.0%; Fig. 5). However, the vast majority of unassigned sequences in WP99 was represented by a single dominant OTU (denovo99; 86% of all sequences without species assignment) closely related to Chloromonas brevispina K-2 (Fig. 4), and only one minor OTU with five ITS2 reads was matching Chloromonas brevispina K-2. In contrast, in WP79 Chloromonas brevispina K-2 represented the dominant abundant OTU and denovo99 was much less abundant.

Fig. 5.

Snow algal community composition derived from performing all optimization steps comprising all taxa with a minmum relative abundance of 5%. Full details can be found in the Tables S4 and A6. Some discrepancies still prevailed between the 18S and ITS2 rDNA data sets, in particular in terms of the percentage of sequenes with an unambiguous species assignment. Soil algae, which were only present in WP79, due to a different sampling strategy in 2015, are highlighted with an asterisk. Less abundant taxa are summarised in ‘Other’. Species abbreviations in brackets are referring to cells identified in the light micrographs (Fig. 1, Fig. S3).

The presence of allochthonous soil algae like Botrydiopsis constricta (AJ579339.1), Botrydiopsis callosa (AJ579340.1), Heterococcus pleurococcoides/fuornensis/chodatii (Xanthophyceae; Broady 1976; Negrisolo et al. 2004), Chloroidium saccharophilum (KX024691.1; Darienko et al. 2010), Lobosphaera sp. (KT119889.1), Lobosphaera incisa (KM020046.1) and Lobosphaera tirolensis (Chlorophyta; AB006051.1; Karsten et al. 2005) (Tables S4 and A7) in WP79, and their absence in WP99 (when surface and soil-near snow were avoided during sample collection), show the importance of a consistent sampling strategy when the aim is to compare species composition and abundaces between different sites (Fig. S6).

Discussion

Monospecific field blooms are generally targeted for studies of snow algal population ecology, physiology and morphology (e.g., Chlainomonas kolii (Novis 2002), Chlamydomonas nivalis (Remias et al. 2005), Chloromonas brevispina (Hoham et al. 1979), Chloromonas nivalis subsp. tatrae (Procházková et al. 2018b) or Chloromonas polyptera (Remias et al. 2013)). HTS techniques recently revealed that snow algae commonly show a spatially structured distribution and clones of a single species can be dominant in one location (Brown et al. 2016). For this study, we intentionally investigated a non-permanent snow field characterized by a higher biodiversity of cryoflora and with at least three dominant species usually present every year. We used these more complex samples as a case study to evaluate the application of HTS techniques on this ‘exotic’ (less studied) kind of ecosystem and discuss below the important steps and criteria for these types of samples.

Generation of a custom reference database of locally abundant taxa and light microscopy observations

The intentional use of monospecific blooms of snow algae is advisable in order to obtain long reference sequences of multiple DNA regions by Sanger sequencing for at least the dominant key organisms in the studied environment. A polyphasic approach (i.e., collectively using genetic, chemotaxonomic and phenotypic methods) is required to accurately determine the taxonomic identity of species found in field samples (Matsuzaki et al. 2015). Alternatively, single cells, with identifiable different morphologies, can be picked out of mixed samples for single-cell sequencing of different markers if one aims to link morphology to genotype (Bock et al. 2014). Light microscopy observations of snow algal cell morphologies present in the studied field samples should be conducted each time (and as soon as possible) to gain first insights into which species might be expected in the sequencing results. For instance, the qualitative light microscopy performed in this study revealed that the most abundant taxon in both HTS samples was Chloromonas brevispina. The ITS2 rDNA data set confirmed the dominance of this species in WP79, but not initially in WP99. However, a detailed evaluation of the light micrographs (Fig. 1, Fig. S2) and the ITS2 data from WP99 suggested that the dominant OTU denovo99 (“no blast hit” in Figure 5) is closely related to Chloromonas brevispina K-2, and that they share most likely similar morphologies. Thus, the dominant taxon was finally recovered in both ITS2 data sets. Very low haplotype diversity was shown for Chlamydomonas nivalis DL07, Chloromonas brevispina K-2 and Scotiella cryophila K-1 within the same algal bloom (with only low abundant subdominant haplotypes). This is in par with previous findings on red snow from North Amterica (Brown et al. 2016).

A two-marker approach is mandatory

Currently, a considerably higher number of 18S rDNA reference sequences exists in public databases compared to ITS2 rDNA. This is especially true for cryophilic microalgae. Moreover, in more frequently investigated habitats such as high alpine lake phytoplankton, the same problem may arise, but can be successfully solved by a two marker approach (in combination with light microscopy observations) (Bock et al. 2014). A lack of ITS2 reference sequences is also likely to be encountered in long-term studied algal habitats in lowlands close to the settlement (e.g., artificial water bodies such as fish ponds and water dams). Therefore, 18S rDNA seems to be the obvious marker of choice for high-throughput studies. Several other gene loci, e.g. tufA (Vieira et al. 2016) and rbcL (Hall et al. 2010, Zou et al. 2016), have been recommended as the promising DNA barcode for some green algae (Hall et al. 2010). However, tufA records for the genus Chloromonas, which is often found in snow, has been scarce so far in NCBI database. The HTS technology currently limits the chosen marker to only a fraction of its actual length. The chosen V4-V5 region is the most variable region of the 18S rDNA gene for snow algal taxa, yet, the variability was not sufficient and unambiguous species assignments were often not feasible (Table 3). The most abundant 18S rDNA OTU was assigned to Chloromonas brevispina K-2. However, there was likely a “hidden” diversity to a certain extent, since a large number of different Chloromonas species can share up to 100% identity of this marker. An oligotyping approach could resolve some of the “hidden” diversity for species with a difference of 1 bp in the sequenced amplicon and is therefore highly recommended for further refinements of data sets that are characterized by very low variability, which is not detectable by conventional clustering of operational taxonomic units. Furthermore, due to the possibility of absolutely identical reference sequences, it is essential to manually check alignments to the reference sequences. For instance, Chloromonas polyptera has been reported from HTS studies in several places in the northern hemisphere (Lutz et al. 2015a, Terashima et al. 2017), yet, these studies seem to result in ambiguous species assignments. For example, Terashima et al (2017) reported 100% identity in the 18S rDNA of their OTU44 with Chloromonas polyptera (JQ790556). However, the sequenced gene fragment also shares 100% identity with Chloromonas sp. Gassan B (LC012714.1), Chloromonas sp. ANT1 (AB903007.1), Chloromonas sp. TA8 (AB902996.1) and an uncultured ‘Viridiplantae’ clone (HQ188979.1). In summary, 18S rDNA amplicons do not adequately identify taxa to species level in several cases including a study of freshwater phytoplankton (Xiao et al. 2014).

In contrast, Illumina reads are able to span the entire region of the hypervariable ITS2 rDNA. This molecular marker provides a much higher resolution than 18S rDNA. The prediction of the ITS2 rDNA transcript secondary structures allowed a thorough identification of the haplotype diversity. It is therefore a powerful tool to delete wrong OTU assignments and therefore represents a more appropriate way of describing the true biodiversity (e.g., in the sample WP99 the most abundant OTU ‘denovo99’ is not Chloromonas brevispina K-2). Nevertheless, this methodological approach (ITS2 rRNA secondary structure prediction of each OTU) is currently immensely time-consuming. The process is partly automatized (e.g., using Mfold, 4SALE), nevertheless, significant input of manual validation and correction is still required. Wolf et al. (2013) reported probability of ~0.99 that there is no intragenomic CBC based on comparison of ITS2 of 178 species of land plant. However, some snow Chloromonas species possess intragenomic CBCs (Matsuzaki et al. 2015), which might cause an overestimation of biodiversity in the currently proposed HTS approach. Evaluation of CBCs detection should be carried out carefully: not only pure presence/absence of CBC but also the exact position of CBC in the ITS2 molecule (i.e., whether it is in or outside the most conserved part) and the level of genetic difference of ITS2 between OTUs should be taken into account. For instance, among three OTUs from aplanozygotes and strains of C. miwae, one CBC was observed between specimen Gassan-C/strain NIES-2379 and strain NIES-2380 (Matsuzaki et al. 2015). This CBC was located outside the most conserved branch of helix III and genetic differences in the nuclear rDNA ITS2 region between the aplanozygote specimen and the C. miwae strain were only 0.0–0.4% (see S15 in Matsuzaki et al. 2015), and thus, these specimens and strain regarded as one species.

Finally, none of the two markers adequately described the community composition, either due to their low resolution (18S) or the lack of reference sequences (ITS2). Moreover, the output can also be partly influenced by library production biases and primer inefficiencies.

Combining the strengths of both markers using a two-marker approach is thus mandatory, in particular in less-well studied environments. In addition, one marker can guide the data optimization of the other marker. Despite its low resolution, the 18S marker can provide guidance, which ITS2 rDNA reference sequences need to be generated and vice versa.

Specific to the samples analysed here, the differences in species abundances between samples from 2015 and 2016 could be explained by the differences in sampling strategies (complete snow column in 2015, surface and ground snow discarded in 2016) or the patchy nature of snow algal communities. Moreover, in 2016 sampling was carried out about two weeks earlier in the season and by the fact that weather conditions can generally differ from melt season to melt season.

Advantages, limitations and perspectives of HTS for community composition analyses

HTS allowed a more comprehensive assessment of the prevailing biodiversity than traditional Sanger sequencing and microscopic observations. In addition to the detection of low- abundant taxa, a multitude of sequences, which did not match any references in the databases were generated. Thus, some of these presumably represent new species (e.g., OTU ‘denovo99’). Strain-based taxonomic studies or accurate species determination of monospecies field blooms by Sanger sequencing to gain complete reads of the target marker are required to increase the number of reference sequences in the databases. On the other hand, Sanger sequencing can be problematic for mixed communities in terms of chromatogram corrections unless cloning is involved. Alternatively, presumably monospecies snowfields can be sampled for HTS studies to evaluate the biodiversity and also for detection of any intragenomic variations of ITS2.

However, as we have shown here, the current ability to interpret HTS data is poor when applied on environments that are highly underrepresented in public databases, unless the employed, usually automatic assignment method is carefully evaluated and optimized through additional manual verifications. The quality of the reference databases is crucial and new entries have to be continuously updated. The Qiime-compatible Silva database delivers reference sequences in one batch, yet, an updated version is only released about once a year. In contrast, NCBI is under continuous revision and therefore we recommend that new potential reference sequences are added manually to the Qiime-compatible Silva database prior to use. Only such optimized data can then be used for further evaluations including phylogeography and phylogenetic studies based on the generation of multi-locus sequence data in a fast and cost-effective way (McCormack et al. 2013).

In this case study, we evaluated the application of high-throughput sequencing on an unconventional ecosystem of melting European Alpine snowfields. Based on light microscopic observations, the investigated snowfields were dominated by three algal species, which were however not always reflected in the sequencing dataset. Consequently, HTS data need to be handled with care if applied on habitats or groups of organisms that are (highly) underrepresented in molecular databases. Currently, the need to generate appropriate reference sequences for the key taxa in the studied environment is an inevitable task for such studies. Furtheremore a two-marker approach, consistent sampling strategy, light microscopy-based guidance and final manual verification of all taxonomic assignments are strongly recommended.

List of abbreviations

CBC

Compensatory Base Changes

HTS

High-throughput sequencing

OTU

Operational Taxonomit Unit