Environmental Sequencing Provides Reasonable Estimates of the Relative Abundance of Specific Picoeukaryotes

ABSTRACT

High-throughput sequencing (HTS) is revolutionizing environmental surveys of microbial diversity in the three domains of life by providing detailed information on which taxa are present in microbial assemblages. However, it is still unclear how the relative abundance of specific taxa gathered by HTS correlates with cell abundances. Here, we quantified the relative cell abundance of 6 picoeukaryotic taxa in 13 planktonic samples from 6 European coastal sites using epifluorescence microscopy on tyramide signal amplification-fluorescence in situ hybridization preparations. These relative abundance values were then compared with HTS data obtained in three separate molecular surveys: 454 sequencing of the V4 region of the 18S ribosomal DNA (rDNA) using DNA and RNA extracts (DNA-V4 and cDNA-V4) and Illumina sequencing of the V9 region (cDNA-V9). The microscopic and molecular signals were generally correlated, indicating that a relative increase in specific 18S rDNA was the result of a large proportion of cells in the given taxa. Despite these positive correlations, the slopes often deviated from 1, precluding a direct translation of sequences to cells. Our data highlighted clear differences depending on the nucleic acid template or the 18S rDNA region targeted. Thus, the molecular signal obtained using cDNA templates was always closer to relative cell abundances, while the V4 and V9 regions gave better results depending on the taxa. Our data support the quantitative use of HTS data but warn about considering it as a direct proxy of cell abundances.

IMPORTANCE Direct studies on marine picoeukaryotes by epifluorescence microscopy are problematic due to the lack of morphological features and due to the limited number and poor resolution of specific phylogenetic probes used in fluorescence in situ hybridization (FISH) routines. As a consequence, there is an increasing use of molecular methods, including high-throughput sequencing (HTS), to study marine microbial diversity. HTS can provide a detailed picture of the taxa present in a community and can reveal diversity not evident using other methods, but it is still unclear what the meaning of the sequence abundance in a given taxon is. Our aim is to investigate the correspondence between the relative HTS signal and relative cell abundances in selected picoeukaryotic taxa. Environmental sequencing provides reasonable estimates of the relative abundance of specific taxa. Better results are obtained when using RNA extracts as the templates, while the region of 18S ribosomal DNA had different influences depending on the taxa assayed.

INTRODUCTION

Protists are key components of marine ecosystems, being major players in the global respiration and production budgets (1, 2) and playing central roles in marine food webs (3). Despite their importance and ubiquity, it was only during the past decade that environmental studies, based on molecular (i.e., culture-independent) techniques, revealed an unsuspected protist diversity in a large variety of marine ecosystems (4–13). These studies were based on the analysis of 18S rRNA genes retrieved directly from natural assemblages by PCR amplification, cloning, and sequencing. Now, the development and use of high-throughput sequencing (HTS) tools, e.g., 454 or Illumina, which produce thousands of sequences from a single sample, have revolutionized the field, allowing deeper assessments of diversity (14) as well as better estimates of specific relative abundances. One of the main challenges of this approach, however, is to understand the correspondence between the relative abundances of sequences and cells—that is, how close the specific diversity detected in molecular surveys is to the true species composition of natural assemblages.

Few studies have analyzed the relationship between direct microscopic inspections and sequencing data in protists. One of the first studies compared cloning and sequencing results with an accurate list of protist species (5- to 100-μm size range) identified by microscopy (15). In that case, as the sequencing effort was very limited (fewer than 100 clones), few of the protists identified by morphology were detected in the sequencing set. In addition, the few sequences obtained did not represent the dominant observed species, a clear sign of the biases in this molecular approach. More recent comparative studies used HTS, and therefore were not limited by the sequencing effort but focused on specific taxa, in particular marine and freshwater ciliates (2, 16–18). Ciliate species have the advantage of having conspicuous morphological traits that allow proper identification by inverted microscopy. In most cases, the same species were found in microscopic and molecular data sets, but the relative abundances of sequences and morphotypes were not in agreement, so each approach revealed a different community structure. Other studies prepared mock communities, and the results obtained were similar: all individual taxa were detected, but the relative proportion of sequence types was different from cell mixes (19, 20). Overall, the popularization of HTS now allows a high-resolution exploration of protist richness present in natural samples; yet, when it comes to evenness, the picture obtained is still limited.

Among protists, picoeukaryotes (protists up to 3 μm in size) are known to be very diverse, widely distributed, and ecologically important in the marine plankton realm (21). Picoeukaryotes are counted as a group by epifluorescence microscopy using a general DNA stain (22) or by flow cytometry (23), but, due to their small size and lack of morphological traits (24), they cannot be taxonomically identified by these tools. This can be achieved with fluorescence in situ hybridization (FISH), which enables the visualization and quantification of specific cells in natural assemblages by using oligonucleotide probes as phylogenetic stains (25). FISH has served to identify the cells from novel environmental clades (11, 26, 27) and has been applied in a few marine surveys (28–31). However, this approach is relatively time consuming and targets only one taxon at a time.

In this study, we assessed the feasibility of using HTS data as a quantitative metric in picoeukaryote diversity studies by comparing relative HTS read abundances with relative FISH cell counts in selected picoeukaryotic taxa. Unlike the previous studies, in which a single taxon (ciliates) or artificial communities were analyzed, this study focused on a set of highly divergent lineages found in geographically separated and unrelated microbial assemblages. Any pattern emerging from this heterogeneous and noisy data set was expected to be rather robust. We also investigated if there was a difference in community composition assessed by using environmental DNA or RNA extracts as the templates (DNA and cDNA reads, respectively), sequencing different regions of the 18S ribosomal DNA (rDNA) (V4 versus V9) or using different HTS platforms (454 versus Illumina). To address these questions, we used published sequencing data sets from several European coastal samples (Massana et al., 2015, for DNA and cDNA-V4 [32] and Logares et al., 2014, for cDNA-V9 [33]) and chose 6 picoeukaryote taxa (<3 μm) for which we had specific FISH probes for quantification.

MATERIALS AND METHODS

Sampling.

Samples were taken during the BioMarKs project (http://www.biomarks.eu/) in six European coastal sites: Blanes (Spain, 41°40′N, 2°48′E), Gijon (Spain, 43°40′N, 5°35′W), Naples (Italy, 40°48′N, 14°15′E), Oslo (Norway, 59°16′N, 10°43′E), Roscoff (France, 48°46′N, 3°57′W), and Varna (Bulgaria, 43°10′N, 28°50′E) (Table 1). Seawater was collected with Niskin bottles attached to a conductivity-temperature-depth rosette at the surface and the deep chlorophyll maximum (DCM) depths. For molecular surveys, ∼20 liters of seawater was prefiltered through a 20-μm-pore-size metallic mesh and then sequentially filtered through 3- and 0.8-μm-pore-size polycarbonate filters (142-mm diameter). The 0.8-μm-pore-size polycarbonate filter contained the picoplankton (0.8- to 3-μm size fraction) and was flash frozen and stored at −80°C. The filtration time was less than 30 min to avoid RNA degradation.

TABLE 1.

Planktonic samples analyzed (sampling site, date, depth, and seawater temperature) and cell counts (cells · ml⁻¹) in these samples^a

Sampling site	Date	Depth (m)	Temp (°C)	DAPI count		Flow cytometry count of phototrophs	% phototrophsb	% heterotrophsb
				Phototrophs	Heterotrophs
Blanes	Feb 2010	1 (surface)	12.5	9,273	445	9,215	48.6	53.7
Gijon	Sept 2010	1 (surface)	20.2	1,606	2,503	2,990	14.5	20.2
Naples	Oct 2009	1 (surface)	22.8	–c	–c	2,714
		26 (DCM)	22.4	–c	–c	2,049
	May 2010	1 (surface)	19.2	4,376	4,372	4,700	1.1	54.6
		34 (DCM)	15.5	1,808	1,331	1,802	8.3	28.8
Oslo	Sept 2009	1 (surface)	15.0	12,342	4,470	9,540	12.4	21.9
		20 (DCM)	15.0	8,773	2,807	8,930	17.9	38.4
	June 2010	1 (surface)	15.0	7,727	2,893	13,295	25.5	7.9
		10 (DCM)	12.5	21,523	2,823	17,900	22.9	40.7
Roscoff	Apr 2010	1 (surface)	9.9	7,203	1,034	8,240	43.9	68.9
Varna	May 2010	1 (surface)	21.5	–c	–c	3,861
		40 (DCM)	9.5	7,043	731	9,487	24.9	24.6

^aThe total picoeukaryote abundance (cells ≤3 μm) was determined by DAPI (phototrophs and heterotrophs), and the photosynthetic picoeukaryote abundance was determined by flow cytometry.

^bThese columns show the percentages of phototrophic and heterotrophic cells targeted by the utilized probes.

^c–, DAPI counts were not performed, so picoeukaryotes could not be differentiated between phototrophs and heterotrophs. In these samples, total picoeukaryote counts were done on FISH filters and were 4,272 cells · ml⁻¹ in Naples-2009 Surface, 1,834 cells · ml⁻¹ in Naples-2009 DCM, and 4,656 cells · ml⁻¹ in Varna Surface. These values were used in the correlations.

Unfiltered seawater was taken for direct cell counts. For total microscopic counts, seawater samples were fixed with glutaraldehyde (1% final concentration) and left for 1 to 24 h at 4°C. Then, aliquots of 20 ml were filtered through 0.6-μm-pore-size polycarbonate black filters and stained with DAPI (4′,6-diamidino-2-phenylindole) at 5 μg · ml⁻¹. Filters were mounted on a slide and stored at −20°C until processed. For specific counts with tyramide signal amplification (TSA)-FISH, aliquots of 100 ml were fixed with filtered formaldehyde (3.7% final concentration), incubated for 1 to 24 h in the dark at 4°C, and filtered through 0.6-μm-pore-size polycarbonate filters (25-mm diameter). Filters were kept at −80°C until processed. For flow cytometry counting of photosynthetic picoeukaryotes, aliquots of 1.5 ml were fixed with a mix of paraformaldehyde and glutaraldehyde (1% and 0.25% final concentrations, respectively), frozen in liquid nitrogen, and stored at −80°C until processed.

Picoeukaryote cell abundance by DAPI staining and flow cytometry.

The total cell abundance of picoeukaryotes was estimated in DAPI-stained filters. Cells were counted with an epifluorescence microscope (Olympus BX61) at 1,000× under UV excitation, changing to blue light excitation to verify the presence or absence of chlorophyll autofluorescence (phototrophic and heterotrophic cells, respectively). A transect of about 13 mm was inspected, and cells were classified in size classes: 2 μm, 3 μm, 4 μm, 5 μm, and >5 μm. All data reported in the study referred to cells within the two smaller size classes (2 to 3 μm), which accounted on average for 82% of the cells.

Cell abundance of photosynthetic picoeukaryotes was determined in a FACSort flow cytometer by using the red fluorescence signal (chlorophyll) after excitation in a 488-nm laser and the side-scattered light of each particle. Fluorescent microspheres (0.95-μm beads) were added as an internal standard (at 10⁵ beads · ml⁻¹). Data were acquired for 2 to 4 min with a flow rate of 50 to 100 μl · min⁻¹ using the settings previously described (34).

Cell abundance of specific picoeukaryote taxa by TSA-FISH.

The specific oligonucleotide probes used targeted several picoeukaryote taxa: NS4 and NS7 targeted the uncultured clades MAST-4 and MAST-7; CRN02 and MICRO01, the species Minorisa minuta and Micromonas spp.; PELA01, the class Pelagophyceae; and ALV01, the environmental clade MALV-II (Table 2). These probes have been published in other studies (see references cited in Table 2 [26–28, 35–37]) except NS7. Probe NS7 was designed here with ARB (38) and targeted 91% of the 192 sequences from MAST-7 available in GenBank; it had 1 mismatch with the remaining MAST-7 sequences and had at least 2 central mismatches with nontarget sequences. Probe NS7 gave a better signal when combined with oligonucleotide helpers contiguous to the probe region (NS7 helper A: AACCAACAAAATAGCAC; NS7 helper B: CCCAACTATCCCTATTAA) that were added to the hybridization buffer at the same concentration as the probe. We tested a range of formamide concentrations to find the best hybridization condition, and we checked that the probe gave a negative signal with a variety of nontarget cultures. Finally, a probe targeting all eukaryotes (EUK502, 37) was also used. All probes were labeled with horseradish peroxidase (HRP).

TABLE 2.

List of oligonucleotide FISH probes used and effectiveness of the probes against reads from this study (% reads by probe)^a

Probe name	Target group	Probe sequence (5′ to 3′)	Probe reference	No. of reads per taxon		% reads by probe
				In OTU table	From the raw reads
NS4	MAST-4	TACTTCGGTCTGCAAACC	Massana et al., 2002 (26)	2,082	2,082	98.0
NS7	MAST-7	TCATTACCATAGTACGCA	This study	2,842	2,833	95.7
CRN02	Minorisa minuta	TACTTAGCTCTCAGAACC	del Campo et al., 2013 (35)	1,853	1,853	99.8
PELA01	Pelagophyceae	ACGTCCTTGTTCGACGCT	Not et al., 2002 (36)	4,440	3,169	98.5
MICRO01	Micromonas spp.	AATGGAACACCGCCGGCG	Not et al., 2004 (28)	11,166
ALV01	MALV-II	GCCTGCCGTGAACACTCT	Chambouvet et al., 2008 (27)	35,359	29,894	83.0
EUK502	Eukaryotes	GCACCAGACTTGCCCTCC	Lim et al., 1999 (37)

^aThe table shows the number of 454 reads from each phylogenetic group extracted from the OTU table or from raw reads by local BLAST. The last column shows the percentage of raw reads in each group that have the probe target region with 0 mismatches.

Hybridizations were performed as previously described (39). Filter pieces (about 1/10) of the 0.6-μm-pore-size polycarbonate filters were covered with 20 μl of hybridization buffer (40% deionized formamide [except 30% for probe CNR01], 0.9 M NaCl, 20 mM Tris-HCl [pH 8], 0.01% SDS) and 2 μl of HRP-labeled probes (stock at 50 ng · μl⁻¹) and incubated overnight at 35°C. After the hybridization, filter pieces were washed twice for 10 min at 37°C with a washing buffer (37 mM NaCl [74 mM NaCl when hybridizing with 20% formamide], 5 mM EDTA, 0.01% SDS, and 20 mM Tris-HCl [pH 8]) and transferred to phosphate-buffered saline (PBS) for 15 min at room temperature. TSA was carried out in a solution (1× PBS, 2 M NaCl, 1 mg · ml⁻¹ blocking reagent, 100 mg · ml⁻¹ dextran sulfate, and 0.0015% H₂O₂) containing Alexa 488-labeled tyramide (4 μg · ml⁻¹) by incubating in the dark at room temperature for 30 to 60 min. Filter pieces were transferred twice to a PBS bath in order to stop the enzymatic reaction and air dried at room temperature. Cells were countersained with DAPI (5 μg · ml⁻¹), and filter pieces were mounted on a slide. Targeted FISH cells were counted by epifluorescence under blue light excitation and checked with UV radiation (DAPI staining) for the presence of the nucleus. Cells labeled with the probe EUK502 were counted using the same size classes as for DAPI counts. Data reported refer to cells of 2- to 3-μm sizes, which accounted on average for 84% of the cells.

HTS by 454 and Illumina.

HTS data derive from papers published during the BioMarKs project (http://www.biomarks.eu/). Total DNA and RNA from 13 picoplankton samples were extracted simultaneously from the same filter. For RNA extracts, contaminating DNA was removed, and RNA was immediately reverse transcribed to cDNA. Data for the 454 sequencing are derived from the work by Massana et al. (32) and used the eukaryotic universal primers TAReuk454FWD1 and TAReukREV3 (40), which amplified the V4 region of the 18S rDNA (∼380 bp). Amplicon sequencing from DNA and cDNA templates was carried out on a 454 GS FLX Titanium system (454 Life Sciences, USA) in Genoscope (http://www.genoscope.cns.fr). The complete sequencing data set is available at the European Nucleotide Archive (ENA) under the accession number PRJEB9133. Data for the Illumina sequencing are derived from the work by Logares et al. (33) and used the eukaryotic universal primers 1398f and 1510r (41), which amplified the V9 region of the 18S rDNA (∼130 bp). Paired-end 100-bp sequencing was performed using a Genome Analyzer IIx (GAIIx) system located at Genoscope. Only RNA (cDNA) samples were sequenced with Illumina. Sequences are publicly available at MG-RAST (http://metagenomics.anl.gov) under accession numbers 4549958.3, 4549965.3, 4549959.3, 4549945.3, 4549943.3, 4549927.3, 4549941.3, 4549954.3, and 4549922.3.

Sequence analysis of HTS reads.

HTS reads by 454 and Illumina were quality checked following criteria similar to those detailed in the original papers (32, 33). After the quality control, chimera detection was run with UCHIME (42) and ChimeraSlayer (43) using SILVA108 and PR2 (44) as reference databases. The final curated reads were clustered into operational taxonomic units (OTUs) by using UCLUST 1.2.22 (45), with similarity thresholds of 97% for V4 reads and 95% for V9 reads. Representative reads of each OTU were taxonomically classified by using BLAST against SILVA108, PR2, and a marine microeukaryote database (46). After the taxonomic assignment, metazoan OTUs were removed. From the complete OTU tables for 454 (32) and Illumina (33) data sets, the samples targeting the picoplankton were extracted: 13 samples for DNA-V4, 13 samples for cDNA-V4, and 9 samples for cDNA-V9. Then, OTUs corresponding to taxa typically larger than 3 μm (Dinophyceae, Ciliophora, Acantharia, Diatomea, Polycystinea, Raphidophyceae, Ulvophyceae, Rhodophyta, and Xanthophyceae; in this order of relative abundance) were removed. These groups accounted for 8.0% to 87.7% (average, 36.9%) of the 454 data set and 11.5% to 73.5% (average, 33.9%) of the Illumina data set. The read numbers in the final OTU tables of picoeukaryotes were 110,258 for DNA-V4, 77,554 for cDNA-V4, and 1,753,600 for cDNA-V9.

The relative abundance of the picoeukaryotic groups of interest was retrieved from these taxonomically classified OTU tables, by dividing the number of reads of the specific OTUs corresponding to the groups of interest by the total number of reads in the sample. Altogether, the six taxa of interest accounted for 36.4% of the DNA-V4 reads, 23.5% of the cDNA-V4 reads, and 32.4% of the cDNA-V9 reads. In addition to the taxonomic classification of OTUs in the OTU table, we classified the unclustered 454 and Illumina reads to obtain the raw reads for probe checking (see Results) and to double-check the taxonomic classification. For this second classification, we downloaded GenBank sequences representative of each picoeukaryotic group of interest and used this specific taxon database to retrieve HTS reads by local BLAST (sequence similarity, >97%).

RESULTS

An overview of total picoeukaryote counts in marine coastal waters.

We estimated the total cell abundance of picoeukaryotes by epifluorescence microscopy and flow cytometry in 13 planktonic samples taken in 6 geographically separated European coastal sites and different depths (Table 1). Total picoeukaryote counts (cells <3 μm) by epifluorescence microscopy of DAPI-stained samples revealed a wide range of cell abundances, from 3,139 cells · ml⁻¹ in Naples-2010 DCM to 24,346 cells · ml⁻¹ in Oslo-2010 DCM (average in all samples, 10,500 cells · ml⁻¹). Phototrophic and heterotrophic cells were differentiated while counting the DAPI samples. The total abundance of phototrophic cells was generally higher than that of heterotrophic cells (average, 8,200 and 2,400 cells · ml⁻¹, respectively), with the exception of Naples-2010 surface, where the two assemblages had similar abundances. In some cases (Blanes, Oslo-2010 DCM, Roscoff, and Varna DCM), phototrophic cells were >6 times more abundant than heterotrophic cells. Counts of phototrophic picoeukaryotes obtained by flow cytometry correlated well with the microscopic counts in the 10 samples analyzed (linear slope, 0.74; Pearson's r = 0.9; P < 0.001). When the regression line was forced to intercept at 0, the slope was 0.90.

The general eukaryotic probe EUK502 was also used to estimate total picoeukaryotic abundance. Cell counts by TSA-FISH were always lower than the DAPI counts (60% on average) (Fig. 1). In fact, the sample with the highest total cell abundance was different if estimated by DAPI (Oslo-2010 DCM) or by TSA-FISH (Oslo-2009 surface). The regression between the two data sets was significant but had a slope very distant from 1 (linear slope, 0.26; Pearson's r = 0.74; P < 0.05). When the line was forced to intercept at 0, the slope was still very low (0.43). There was some tendency to this discrepancy, as TSA-FISH seemed to underestimate more severely the total cell counts in samples dominated by very small cells. Clearly, DAPI counts provided a better estimate than TSA-FISH counts of total picoeukaryotic abundance; therefore, DAPI counts were used to calculate the relative cell abundances of each of the 6 specific picoeukaryotic groups. TSA-FISH counts of each group were in the numerator, and total DAPI counts were in the denominator.

FIG 1.

Comparison of total picoeukaryotic abundance (cells ≤3 μm) by DAPI counts and TSA-FISH counts using the eukaryotic probe EUK502 in all planktonic samples.

Abundance of specific picoeukaryotic taxa.

We used TSA-FISH to estimate the total abundance of six groups of picoeukaryotes, chosen because they were well represented in the sequencing data sets of the picoplankton from the studied samples (and poorly represented in the nanoplankton; see Table S1 in the supplemental material). They belonged to different eukaryotic supergroups: the Stramenopiles (MAST clades and Pelagophyceae), Alveolates (the parasite clade MALV-II), Archaeplastida (Micromonas spp.), and Rhizaria (Minorisa minuta). The taxonomic coverage of the probes used varied from being very narrow, targeting a species (Minorisa minuta) or a constrained phylogenetic clade (Micromonas spp. and the MAST lineages), to being very wide, targeting an algal class (Pelagophyceae) or the diverse MALV-II group (formed by 44 phylogenetic clades). The sum of heterotrophic cells (MASTs, M. minuta, and MALV-II) represented, on average, 36% of heterotrophic picoeukaryotes counted by DAPI, whereas the phototrophic cells targeted (Micromonas and Pelagophyceae) represented, on average, only 22% of phototrophic picoeukaryotes (Table 1).

The cell abundances of the six targeted groups varied strongly among the different samples (see Table S2 in the supplemental material). We found that Micromonas, MAST-4, MAST-7, and MALV-II were the most abundant taxa (average cell abundances of 1,492, 279, 160, and 127 cells · ml⁻¹, respectively) and were detected in all samples. Minorisa minuta was very abundant in some sites but absent in others. In contrast, Pelagophyceae was the least abundant taxon (average cell abundance of 59 cells · ml⁻¹). These cell counts pointed out that each sample contained a different community. Micromonas was the most abundant taxon in 7 samples; MAST-4, in 4 samples; and Minorisa and MALV-II, in the other two samples (see Table S2).

In silico validation of the FISH probes against raw V4 reads.

Before applying TSA-FISH, we evaluated the effectiveness of the probes against the V4 reads obtained from the same samples. This analysis was done with raw reads (extracted from the initial data set by using GenBank sequences of each group as search templates) to take into account all sequence variants. The number of raw reads per group obtained from this way was very similar to the number derived from the OTU table (Table 2). About 1,000 to 3,000 reads were extracted per group (except MALV-II, which had about 30,000 reads). Then, we calculated the percentage of raw reads that had a 100% match with the probes (Table 2). The five specific probes validated this way retrieved a very high percentage of reads, more than 95%, in all cases except MALV-II (83%). Therefore, the vast majority of reads from these five groups in our samples had the target region of the probes.

The probe targeting Micromonas was not designed at the V4 region of the 18S rDNA, so it could not be directly evaluated with V4 reads from this study. Therefore, we took the OTUs affiliating with Micromonas (7 OTUs and 11,166 reads), retrieved the closest GenBank complete sequence from these OTUs (nearly identical at the V4 region), and verified the effectiveness of the probe against these 7 GenBank sequences. Only 3 sequences (accounting for 30% of the reads) exhibited a perfect match, whereas the remaining 4 sequences had a mismatch in the first position of the probe. Thus, probe MICRO01 could be improved perhaps by removing the first base, but since this mismatch is located in the first position, it likely does not affect the FISH counts.

Comparison of group-specific read abundances and TSA-FISH counts.

The relative abundances of 454 V4 reads (from DNA and cDNA templates) and Illumina V9 reads (from cDNA templates) of each group of interest were compared with the relative cell abundance assessed by epifluorescence microscopy (specific TSA-FISH counts relative to total DAPI counts) in 13 samples for the V4 reads, and in 9 samples for the V9 reads (DCM samples from Naples and Oslo were excluded) (Fig. 2). The statistics of these plots are shown in Table 3. For the DNA-V4 survey, the correlation of the relative abundance of cells and the DNA reads was significant for all groups (P < 0.05) except for MAST-4 and Pelagophyceae, and the goodness of these correlations varied among groups; goodness was strongest for Minorisa minuta (R² = 0.97) and weakest for MALV-II (R² = 0.29). Despite these good correlations, linear slopes of the plots were always different from 1 except for MAST-7. In most cases, slopes were below 0.5, indicating an underestimation of cell abundance by 454 reads, while the slope for MALV-II was very high (4.46), indicating a severe overestimation of the molecular signal in this group.

FIG 2.

Comparison of relative abundance of HTS reads against TSA-FISH cell counts in the 13 planktonic samples (9 samples for cDNA-V9 reads) for 6 picoeukaryotic taxa: MAST-4 (a), MAST-7 (b), Minorisa minuta (c), Pelagophyceae (d), Micromonas spp. (e), and MALV-II (f). Dark blue symbols indicate DNA-V4 reads; light blue, cDNA-V4 reads; and green, cDNA-V9 reads. Regression lines are shown, and their statistics are presented in Table 3.

TABLE 3.

Statistics (R², slope value, and P value) of the correlations between relative abundance of reads and cells in three molecular surveys^a

Group	V4-454 survey								V9-Illumina survey (cDNA)
	DNA				cDNA
	R2	Slope	P value	p1b	R2	Slope	P value	p1b	R2	Slope	P value	p1b
MAST-4	0.18	0.14	nsc		0.31	0.21	<0.05	<0.001	0.3	0.84	ns
MAST-7	0.33	0.75	<0.05	ns	0.31	1.16	<0.05	ns	0.36	2.79	ns
Minorisa minuta	0.97	0.24	<0.001	<0.001	0.98	1.01	<0.001	ns	0.99	1.13	<0.001	<0.001
Pelagophyceae	0.06	0.14	ns		0.94	2.78	<0.001	<0.001	0.68	5.68	<0.01	<0.01
Micromonas spp.	0.87	0.47	<0.001	<0.001	0.73	0.83	<0.001	ns	0.87	0.2	<0.001	<0.001
MALV-II	0.29	4.46	<0.05	<0.05	0.39	1.68	<0.05	<0.05	0.60	0.89	<0.05	ns

^aThe three molecular surveys are 454 DNA-V4 (Fig. 2, dark blue), 454 cDNA-V4 (Fig. 2, light blue), and Illumina cDNA-V9 (Fig. 2, green).

^bp1 compares the slopes against the desired value of 1 (i.e., ns indicates that the slope is not significantly different from 1).

^cns, nonsignificant.

In contrast, the correlations between relative cell and read abundances in the cDNA-V4 survey were generally better for all groups and also were significant for Pelagophyceae and MAST-4 (Table 3). Similar to the DNA-V4 survey, each group had a different slope, but, in this case, there were three taxa (MAST-7, M. minuta, and Micromonas) with slopes statistically not different from 1, indicating that their relative abundances obtained by cell counts and 454 reads were comparable. In the 6 groups analyzed, the slopes obtained in the cDNA survey were closer to 1 than the slopes derived from the DNA survey, showing a better performance of the cDNA approach.

For the Illumina cDNA-V9 survey, the correlations were slightly worse than for the cDNA-V4 survey (Fig. 2; Table 3), as they were nonsignificant (P > 0.05) for MAST-4 and MAST-7. Regarding the linear slopes, the three groups with good performances at the cDNA-V4 survey (M. minuta, Pelagophyceae, and Micromonas) had slopes statistically different from 1, indicating that, in these groups, the V4 region (and not the V9 region) could be used as a proxy of cell counts. In contrast, MALV-II had a better correlation with the V9-cDNA reads than with the V4 reads, and its slope was not statistically different from 1. This highlights that there is not a best region that applies to all taxa.

Differences when targeting V4 and V9 regions of the 18S rDNA.

To discard the possibility that the differences observed between the V4 and the V9 regions were due to the use of different sequencing platforms (454 for V4 and Illumina for V9), we sequenced with Illumina (MiSeq platform) the V4 region of one sample of the data set (Oslo-2009 DCM) using both templates (DNA and cDNA). The relative abundances of ∼60 taxonomic groups inferred from the same targeted region (V4) in the 2 platforms displayed a very good agreement, with an R² of 0.97 and of 0.91 (for DNA and cDNA, respectively), and linear slopes of 0.92 to 1.02. Both slopes were not significantly different from 1. Furthermore, this analysis was done in an additional set of 14 samples (from other planktonic size fractions and sediments; data not shown), and the two platforms performed similarly, with R² results ranging from 0.57 to 1.00 (average, 0.91) and slopes ranging from 0.73 to 1.21 (average, 0.99). Therefore, sequencing the same 18S rDNA region with 454 or Illumina (MiSeq) gave highly consistent results.

Therefore, the differences outlined above between V4-454 and V9-Illumina sequencing (Table 3) were due to targeting different 18S rDNA regions and not due to the sequencing platform. In order to observe these differences in more detail, we compared the relative abundances of cDNA-V4 reads and cDNA-V9 reads for the six picoeukaryotic taxa studied here (Fig. 3). Clear and consistent differences were identified in each case. As before, the correlations were good and significant, with R² results ranging from 0.68 to 0.98 (lower in MALV-II: 0.45), but the slopes deviated significantly from 1 (P < 0.05). The V9 analysis significantly increased the relative abundance of the stramenopile groups (the two MAST clades and Pelagophyceae), with slopes ranging from 2.3 to 3.4, while the opposite occurred for Micromonas and MALV-II, which had slopes of 0.2 and 0.3, respectively, and the same occurred for Minorisa minuta (slope, 1.1).

FIG 3.

Comparison of relative abundance of V9-Illumina reads and V4-454 reads (cDNA surveys in both cases) in 9 planktonic samples for 6 picoeukaryote taxa: MAST-4 (a), MAST-7 (b), Minorisa minuta (c), Pelagophyceae (d), Micromonas spp. (e), and MALV-II (f).

DISCUSSION

Identifying marine picoeukaryotes by direct microscopy is problematic because of their small sizes, and, as a consequence, there is an increasing interest in using high-throughput sequencing (HTS) technologies to explore their diversity. HTS surveys provide a detailed picture of the taxa present in the community, including rare species in the assemblage (18, 33), and reveal diversity not evident using other methods. However, the interpretation of the HTS signal in terms of total cell abundances is not straightforward. Interestingly, TSA-FISH is able to bridge microscopic and sequencing approaches by using specific phylogenetic probes to estimate true cell abundances (28, 36). FISH, besides being very laborious, is limited by the number of taxon-specific probes available as well as by the phylogenetic resolution of the probes (47). Moreover, TSA-FISH could be inaccurate due to putative mismatches of the probes with the target group, which would result in cell count underestimates. We addressed this issue by evaluating the six probes against sequences obtained from the same samples, and we found an acceptable performance (very good in four cases: 83% of reads for MALV-II and only one terminal mismatch for Micromonas). This validated that the TSA-FISH cell counts performed here were accurate and supported the main objective of this study, which was to evaluate how well the HTS signal estimates community structure in terms of specific abundance.

More sequences imply more cells.

Since the HTS signal is always relative (number of reads of a given taxon with respect to the total read number), we needed the total picoeukaryote abundance to calculate relative cell abundances. In principle, using TSA-FISH with a universal eukaryotic probe would be consistent with the study and would also provide an extra layer of certainty, since it allows an easier differentiation of eukaryotic cells from fluorescent particles and large bacteria. However, TSA-FISH counts systematically resulted in fewer cells than direct DAPI counts, and we noticed protists that were not labeled with the EUK502 probe. Moreover, this discrepancy was particularly critical in samples dominated by very small cells. The wide size spectra of protist cells in natural samples implied a large variation in the fluorescent signal, so small cells with dim fluorescence may remain unnoticed when close to large fluorescent cells and may easily fade away while counting a field having many cells with diverse sizes and morphologies. This problem did not happen when using specific probes, since we focused on counting a defined cell type (even with dim fluorescence). Therefore, we used the direct DAPI counts to calculate relative cell abundances.

When comparing the relative abundance of HTS reads against the relative cell abundance obtained by TSA-FISH for the different taxa, we generally found a good correlation between the two methods. The R² coefficients of each picoeukaryotic taxon were similar in the three comparisons conducted (DNA-V4, cDNA-V4, and cDNA-V9 versus TSA-FISH), except that there was a very poor correlation for Pelagophyceae in the DNA-V4 survey. Nevertheless, the statistical significance was always better for the cDNA survey than for the DNA survey. These correlations imply that relative read abundance was proportional to relative cell abundance, i.e., an increase in the HTS signal from a particular taxon is the result of an increase of the proportion of targeted cells in the sample. However, the correlation coefficients were far from 1 in most cases, and this noisy signal was probably related to molecular biases plus the large differences in the picoeukaryotic composition of each sample.

Molecular surveys based on a single gene are affected by the widely discussed PCR biases (48). During PCR, some phylotypes can be amplified preferentially, some groups can remain undetected due to primer mismatches (49), or there could be biases due to the number of PCR cycles (50). Thus, it has been suggested that the relative read abundance can no longer reflect the real composition of the original community, biasing diversity estimates and producing over- or underestimations of specific groups (2). Furthermore, sequencing errors may create false or chimeric taxa (16, 51, 52). Our results indicate that PCR biases and putative sequencing artifacts do not affect proportionality between relative read and cell abundance: more reads imply a higher proportion of cells. The significant correlations detected here using this sample data set, where each sample had large differences in the picoeukaryotic composition because they were taken in distant sites and different times of the year, justify the use of relative read abundance as a proxy of community composition for comparative purposes.

Relative abundances of sequences and cells often disagree.

Despite the significant correlations discussed above, HTS and TSA-FISH surveys did not give the same quantitative information, as the regression line often was statistically different from 1. Moreover, these slopes varied strongly among the 3 HTS surveys. In order to compare these surveys, we analyzed the relative abundances of the 6 picoeukaryotic groups (among themselves) in the different samples (Fig. 4). This showed a general agreement between TSA-FISH and the two cDNA surveys, but, depending on the composition of the sample, the agreement was better using the V4 region or the V9 region. In samples dominated by Micromonas (e.g., Blanes, Oslo-2010, Roscoff, Varna DCM), the picture obtained with the V4 region better matched the cell abundance, while the V9 region performed better in samples dominated by stramenopiles (MAST-4, MAST-7, Pelagophyceae). In our samples, the cDNA-V4 survey gave a better representation of the true species composition for 5 of the samples, while cDNA-V9 performed better in 4 of the samples.

FIG 4.

Relative abundance of the different groups (among themselves) shown by the four approaches (TSA-FISH, cDNA-V4, DNA-V4, cDNA-V9) in all planktonic samples. Gray bars indicate the absence of the sample.

In all cases, the DNA survey gave a more biased perspective of the relative abundance of the 6 picoeukaryotic taxa, being influenced by a very high abundance of MALV-II reads in all samples. This is probably due to a particularly high number of rDNA-operon copies in MALV groups (2, 30, 32). The 18S rDNA copy number can vary by orders of magnitude among protist taxa, from a few copies per cell in some green algae (53) to about 30 copies in MAST-4 (54) or several thousand copies in some dinoflagellates (53), depending on the cell size and genome size (55). Large differences in the copy number of the targeted gene will affect the abundance estimates in DNA surveys (2). Moreover, reads retrieved in DNA surveys could derive from dead organisms or dissolved extracellular DNA. It is known that dissolved DNA is preserved in marine waters (56), escaping from degradation and persisting for different periods of time, from hours to days (57). In contrast, reads from cDNA surveys derive from ribosomes and represent metabolically active taxa in the community, as ribosomes are needed to perform the RNA translation in metabolically active cells (58, 59). This, in addition to the 18S rDNA copy number, could explain the differences observed between DNA and cDNA surveys. Moreover, our data also highlighted the impact of targeting different regions of the 18S rDNA gene for estimating relative abundances. For example, the cDNA-V9 survey showed a higher signal (more reads) for MAST taxa and a lower signal for Micromonas compared with cDNA-V4. It is known that the ranges of taxonomic groups detected by V4 and V9 are different (40, 60, 61) and that some groups can be over- or underrepresented. In particular, in our samples, the V4 region gave good estimates of cell counts for MAST-7 and Micromonas spp.; the V9, for MALV-II; and both regions, for Minorisa minuta. Thus, the region targeted (and the primers used) is fundamental to interpret any existing molecular data.

Concluding remarks.

To our knowledge, this is the first study investigating the correspondence between HTS and cell counts for selected and relevant taxa of marine picoeukaryotes. Indeed, true cell abundances of picoeukaryotic taxa require the TSA-FISH approach, but, as this approach has inherent limitations (it is time consuming, few probes are available, and fine resolution cannot be provided), we see the need to pursue HTS studies. Our results indicate a good correlation between the two methods, implying that more cells result in more sequences, although they give different quantitative information, i.e., the relative read abundance cannot be directly related to relative cell abundance. The cDNA-V4 survey showed the best agreement with TSA-FISH abundance, providing 1:1 relationships in half of the assayed taxa, but the cDNA-V9 was best for other taxa. Thus, the targeted region of the 18S rDNA gene clearly affected the relative abundance of specific taxa. Finally, based on the data mentioned here, we suggest that the sequencing platform used (454 or Illumina) does not produce major biases in diversity. In conclusion, the most quantitative option is to use cDNA templates rather than DNA, while the choice of the targeted region will result in different relative abundances in each particular taxon.