Primer Design for an Accurate View of Picocyanobacterial Community Structure by Using High-Throughput Sequencing

The molecular diversity of the aquatic picocyanobacterial community cannot be accurately described using only the available universal 16S rRNA gene primers that target the whole bacterial and archaeal community. We show that the hypervariable regions V5, V6, and V7 of the 16S rRNA gene are better suited to study the diversity, community structure, and dynamics of picocyanobacterial communities at a fine scale using Illumina MiSeq sequencing. Due to its variability, it allows reconstructing phylogenies featuring topologies comparable to those generated when using the complete 16S rRNA gene sequence. Further, we successfully designed a new set of primers flanking the V5 to V7 region whose specificity for picocyanobacterial genera was tested in silico and validated in several freshwater and marine aquatic communities. This work represents a step forward for understanding the diversity and ecology of aquatic picocyanobacteria and sets the path for future studies on picocyanobacterial diversity.

ABSTRACT

High-throughput sequencing (HTS) of the 16S rRNA gene has been used successfully to describe the structure and dynamics of microbial communities. Picocyanobacteria are important members of bacterioplankton communities, and, so far, they have predominantly been targeted using universal bacterial primers, providing a limited resolution of the picocyanobacterial community structure and dynamics. To increase such resolution, the study of a particular target group is best approached with the use of specific primers. Here, we aimed to design and evaluate specific primers for aquatic picocyanobacterial genera to be used with high-throughput sequencing. Since the various regions of the 16S rRNA gene have different degrees of conservation in different bacterial groups, we therefore first determined which hypervariable region of the 16S rRNA gene provides the highest taxonomic and phylogenetic resolution for the genera Synechococcus, Prochlorococcus, and Cyanobium. An in silico analysis showed that the V5, V6, and V7 hypervariable regions appear to be the most informative for this group. We then designed primers flanking these hypervariable regions and tested them in natural marine and freshwater communities. We successfully detected that most (97%) of the obtained reads could be assigned to picocyanobacterial genera. We defined operational taxonomic units as exact sequence variants (zero-radius operational taxonomic units [zOTUs]), which allowed us to detect higher genetic diversity and infer ecologically relevant information about picocyanobacterial community composition and dynamics in different aquatic systems. Our results open the door to future studies investigating picocyanobacterial diversity in aquatic systems.

IMPORTANCE The molecular diversity of the aquatic picocyanobacterial community cannot be accurately described using only the available universal 16S rRNA gene primers that target the whole bacterial and archaeal community. We show that the hypervariable regions V5, V6, and V7 of the 16S rRNA gene are better suited to study the diversity, community structure, and dynamics of picocyanobacterial communities at a fine scale using Illumina MiSeq sequencing. Due to its variability, it allows reconstructing phylogenies featuring topologies comparable to those generated when using the complete 16S rRNA gene sequence. Further, we successfully designed a new set of primers flanking the V5 to V7 region whose specificity for picocyanobacterial genera was tested in silico and validated in several freshwater and marine aquatic communities. This work represents a step forward for understanding the diversity and ecology of aquatic picocyanobacteria and sets the path for future studies on picocyanobacterial diversity.

INTRODUCTION

Picocyanobacteria are common components of plankton and, together with other groups of cyanobacteria, are the only prokaryotes capable of performing oxygenic photosynthesis. They have been present on Earth for 3.5 billon years (1), and during their history they have evolved ecological strategies to adapt to various environmental conditions from the poles to the tropics (2–7). Picocyanobacteria contribute substantially to the primary production of numerous aquatic systems and play a major role in aquatic ecosystem energy flow, as they are at the base of food webs (8–11). Phenotypically, they are single or colonial rod-shaped cells ranging from 0.2 to 3 µm and, based on their phylogeny, they belong to three genera, Synechococcus, Prochlorococcus, and Cyanobium (12). Prochlorococcus is the sister group of the marine Synechococcus and has been divided into distinct clades with ecologically relevant differences in light adaptation (high light [HL] or low light [LL] intensities) and nutrient utilization (13, 14). Synechococcus is a polyphyletic genus containing marine and freshwater taxa (1, 15). Phylogenetic reconstructions resolved three subclusters (named 5.1, 5.2, and 5.3) with marine representatives (16–18), and at least 13 clades of nonmarine picocyanobacteria (19). The 5.1 cluster encompasses most marine groups, and some freshwater strains have also been affiliated to clusters 5.2 and 5.3 (15, 19). Less information is available about the Cyanobium clade. It is a cosmopolitan group represented by strains from systems with contrasting limnological characteristics (i.e., from shallow hypertrophic to deep and oligotrophic lakes), and, based on 16S rRNA gene phylogenetic reconstruction, is closely related to some freshwater Synechococcus clusters (19, 20).

As it occurs with most prokaryotes, picocyanobacteria lack obvious morphological differences and, consequently, the use of molecular techniques is essential to discriminate the distinct taxa (13, 21, 22). During the last decade, new molecular and bioinformatic techniques have changed our perspective on microbial diversity. In particular, high-throughput sequencing (HTS) techniques applied to genetic markers, such as the 16S rRNA gene, have allowed a better understanding of the structure of bacterial communities and their responses to environmental changes (for examples, see references 23–27). To date, most researchers have used the 16S rRNA gene universal primers targeting the whole bacterial community to explore bacterial diversity. However, in these works, the number of reads recovered from the phylum Cyanobacteria hardly exceeds 20% of the total read abundance (25, 27–31). Given that natural environments are composed of a few dominant taxa, using universal primers may overlook the diversity and population genetic structure of less abundant organisms even when using deep sequencing. To increase such genetic resolution, the use of specific primers is the best approach to the study of a particular target group (32). This was the main goal of our work, in order to get a more realistic picture on the diversity and structure of picocyanobacteria in natural communities. Accurate information of picocyanobacterial genetic diversity has been obtained using high-resolution marker genes, such as petB, rbcL, rpoC, cpcBA, mpeBA, and mpeW, and the 16S-23S rRNA internal transcribed spacer (ITS) (4, 6, 15, 17, 18, 33–36). However, this information is strongly biased toward marine picocyanobacterial strains, leaving the freshwater species still undersampled in databases; thus, a comprehensive study of all aquatic cyanobacterial using such functional markers is not currently possible. As an alternative, using the 16S rRNA gene as a marker is still the best-suited approach.

In fact, 16S rRNA gene primer pairs specific for coccoid Cyanobacteria, CYA359F/CYA781(b)R (37), were developed and used in several denaturing gradient gel electrophoresis (DGGE) fingerprinting studies, in which the specificity was verified (38–41). In particular, Boutte et al. (42) observed that the primers designed by Nübel et al. (37) were the most specific for coccoid Cyanobacteria. These primers flank the V2 to V4 region of the 16S rRNA gene, encompassing ca. 500 bp, and could potentially be used in HTS. However, the 16S rRNA gene is a mosaic of conserved and variable regions with different degrees of conservation in different prokaryotic groups (43), and consequently the choice of the region for diversity studies requires careful consideration. The most used variable regions for the characterization of prokaryotic communities are the V4, V3 to V4, V4 to V5, and V6 regions (for examples, see references 26 and 44), but which hypervariable region is the most suitable for use in picocyanobacterium-specific diversity studies has never been evaluated.

With the aim of unveiling the hidden diversity within the picocyanobacteria and in order to establish a systematic approach based on 16S rRNA gene iTags that would facilitate evaluation of picocyanobacterial community structure, we (i) determined which hypervariable region of the 16S rRNA gene provides the most taxonomic and phylogenetic information for the genera Synechococcus, Prochlorococcus, and Cyanobium, (ii) designed specific 16S rRNA gene primers for these genera to be used in the Illumina MiSeq sequencing platform, and (iii) evaluated the specificity and performance of the newly designed primers in marine (Blanes Bay, NW Mediterranean) and freshwater (Chascomús, Carpincho, Bragado, and Monte lakes, the Pampa plain, Argentina) systems.

RESULTS

Defining the most adequate hypervariable region.

To evaluate which 16S rRNA gene region is the most informative in picocyanobacterial, we first created in silico libraries using public picocyanobacterial sequences. Two 16S rRNA gene libraries were constructed: full-length sequence (FLS) and short-length sequence (SLS) libraries; the first contained almost the complete 16S rRNA gene (ca. 1,400 bp). These FLS were then used as the seed to construct the SLS libraries. Two types of SLS libraries were generated, one with 16S rRNA gene fragments containing the hypervariable region V2 to V4 (SLS_V2–V4) and another one with fragments containing the V5 to V7 regions (SLS_V5–V7) (Table S1).

To test for different degrees of variability in the hypervariable regions for different clades, we organized the sequences into five phylogenetic groups. Three groups contained sequences of Synechococcus (designated Syn and listed with consecutive numbers), one contained sequences of Cyanobium (Cyab), and another one contained sequences of Prochlorococcus (Proch) (Table S1). Thus, a total of six FLS libraries were constructed, one containing all of the sequences (FLS-Pcy) and one for each phylogenetic group (Table S1). The 16S rRNA gene fragments of SLS were extracted from each of the 6 FLS libraries, resulting in a total of 12 SLS libraries (Table S1). Once the FLS and SLS libraries were ready, we constructed maximum likelihood (ML) phylogenetic trees and analyzed (i) the genetic variability and (ii) the phylogenetic information contained in each of the V2 to V4 and the V5 to V7 variable regions.

The genetic variability along the different 16S rRNA gene regions was explored through the Faith phylogenetic diversity index (PD_Faith) (45) and the mean-nearest-taxon-distance (MNTD) (46). The PD_Faith value is defined as the sum of the minimum lengths of all branches in a phylogenetic tree (45), while the MNTD value calculates the phylogenetic distance between each taxon within a community and its closest relative (46). Higher values of both metrics indicate greater phylogenetic diversity but MNTD is more sensitive to variations of closely related taxa than PD_Faith (45). In this study, the higher values of PD_Faith were observed in the trees constructed with the short-length sequence (SLS) libraries containing the V5 to V7 regions (i.e., SLS_V5–V7) and were significantly higher than those from libraries containing the V2 to V4 regions (i.e., SLS_V2–V4) (Fig. 1a), indicating larger genetic variability in the V5 to V7 16S rRNA gene hypervariable regions for picocyanobacterial taxa. The same pattern was observed when phylogenetic groups were analyzed separately (except for the Syn2 group) (Fig. 1a). The results obtained with the MNTD index were in agreement with those obtained with the PD_Faith index (Fig. 1b). The highest values were observed in the SLS_V5–V7 libraries, being significantly higher for SLS-Pcy (all picocyanobacterial sequences) and SLS-Proch (Prochlorococcus sequences). Thus, these results confirm that the fragment containing the V5 to V7 hypervariable region features more genetic variability than that of the V2 to V4 hypervariable region.

FIG 1.

Values of PD_Faith (a) and MNTD (b) indices obtained from phylogenetic trees constructed from each short-length sequence library (SLS), consisting of one containing all the sequences (Pcy) and one for each phylogenetic group (Cyab, Cyanobium; Syn, Synechococcus; Proch, Prochlorococcus). Significant differences between the compared SLSs are indicated with an asterisk.

To evaluate the phylogenetic information contained in the different regions, we compared the topology of the phylogenetic trees constructed from the short- and full-length sequence libraries. First, we calculated the patristic distances (i.e., the sum of branch lengths that link two tips in the tree [PDist]) between all possible pairs of sequences in each constructed phylogeny and correlated each PDist of the SLS trees with those of the FLS trees (see methodology). The PDist correlations were always higher for FLS and SLS containing the V5 to V7 hypervariable region (SLS_V5–V7) than for those in the SLS_V2–V4. This occurred when using all the sequences and also when each phylogenetic group was studied separately (Table 1 and Fig. 2). We also calculated the Robinson-Foulds (RF) and weighted Robinson-Foulds (WRF) indices (47). Both metrics count the number of different bipartitions between two trees. The lower the values, the more similar the trees are with respect to their topology. The difference between the RF and WRF indices is that the latter takes into account the bootstrap support values of the bipartitions instead of looking only at their presence or absence, and thus bifurcations with lower support are less penalized. In all cases, the RF and WRF indices were significantly lower for the V5 to V7 fragment than for the V2 to V4 fragment (Table 1), indicating that the number of different bipartitions between the FLS and SLS trees occurred to a lesser degree in FLS versus SLS_(V5–V7) than in FLS versus SLS_(V2–V4) trees (Table 1). In addition, the WRF values were much lower than the values of their respective RFs, indicating that the different bipartitions occurred in nodes with low bootstrap values. Although this was observed for both regions, the lowest values were between FLS and SLS_(V5–V7). To further strengthen our results, we evaluated the congruency between FLS and SLS tree topologies by the F measure, defined as the harmonic mean of precision and recall (see Materials and Methods). The F values range from zero for fully dissimilar trees to a maximum value for fully similar trees, value approaches one as the topologies of both trees become similar. For all compared trees, the SLS_(V5–V7) libraries showed high F measure values compared to those of SLS_(V2–V4) libraries, supporting the results described above (Table 1).

TABLE 1.

Summary of measures and indices used to compare the phylogenetic information in trees constructed using FLS and SLS libraries

Library	Region	Replicate of SLS	Patristic PCa	RF distanceb	WRF distancec	F measured
FLS-Pcy	SLS-Pcy(V2–V4)	a	0.546	2,460	46.777	0.072
		b	0.544	2,462	47.606	0.077
		c	0.548	2,433	46.880	0.080
	SLS-Pcy(V5–V7)	a	0.612	2,402	44.327	0.096
		b	0.618	2,418	45.285	0.091
		c	0.712	2,412	43.970	0.089
FLS-Cyab	SLS-Cyab(V2–V4)	a	0.442	132	1.500	0.083
		b	0.542	133	1.580	0.082
		c	0.426	134	1.600	0.069
	SLS-Cyab(V5–V7)	a	0.541	120	1.470	0.142
		b	0.441	122	1.460	0.115
		c	0.440	122	1.500	0.102
FLS-Syn1	SLS-Syn1(V2–V4)	a	0.190	1,116	9.191	0.046
		b	0.197	1,114	8.890	0.044
		c	0.207	1,110	8.862	0.039
	SLS-Syn1(V5–V7)	a	0.318	1,104	8.620	0.049
		b	0.311	1,107	8.624	0.053
		c	0.323	1,102	8.590	0.054
FLS-Syn2	SLS-Syn2(V2–V4)	a	0.712	418	6.967	0.087
		b	0.6936	417	6.780	0.062
		c	0.7121	420	6.770	0.054
	SLS-Syn2(V5–V7)	a	0.763	402	6.480	0.122
		b	0.7631	408	6.210	0.105
		c	0.8031	404	6.490	0.098
FLS-Syn3	SLS-Syn3(V2–V4)	a	0.456	46	0.360	0.281
		b	0.569	43	0.340	0.317
		c	0.490	45	0.330	0.285
	SLS-Syn3(V5–V7)	a	0.725	0	0.000	1.00
		b	0.689	2	0.090	0.968
		c	0.698	1	0.087	0.984
FLS-Proch	SLS-Proch(V2–V4)	a	0.544	606	10.834	0.087
		b	0.573	603	11.021	0.087
		c	0.551	607	10.720	0.084
	SLS-Proch(V5–V7)	a	0.737	576	10.272	0.132
		b	0.739	575	10.190	0.127
		c	0.737	577	10.212	0.124

^aPatristic PC, Pearson correlation of patristic distances assessed by the Mantel test. In all cases, P < 0.05.

^bRF, Robinson-Fould.

^cWRF, weighted Robinson-Fould.

^dF measure = 2 × [(precision × recall)/(precision + recall)].

FIG 2.

Patristic distances of FLS and SLS_(V2–V4) trees (upper panel) and FLS and SLS_(V5–V7) trees (lower panel). The distances were normalized to a maximum value of one and ordered at intervals of 0.01according to the distance from the FLS tree. Each point represents the average of the FLS distance and corresponding averaged SLS distances in each interval. The standard deviations of the averaged SLS distances were used as error bars on the chart.

Furthermore, to test whether the topologies of the constructed SLS trees were a methodological artifact generated by sequence length, we correlated the PDist values of the three trees constructed from each SLS library. In all cases, the correlations between PDist values were close to one (Fig. S1), confirming that the differences in topology between the SLS and FLS trees described above are reliable.

Design of specific primers for picocyanobacteria.

Based on the previous results, and in order to cover Cyanobacteria of the genera Synechococcus, Prochlorococcus, and Cyanobium, we designed a set of primers flanking the hypervariable region V5 to V7 of the 16S rRNA gene (fragment size, 520 bp), Cya-771F (5′-AGGGGAGCGAAAGGGATTA-3′) and Cya-1294R (5′-GCCTACGATCTGAACTGAGC-3′). In silico analysis (Table 2) demonstrated a high specificity for the target sequences, i.e., more than 99% of the total sequences recovered by the primers Cya-771F/1294R belong to the three cyanobacterial picoplanktonic genera. Additionally, from 90.67% to 94.28% of total sequences within each genus (i.e., Prochlorococcus, Synechococcus, and Cyanobium) could be recovered with our primer set, indicating a high coverage rate for these specific picocyanobacterial genera (Table 2). We also compared our results with the coverage range and specificity obtained with the primers proposed by Nübel et al. (37) specific for coccoid Cyanobacteria. When perfect matches were considered (i.e., 0 mismatches), our set of primers displayed higher specificity and coverage for the sequences of the targeted organisms (Table 2), while a greater number of sequences for other Cyanobacteria genera (e.g., Leptolyngbyales, Nostocales, Oxyphotobacteria, and Phormidesmiales) were recovered using the Nübel et al. (37) primers. Furthermore, when 1 and 2 mismatches were allowed, both sets of tested primers increased their target group sequence coverage and decreased their specificity, as expected (Table S2 and Table S3). However, when considering 2 mismatches, Nübel et al. (37) primer specificity declined from 99.7% to 97.1%, while the change in specificity for our primers was lower (from 99.9% to 99.2%) (Table S3). Finally, we calculated the coverage and specificity of two sets of universal prokaryotic primers broadly used in microbial community studies, Bact_341F (5′-CCTACGGGNGGCWGCAG-3′) and Bact_805R (5′-GACTACHVGGGTATCTAATCC-3′), proposed by Herlemann et al. (48), and 515F-Y (5′-GTGYCAGCMGCCGCGGTAA-3′) and 926R (5′-GTGYCAGCMGCCGCGGTAA-3′) from Parada et al. (26). Regardless of the number of mismatches allowed (0,1, or 2), the coverage ranges for the targeted organisms were similar to those obtained with the specific primers; however, the specificity values obtained with universal primers were, in both cases, considerably lower (Tables S4 to S9 and Fig. S2).

TABLE 2.

In silico evaluation of specificity and coverage of the primer pairs designed in this work compared with those designed by Nübel et al. (37), with zero mismatches allowed

Taxonomya	Coverage (%)b		Matchc		Eligibled		Specificity (%)e		Outgroup_matchedf		Outgroup_matchableg
	This work	Nübel et al. (1997)	This work	Nübel et al. (1997)	This work	Nübel et al. (1997)	This work	Nübel et al. (1997)	This work	Nübel et al. (1997)	This work	Nübel et al. (1997)
Archaea	0	0	0	0	15,292	23,948	99.74	99.49	1.650	3.313	645,343	647,636
Bacteria	0.29	0.58	1.650	3.314	572,584	575,263	100.00	100.00	0	0	88,051	96,341
Bacteria; Actinobacteria	0.00	0.00	1	0	59,305	59,486	99.73	0.54	1.649	3.313	601,330	612,098
Bacteria; Proteobacteria	0.00	0.00	1	1	228,791	229,791	99.62	99.25	1.649	3.312	431,844	441,793
Bacteria; Cyanobacteria	12.27	24.57	1.648	3.313	13,434	13,486	99.99	100.00	2	0	647,201	658,118
Bacteria; Cyanobacteria; Oxyphotobacteria	12.83	25.69	1.648	3.313	12,846	12,894	99.99	100.00	2	0	647,789	658,710
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales	90.15	89.73	1.638	1.652	1,817	1,841	99.99	99.75	12	1.661	658,818	669,763
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Cyanobiaceae	91.97	90.76	1.638	1.620	1,781	1,785	99.99	99.75	12	1.693	658,854	669,799
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Cyanobiaceae; Prochlorococcus MIT9313	90.67	89.85	408	602	450	670	99.81	99.60	1.242	2.711	660,185	670,934
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Cyanobiaceae; Synechococcus CC9902	90.88	89.14	578	402	636	451	99.84	99.57	1.072	2.911	659,999	671,133
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Cyanobiaceae; Cyanobium PCC-6307	94.28	92.62	610	602	647	650	99.84	99.60	1.040	2.711	659,988	670,934
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Cyanobiaceae; Synechococcus MBIC10613	100.00	91.04	5	579	5	636	99.75	99.59	1.645	2.734	660,630	670,948
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Cyanobiaceae; spongiarum group	85.71	76.19	36	32	42	42	99.76	99.51	1.614	3.281	660,593	671,542
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Prochlorotrichaceae	0.00	80.00	0	4	11	5	99.75	99.51	1.650	3.309	660,624	671,579
Bacteria; Cyanobacteria; Oxyphotobacteria; Synechococcales; Synechococcaceae	0.00	100.00	0	11	15	11	99.75	99.51	1.650	3.302	660,620	671,573
Bacteria; Cyanobacteria; Oxyphotobacteria; Chloroplast	0.00	0.12	0	7	6,031	6,040	99.75	99.50	1.650	3.306	654,604	665,544
Bacteria; Cyanobacteria; Oxyphotobacteria; Eurycoccales	0.00	22.22	0	10	45	45	99.75	99.51	1.650	3.303	660,590	671,539
Bacteria; Cyanobacteria; Oxyphotobacteria; Leptolyngbyales	3.06	89.34	6	176	196	197	99.75	99.53	1.644	3.137	660,439	671,387
Bacteria; Cyanobacteria; Oxyphotobacteria; Limnotrichales	5.88	82.35	1	14	17	17	99.75	99.51	1.649	3.299	660,618	671,567
Bacteria; Cyanobacteria; Oxyphotobacteria; Neosynechococcales	0.00	100.00	0	1	1	1	99.75	99.51	1.650	3.312	660,634	671,583
Bacteria; Cyanobacteria; Oxyphotobacteria; Nostocales	0.03	18.58	1	700	3,757	3,768	99.75	99.61	1.649	2.613	656,878	667,816
Bacteria; Cyanobacteria; Oxyphotobacteria; Oxyphotobacteria incertae sedis	0.41	85.45	2	423	493	495	99.75	99.57	1.648	2.890	660,142	671,089
Bacteria; Cyanobacteria; Oxyphotobacteria; Phormidesmiales	0.00	86.60	0	252	291	291	99.75	99.54	1.650	3.061	660,344	671,293
Bacteria; Cyanobacteria; Oxyphotobacteria; Pseudanabaenales	0.00	4.90	0	5	102	102	99.75	99.51	1.650	3.308	660,533	671,482
Bacteria; Cyanobacteria; Oxyphotobacteria; RD011	0.00	87.50	0	7	8	8	99.75	99.51	1.650	3.306	660,627	671,576
Bacteria; Cyanobacteria; Oxyphotobacteria; RD017	0.00	83.33	0	10	12	12	99.75	99.51	1.650	3.303	660,623	671,572
Bacteria; Cyanobacteria; Oxyphotobacteria; SepB-3	0.00	87.50	0	7	8	8	99.75	99.51	1.650	3.306	660,627	671,576

^aSILVA taxonomy, SSU r132 database; Ref NR sequence collection.

^bCoverage % = (match/eligible) × 100.

^cNumber of match hits in the taxonomic path.

^dNumber of hits with sequence data at the primers’ position in the taxa path.

^eSpecificity %= 100 − (outgroup_matches/outgroup_matchable) ×100.

^fNumber of matched hits outside the taxonomic path.

^gNumber of matchable hits outside the taxonomic path.

Testing the primers with environmental samples.

The specificity and usefulness of the designed primers were tested in both freshwater and marine samples using Illumina sequencing. We obtained totals of 21,477 and 56,610 high quality reads from the marine (Blanes Bay) and freshwater (Argentinian lakes) samples (see Materials and Methods), respectively. All sequences from Blanes Bay were assigned to either the genus Synechococcus or Prochlorococcus, and 96.3% of the reads retrieved from lakes belonged to the genera Synechococcus (32%), Cyanobium (12%), and environmental clones closely related to them (51%) (Table S10). The remaining percentage was assigned to filamentous Cyanobacteria (2.3%) of the Cylindrospermosis and Leptolynbya genera, as well as to heterotrophic bacteria (1.3%), particularly Planctomycetales and Verrucomicrobiales. Rarefaction curves showed an asymptotic tendency only in two out of the ten samples (in Chascomús lake during the spring of 2012 and in the summer sample from Blanes Bay), while in the rest of the samples, picocyanobacterial diversity did not reach saturation with the sequencing effort used in this work (Fig. S3). A total of 113 zero-radius operational taxonomic units (zOTUs) within the picocyanobacterial genera were defined, with 87 zOTUs being exclusive to fresh waters and 26 zOTUs being exclusive to the marine site (Blanes Bay).

The samples from Blanes Bay correspond to two different periods of the year (spring and summer) when peaks of picocyanobacteria are frequently registered. Out of 26 zOTUs, 24 belonged to Synechococcus, and only 2 displayed high similarity with sequences of Prochlorococcus (Fig. 3), which contributed with less than 2% to the total abundance in each sample (Prochlorococcus spp. are more abundant in the fall-winter season in this site [49]). The taxonomic composition was very similar in both samples, including only three zOTUs exclusive to one season (zOTU-91 and zOTU-117 from summer, and zOTU-101 from spring), and the community in summer showed a higher diversity (Shannon index = 3.69; PD_Faith = 1.8) than in spring (Shannon index = 3.44; PD_Faith = 1.6) (Table 3). Despite community structure being significantly different between seasons (Student t test, P < 0.001), there was a single sequence (zOTU-9) that dominated in both seasons, contributing 63.4% of the total number of reads in spring and 30.6% in summer (Fig. 3). All Synechococcus zOTUs belonged to the marine subcluster 5.1, clustering into five phylogenetic clades defined in the reference tree (RT), I, IV, V, CRD1, and WPCI (Fig. 3 and Fig. S4). In both samples, the most abundant lineage was clade I, followed by clades WPC1 and IV, which exhibited a larger relative contribution in summer (Fig. 4a). Regarding the Prochlorococcus zOTUs, both of them belonged to clade HLI and were closely related to the MED4 strain retrieved from the Mediterranean Sea (Fig. 3 and Fig. S4).

FIG 3.

Relative abundance of zOTUs determined for the two samples from Blanes Bay (left. summer; right, spring). For each zOTU, the GenBank accession number and the taxonomic and phylogenetic clade assignments are presented.

TABLE 3.

Number and abundance of zOTUs per sample and values of the different indices used for the characterization of the communities

Sample type	System	Season (sample date)	No. of zOTUs	No. of zOTUs >1%a	% of reads >1%b	Index value
						Shannon	Chao1	PDFaithc
Marine	Blanes Bay	Spring (April 13)	24	10	96.5	3.44	21	1.6
	Blanes Bay	Summer (August 13)	25	14	94.2	3.69	22	1.8
Freshwater (temporal)	Chascomús	Spring (October 12)	66	24	86.5	3.32	71	0.28
	Chascomús	Winter (June 13)	72	22	88.9	3.12	85	0.46
	Chascomús	Spring (November 13)	76	24	84.5	3.37	87	0.43
	Chascomús	Summer (February 14)	83	25	85.2	3.48	86	0.49
	Chascomús	Autumn (May 14)	84	25	80.2	3.59	86	0.49
Freshwater (spatial)	El Carpincho	Spring (December 09)	59	30	87.0	2.79	73	0.26
	Bragado	Spring (December 09)	78	13	84.2	3.37	80	0.44
	Monte	Spring (December 09)	68	24	91.3	3.39	71	0.45
	Chascomús	Spring (November 13)	65	23	89.4	3.12	69	0.36

^aNumber of zOTUs composing the “abundant fraction” in each sample (i.e., abundance of >1%).

^bPercentage contribution of zOTU abundant fraction to the number of total reads in each sample (i.e., abundance of >1%).

^cPD_Faith, Faith phylogenetic diversity index.

FIG 4.

Bar graph showing the relative abundance of zOTUs per clades in (a) Blanes Bay, (b) Chascomús lake, and (c) Pampean lakes, based on the phylogenetic reconstruction shown in Fig. S4 to S6.

A total of 86 zOTUs were obtained from the five seasonal samples taken from Chascomús lake (CH). Richness varied between 66 (spring 2012) and 84 zOTUs (fall 2014), and the Shannon index varied little between samples (Table 3). However, the phylogenetic diversity (PD_Faith) was notably lower in spring 2012 (Table 3). A cluster analysis revealed higher similarity in the picocyanobacterial community structure in samples within the same year, independent of the season (Fig. 5a). None of the environmental variables measured were significantly associated with community structure. In each sample, the abundant fraction of the communities was represented by 22 to 25 zOTUs (Table 3), accounting for more than 80% of total reads. Most zOTUs (67.8%) were found all year long. In particular, zOTU-2 was always present at >5% of relative abundance (Fig. 5a). Most zOTUs clustered into 11 clades previously defined in the reference tree (Fig. 5a and Fig. S5), while 4 new clades were formed exclusively by the newly obtained zOTUs (the new clades were named 1 to 4 in Fig. S5). zOTUs composing the core group (those that contributed with more than 5% of the reads) belonged to different phylogenetic clades, and their contribution changed markedly among samples (Fig. 4b and 5a). For instance, clade P_I was prevalent in spring and winter and clade P_IV in summer 2014, and clade A was more represented in 2014 (Fig. 4b).

FIG 5.

Cluster analysis of zOTU abundance from (a) Chascomús lake and (b) the four Pampean shallow lakes studied. The heat maps represent the relative abundances of the abundant zOTUs (i.e., abundance of ≥1%). The asterisks indicate the zOTUs with a relative abundance of >5%. For each zOTU, the GenBank accession number and the taxonomic and phylogenetic clade assignments are presented.

In the spatial comparison between the four Pampean lakes, zOTU richness varied between 78 in Chascomús and 59 in Bragado lakes. Bragado lake also had the lowest taxonomic and phylogenetic diversity (Table 3). Only 10 out of the total of 85 zOTUs were exclusive to any particular environment. The cluster analysis showed that the Monte and El Carpincho lakes presented a more similar community composition to each other than the other lakes (Fig. 5b), and the structure of the picocyanobacterial community was highly correlated with the quality of light (expressed as the relation between the vertical diffuse light attenuation coefficient for red [675 nm] and green [440 nm] [Kd_red/Kd_green]; Table S11) and total phosphorus concentration (Mantel test, r = 0.83, P < 0.05). The number of abundant zOTUs (i.e., relative abundances of >1%) varied between 13 (Bragado) and 30 (El Carpincho), and in all cases they contributed with more than 84% to the total reads of each lake (Table 3). The zOTUs were affiliated with the 13 clades previously described in the temporal study of Chascomús (Fig. S6). The dominant zOTUs of each environment belonged to different clades (Fig. 5b and Fig. S5). The most notable case was clade 4, which was almost exclusively represented by sequences from Bragado (Fig. 4c). Clade P_VII was also more prevalent in this lake. Clade P_I showed the highest relative abundance in Chascomús and El Carpincho lakes, and clade A was dominant in Monte Lake (Fig. 4c).

DISCUSSION

We found that the hypervariable region V5 to V7 of the 16S rRNA gene is the most appropriate for studying the diversity, structure, and dynamics of picocyanobacterial communities using high-throughput sequencing. Specifically, we found that (i) the V5 to V7 region features more variability in picocyanobacteria than other regions and (ii) that the phylogenies based on the V5 to V7 region mirror to a larger extent the phylogenies obtained with the complete 16S rRNA gene. Additionally, we successfully designed and used a set of primers flanking the V5 to V7 region, which target almost exclusively Synechococcus, Prochlorococcus, and Cyanobium, thus allowing the accurate characterization of natural picocyanobacterial communities.

We explored the taxonomic information contained in the hypervariable regions of the 16S rRNA gene using indices with different sensitivity to phylogenetic relatedness (i.e., PD_Faith and MNTD). The results indicate that the V5 to V7 region allows for a better discrimination of both phylogenetically distant and closely related picocyanobacteria (46). Which region is suitable for studying prokaryotic diversity has mostly been explored for entire bacterial communities and other taxonomic groups (26, 50–52); however, to our knowledge, this is the first time that it has been evaluated for picocyanobacterial genera. Reviewing the information present in the literature (26, 50–52), we detected contrasting results about the suitability of 16S rRNA gene hypervariable regions for the study of prokaryotic communities, which alerted us about the importance of selecting a particular hypervariable region of 16S rRNA according to the taxonomic level/group at which the results need to be inferred (53).

The study of phylogenetic relationships (i.e., patristic distance, RF and WRF indices, and F value) indicated that the information present in fragments of the 16S rRNA gene containing the V5 to V7 region allows the inference of the phylogenetic relationships obtained from the full-length 16S rRNA gene for the picocyanobacterial groups. In previous studies, the V4 fragment was found to be the best for establishing phylogenetic relationships when the entire bacterial community was considered (26, 52, 54). However, this contrasts with our results, which suggest that the hypervariable region of the 16S rRNA gene that best reflects the full-length gene phylogeny depends on the taxonomic groups being studied. Additionally, we observed a high correlation between tree topologies constructed from the same SLS, which suggests that (i) the difference in the correlations of SLS_(V2–V4) and SLS_(V5–V7) with the FLS may be attributable to the phylogenetic information present in the fragments instead of the size of the sequences and (ii) the phylogenies obtained from partial sequences have a good reproducibility power.

The specificity of the new set of primers analyzed in silico was validated in natural communities. Nevertheless, some discrepancies in the recovered taxa were detected, indicating the importance performing primer tests on environmental samples. For example, some reads were assigned to genera that appeared in the in silico analyses allowing 1 or 2 mismatches (e.g., Planctomycetes and Verrucomicrobia), while some genera typical of aquatic systems (e.g., Pseudanabaenales, Actinobacteria, and Firmicutes) that were recovered in the in silico analysis did not appear in the zOTU table.

On the other hand, by defining operational taxonomic units as exact sequence variants (zOTUs) (55) we could achieve a fine-scale taxonomic resolution (55, 56), allowing us to detect greater genetic diversity among the picocyanobacteria, as well as specific patterns of community composition in different systems and seasons. In Blanes Bay, Synechococcus reads were much more abundant and diverse than those of Prochlorococcus, as was also observed in the cell abundances (Table S11). The number of operational taxonomic units defined here as zOTUs was greater than those established in previously studies, where bacterial primers 341F (5′-CCTACGGGNGGCWGCAG-3′) (48) and 806RB (5′-GGACTACNVGGGTWTCTAAT-3′) (44) were used to amplify partial 16S rRNA genes (i.e., 10 zOTUs of Synechococcus and 1 zOTU of Prochlorococcus; I. Ferrera and J. M. Gasol, personal communication). Picocyanobacterial community composition differed between spring and summer, yet there was a unique zOTU that dominated in both seasons. This supports observations from other sites that also revealed the presence of a single strain of Synechococcus throughout the year (57, 58), and points toward the existence of a single strain adapted to grow in different seasons characterized by quite different temperatures and nutrients (for example, see reference 49). When analyzing the phylogenetic structure of the communities, we observed that most Synechococcus zOTUs correspond to clades typical of coastal zones, clades I, IV, V, and WPC1 (15, 18, 59), which tend to cooccur (for examples, see references 4, 15, 29, 33, 59). However, zOTU-17 was assigned to clade CRD1, which has been described from the Costa Rica upwelling (60), as well as from the Pacific, Indian and Atlantic Oceans, although it is seemingly absent from the Mediterranean Sea (6). Regarding Prochlorococcus zOTUs, both belonged to the HLI clade, a dominant clade in surface waters with moderate stratification (17, 61–63).

In Chascomús lake there was a core of zOTUs (i.e., relative abundance of >5%) that changed their relative contribution between seasons. This pattern is usually observed in temperate lakes regardless of their limnological characteristics (22, 64) and suggests different ecological roles within genetically close strains. However, as in previous reports (22, 64), we failed to relate the distinct niches among the dominant zOTUs with environmental factors, such as nutrient concentrations, pH, underwater light, or temperature (Table S11). This may suggest that biotic feedback mechanisms, such as competition, viral infection, and prey-predator interactions are significant factors driving the temporal dynamics of the picocyanobacterial community, and further studies are needed to shed light on this issue. Regarding the spatial presence of the different zOTUs, each Pampean shallow lake was represented by a particular assemblage associated with local environmental factors. These results lend support to previous evidence about the importance of underwater light conditions and the availability of nutrients in governing freshwater picocyanobacterial distribution (35, 64–68).

Finally, the incorporation of zOTUs into the phylogenetic tree allowed the observation of a large number of coexisting clades with specific abundance dynamics changing over space and time. This has been well reported by other authors (20, 35, 69, 70) and seems to be a common feature in picocyanobacterial communities. What allows the coexistence of these very similar organisms? The great genetic microdiversity within picocyanobacterial genera may effectively be a good explanation. However, how environmental factors drive changes in clade abundances and whether cooccurring phylotypes within the same and different clades affect each other in beneficial or detrimental ways are questions that still need to be posed.

In summary, we determined that the V5 to V7 hypervariable region of the 16S rRNA gene provides more taxonomic and phylogenetic information for Synechococcus, Prochlorococcus, and Cyanobium than do other hypervariable regions, and we designed a set of specific primers for these genera. The results obtained from marine and freshwater environments showed the usefulness of the present approach and tools to explore the diversity and successional patterns in the picocyanobacterial assemblage via fine (i.e., zOTU) or coarse-grained (clade and subclade) taxonomic resolutions.

MATERIALS AND METHODS

Defining the most adequate hypervariable region.

To determine the most informative hypervariable 16S rRNA gene region, we followed the methodology proposed by Ghyselinck et al. (52). Genetic libraries were constructed using 1,334 sequences from representative picocyanobacterial clades selected from the public database SILVA 119 (https://www.arb-silva.de/download/arb-files/) (Table S1). The full-length sequence (FLS) libraries contained complete sequences with the information of all hypervariable regions (V1 through V9). To generated the short-length sequence (SLS) libraries, we searched for the V4 and V6 hypervariable regions from each FLS and trimmed out fragments of 500 bp in length containing either the V4 or the V6 region using V-Xtractor v 2.0 (71).

For each FLS and SLS library constructed, we aligned the sequences using Multiple Alignment using Fast Fourier Transform (MAFFT) v.7 with default options (72) and constructed maximum likelihood (ML) phylogenetic trees using FastTree v.2.1.5 (73) with default options using the Geneious v 9.1. interface (74). A single ML tree was constructed for each of the 6 FLS libraries, along with 3 trees for each of the 12 SLS libraries, using the general time reversible (GTR) model with a gamma-distributed rate of variation across sites. To estimate statistical support, 1,000 pseudoreplications were performed using the “rapid bootstrapping” algorithm. Three sequences of Gloeobacter sp. were used as an outgroup. In those analyses in which rootless trees were required, these outgroup sequences were discarded. The generated trees were used in downstream analyses.

Genetic variability was explored using the Faith phylogenetic diversity index (PD_Faith) (45) and the mean nearest taxon distance (MNTD) (46). PD_Faith is defined as the sum of the phylogenetic branch lengths connecting all taxa present in a community; higher values of PD_Faith would indicate greater phylogenetic diversity and therefore greater genetic variability (45). The MNTD quantifies the phylogenetic distance between each taxon within a sample and its closest relative in the sample, returning an average (46). Thus, higher MNTD values indicate greater phylogenetic diversity among closely related taxa. To calculate PD_Faith and MNTD values we used the package Picante (v.1.6) (75) within the R environment.

The phylogenetic information present in the different regions of the 16S rRNA gene was evaluated comparing the tree topology through the patristic distances (PDist), the Robinson-Foulds (RF) distances and the Robinson-Foulds weighted (WRF) distances (47). The PDist values were calculated between all possible pairs of sequences in each phylogeny constructed and were presented as distance matrices. Then, correlations between each SLS (in triplicate) and FLS distance matrix were tested using Mantel tests based on Pearson correlations. In order to visualize these data, the distances were normalized to a maximum value of one and ordered at intervals of 0.01 according to the distance from the FLS tree. For each interval, we calculated the average of the FLS distance and corresponding averaged SLS distances and then plotted them in a graph. The standard deviations of the averaged SLS distances were used as error bars in the chart (54).

Similarly, the RF distances allowed us to compare topologies of phylogenetic trees by counting the number of bipartitions that occur in one tree but not in the other, so a lower RF value indicates that a pair of trees is topologically more similar. We also calculated the WRF distance (47). This measure, unlike the RF distance, takes into account the bootstrap value of each bipartition instead of considering presence/absence only. For example, if a bipartition is present in one tree but not in another and has a bootstrap value of 0.3, then the WRF distance counts it as 0.3 instead of 1. Thus, by using the WRF distance, the bifurcations with low support are less penalized. When the WRF values are lower than the respective RF values, it means that the differences between trees occurred in clades with low bootstrap values. The patristic distances and RF/WRF metrics were calculated in the R environment using the stats and phangorn v.2.4 R packages (76).

Finally, to further strengthen our results, we compared the topologies between SLS and FLS trees using the precision and recall metrics. These measures were defined as follows: precision = TP/(TP + FP) and recall = TP/(TP + FN), where true positives (TP) are the set of bipartitions present in both the SLS tree and the FLS tree, false positives (FP) are the number of bipartitions present in the SLS tree but not in the FLS, tree and FN is the number of bipartitions present in the FLS tree but not in the SLS tree (77). These metrics were calculated in the R environment using the phangorn v.2.4 R package (76). The results were expressed as the F measure, which is defined as the harmonic mean of precision and recall and balances false positives and false negatives [F = 2 × (precision × recall)/(precision + recall)].

Primer design.

We designed a set of specific primers for each of the Synechococcus, Cyanobium, and Prochlorococcus genera. Using the FSL-Pcy library as a reference, we searched possible primers with Geneious v.9.1.5 (74), targeting a region that would start in the conserved area at the beginning of the V5 hypervariable region up to the end of the V7 hypervariable region, with an amplicon size of approximately 540 bp, suitable for Illumina MiSeq sequencing using 2 × 300 bp. Once the candidate primers were defined, we tested their specificity and coverage in silico against the SSU r132 RefNR database allowing 0, 1, and 2 mismatches using the Test Probe 1.0 feature from SILVA (http://www.arb-silva.de/?id=650).

Testing the primers with environmental samples: field sampling, PCR, and sequencing.

The specificity and usefulness of the designed primers were then tested in two sets of samples, as described below.

(i) Two marine samples (Table S11) were taken from the coastal Blanes Bay Microbial Observatory in the Mediterranean Sea (41°39.90′N, 2°48.03′E) (49). We selected one sample taken in spring and one in summer 2013 (April and August) based on previous knowledge that during these seasons the abundance of Synechococcus reaches maximum values (up to 8 × 10⁴ cell per milliliter) in this coastal station (e.g., 49, 57, 58).

(ii) Eight freshwater samples were taken from four hypertrophic shallow lakes (Table S11) of the Salado River floodplain basin, located in the Pampa plain in Argentina (35°41′S, 59°35′W) (22, 68). The samples were selected in order to cover spatial and temporal variation. Three samples were taken during spring 2009 from the Bragado, El Carpincho, and Monte lakes. In addition, five samples from Chascomús lake were collected between October 2012 and May 2014 (Table S11). Site features and sampling details are provided in detail in Fermani et al. (68) and Huber et al. (22).

Sample collection and DNA extractions are described in Ferrera et al. (58) for the marine samples and in Huber et al. (22) for the freshwater samples. The volume of water processed is indicated in Table S11.

Sequencing was performed by the RTL Genomics division of Research and Testing Laboratories of the South Plains (Lubbock, TX) using Illumina MiSeq 2 ×300 paired-end reads (sequencing runs 1 and 2), with a sequencing depth of approximately 20,000 sequences per sample. The V5 to V7 region of the 16S rRNA gene was amplified using the primers designed in this work, Cya-771F and Cya-1294R (see above). The PCR conditions for these primers were as follows: 5 min preheating step at 94°C; 10 cycles of denaturation at 94°C for 1 min; an annealing touchdown procedure for 1 min in which the temperature decreased by 1°C every second cycle from 65 to 56°C; followed by 25 cycles of 1 min of denaturation at 94°C; and annealing at 55°C for 1 min and extension at 72°C for 3 min, with a final extension step of 10 min at 72°C.

Data analysis.

Raw sequences were processed using a modified version of the pipeline proposed by Logares (78) (https://github.com/ramalok). The reads were first analyzed for error correction using the algorithms based on Hamming graphs and Bayesian subclustering (BAYES HAMMER tool) (79) implemented in SPAdes v3.5.0 (80). Then, the forward and reverse sequences were assembled using the function fastq_mergepairs from USEARCH-v10 (81). The minimum overlap length was set to 20 bp and assemblage sequences with less than 100 nucleotides were discarded; the rest of the parameters were used as the default. Quality checking was performed using fastq_filter in USEARCH-v10 (81). Reads that passed the quality control were then analyzed using UNOISE2 (82) to define operational taxonomic units (OTUs) with no clustering (zero-radius OTUs [zOTUs] or amplicon sequence variants [ASVs]). These tool-defined zOTUs provide a higher accuracy than OTUs by achieving single-nucleotide resolution after correcting for Illumina sequencing errors and chimeras (55, 56). Finally, taxonomy assignment of zOTUs was done by BLAST (83), using the SILVA database (SSU Ref 132 NR 99) as a reference, and the zOTU table was created with the function otutab in USEARCH-v10 (81).

zOTUs with a relative abundance below 1% per sample were considered representative of the “rare biosphere,” while those whose abundances exceeded 1% were considered part of the “abundant fraction” (84). We constructed zOTU rarefaction curves for each environment, using vegan v.2.5 to evaluate whether we saturated richness. The taxonomic diversity and richness of the picocyanobacterial communities were estimated with the Shannon (85) and Chao1 (86) indices, respectively (Table 3). Bray-Curtis dissimilarity matrices were constructed to analyze the similarities in the composition of the assemblages. The abundance of each zOTU was previously transformed using Hellinger transformation, as recommended by Legendre and Gallagher (87). We also calculated phylogenetic diversity using the PD_Faith index, based on maximum likelihood trees constructed using the Randomized Axelerated Maximum Likelihood (RAxML) v.8 program (88).

We also tested the environmental factors that determine community structure. For that purpose, we first selected possible drivers using the function BIOENV (89) within the R environment (90). After that, we constructed a Euclidian distance matrix with the selected variables and determined its correlation with the picocyanobacterial dissimilarity matrix using a Mantel test. The physical, chemical, and biological variables that were used in this analysis are summarized in Table S11.

To infer the phylogenetic positions of the zOTUs, we incorporated the amplicon short sequences into freshwater and marine phylogenetic reference trees (RT). First, we aligned the query sequences against the reference alignments using PaPaRa v.2.5 (91), and then we placed the queries onto the trees using the Evolutionary Placement Algorithm (EPA) (92) as implemented in RAxML (88). The RTs were constructed using sequences of the 16S rRNA gene (minimum length 700 bp) from the NCBI public database, ensuring that the main clades of picocyanobacteria were represented (6, 18–20, 22, 93) (Table S12). The marine RT included 94 sequences, 22 assigned to seven Prochlorococcus clades (HL and LL) and 72 to Synechococcus subclusters 5.2, 5.3, and 5.1 (with 11 clades assigned to 5.1). For the freshwater RT, we selected 181 published sequences and also incorporated 82 unpublished sequences corresponding to environmental clones of the complete 16S rRNA gene retrieved from two Pampean lakes (see Supplemental Material). Twenty-two clades were defined; 13 have been described in previous studies and 9 were novel, being constituted mostly of taxa retrieved from Pampean systems (designated by the letter P and listed with Roman numerals, i.e., PI to PIX); Synechococcus strain PCC 6301 was used as external group. The sequences were aligned using MAFFT v.7, and the RTs were constructed using the maximum likelihood method in RAxML v.8 (88) with the GTR+GAMMA model for freshwater taxa and the GTR+CAT+I model for marine taxa, considering 1,000 bootstraps in both cases. Trees were edited using the Interactive Tree Of Life (iTOL; http://itol.embl.de) (94).

Data availability.

All sequences generated in this study can be accessed through European Nucleotide Archive (ENA) under the study accession number PRJEB27291 and accession numbers ERR2639355 to ERR2639364.