Microbial diversity in tropical marine sediments assessed using culture-dependent and culture-independent techniques

Summary

The microbial communities associated with marine sediments are critical for ecosystem function yet remain poorly characterized. While culture-independent (CI) techniques capture the broadest perspective on community composition, culture-dependent (CD) methods can select for low abundance taxa that are missed using CI approaches. This study aimed to assess microbial diversity in tropical marine sediments at five shallow-water sites in Belize using both CD and CI techniques. The CD methods captured approximately 3% of the >800 genera detected across all sites using the CI approach. Additionally, 39 genera were only detected in culture, revealing rare taxa that were missed with the CI approach. Significantly different communities were detected across sites, with rare taxa playing an important role in distinguishing among communities. This study provides important baseline data describing shallow-water sediment microbial communities, evidence that standard cultivation techniques may be more effective than previously recognized, and the first steps towards identifying new taxa that are amenable to agar plate cultivation.

Introduction

Advances in sequencing technologies and bioinformatics have led to major improvements in our ability to assess the diversity of environmental microbes (Lynch and Neufeld, 2015; Hug et al., 2016; Thompson et al., 2017). The application of culture-independent (CI) methods has transformed our understanding of microbial diversity, while metagenome-assembled genomes and single-cell genomics have provided insight relevant to functional traits in as-yet uncultured organisms (Kalisky and Quake, 2011; Evans et al., 2015; Parks et al., 2017). Despite these advances, the microbial diversity associated with marine sediments remains poorly characterized relative to other major biomes such as soil and seawater (Lloyd et al., 2018; Martiny, 2019; Baker et al., 2021). Sediment microbial communities are diverse (Thompson et al., 2017), densely populated (Dale, 1974; Musat et al., 2006), play integral roles in fundamental ecosystem processes (Snelgrove et al., 1997; Baker et al., 2021) and can exhibit extraordinary levels of fine-scale spatial structure (Probandt et al., 2018). While marine sediments cover ∼70% of the earth’s surface (Parks and Sass, 2009), the average ocean depth of >3 km makes access difficult and likely contributes to the lack of data. Nonetheless, the microbial communities associated with more readily accessible shallow-water sediments also remain poorly described (Baker et al., 2021). Baseline information describing sediment microbial diversity provides an important mechanism to understand community structure over time and across environmental gradients.

The gains afforded by CI diversity estimates can overshadow the intrinsic value of microbial cultivation. CD methods provide opportunities to assess microbial metabolism and contributions to ecosystem function in ways that cannot be achieved using CI approaches. For instance, Nitrospira cultures revealed the first bacterium known to perform complete nitrification (Daims et al., 2015) while cultured Thermosulfidibacter takaii resulted in the discovery of a reversible TCA cycle that was not detected with metagenomics (Nunoura et al., 2018). Additionally, testing for inhibition among marine Vibrio strains revealed that competition is greater between than within ecologically cohesive populations (Cordero et al., 2012) while two closely related species of Salinispora demonstrated ecological trade-offs in competitive strategies (Patin et al., 2015). In general, CI techniques provide more comprehensive diversity estimates while CD methods provide a clearer path to taxonomic resolution (Orphan et al., 2000; Chen et al., 2008; Shivaji et al., 2011; Vaz-Moreira et al., 2011; Dickson et al., 2014) and functional assessments. While CI techniques can also be used to infer metabolic requirements (Tripp et al., 2008), the two approaches can be considered complementary. Outside of a recent example (Bech et al., 2020), surprisingly few studies have applied both methods to marine sediments.

While a majority of microbial taxa have yet to be obtained in culture (Lloyd et al., 2018), major advances with bacterioplankton in the SAR11 clade (Rappé et al., 2002; Henson et al., 2018), archaea in the Asgard superphylum (Imachi et al., 2020), and the application of reverse genomics (Cross et al., 2019), among others, suggest that many if not most microbes can ultimately be cultured (Lewis et al., 2020). These advances, along with numerous interpretations of culturability paradigms, have revived discussion of the ‘great plate count anomaly’ and the theory that fewer than 1% of bacterial taxa have been cultured (Martiny, 2019; Steen et al., 2019). Such contrasting views on the proportions of bacteria that remain uncultured highlight the need to reassess frequently cited paradigms such as the great plate count anomaly.

This study aimed to assess microbial diversity in marine sediments using both CD and CI techniques. Cultured diversity was assessed by sequencing DNA from agar plates that had been inoculated with sediments while CI diversity was assessed from environmental DNA derived directly from sediments. The results reveal culturing efficiencies of 1%–2% based on the number of 16S rRNA amplicon sequence variants (ASVs) detected and 3%–4% in terms of the number of genera detected. Additionally, the CD method detected 39 genera that were not detected using the CI approach, highlighting opportunities to capture rare members of the community in culture. These results emphasize the complementarity of CD and CI methods for assessments of microbial diversity in marine sediments.

Results

Culture-dependent sediment microbial diversity

Five replicate sediment samples collected from each of five diverse, shallow-water sites around Carrie Bow Cay, Belize were inoculated onto two types of agar media and incubated for 8 days after which DNA was extracted from the plates to assess CD microbial diversity. Across all five sites, CD alpha diversity averaged 54 ± 3 16S rRNA ASVs on seawater agar (SWA) and 39 ± 2 ASVs on marine agar (MA) (Fig. S1a). SWA yielded significantly greater ASV richness (Fig. S2a; Faith’s PD, Kruskal-Wallis H = 8.870, p = 0.003) and evenness (Fig. S2b; Pielou’s Kruskal–Wallis H = 7.645, p = 0.006) when compared to the ASVs detected on the more nutrient-rich medium MA. The microorganisms detected on the SWA and MA plates were classified into five bacterial phyla (Figs 1A and S3a). The same three phyla were observed in the highest abundance on both media, with Proteobacteria averaging 80% and 96% of the communities on SWA and MA plates respectively. Major differences between the two media were observed within the Bacteroidetes, which represented 20% of the SWA communities compared to 1.6% for MA, and Firmicutes, which represented 2.6% of the communities on MA and 0.13% on SWA. The other two phyla detected in culture were Epsilonbacteraeota (formerly Epsilonproteobacteria) and Actinobacteria, both of which averaged <1% of the communities on both media (Figs 1 and S3a).

Fig. 1.

Rank abundance of phyla detected using (A) culture-dependent methods and (B) culture-independent methods. Phyla are divided into separate plots based on relative abundance to better visualize the variation observed. Light grey bars represent taxa detected on marine agar (MA) while white bars represent taxa detected on seawater agar (SWA). Archaea are denoted as [A] and error bars indicate standard error.

The ASVs detected on SWA and MA were assigned to 102 and 76 genera respectively (Table S1). Of the 128 different genera identified, 50 were detected on both media, 52 were unique to SWA and 26 were unique to MA (Fig. 2). Vibrio was the most abundant genus observed on both media, accounting for 27% and 44% of the ASVs detected on SWA and MA respectively. Other relatively abundant genera included Ruegeria (SWA 25.4%, MA 35.1%), Persicobacter (SWA 18.8%, MA 0.7%), Micro-bulbifer (SWA 6.0%, MA 3.5%) and Alteromonas (SWA 3.9%, MA 1.7%) (Fig. S3b). The vast majority of genera (92 on SWA and 62 on MA) averaged <1% of the community (Table S1)

Fig. 2.

Proportionally scaled Venn diagram of microbial genera detected using culture-dependent (SWA and MA) and culture-independent methods.

Culture-independent microbial diversity

CI microbial diversity was assessed from the same sediment samples used for the CD analyses. CI alpha diversity averaged 2942 ± 133 ASVs (Fig. S1b) with all sites displaying fairly high ASV richness (Fig. S4a; Faith’s PD range 79.15–222.88) and even distributions (Fig. S4b; Pielou’s evenness index range 0.81–0.93). These communities were classified into 68 phyla (57 Bacteria, nine Archaea and two Eukarya) (Figs 1B and S3a), with about half of the sequences assigned to Proteobacteria. The next three most abundant phyla were the Bacteroidetes, Planctomycetes and Cyanobacteria. On average across all samples, 81% of the phyla detected (55 of 68) represented less than 1% of the community and thus can be considered rare. Nonetheless, these rare taxa averaged ∼7% of the community when combined, highlighting their importance to community composition. More than 15 phyla, including some that were well represented in the communities, were assigned to candidate phyla or unannotated taxa, indicating that microbial diversity in marine sediments remains poorly characterized even at the highest taxonomic levels (Fig. 1B).

At a finer taxonomic scale, 1844 genera (1728 Bacteria, 113 Archaea and three Eukarya) were observed using the CI technique across all five sites (Fig. 2). Of these, only 16 averaged ≥1% of the community, with the most common genus being Woeseia, which averaged 4.96% of the community (Fig. S3b). Many of the sequences could not be annotated at the genus level or were simply annotated as ‘uncultured’, suggesting they may belong to new taxa. In addition to Woeseia, those that averaged ≥1% of the community and could be identified at the genus level included Xenococcus (2.49%), Zeaxanthinibacter (2.18%), Candidatus Nitrosopumilus (1.65%), Pleurocapsa (1.46%), Chroococcidiopsis (1.10%) and Rhodopirellula (1.07%) (Table S1). Fewer than 0.001% of the sequences were annotated as unassigned eukaryotic or ciliate associated (Table S1).

Method comparisons

As expected, the CI communities were significantly richer than the cultured communities (Fig. S2a; Faith’s PD Kruskal–Wallis H = 52.076, p < 0.001). ASV relative abundance was also more even in the CI communities (Fig. S2b; Pielou’s Evenness Kruskal–Wallis H = 51.662, p < 0.001), likely reflecting the biased nature of cultivation techniques. In total, 1755 genera were unique to the CI approach while an additional 89 were detected using both the CI and CD approaches (Fig. 2). Of those, 45 were detected on both media types while 28 and 16 were unique to SWA and MA respectively (Fig. 2). Surprisingly, 39 genera detected in culture were not detected using the CI technique (Fig. 2; Table 1). Of these, 24 were unique to SWA, 10 were unique to MA and five were detected on both media (Fig. 2). The two most abundant genera that were not detected using the CI approach were Celeribacter, which averaged 1.92% and 0.05% of the SWA and MA communities respectively, and Halomonas, which averaged 0.71% and 1.14% of the SWA and MA communities respectively. The remaining 37 genera averaged <1% of the community (Table 1). Taxa specific to SWA included Marinomonas (Proteobacteria), Mesonia (Bacteroidetes) and Streptomyces (Actinobacteria). Taxa specific to MA included Fictibacillus (Firmicutes), Taeseokella (Bacteroidetes) and Fangia (Proteobacteria) (Table 1). Of the 39 genera that were unique to the CD technique, all but Aureimonas have been reported from marine sources (Table 1).

Table 1.

Amplicon sequence variants (ASVs) annotated to the genus level that were detected using culture-dependent methods but not detected using culture-independent methods (n = number of replicates).

Genus	Site(s)	Medium (n)	Mean relative percent MA ± standard error	Mean relative percent SWA ± standard error	Named species reported from marine environments
Agrococcus	1	SWA (1)	NA	5.45E-03 ± 5.45E-03	Yes (Lee, 2008)
Amaricoccus	5	SWA (1)	NA	2.61E-03 ± 2.61E-03	No marine isolates, but see Pohlner et al. (2019)a
Aureimonas	1	SWA (1)	NA	6.46E-02 ± 6.46E-02	Nob
Blastomonas	3	SWA (1)	NA	2.73E-02 ± 2.73E-02	Yes (Meng et al, 2017)
Brevundimonas	1	SWA (1)	NA	7.22E-03 ± 7.22E-03	Yes (Fritz et al., 2005)
Celeribacter	1 and 5	MA (2) and SWA (3)	4.90E-02 ± 3.57E-02	1.92E+00 ± 1.91E+00	Yes (Lee et al., 2012)
Colwelliaceae uncultured	1, 2, 3, 4 and 5	MA (11) and SWA (16)	8.48E-02 ± 2.88E-02	1.74E-01 ± 8.17E-02	Yes (Jean et al., 2006)
Corynebacterium 1	2	MA (1)	1.14E-02 ± 1.14E-02	NA	Yes (Ben-Dov et al., 2009)
Fangia	3	MA (1)	2.90E-03 ± 2.90E-03	NA	Yes (Lau et al., 2007)
Fictibacillus	5	MA (1)	4.58E-02 ± 4.58E-02	NA	Yes (Dastager et al., 2014)
Halobacillus	5	MA (1)	3.64E-03 ± 3.64E-03	NA	Yes (Teasdale et al., 2009)
Halomonas	1, 2, 3, 4 and 5	MA (15) and SWA (15)	1.14E+00 ± 3.93E-01	7.07E-01 ± 2.57E-01	Yes (Romanenko et al, 2002)
Hoe flea	1	SWA (1)	NA	8.32E-03 ± 8.32E-03	Yes (Biebl et al., 2006)
Kocuria	1	SWA (1)	NA	5.91E-02 ± 5.91E-02	Yes (Kim et al., 2004)
Luteivirga	5	SWA (1)	NA	2.37E-03 ± 2.37E-03	Yes (Haber et al., 2013)
Marinomonas	1	SWA (1)	NA	3.53E-02 ± 3.53E-02	Yes (Romanenko et al., 2009)
Mesofiavibacter	3	SWA (1)	NA	3.42E-03 ± 3.42E-03	Yes (Asker et al., 2007)
Mesonia	3	SWA (1)	NA	1.43E-02 ± 1.43E-02	Yes (Choi et al., 2015)
Microbacteriaceae UA	1	SWA (1)	NA	6.14E-03 ± 6.14E-03	Yes (Lee, 2008)
Neiella	1,2 and 4	SWA (3)	NA	1.28E-02 ± 7.43E-03	Yes (Du et al., 2013)
Nesiotobacter	3	MA (1) and SWA (1)	2.48E-02 ± 2.48E-02	1.62E-02 ± 1.62E-02	Yes (Ganesh Kumar et al., 2019)
Nitrincolaceae UA	2	SWA (1)	NA	2.77E-03 ± 2.77E-03	Yes (Arahal et al., 2007)
Oleibacter	5	SWA (1)	NA	1.46E-03 ± 1.46E-03	Yes (Teramoto et al., 2011)
Paenibacillus	5	MA (1)	7.88E-03 ± 7.88E-03	NA	Yes (Lee et al., 2013)
Paracoccus	1	SWA (1)	NA	5.24E-03 ± 5.24E-03	Yes (Liu et al., 2008b)
Pontibacter	5	SWA (1)	NA	4.53E-03 ± 4.53E-03	Yes (Nedashkovskaya et al., 2005)
Pseudofulvibacter	1	MA (1)	6.12E-03 ± 6.12E-03	NA	Yes (Yang et al., 2016b)
Psychrobacter	2, 3 and 4	MA (3)	9.14E-03 ± 5.96E-03	NA	Yes (Romanenko et al., 2002b)
Psychromonadaceae UA	1 and 5	SWA (3)	NA	1.20E-02 ± 8.22E-03	Yes (Li et al., 2013)
Robertkochia	4	SWA (1)	NA	2.17E-03 ± 2.17E-03	Yes (Hameed et al., 2014)
Salegentibacter	5	SWA (1)	NA	1.84E-03 ± 1.84E-03	Yes (Nedashkovskaya et al., 2006)
Skermanella	1	SWA (1)	NA	1.50E-02 ± 1.50E-02	No marine isolates, but see Maldonado et al. (2009)c
Sphingomonas	1	SWA (1)	NA	5.43E-03 ± 5.43E-03	Yes (Schut et al., 1997)
Staphylococcus	2	MA (1)	5.33E-03 ± 5.33E-03	NA	Yes (Arora, 2013)
Streptomyces	5	SWA (3)	NA	1.20E-02 ± 7.85E-03	Yes (Gallagher and Jensen, 2015)
Sulfitobacter	3	SWA (1)	NA	3.53E-03 ± 3.53E-03	Yes [Park et al. (2007)]
Taeseokella	5	MA (1)	3.04E-02 ± 3.04E-02	NA	Yes (Li et al., 2019)
Thalassobius	2 and 4	MA (1) and SWA (1)	2.74E-03 ± 2.74E-03	2.75E-03 ± 2.75E-03	Yes (Yi and Chun, 2006)
Zunongwangia	5	MA (1)	4.50E-03 ± 4.50E-03	NA	Yes (Shao et al., 2014)

Taxonomic assignments made with the SILVA v132 database through QIIME2–2020.2. If the genus was not annotated (UA), the lowest taxonomic rank is indicated. NA = not applicable. If the genus has named species from the marine environment, at least one example is cited.

^aAll Amaricoccus species have been isolated from sludge, but Pohlner et al. (2019) also found OTUs that hit to Amaricoccus from marine sediment (Pohlner et al., 2019).

^bAureimonas is the sister genus of Aurantimonas (Rathsack et al., 2011). Aurantimonas that has been isolated from marine sources, however, it has also been identified as a common contaminant (Rathsack et al., 2011; Salter et al., 2014).

^cMaldonado et al. (2009) isolated a strain from the marine environment where the closest 16S hit was Skermenella (Maldonado et al., 2009).

Not surprisingly, taxa with few or no cultured representatives were detected using the CI approach. These included the phylum Latescibacteria (aka WS3), which is commonly detected in CI studies (Youssef et al., 2015; Farag et al., 2017; Lloyd et al., 2018) and represented on average 1.07% of the sediment communities. To our knowledge, this phylum does not have a cultured representative. Additionally, bacteria from the widely distributed and diverse phyla Acidobacteria, Patescibacteria and Gemmatimonadetes averaged 2.7%, 0.95% and 0.6% of the CI community respectively, yet have few strains in culture (Hugenholtz et al., 2001; Ward et al., 2009; DeBruyn et al., 2011; Soro et al., 2014; Lemos et al., 2019). The median sequence similarity between the CI ASVs and their nearest cultured representatives in the SILVA database was 85.4% (Fig. 3 and S5).

Fig. 3.

Scatterplot denoting the culture-dependent (CD) and culture-independent (CI) amplicon sequence variant (ASV) similarity to the nearest cultured or type strain (y-axis) versus the nearest sequence identified with the full non-redundant SILVA v138.1 database (x-axis). Points above the dashed red line (slope of 1) represent ASVs with the nearest relative in culture while points below the red line indicate the nearest relative was identified with CI methods.

In contrast to the CI ASVs, the cultured ASVs had a median similarity of 99.6% with cultured strains listed in the SILVA database. Yet, there were distinct outliers (Figs 3 and S5) including one with a top match of only 79% suggesting that close relatives had not previously been cultured. This sequence shared c. 90% with both an uncultured Chitinophagales in the SILVA database and an ‘uncultured prokaryotic clone’ from a marine intertidal sample based on an NCBI BLAST analysis, further supporting the novelty of this cultured ASV (Table S2). In contrast, an NCBI BLAST analysis of the ASV that was 88% similar to any culture in the SILVA database (Fig. 3) revealed 99.6% sequence similarity to the recently cultured sponge microbe Xanthovirga aplysinae (Goldberg et al., 2020) (Table S2). While 10 ASVs shared <95% sequence similarity to cultured strains in the SILVA database, subsequent NCBI BLAST analysis revealed that seven ASVs had >95% similarity to recent isolates from marine sources such as holothurians and corals (Table S2).

Beta diversity assessed using both unweighted and weighted UniFrac analyses revealed significant differences between the CD and CI communities (Fig. 4A; PERMANOVA pseudo-F = 39.826, p = 0.001; Fig. 4C; PERMANOVA pseudo-F = 125.15, p = 0.001). A PCoA analysis revealed one CI replicate that was distinct from the rest of the samples (Fig. 4), likely due to the large (>20%) relative contribution of an unknown genus in the Flavobacteriales. When considering site, the cultured communities showed no clear pattern while the CI communities showed some clustering in the Unifrac PCoA plots (Fig. 4B and D) and thus were analysed in more detail.

Fig. 4.

UniFrac analysis of culture-independent and culture-dependent sediment microbial communities visualized via principal coordinates analysis (PCoA). Unweighted results coloured by (A) sample type and (B) site. Weighted results coloured by (C) sample type and (D) site.

Culture-independent site comparisons

The five sites were within 5 km of each other (Fig. S6; Table S3), but represented different habitats: an 8 m deep spur and groove reef (site 1), a 20 m deep reef slope (site 2), a 1 m deep sand patch in a seagrass bed at the mouth of a mangrove island (site 3), a 6 m deep seamount in a lagoon (site 4) and a 1 m deep sand and rubble patch near the marine station’s dock (site 5). Both weighted and unweighted UniFrac analyses indicate significant differences in community composition among sites (Fig. 5; Unweighted UniFrac PERMANOVA pseudo-F = 2.982, p = 0.001; Weighted UniFrac PERMANOVA pseudo-F = 7.987, p = 0.001). In the unweighted UniFrac, site 3 appeared the most distinct from the other communities (Fig. 5A). This site displayed the highest species richness (Fig. S4a) and had about twice the average relative abundance of Chloroflexi compared to the other sites, which likely contributed to its separation. Additionally, a genus-level analysis of composition of microbiomes (ANCOM) identified six taxa that were significantly different at site 3. These were Marixanthomonas, an unidentified genus of the BD2–7 (Family Cellvibrionales), an uncultured gammaproteobacterium, and three taxa that were absent from site 3: Stanieria, an unknown genus of Xenococcaceae, and uncultured MBAE14 gammaproteobacterium. Other among site differences include relatively more Proteobacteria at site 2 and relatively fewer Cyanobacteria at sites 3 and 4.

Fig. 5.

Unifrac analysis of culture-independent sediment microbial communities across five sites. A. Unweighted UniFrac and (B) Weighted UniFrac. Communities visualized via principal coordinates analysis (PCoA). PERMANOVA results indicate sites are significantly different from each other in both unweighted (pseudo-F = 2.982, p = 0.001) and weighted (pseudo-F = 7.987, p = 0.001) analyses.

Culturing efficiency

We first assessed culturing efficiency by comparing the total number of ASVs detected using the CD and CI techniques, regardless of taxonomic affiliations. Importantly, the ASV rarefaction curves reached saturation using both techniques (Fig. S1), thus ensuring that effective comparisons could be made. On average, the numbers of cultured ASVs represented 1.82% and 1.33% (for SWA and MA respectively) of those detected using the CI approach. We then calculated taxonomic overlap at various levels from domain to ASV (Table 2; Fig. S7). Overall, 3.95% of the genera detected using the CI approach were also detected in culture on medium SWA while 3.31% were detected on MA. The numbers declined slightly at the species level to 3.16% and 2.64% for SWA and MA respectively but nonetheless remain well above the <1% value commonly cited for bacterial culturability (Staley and Konopka, 1985; Martiny, 2019; Steen et al., 2019).

Table 2.

Percent cultured across taxonomic levels.

Taxonomic level	Percent cultured		Percent taxa unique to culture-dependent samples
	SWA	MA	SWA	MA
Domain	33.33	33.33	0	0
Phylum	7.35	7.35	0	0
Class	2.39	3.19	0	0
Order	2.91	2.75	0.46	0.15
Family	3.36	2.76	0.51	0.34
Genus	3.95	3.31	1.54	0.80
Species	3.16	2.64	1.77	1.22
ASV	0.39	0.33	0.86	0.68

Percent cultured calculations were based on taxonomic overlap in culture-dependent (CD) and culture-independent (CI) samples. Percent taxa unique to CD samples represent the number of taxa only identified with CD methods as a proportion of the total taxa identified at the corresponding level.

Discussion

It is widely recognized that CD approaches underestimate microbial diversity. Early comparisons of colony to cell counts indicated that <1% of environmental bacteria were cultured, a phenomenon referred to as the ‘great plate count anomaly’ (Razumov, 1932; Staley and Konopka, 1985). The advent of CI techniques has brought the extent of uncultured microbial diversity into better perspective (Lynch and Neufeld, 2015; Hug et al., 2016; Lloyd et al., 2018) and further driven efforts to obtain new taxa in culture (Kaeberlein et al., 2002; Tamaki et al., 2009; Tanaka et al., 2014; Rygaard et al., 2017). Despite these advances, CI techniques also have limitations (Kennedy et al., 2014; Brooks et al., 2015; Fischer et al., 2016; Laursen et al., 2017; Wear et al., 2018; Willis et al., 2019) and can miss up to 10% of environmental sequences (Eloe-Fadrosh et al., 2016). Additionally, denoising parameters, the level of sequence identity selected for clustering and the compositional nature of amplicon data can also affect diversity estimates (Patin et al., 2013; Callahan et al., 2017; Edgar, 2017, 2018; Gloor et al., 2017; Nearing et al., 2018; Straub et al., 2019).

Given the opportunities afforded by next-generation sequencing, it is surprising that CD and CI approaches are seldom combined. Recent exceptions include studies of river (Pédron et al., 2020) and lake sediment (Elfeki et al., 2018), seawater (Rygaard et al., 2017), cheese (Perin et al., 2017), lungs (Dickson et al., 2014) and raccoon microbiomes (Junkins and Stevenson, 2021). Here, we used both techniques to explore microbial diversity in marine sediments and to estimate culturing efficiency by sequencing bacteria directly from agar plates, in effect a type of enrichment culture, as opposed to the more traditional approach of strain isolation or colony counting. The 16S rRNA gene sequences amplified from sediment environmental DNA required ∼50 000 post quality control and denoising reads per sample to approach saturation in the rarefaction curves. This relatively deep sequencing was needed to capture the diversity present in these complex communities, where the majority of taxa were detected in low relative abundance (<1%). In total, we detected >27 000 ASVs in sediments collected from five sites around Carrie Bow Cay, Belize. A recent and geographically broader study assessing microbial diversity in marine sediments detected >34 000 bacterial and archaeal ASVs across 299 sites (Hoshino et al., 2020), with the apparent reduction in diversity per site likely due to saturation not being achieved.

For the purposes of this study, we defined cultured taxa as those detected after removing ASVs with <70 reads per sample. This cut-off was selected based on inoculum controls where the maximum ASV frequency observed from agar plates immediately after inoculation was 67 reads. As expected, the richness estimates for the CD technique, which averaged 54 and 39 ASVs per sample for SWA and MA respectively, were considerably lower than the >2000 ASVs detected per sample using the CI technique. Nonetheless, this translates to culturing efficiencies of 1.82% and 1.33% for SWA and MA respectively. While we did not perform colony counts, the number of cultured ASVs appeared higher than the number of colonies readily visualized by eye on the plates after eight days of incubation. This suggests that some of the cultured diversity is associated with micro-colonies or mixed colonies that would not be easy to count or isolate using standard practices. It would be beneficial in the future to examine plate surfaces under magnification and perform colony counts in parallel with plate sequencing to assess the presence of micro-colonies. We also determined culturing efficiency based on the taxa detected, with a focus on genera given the limitations of species assignments for short regions of the 16S rRNA gene (Liu et al., 2008; Vĕtrovský and Baldrian, 2013; Yang et al., 2016; Johnson et al., 2019). In this case, the culturing efficiency was 3.95% and 3.31% for SWA and MA respectively (Table 2). While these numbers are reduced with species-level assignments, they nonetheless exceeded expectations given that only two media and one time point were tested. The cultured genera included some denoted as ‘uncultured’ or unassigned, highlighting the lack of information describing marine sediment communities (Baker et al., 2021) and suggesting that traditional agar plating techniques may provide unexpected opportunities to capture new taxa in culture.

Proteobacteria was one of the three most abundant phyla cultured and also represented the largest fraction of the CI community, indicating some broad commonalities between the CD and CI techniques. Differences between the taxa enriched on the two culture media were also apparent, with SWA yielding a considerably larger fraction of Bacteroidetes relative to MA (∼20% compared to∼2%) and MA yielding more Firmicutes than SWA (∼3% compared to∼0.1%). The increased microbial richness on the relatively nutrient-poor medium SWA supports previous results (Jensen et al., 1996;Watve et al., 2000) and suggests that, even for sediment communities, high nutrient concentrations can be inhibitory. Interestingly, the candidate phylum PAUC34f was detected in every CI replicate, yet this phylum remains uncultured (Chen et al., 2020). While PAUC34f was not detected in our post-filtering CD data, it was detected in one SWA replicate prior to correcting for the inoculum control, suggesting that members of this phylum may be amenable to agar plate cultivation.

Approximately 20% of the ASVs detected using the CD method could not be classified at the genus level. Given that the error filtering and denoising pipelines employed improve taxonomic resolution (Callahan et al., 2016, 2017; Amir et al., 2017; Nearing et al., 2018; Prodan et al., 2020), it is likely that these ASVs represent new taxa. Our inoculum controls further support this suggestion. In addition to culturing what appear to be new taxa, we also cultured genera that contained few named species. For instance, the genus Ascidiaceihabitans was detected in six of the SWA replicates at up to 27% of the community and in three of the MA replicates at up to 2% of the community, yet contains only one species isolated from the tunicate Halocynthia aurantium (Kim et al., 2014). We also observed the genus Endozoicomonas, which has been found in association with a wide variety of marine organisms including corals (Bayer et al., 2013), tunicates (Schreiber et al., 2016) and sea slugs (Kurahashi and Yokota, 2007). A recent CI study revealed potentially diverse functional roles for Endozoicomonas as a symbiont while also postulating that this bacterium has a free-living stage based on its large genome size (Neave et al., 2017). Our detection of Endozoicomonas at up to 2% of the community in two MA replicates potentially represents the free-living stage of this taxon.

The CI analyses revealed some clustering by site, with a seagrass bed (site 3) seemingly the most distinct. Gribben et al. (2017) found that seagrass sediment microbial communities can reduce the success of an invasive macrophyte, indicating their importance for the overall health of these ecosystems (Gribben et al., 2017). The ANCOM results identified seven genera that significantly differed between sites, with three having the highest relative abundance at site 3. Further studies are needed to determine what, if any, role these taxa may play in seagrass communities. Additional metadata could also provide valuable insight into community differences between sites. The unweighted UniFrac analysis, which does not account for relative abundance, showed clearer site separation for the CI data, suggesting there are differences in taxonomic composition across sites yet commonalities in the dominant community members. Additionally, all significant genera identified through ANCOM were rare members of the community, with a maximum relative abundance of 0.2%. Thus, rare community members are likely driving differences among sites. While we have not assessed the functional implications of these taxa, previous research has demonstrated the disproportionate role of rare microbes in community ecology (Jousset et al., 2017; Bech et al., 2020).

One surprising result was the detection of cultured genera that were not detected using CI methods. Similar results were reported from Mediterranean water samples (Crespo et al., 2016), recent work in a freshwater system (Pédron et al., 2020) and a comparative study on raccoon microbiomes (Junkins and Stevenson, 2021). Almost all genera unique to the CD methods are likely members of the rare biosphere. To the best of our knowledge, all of these genera, with the exception of Aureimonas, have been reported from the marine environment (Table 1), suggesting they are not contaminants. Aureimonas is a sister genus to Aurantimonas, which does include marine representatives (Rathsack et al., 2011), suggesting it may have been misannotated. Alternatively, this may be the first report of the genus Aureimonas from the marine environment.

One limitation of estimating cultured diversity by extracting DNA directly from agar plates is that strains are not obtained in pure culture. In some cases, bacteria may fail traditional isolation attempts due to obligate associations with co-occurring microbes or metabolic needs that are not met when isolated from the community. Techniques such as reverse genomics (Cross et al., 2019) and metagenomics may help address these issues. Additionally, the use of diffusion chambers (Bollmann et al., 2007) or the ichip (Berdy et al., 2017) provide methods to introduce environmental factors that can increase culturing success. It is likely that many of the taxa detected in culture would not be recognized as traditional colonies and thus new approaches to isolate bacteria from agar plates likely need to be developed. Nonetheless, one of the more unusual ASVs cultured was identified as X. aplysinae. Given that this recently described genus was isolated on MA (Goldberg et al., 2020), it is likely that culturing efforts in under-explored environments, even using traditional approaches, will continue to yield novel diversity.

It is possible that our threshold for defining culturability was not stringent enough. Differences between sample and control sediments, and variation in the sequencing platforms used for each, may have affected the cut-off selection. Nonetheless, these values were determined empirically and removed approximately 50% of the low abundance ASVs detected on both media. Interestingly, a recent study comparing plate-washed cultures to CI, oral and gut microbiome sequences found that >50% of the cultured ASVs were absent from the CI analyses (Junkins and Stevenson, 2021). Thus, culturing efficiency can vary widely depending on sample type, with microbial interactions on agar plates likely playing important and largely undefined roles in bacterial culturability.

In conclusion, the sediment microbial communities analysed here were highly diverse, with rare genera (<1%) representing the majority of the community and driving differences among sites. The cultivation of taxa that were not detected using CI methods supports the use of both techniques for assessments of community composition and adds to our growing understanding of the limitations of CI diversity assessments. The taxa detected by extracting DNA directly from agar plates suggest that traditional colony picking approaches may not adequately reflect cultured diversity and provide new opportunities for future targeted cultivation efforts.

Experimental procedures

Sample collection and processing

In September 2015, divers collected marine sediment samples from five locations around the Smithsonian Field Station at Carrie Bow Cay, Belize (Fig. S6; Table S3). At each site, five replicate sediment samples were collected in Whirl-Pak^® (Nasco) bags from an c. 3 m² area. Upon return to the field station, 20 ml of wet sediment from each Whirl-Pak was transferred into 50 ml falcon tubes with 20 ml of RNAlater^® and stored at 4°C before transport on dry ice to Scripps Institution of Oceanography (SIO) where they were stored at -40°C prior to DNA extraction. For cultivation, freshly collected sediment samples were diluted 1:2 with autoclaved seawater, vigorously shaken, and further diluted 1:10 and 1:100 after which 50 μl of each dilution was inoculated onto two culture media, spread with a sterile glass rod, allowed to dry in a laminar flow hood, then sealed with parafilm. The culture media were SWA (16 g agar, 1 L natural seawater) and 50% MA (0.5 g yeast extract, 2.5 g peptone, 16 g agar, 1 L natural seawater) both amended with the antifungal agent cyclohexamide (final concentration of 200 μg ml^-1). The resulting 150 plates (3 dilutions × 5 replicate sediments × 2 media × 5 locations) were allowed to incubate at ambient temperatures for 8 days including transport back to SIO after which they were stored at -40°C prior to DNA extraction. To detect potential microbial contamination associated with the media, reagents and extraction method, negative controls were made by spiking plates of both media types with ∼3 × 10⁷ cells of Vibrio coralliilyticus to ensure adequate DNA concentrations for sequencing, then immediately parafilmed and stored at -40°C prior to extraction. Similarly, to control for DNA associated with the sediment inoculum, three replicate dilution series prepared from sediments collected in San Diego were plated onto each medium, spiked with ∼3 × 10⁷ cells of V. coralliilyticus, parafilmed, and immediately stored at -40°C prior to DNA extraction. Given that the inoculum controls were from a different location, they were used to set a quantitative threshold for actively growing cells.

DNA extraction

Environmental DNA was extracted from approximately 1 g of freshly thawed sediment per sample following physical (bead beating) and chemical (phenol-chloroform) DNA extraction methods (Patin et al., 2013). One replicate from site 5 was lost resulting in a total of 24 sediment DNA samples. DNA extractions were performed in duplicate for each sediment sample (2 g of sediment in total extracted per sample) and the extracts combined prior to purification. To extract DNA from cultured bacteria, the agar plates were left to thaw at room temperature for 30 min and 3 ml molecular grade water was added to the surface. A heat sterilized metal loop was used to scrape the surface of each plate and the resulting suspension was pipetted into a 15 ml falcon tube. The three dilutions plated for each sediment were combined into a single falcon tube and centrifuged at 8000 RPM (9803 RCF) and 4°C for 5 min generating 50 samples (5 replicate sediments × 2 media × 5 locations). The supernatant was removed and the bottom 2 ml including the cell pellet were subject to bead-beating prior to DNA extraction following the protocol applied to the sediments (Patin et al., 2013). Negative and inoculum controls were extracted using the same protocols with the replicates pooled yielding four samples: SWA negative, SWA inoculum, MA negative and MA inoculum all spiked with V. coralliilyticus.

PCR and sequencing

The v4 region of the 16S rRNA gene was PCR amplified using the primers 515F (TCGTCGGCAGCGTCAGATG TGTATAAGAGACAGGTGYCAGCMGCCGCGGTAA) and 806Rb (GTCTCGTGGGCTCGGAGATGTGTATAAGA GACAGGGACTACNVGGGTWTCTAAT) (Caporaso et al., 2012). PCR was performed using Phusion Hot Start Flex 2 × Master Mix and the following program: 98°C for 30 s followed by 25 cycles of 98°C for 10 s, 60°C for 30 s and 72°C for 30 s with a final 5 min extension at 72°C. Reactions were performed in triplicate with 1 μl of 5 ng μl^-1 DNA in a total volume of 25 μl/reaction and replicates pooled. Products were cleaned using ExoSap-IT^® before adding Nextera XT (Illumina) indices and sequencing adapters onto 5 μl of sample in 50 μl reactions with the following PCR program: 98°C for 1 min followed by five cycles of 98°C for 10 s, 65°C for 20 s and 72°C for 30 s with a final extension at 72°C for 2 min. Gel electrophoresis was used to confirm the presence of a PCR product of the predicted size. Sequences were normalized based on DNA concentration, pooled and cleaned with AMPure XP beads. The purified libraries for experimental samples were sequenced at the Institute for Genomic Medicine (IGM), University of California, San Diego (UCSD) on an Illumina MiSeq v2 500 cycle. For control samples, DNA was sent to Novogene (South Plainfield, NJ) for library preparation following their proprietary methods and sequenced with the primers 515F & 806Rb on an Illumina NovasSeq. For both experimental and control samples, 2 × 250 paired-end sequencing was performed with a targeted depth between 100k and 150k reads per sample.

Analysis

Raw sequences from experimental and control samples were imported into QIIME2–2020.2 (Bolyen et al., 2018) and denoised using DADA2 (Callahan et al., 2016) with an input of p-trim-left-f of 19 and p-trim-left-r of 20 to remove primers. Based on the raw files, p-trunc-len-f and p-trunc-len-r were set to 250 and 155 base pairs respectively and chimaeras removed with the default consensus method. Post denoising, experimental samples averaged 152 243 reads while controls averaged 118 504 reads. To control for inoculum DNA that was not associated with actively growing cells, the relative read abundances of all ASVs, after accounting for the V. coralliilyticus spikes, were quantified. The most abundant ASV, identified as an unknown bacterium, accounted for 67 reads (Table S4). Based on this, a threshold of 70 reads was used to identify cultured ASVs. This cut-off value removed approximately 50% of the cultured ASVs (Table S5). Additionally, we checked each ASV associated with the four control samples and determined that none of the ASVs remained in the experimental samples after applying the filtration step. Taxonomy was assigned using the SILVA v132 database (Quast et al., 2013) and the results were filtered to remove chloroplast and mitochondria sequences (Table S5) in QIIME2–2020.2.

To assess how similar ASV sequences were to cultured representatives, an approach based on Steen et al. (2019) was employed. Using the align.seqs command in Mothur (Schloss and Westcott, 2011), both CD ASVs and CI ASVs were aligned with sequences from the SILVA database. To compare to cultured strains, type strains ‘[T]’ and cultured ‘s[C]’ strains were down-loaded from SILVA v138.1 as an aligned fasta file including gaps. For comparison to SILVA CI sequences, the corresponding non-redundant full library was used (SILVA_138.1_SSURef_NR99_tax_silva_full_align_trunc. fasta) after removal of type and cultured strains. The align. report files were subsequently filtered to remove any sequences with inadequate alignments (pairwise alignment lengths <250). Histograms based on similarity to the nearest cultured representative and scatterplots with the nearest cultured relative compared to the most similar CI sequences were generated using ggplot2 (Wickham, 2016) in R. Some of the ASVs were then further interrogated by using the NCBI BLAST tool by selecting to exclude uncultured sequences.

For alpha and beta diversity analyses, samples were rarefied to a depth of 62, 830 reads and analysed at the ASV level with QIIME2–2020.2. Associated statistical analyses were also performed with QIIME2–2020.2 at the ASV level using the non-parametric Kruskal–Wallis test (Kruskal and Wallis, 1952) for method and site comparisons in relation to alpha diversity indices (Faith’s Phylogenetic Diversity and Pielou’s Evenness) and multivariate PERMANOVA tests with 999 permutations (Anderson, 2001) for beta diversity [both weighted and unweighted UniFrac (Lozupone and Knight, 2005)] comparisons across methods and sites. ANCOM analysis (Mandal et al., 2015), which was specifically designed to address compositional microbial data, was performed at the genus level using QIIME2–2020.2 to determine genera that significantly differed across sites in CI samples. Figures were generated using QIIME2–2020.2, RStudio version 3.6.2 (R Core Team, 2019), ggplot2 (Wickham, 2016), nVennR (Quesada, 2020) and Excel version 16.36.

Data availability

All raw sequences files are available through NCBI’s Sequence Read Archive (SRA). Accession numbers for CI files are SAMN08824420–SAMN0882443, CD files are SAMN15932210–SAMN15932259 and controls are SAMN15932260–SAMN15932263.

Supplementary Material

suppl figures

Fig. S1. Mean alpha rarefaction curves across sediment samples from five sites in Belize. a) Culture-dependent results obtained using seawater agar (SWA) and marine agar (MA) media and b) Culture-independent results. Error bars represent standard error among replicates.

Fig. S2. Boxplots of marine sediment alpha diversity assessments from culture-dependent and culture-independent methods using a) Faith’s Phylogenetic Diversity Index and b) Pielou’s Evenness. The culture-dependent method employed two media: marine agar (MA) and seawater agar (SWA). Data points are overlayed on the boxplot to show variation.

Fig. S3. Relative abundance of microbial communities in marine sediments. a) Phylum level culture-dependent diversity on two growth media (MA and SWA, left) and culture-independent diversity (right). b) Genus level culture-dependent diversity on two growth media (MA and SWA, left) and culture-independent diversity (right). Legends lists a) all phyla and b) the top 50 genera in order from most to least abundant and six additional genera that had a notable percentage in at least one replicate. Note that bar colours repeat for some rare taxa. Bars are ordered from left to right in each section by site as follows: Site 1 (bars 1–5), Site 2 (bars 6–10), Site 3 (bars 11–15), Site 4 (16–20) and Site 5 (CD bars 21–25 and CI bars 21–24).

Fig. S4. Culture-independent alpha diversity boxplots of marine sediment microbial communities from five sites around Carrie Bow Cay, Belize determined using a) Faith’s Phylogenetic Diversity Index and b) Pielou’s Evenness.

Fig. S5. Similarity of amplicon sequence variants (ASVs) detected using culture-dependent (CD) and culture-independent (CI) methods to previously cultured strains. Cultured representatives included both type and cultured strains extracted from SILVA v138.1.

Fig. S6. Collection site map.

Fig. S7. Proportional Venn Diagrams denoting the number of taxa detected with culture-dependent (marine agar and seawater agar) and culture-independent methods across different taxonomic levels.

suppl tables

Table S2. Amplicon sequence variants from cultured samples that had <95% similarity with the SILVA v138.1 cultured and type strains. Nearest cultured (CD) and culture-independent (CI) matches are indicated with corresponding accession numbers in parenthesis. Percent similarity between the match as determined by SILVA and NCBI BLAST are also listed.

Table S3. Site information for samples collected around Carrie Bow Cay, Belize.

Table S4. Amplicon sequence variants (ASVs) detected in control samples sorted by decreasing frequency. Negative controls were marine agar (MA) or seawater agar (SWA) plates spiked only with Vibrio. Inoculum controls were MA or SWA plates spiked with Vibrio and fresh sediment inoculum. All controls were done in triplicate, but DNA was pooled prior to 16S sequencing resulting in one replicate per control sample type. ASVs were assigned after denoising with DADA2. Taxonomy was determined with SILVA v132. ASVs associated with the Vibrio spike-in were omitted from the table. See methods section for further details.

Table S5. Number of amplicon sequence variants (ASVs) associated with culture-dependent methods. Control samples included negative controls (MA Negative (n = 1) and SWA Negative (n = 1)) and sediment inoculum controls (MA Inoculum (n = 1) and SWA Inoculum (n = 1)). Initial control ASV counts are indicative of ASVs identified after excluding the known Vibrio spike-in. Culture-dependent samples were MA (n = 25) and SWA (n = 25) plates inoculated from Belizean sediments. One ASV from the culture-dependent SWA samples was identified as a chloroplast sequence and removed. ASVs were generated from denoising with DADA2 and taxonomy was assigned with SILVA v132. All steps were performed in QIIME2–2020.2.

Table S1

Table S1. All genera assignments from QIIME2–2020.2 analysis with the SILVA v132 Database and the average relative percent of the community identified to each genus across CD and CI methods.