All experiments were carried out using sporophytes of the E. subulatus freshwater strain (EC371, accession CCAP 1310/196, West and Kraft, 1996). This culture was obtained from Bezhin Rosko (Santec, France) in 2007 and maintained in our laboratory under the following conditions since then: cultures of EC371 were grown in Petri dishes (90 mm Ø) in natural seawater (NSW; collected in Roscoff 48°46′40′′ N, 3°56′15′′ W, 0.45 μm filtered, autoclaved at 120°C for 20 min), or in diluted seawater-based medium (DNSW; by 20-fold dilution of natural seawater with distilled water). Both media were enriched with Provasoli nutrients (Starr and Zeikus, 1993) and cultures kept at 13°C with a 12 h dark-light cycle (photon flux density 20 μmol m^-2 s^-1).

A range of cultivation strategies as well as bacterial growth media was exploited. The starting material for bacterial isolation was EC371 grown with its full microbial flora (direct plating and dilution-to-extinction cultivation) and EC371 with a reduced microbial flora (size-fractionation and antibiotic treatment), both originating from the same algal culture. Algal subcultures were sampled 5–10 days after the last change in medium. Algal growth medium and ground algal tissue, in NSW and DNSW were used. The isolation experiments took place from November 2013 to September 2016. Three selected cultivation experiments (dilution-to-extinction cultivation, direct plating with antibiotics; direct plating without pretreatment) were repeated under identical conditions after 6 months, 1 year, or 3 years, respectively, to assess reproducibility of the results over time. An overview of the isolation methods and cultivation strategies is provided in Figure 1.

To estimate the proportion of cultivable bacteria in our algal cultures, the collection of bacterial isolates was compared with 16S rRNA gene libraries from the same algal culture used for our isolation experiments. These libraries served as a reference to assess the total microbiome, including cultivated and non-cultivated bacteria; they were not used to infer diversity per se. EC371 cultures were grown for 9 weeks in seawater-based culture medium (changed on a monthly basis, last 1 week prior to sampling). Algal tissue was filtered with sterile coffee filters, dried on a paper-towel, and snap-frozen in liquid nitrogen. Four technical replicates were pooled two by two. Total DNA was isolated (NucleoSpin^® Plant II, Machery-Nagel; standard protocol) and purified with Clontech CHROMA SPIN^TM-1000+DEPC-H2O Columns. The V3–V4 region of the 16S rRNA gene was amplified and sequenced with Illumina MiSeq technology by MWG Eurofins Biotech (Ebersberg, Germany) using their proprietary protocol. The first preliminary quality control was done with FastQC³, and fastq_quality_trimmer from the FASTX Toolkit⁴ was used to quality-trim and filter the 568,100 reads (quality threshold 25; minimum read length 200). The resulting 553,896 sequences (2.5% removed) were analyzed with Mothur (V.1.38.0) according to the MiSeq Standard Operating Procedures⁵ (Kozich et al., 2013). Filtered reads were assembled into 270,522 contigs, preclustered (allowing for four mismatches), and aligned with the non-redundant Silva SSU reference database version 123 (Quast et al., 2013). Chimeric sequences were removed using the Uchime algorithm (Edgar et al., 2011) implemented in Mothur, and the remaining sequences classified taxonomically using the method of Wang et al. (2007). Non-bacterial sequences were removed. The sequences were then clustered into operational taxonomic units (OTUs) at a 97% identity level and each OTU was classified taxonomically. All OTUs with n ≤ 5 sequences were removed (0.02%) resulting in a final data matrix with 217,923 sequences. The sequences obtained from cultivable isolates were compared with the 16S rRNA gene metabarcoding data using BLASTn searches against raw reads (99% identity) and consensus OTU sequences (97% identity). In addition, the current dataset (META2016-NSW) was compared to previous datasets (META2013-NSW, META2013-DNSW) obtained from the same algal strain 3 years earlier (Dittami et al., 2016). All counts for each individual OTU were normalized against the total number of sequences in the corresponding dataset. Raw Illumina reads were deposited at the European Nucleotide Archive under project accession number PRJEB22665. To compare the cultivable sequences and their abundance in the 16S metabarcoding data (META13-NSW and META16-NSW) a heatmap was created using the iTOL web application⁶ (Letunic and Bork, 2016). Log(x+1)-transformed data was used for OTU sequence counts and cultivation abundances. All datasets (three for metabarcoding, 10 for cultivation) were grouped by hierarchical clustering using Euclidean distance calculations and the average linkage method implemented in the pvclust R-package⁷ (Suzuki and Shimodaira, 2006). The resulting tree was tested using bootstrap analysis (500 replications). OTUs that did not correspond to cultivable strains are not shown in the graphical representation. In this manuscript, “isolate” refers to every bacterial culture for which a 16S rRNA sequence was obtained. All isolates with identical 16S rRNA sequences are considered to belong to the same “strain”.

16S metabarcoding experiments were carried out with the same algal culture also used for the isolation of bacteria and the libraries were used as a reference for the cultivated and non-cultivated microbiome as a whole. After cleaning and filtering of the data, the sequences were clustered into 48 OTUs. The most abundant OTU belonged to the genus Alteromonas (OTU1) and accounted for 41.6% of the reads, which makes Gammaproteobacteria the most abundant class (42.3%). Other abundant OTUs corresponded to an unclassified Rhodobacteraceae (OTU3, 11%) and an unclassified Bacteroidetes (OTU4, 10%). Together these three OTUs correspond to 62% of all sequences (Figure 3A). Alphaproteobacteria make up 32.8% of the sequences and Bacteroidetes 15.3% (Figure 3B). Other phyla identified are Actinobacteria (2.1%) and Deltaproteobacteria (7.5%). Of the 48 OTUs, 10 corresponded to strains cultivated in our experiments. These 10 OTUs accounted for 47% of the reads in the metabarcoding data. Purely based on absence/presence of OTUs the culturability was 21%. Three additional cultivable strains corresponded to OTUs with sequence abundance below the threshold (n ≤ 5, Staphylococcus, Hyphomonas, Oceanicaulis; Figure 3). Furthermore, taking into account all 16S rRNA gene libraries and rare reads, 22 of the 46 cultivable strains were detected. Among the 24 undetected strains that were not found in any of the barcoding libraries, 11 were isolated exclusively from DNSW and 8 exclusively from NSW. In the same vein, 11 strains were cultured exclusively from algal medium, and 7 only from algal tissue (Supplementary Table S4).

The 16S rRNA gene metabarcoding data obtained in this study (META2016-NSW) differed strongly from that obtained 3 years earlier (META2013-NSW) from the same Ectocarpus strain. Several OTUs that were present in the 2013 samples were no longer present in 2016 or declined in abundance below the detection limit. However, there were still 50 OTUs (corresponding to 90% of the sequences) shared between the 2013 and 2016 samples. The most abundant OTU in 2013 belonged to the genus Hoeflea (29% of reads) while the most abundant OTU in 2016 (Alteromonas sp.; 42% of reads) accounted for only 2.4% of the reads in 2013.

The main aim of this study was to establish a diverse collection of cultivable Ectocarpus-associated bacteria that can be used to perform functional studies of brown macroalgal–bacterial interactions in this model organism. We applied different cultivation strategies to facilitate the growth of less abundant or slow-growing bacteria and thus increased the variety of cultivable bacteria.

Among the cultivable taxa frequently found on brown macroalgal surfaces (such as Laminaria, Saccharina, Fucus, Ascophyllum) are Proteobacteria, Firmicutes, Bacteroidetes, and Actinobacteria, where the latter three phyla are generally less abundant (Ivanova et al., 2002; Wang et al., 2008; Wiese et al., 2009; Salaün et al., 2010; Dong et al., 2012; Goecke et al., 2013a,b; Hollants et al., 2013; Martin et al., 2015). Two cultivation studies in Ectocarpus species showed the presence of Gammaproteobacteria, Actinobacteria, and Flavobacteria (Kong and Kwong-yu, 1979; Tapia et al., 2016). The results of our study, with Proteobacteria being most abundant followed by Bacteroidetes, Firmicutes, and Actinobacteria, largely agree with these findings, except that Alphaproteobacteria were the most abundant proteobacteria in this study (Figure 2 and Supplementary Table S3), compared to Gammaproteobacteria in the previous studies.

The three dominant genera obtained were Limnobacter, Bosea, and Halomonas (Figure 2 and Supplementary Table S3). Limnobacter sp. are oligotrophic freshwater sulfur-oxidizing bacteria (Spring et al., 2001) that occur naturally in aquatic environments (Lu et al., 2011; Zhang et al., 2014) and drinking water reservoirs (Wu et al., 2014) and are generally considered rare in marine settings (Staufenberger et al., 2008; Wiese et al., 2009). The majority of the isolation experiments in this study were indeed carried out with low-salinity culture media, and the algal source also came from fresh water (West and Kraft, 1996), providing two possible explanations for the presence of Limnobacter in our experiments. As these experiments were always complemented with negative controls (i.e., non-inoculated bacterial growth media), a contamination with Limnobacter from the water source used to prepare the bacterial growth media is unlikely.

Members of the genus Bosea are known to be (multi)drug-resistant (Falcone-Dias et al., 2012; Zothanpuia et al., 2016). The results from our study agree with these observations since 92% of the Bosea isolates came from antibiotic-treated algae (Figure 2 and Supplementary Table S3). In addition, several other strains were uniquely isolated from antibiotic-treated algal tissue suggesting that part of the algae-associated microbiome is (multi)drug-resistant, or otherwise protected by the algal cell wall/inside the cell, where drug concentrations may be too low to be effective. In addition, some antibiotics employed in this study, such as chloramphenicol and erythromycin, may have had only temporal bacteriostatic effects.

The genus Halomonas comprises cultivable isolates from various saline environments (Eilers et al., 2000; Donachie et al., 2004; Arahal and Ventosa, 2006; Poli et al., 2009), including microalgal (Keshtacher-Liebson et al., 1995; Croft et al., 2005; Baggesen et al., 2014) and macroalgal surfaces (Ivanova et al., 2002; Wang et al., 2008; Hollants et al., 2013; Tapia et al., 2016). Halomonas-algae associations are potentially beneficial for the alga, since the bacteria may provide vitamins (Croft et al., 2005), release siderophores (Keshtacher-Liebson et al., 1995; Baggesen et al., 2014), or excrete morphogenetic compounds (Spoerner et al., 2012; Tapia et al., 2016) that are essential for algal growth. Symbiotic associations with algae may be linked to the capacity of Halomonas to degrade algal excreted polysaccharides and/or the presence of alginate lyases (Wong et al., 2000; Tang et al., 2008; Wang et al., 2008; Goecke et al., 2012) and indeed, bacterial cells can be closely attached to algal cell walls (Croft et al., 2005; Tapia et al., 2016). In this study, Halomonas was the most abundant isolate obtained with direct plating techniques without pretreatment (DP1 and DP2). More isolates were derived from tissue/protoplasts compared to algal growth medium (p = 1.92E-13), suggesting a close association between Ectocarpus and the Halomonas sp.

In summary, each cultivation strategy resulted in the cultivation of unique strains that were not cultivated with any of the other methods. For example, the application of antibiotics to eliminate Halomonas sp. reduced the competitive pressure between antibiotic resistant bacteria and led to the cultivation of 16 additional bacterial strains. Similar observations have been made in sponges (Sipkema et al., 2011; Lavy et al., 2014), lichens (Parrot et al., 2015), and tap water (Vaz-Moreira et al., 2013). Interestingly, direct plating without pretreatment, although dominated by Halomonas sp., also resulted in the isolation of seven unique strains.

Marine pelagic bacteria often have complex growth and nutrient requirements (Stewart, 2012; Zengler, 2013). In addition, they are generally considered to be oligotrophs, since they inhabit a nutrient-poor environment and grow only very slowly, which might also compromise the cultivation process (Amann et al., 1995; Keller and Zengler, 2004; Zengler, 2013). Hence, a large part of the marine environmental microbiome has been considered non-cultivable using standard cultivation techniques (Amann et al., 1995; Morris et al., 2002; Giovannoni and Stingl, 2005).

Here, we aimed to cultivate bacteria that were associated with the algae/algal cell-walls, a carbohydrate-rich environment due to the accumulation of algal (poly)saccharides, i.e., alginates, fucans, and mannitol (Michel et al., 2010; Popper et al., 2011). The divergence in community structure between pelagic and algae-associated microbiomes is well-established (Kong and Kwong-yu, 1979; Staufenberger et al., 2008; Bengtsson et al., 2010; Burke et al., 2011; Wahl et al., 2012; Goecke et al., 2013b; Mancuso et al., 2016), and several algae-associated bacteria are able to digest/decompose algal cell material (Rieper-Kirchner, 1989; Goecke et al., 2013b; Groisillier et al., 2015; Martin et al., 2015), e.g., via the production of alginate lyases (Sawabe et al., 1997; Dong et al., 2012) and glycoside hydrolases/fucanases (Ficko-Blean et al., 2015). It is thus possible that algae-associated bacteria, contrary to pelagic bacteria, are well-adapted to grow on the laboratory cultivation media provided, resulting in relatively high numbers of cultivable bacteria.

Our data support this hypothesis since culturability was between 44 and 68% based on the dilution-to-extinction experiments (Table 1). For pelagic studies, culturability is usually below 15%, with some observation going as low as 0.05% (Connon and Giovannoni, 2002; Page et al., 2004; Stingl et al., 2007, 2008; Yang et al., 2016). In the same vein, dilution-to-extinction cultivation studies on pelagic bacteria generally apply between 1 and 25 bacterial cells/well as inoculum (Connon and Giovannoni, 2002; Stingl et al., 2007). In our study, however, concentrations as low as 0.5 cells/well were required to obtain pure cultures, further demonstrating that a relatively large part of the algae-associated microbiome is cultivable compared to pelagic bacteria.

Culturability was also assessed by comparing the distribution and abundance of taxa obtained in the cultivation study with taxa inferred from16S rRNA gene libraries. In a previous study on Ectocarpus, this type of comparison demonstrated an overall ratio of culturability of 11% based on the presence/absence of OTUs (Tapia et al., 2016). In the present study, this number was further increased with 21% of the OTUs and 47% of all 16S rRNA sequences corresponding to cultivable strains (Figure 3C).

To our knowledge, this is the first study to apply dilution-to-extinction cultivation to macroalgae-associated bacteria, and the standardized cultivation method (Connon and Giovannoni, 2002) was amended by adding the algal metabolites alginate and/or mannitol to the culture media. We assume that it was, therefore, well-adapted to the metabolic needs of the majority of Ectocarpus-associated bacteria, and indeed several bacteria known to be potential cell-wall digesters have representatives in our culture collection, e.g., Alteromonas (Sawabe et al., 1997), Flavobacteria (Groisillier et al., 2015), Maribacter (Martin et al., 2015), Erythrobacter (Goecke et al., 2013b), and Halomonas (Wong et al., 2000). Together these results validate the combination of cultivation approaches chosen to increase culturability in our system.

Of the 46 unique strains that were isolated in this study, 16 were isolated at least once from algal tissue grown in NSW, and could thus be directly compared to META2016-NSW metabarcoding data set generated in this study. Seven of them (44%) were represented in this gene library. To be able to compare also strains isolated only from low salinities with metabarcoding data, we included two further data sets obtained for the same strain in 2013. All data sets taken together, 22 of the 46 (48%) strains were found at least in one of the libraries, while 24 were undetectable or below the detection limit. Whether a strain was isolated directly from algal tissue or from the algal culture medium did not have a strong impact on these numbers (Supplementary Table S4).

There are several hypotheses to explain this observation. First, methodological flaws or biases including the inadequacy to extract DNA from certain bacterial cells due to species-specific characteristics (e.g., gram-positive are generally more difficult to extract than gram-negative cells), primers specificity, or PCR conditions (Suzuki and Giovannoni, 1996; Donachie et al., 2004, 2007). This may explain biases but, is unlikely to account for the complete absence of a taxon, because all cultivable taxa were detectable with standard primers and extraction methods in our cultures. A second explanation is that some “rare” microbes may be laboratory- or human-derived contaminants, e.g., Staphylococcus, sp., and Bacillus sp. All measures to avoid bacterial contamination of our algal/bacterial samples were taken and the controls were included in all cultivation experiments and generally negative for growth. Nevertheless, it is plausible that some of these “rare” bacteria were acquired during the monthly transfers of the algal cultures or during the bacteria cultivation procedures and growth of these bacteria might have been facilitated by the experimental treatments that were applied.

A third explanation is that the sequencing depth or the number of time points examined (one for DNSW two for NSW) may have been too low to identify members of the microbiome that are “rare” (Zilber-Rosenberg and Rosenberg, 2008; Skopina et al., 2016). Bacteria might be present in low abundance in the natural environment but they can amplify rapidly under specific environmental conditions (Epstein, 2009; Buerger et al., 2012; Lindh et al., 2015). Rare bacteria might thus serve as a “seed bank” (Pedrós-Alió, 2012) that contributes to the microbial richness and may form the basis for temporal instability of the microbiome (Sogin et al., 2006; Shade and Gilbert, 2015; Jousset et al., 2017). Recently, it has been suggested that in particular marine macro-organisms, and possibly Ectocarpus as well, might serve as incubators for rare bacteria (Troussellier et al., 2017), since surface-associated microbiomes generally exhibit higher OTU diversity and harbor many rare OTUs compared to the surrounding seawater. In this scenario, the removal of competing microbes via the antibiotic treatments and our other measures to reduce competition during our cultivation experiments, have probably allowed them to increase in abundance. This explanation is supported by the variability of the microbiome observed in this study compared with the previous study of the same strain under the same conditions (Dittami et al., 2016), and by the fact that 4 of 46 cultured OTUs were found only among the rare (n ≤ 5) reads in the available barcoding data.

Similar observations of cultivable isolates not being detected in corresponding gene libraries have been made in human stool samples (Lagier et al., 2012), sponges (Sipkema et al., 2011; Esteves et al., 2016), seawater (Eilers et al., 2000), and soil (Shade et al., 2012); more examples are discussed by Donachie et al. (2007). We put forward the hypothesis that, in analogy to “uncultivable” microbes that become cultivable by improving cultivation conditions, at least part of the undetected strains may, therefore, become “barcodable” merely by significantly increasing sequencing depths (Pedrós-Alió, 2012) and/or the temporal resolution of the study.

In this study, we show that a remarkably high number of bacterial cells (∼50%) associated with Ectocarpus was cultivable using a range of cultivation techniques. Each cultivation strategy resulted in another dominant genus or weed-species (Bosea for antibiotic-treated algae, Limnobacter for dilution-to-extinction cultivation) and each strategy also led to the cultivation of unique isolates that were not found with any other cultivation method. Our results thus emphasize the need to use samples from different environmental/abiotic conditions to obtain rare taxa and thus increase the overall cultivable diversity. To further improve these numbers, a metagenomics approach may be used to predict the specific cultivation requirements of yet uncultured taxa (Garza and Dutilh, 2015). One successful example of this approach is the cultivation of members of the SAR11 clade, for which genomic analysis revealed their requirement for exogenous reduced sulfur (Tripp et al., 2008). Such metagenomic analyses of the Ectocarpus holobiont are currently ongoing.

Regarding the cultivable isolates, genomic data currently in preparation for several strains may be used to predict their metabolic capacities and to generate hypotheses on how they may complement the metabolism of the alga (Dittami et al., 2014). Because the bacteria are cultivable it will be possible to experimentally verify the hypothesis generated using this approach. Sixty-two bacterial isolates and 12 artificial bacterial communities have already been experimentally tested in preliminary algal–bacterial co-culture experiments. They showed interactions ranging from weak beneficial effects on survival of E. subulatus in diluted natural seawater (29 isolates, 15 unique strains; three communities) to growth-inhibition (data not shown). These strains may serve as the first candidates to study the role of algal–bacterial interactions under abiotic stress.

The present bacterial culture collection constitutes a valuable tool to study the Ectocarpus holobiont in vitro and complements the genomic tools available for the model Ectocarpus. Together, they can be used to address fundamental questions regarding the functions of brown macroalgal holobionts during exposure to abiotic stressors, for instance during the acclimation to low salinity in E. subulatus.