Sampling took place at Galé Alta, Armação de Pêra (37° 04’ 09.6”N and 8° 19’ 52.1”W) off the coast of the Algarve, southern Portugal, in June 2010. Four specimens of S. spinosulus and I. variabilis (Demospongiae, Dictyoceratida, Irciniidae), and four samples of surrounding seawater (1 L each, about 1 m above sponge specimens), were collected in sterile Ziploc®; bags by scuba diving at about 15 m depth. Samples were placed in cooling boxes, transported to the laboratory (c. 2 h) and immediately processed as described below. Sponge species were identified based on macro- and microscopic morphological criteria coupled to molecular phylogenetic inference (Hardoim et al., 2012).

For each collected sponge specimen, three procedures of sample processing were undertaken prior to total community DNA (TC-DNA) extraction, hereafter called “direct,” “indirect,” and “plate washing” methods of sponge sample processing. The first two methods lead to cultivation-independent analysis of TC-DNA directly extracted from the sponge body (“direct” method) or from sponge-derived microbial cell pellets (“indirect” method). “Plate washing” involves extracting TC-DNA from washes of Marine Agar culture plates and is therefore a culture-dependent methodology for assessing the sponge-associated microbiome without having to purify and singularize colonies. Details of each procedure are given in Appendix S1 (Supplementary Material). TC-DNA extraction was performed with the UltraClean®; Soil DNA isolation kit (MO BIO, Carlsbad, CA, USA), according to the manufacturer’s protocol. The same kit was used for TC-DNA extraction from bulk seawater samples following Hardoim et al. (2012). Thus, 28 metagenome samples (four seawater replicates; four I. variabilis replicates handled with “direct,” “indirect,” and “plate washing” methods; and four S. spinosulus replicates handled with “direct,” “indirect,” and “plate washing” methods) were subjected to PCR-DGGE and 454-pyrosequencing bacterial community profiling as explained below.

A nested PCR-DGGE approach targeting the V6 hypervariable region of the 16S rRNA gene was used to fingerprint the bacterial communities associated with both sponge species under the three methods of sample processing. The reaction mixture and thermal cycling for both reactions were as described by Hardoim et al. (2012), except for the concentration of primers (0.6 μM, Table 1). PCR-DGGE profiling was then performed using a PhorU-2 gradient system (Ingeny International, Goes, The Netherlands). Gel gradient, marker constituents, electrophoresis and staining procedures were described previously (Hardoim et al., 2012). Multivariate statistical analysis of PCR-DGGE fingerprints followed methods described elsewhere (Costa et al., 2006; Hardoim et al., 2009) and is detailed in Appendix S1.

A barcoded pyrosequencing approach was employed for in-depth analysis of bacterial community composition and diversity. A thorough description of (i) pyrosequencing sample preparation, (ii) data processing and (iii) analyses is provided in Appendix S1. Briefly, the V4 hypervariable region of the 16S rRNA gene was PCR-amplified using the Ribosomal Database Project primer set (Table 1), which generates amplicons of around 248 bp in length. Two PCR mixtures of 25 μL were prepared per sample, each containing ∼20 ng of template DNA. Each sample was tagged by unique 8-mer barcodes attached to the reverse primer. The amplicons were delivered for pyrosequencing on a 454 Genome Sequencer GS FLX Titanium platform (Roche Diagnostics Ltd, West Sussex, UK) at BIOCANT (Biotechnology Innovation Center, Cantanhede, Portugal). Raw data were processed using AmpliconNoise (Quince et al., 2011) and Galaxy (https://main.g2.bx.psu.edu/; Taylor et al., 2007a), enabling noise filtering (e.g., homopolymers), chimera removal, sequence sorting, and trimming. The Quantitative Insights Into Microbial Ecology (QIIME) software package (Caporaso et al., 2010) was then applied to the filtered data set for operational taxonomic units (OTUs) determination and taxonomic assignment, followed by the generation of a samples-OTUs table using customized scripts (Appendix S1). Data analyses encompassed (i) phylum- and class-level bacterial composition in individual and pooled samples, (ii) assessment of specific and shared bacterial symbionts across sample groups via OTU networks and Venn diagrams, (iii) estimates of symbiont richness (Chao1) and diversity (Shannon’s index) and (iv) multivariate analysis of OTU data. The latter was performed via (a) principal coordinate analysis (PCoA) of OTU profiles using the Unifrac metric within QIIME and (b) constrained ordination of OTU profiles and independent variables (i.e., seawater, sponge, sponge species, and sample processing methods) with the software package Canoco for Windows 4.5 using Hellinger-transformed OTU abundance data. Analyses (i) to (iv) were carried out using both full size (whole data set exploration) and size-normalized sample libraries. The analysis of full size libraries was used to determine the absolute number (and the identity) of all OTUs shared by and specific to each sample category (n = 7: seawater; I. variabilis under “direct,” “indirect,” and “plate washing” methods; and S. spinosulus under “direct,” “indirect,” and “plate washing” methods). The analysis of normalized libraries was applied in the quantitative comparison of bacterial richness, diversity and community structure between the sample categories. For size-normalized analyses, two depth thresholds were defined, 1236 and 3688 sequence reads per sample, which allowed the comparison of (i) all four replicate samples of both sponge species under the three methods of sample processing and triplicate seawater samples, and (ii) all sponge-derived libraries (seawater samples excluded), respectively. Similar analyses were performed for the unfiltered data set, disregarding chimera and noise removal procedures. Pyrosequencing data were deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) under the accession number SRP021445.

To determine the spatial distribution and infer the abundance of sponge-associated bacteria at the micro-scale, in-tube FISH-CLSM was performed as described by Cardinale et al. (2008) with slight modifications (Appendix S1). For the detection of all bacteria, an equimolar mixture of Cy3-labeled EUB338, EUB338II, and EUB338III probes was used (Amann et al., 1990, Table 2). Samples were further hybridized with ALEXA488- or Cy5-labeled FISH probes specific for Acidobacteria (SS_HOl1400; Meisinger et al., 2007), Alphaproteobacteria (ALF968; Neef, 1997), and Gammaproteobacteria (Gam42a; Manz et al., 1992; Table 2). These taxa were selected based on their predominance revealed by 454-pyrosequencing. Taxon-specific abundances relative to total bacterial cell density were calculated by averaging the fraction of specifically stained cells from at least 15 randomly selected fields (confocal stacks) retrieved from three independent FISH experiments per specific probe.

Homogeneity of variance tests were used to inspect the normal distribution of the richness and diversity measurements from PCR-DGGE fingerprints and 454-pyrosequencing. Analysis of variance (ANOVA) tested whether the mean values obtained for all sample groups (n = 7: seawater; I. variabilis under “direct,” “indirect,” and “plate washing” methods; and S. spinosulus under “direct,” “indirect,” and “plate washing” methods) were equal. A pairwise t-test which analyses the significance between groups was then carried out. Homogeneity of variance and ANOVA were also employed to compare 454-pyrosequencing relative abundances of the most dominant bacterial phyla and classes found across groups. The analyses were performed with the stat package in R programming (R Development Core Team, 2012). For both PCR-DGGE and 454-pyrosequencing data, Monte-Carlo permutations were performed to test whether the generated sponge symbiont profiles clustered according to the sample groups.

The PCR-DGGE profiles of S. spinosulus obtained with cultivation-independent “direct” and “indirect” sample processing methods were visually very similar whereas much larger band variation was observed between I. variabilis fingerprints generated using the same methods (Figure 1). Profiles obtained for both sponge species via “plate washing” were different from those generated by cultivation-independent methods (Figure 1A). Ordination analysis of PCR-DGGE banding patterns suggested that both the host species and processing methods determined the structures of the surveyed symbiont communities (Figure 1B). A detailed description of PCR-DGGE results and a discussion on how they compare with 454-pyrosequencing data are given as Supplementary Material (Table S1, Appendix S2).

The bacterial groups targeted by FISH were chosen to represent distinct patterns of abundance within the sequencing data set. Whereas Proteobacteria can be regarded as a “generalist” phylum dominant in seawater and sponge communities (both cultured and uncultured), Acidobacteria constitutes a “specialist” phylum with greater abundance within the unculturable sponge microbiome. The detected bacterial cells were mainly cocci found in between sponge cells, with no evidence for a taxon-dependent aggregation of bacteria within the sponge body. In all analyzed samples, bacterial cells were seldom found on spongin filaments. The high abundance of cells precluded discrete counting of cell numbers, and taxon abundance data relative to total bacterial coverage was retrieved instead. The relative abundance of Alphaproteobacteria in both sponge species was similar: 27.14% in S. spinosulus (Figures 6A,E) and 22.85% in I. variabilis (Figures 6B,F), whereas the Gammaproteobacteria were more abundant in S. spinosulus (21.41%, Figures 6C,E) than in I. variabilis (9.55%, Figures 6D,F). Both Alphaproteobacteria and Gammaproteobacteria were more abundant than Acidobacteria in S. spinosulus (9.09%, Figure 6G). In I. variabilis, the abundance of Acidobacteria (6.20%, Figure 6H) was similar to that of Gammaproteobacteria and lower than that of Alphaproteobacteria. Instead, Acidobacteria 16S rRNA gene tags accounted for 16.91 and 20.68% of the total communities in S. spinosulus and I. variabilis, respectively, surpassing numbers obtained for Alpha- (3.90% for S. spinosulus and 4.27% for I. variabilis) and Gammaproteobacteria (S. spinosulus 9.78% and I. variabilis 11.57%) in both sponge species according with 454-pyrosequencing.

The extent to which existing methods of sample preservation, processing and DNA extraction may affect the structure of sponge symbiont communities is currently under-appreciated. Indeed, detailed analyses comparing sponge microbial community data retrieved with different methodologies are very scarce (see e.g., Hardoim et al., 2009; Simister et al., 2011). The “direct” processing method has been widely used to evaluate the diversity of microbial communities in marine sponges (Hentschel et al., 2002; Taylor et al., 2005; Lee et al., 2011; Hardoim et al., 2012) and is, moreover, the fundamental technique used in metatranscriptomic studies (Radax et al., 2012). By contrast, the “indirect” processing method, through which most of the sponge cells are eliminated, is used in metagenomic, metaproteomic, and single cell genomic surveys that target the sponge microbiota (Fieseler et al., 2006; Siegl and Hentschel, 2010; Thomas et al., 2010a; Siegl et al., 2011; Fan et al., 2012; Liu et al., 2012; Bayer et al., 2013). In the present study, both approaches led to similar bacterial diversity and composition results for S. spinosulus, but not for I. variabilis. In fact, we found that several OTUs consistently lost abundance while one OTU classified as Acidobacteria showed increased abundance in I. variabilis replicates processed with the “indirect” as compared to the “direct” method (see Appendix S6 for details), resulting in higher methodologically dependent symbiont community variability in this species than in S. spinosulus. This contrast may relate to the density of the collagenous filaments present in the mesohyl of these sponges, which is higher in I. variabilis making this species exceptionally tough and more difficult to tear or cut in comparison with other species in the Irciniidae family (Cook and Bergquist, 2002). Consequently, variability in the mechanical easy of bacterial cell detachment and disruption from the I. variabilis endosome matrix may have resulted in methodology-dependent, non-corresponding community structures retrieved from the same host. Rapid fingerprinting of sponge microbial communities by traditional methods such as PCR-DGGE or T-RFLP may still be an adequate means of assessing whether microbial cell enrichments usually prepared for metagenomics are representative of the community directly determined from the sponge body. Overall, the ability of each cultivation-independent processing method to accurately estimate the actual in situ sponge bacterial community may differ from species to species; therefore preliminary data acquisition to aid in the choice of methodology is advisable prior to in-depth analyses of sponge microbiome diversity and function.

The degree of conservation of bacterial communities in marine sponges across host species, habitats and oceans has important implications for the management of marine genetic and metabolic resources given the status of these holobionts as the most prolific source of biologically active compounds in the oceans (Piel, 2004). Recent studies have suggested that these communities are host species-specific (Webster et al., 2010; Lee et al., 2011; Schmitt et al., 2012), contrary, in principle, to earlier studies which indicated microbiome conservation across sponge hosts and geographical locations (Hentschel et al., 2002). Through a detailed analysis of sympatric and co-familial sponges this study supports the increasing evidence for host-dependent symbiont community structures among phylogenetically related sponge species, acquired recently with the use of more traditional techniques (Erwin et al., 2012; Hardoim et al., 2012). Qualitatively, the pool of OTUs found exclusively in each species was largely circumscribed by “rare” symbionts: when OTUs deemed as less dominant were discarded from the analysis, the number of “host species-specific” bacterial phylotypes was drastically reduced whereas a strong signal for conservation of the more prevalent symbionts remained. We therefore propose that the views of host-conserved and host-specific bacterial communities in marine sponges are not mutually exclusive, but rather complementary facets of one single biological process. Indeed, the quantitative contribution of the so-called “rare” and often “species-specific” OTUs to the total bacteriome of the surveyed sponges was small (c. 1.9% of all analyzed sequence reads) despite their diversity. Host specificity was therefore found to be chiefly determined by full quantitative profiles including all community members, their corresponding abundance ranks and differences in abundance between host species. Interestingly, the phylum-level taxonomic composition of the uncultivated, “rare” sponge symbionts resembled that of the dominant symbionts. This could suggest maintenance of core, functional attributes across phylogenetically related microbes with varying abundances within the community, a mechanism that could confer functional stability to the marine sponge holobiont in face of changing conditions, be they host-induced or not. However, closely related bacterial species, and even strains within species, may also have widely differing functions that can result from the acquisition of traits via horizontal gene transfer, a well-documented phenomenon within the Proteobacteria (Costa et al., 2009; van Elsas et al., 2011). In light of the current evidence for abundant horizontal gene transfer potential (Thomas et al., 2010a) and whole genome differentiation among close bacterial relatives within the marine sponge microbiome (Esteves et al., 2013), the net contribution of intra- and inter-species bacterial diversity to the spectrum of functions in these communities remains to be understood.

It has been demonstrated that cultivation-independent methods often fail to detect bacteria isolated with culture-dependent procedures, raising concerns about the validity of cultivation-independent studies (Donachie et al., 2007). In our assay, cultivation of sponge-associated bacteria detected about 35–40 bacterial OTUs per species in comparison with 180–220 OTUs identified by cultivation-independent methods (see Table 3 for details). These ratios surpass typical cultivability estimates based on CFU/microscopy cell count ratios, which in the case of our sponge specimens was in the range of 0.01–0.1% (Hardoim et al., 2012). However, about half of the OTUs recovered with the “plate washing” method escaped detection via cultivation-independent approaches, while the remainder was usually detected by the latter procedures in much lower numbers. These results indicate that marine agar culturing enriches for lower abundance symbionts that remain elusive to cultivation-independent methods. Because our methodology enabled direct comparisons between cultured and uncultured microbiomes through a standardized metagenomic DNA analysis pipeline, biases induced by DNA extraction, PCR amplification and colony picking-and-purification procedures are less likely to explain the abrupt differences observed between these communities in our survey. The hypothesis that the cultivatable sponge-associated bacteriome encompasses less abundant phylotypes that are considerably enriched during cultivation, by likely outcompeting more prevalent but less competent and/or slow-growing bacteria, must be assessed with protocols designed to detect rarer populations in the sponge community (Montalvo et al., 2014). In this regard, the in-tube FISH-CLSM approach used here to image highly ranked bacterial taxa may be a powerful tool to localize specific and less abundant symbionts in keratose sponges. In this study, FISH-CLSM and 454-pyrosequencing estimates of bacterial abundance were not always congruent. Variable target specificity and coverage between the techniques may have accounted for these differences, although the additional steps required for 454-pyrosequencing (cell detachment and lysis, DNA extraction and PCR) are expected to add more analytical biases compared to the direct FISH-CLSM method. Copy numbers are unlikely to explain the observed discrepancies. Alphaproteobacteria (more abundant in FISH-CLSM than in 454-pyrosequencing results) have an average of 2.4 copies of the ribosomal operon, whereas Acidobacteria (more abundant in 454-pyrosequencing) have an average of only 1.6 copies (Klappenbach et al., 2001). These results highlight the importance to use multiple methods for a more comprehensive analysis of symbiont abundance in marine sponges, as proposed elsewhere for other dense microbial settings such as soils (Ushio et al., 2014). Coupling nucleic acid sequencing (especially primer-less approaches, e.g., direct metagenome sequencing) and advanced imaging technologies seems an adequate strategy to be used in upcoming studies.

Although we did not anticipate the detection of Acidobacteria, Chloroflexi, and Poribacteria on culture plates, recent research supports their aerobic and heterotrophic growth capacities (Davis et al., 2011; Siegl et al., 2011; Kamke et al., 2013; see Appendix S7 – Supplementary Material – for a Discussion on Poribacteria results). Due to the very low number of sequences recovered, marine agar showed to be inappropriate for the future isolation of these organisms. Intriguingly, these and other lineages, such as Kiloniella sp. and Ferrimonas sp., could not be detected by a colony picking-and-isolation effort previously applied to our samples (Esteves et al., 2013). Beyond the comprehensiveness of the “plate washing-deep sequencing” procedure used here to characterize the cultivation-dependent communities, this outcome could be explained by the further possibility, enabled by this procedure, of uncovering strictly syntrophic microorganisms or micro-colonies that elude detection by the naked eye (Davis et al., 2011). Finally, the direct comparison between cultivation-dependent and -independent communities made in this study shows that traditional cultivation fails at capturing the prevalent bacterial associates of marine sponges. In this context, not only alternative medium recipes and incubation strategies are needed to enhance the cultivability of sponge symbionts in the laboratory, but improvements to our current methods to sample and analyze cultivated sponge microbiomes are desirable to extend our ability to assess the diversity and metabolic capacities of these and other symbiont cohorts in the future.