A potential source for cellulolytic enzyme discovery and environmental aspects revealed through metagenomics of Brazilian mangroves

Sample collection and DNA extraction

Samples were collected from two different Brazilian mangroves: sample RJ was collected from a preserved mangrove in the Rio de Janeiro state Ilha Grande during October of 2010 (23° 10' 11.852" S, 44° 17' 0.2" W), and sample BA was collected from a peri-urban mangrove area with anthropogenic action in the vicinity of the Bahia state Porto Seguro during November of 2010 (16° 35' 17.225" S, 39° 5' 30.642" W) (Additional file 1: Figure S1). In relation to the BA sample, no specific permissions were required according to the Brazilian government rules (resolution n. 21, August 31, 2006 - Ministry of Environment). The RJ sample was collected in accordance with the Brazilian law (IN 154/2007 IBAMA, Brazilian Institute of Environment and Renewable Natural Resources) and we confirm that the field studies did not involve endangered or protected species.

The samples (50 g) were collected from the soil surface (0–10 cm) using sterilized spatulas and sterilized 50 mL plastic tubes. The samples were maintained on ice during transportation until processing. Some physical-chemical and environmental characteristics of the samples were determined, such as their pH, temperature, and salinity. The samples collected were manually shaken for 2 minutes in 100 mL sterile Erlenmeyer flasks and 10 g were taken for DNA extraction. DNA was extracted from 10 g of each soil replicate using the PowerMax™ Soil DNA Isolation Kit (MoBio Laboratories, Inc., Carlsbad, CA, USA) according to the manufacturer’s recommendations. The quantity and quality of the resulting DNA were verified with a NanoDrop (Thermo Scientific, Wilmington, DE, USA) spectrophotometer at an absorbance ratio of 260/280 nm and 260/230 nm and were confirmed by electrophoresis on a 1% agarose gel. The DNA samples were then diluted to a concentration of 50 ng/μL.

The isolation of cellulolytic microorganisms

Approximately 5 g of soil collected from both mangroves were inoculated in 200 mL of minimal medium MM (peptone 0.3%, K₂HPO₄ 0.05%, MgSO₄ 0.05%) containing sugar cane bagasse 3% as the carbon source. The culture flasks were shaken for 5 min and maintained without agitation during 3 days at 28°C; the flasks were then incubated for 10 days at 28°C and 100 rpm. Samples of 50 μL each were collected in two-day intervals and cultured on Petri dish plates containing four different types of media: minimal medium with sugar cane bagasse 1% (M1), minimal medium with carboxymethyl cellulose (CMC) 1% (M2), Luria-Bertani (LB) medium (M3), and Sabouraud (M4) medium. Microorganisms with different colony morphologies were isolated and used to generate pure cultures, which were subsequently maintained at 4°C.

The cellulolytic ability of each isolate was tested by inoculation into 3 mL tubes of minimal medium with CMC plus sugarcane bagasse, pinus (Pinus elliottii) powder, or medium-density fiberboard (MDF) powder at 1%, as the main carbon source. After incubation for 5 days at 28°C and 150 rpm, the isolates that grew in at least in two of these substrates were also tested in a CMC-Congo red plate assay (Teather et al. 1982).

DNA pyrosequencing and sequence processing

The preparation of two libraries was accomplished according to instructions from the Rapid Library Preparation Method Manual - GS FLX Titanium Series (454-Roche). The first library was made from 500 ng of DNA extracted from the RJ mangrove soil sample and another from 500 ng of DNA extracted from the BA mangrove soil sample. The titration, emulsion PCR, and sequencing steps were performed according to the manufacturer’s instructions. A two-region 454 sequencing run was performed on a 70x75 PicoTiterPlate (PTP) using the Genome Sequencer FLX System (Roche). Each region was loaded with one of the library preparations.

The artificially replicated sequences that were an artifact of the 454-based pyrosequencing technique were identified and eliminated using the Replicates software (Gomes-Alvarez et al. 2009). The remaining reads were further filtered with the LUCY program to remove short sequences (less than 180 bp) and sequences with a phred quality ≤ 20 (Chou and Holmes 2001).

The Newbler Assembler software version 2.5.3 was used to perform the assembly procedures, with a “-rip” flag that allows for the output of each read in a contig. Reads identified by the GS De Novo Assembler as problematic (Partial, Repeat, Outlier, TooShort) or with High-Quality Discrepancies (HQD) were filtered from the dataset. Subsequently, a new cycle of assembling and filtering was performed. These steps were repeated until the problematic and high-quality discrepancy reads were decreased below the cutoff threshold of 1% of the total number of reads filtered out at the first assembly step.

Metagenome data reported in this paper have been deposited into the GenBank database (PRJNA186597) and on the MG-RAST server (4485218.3, 4485219.3, 4485981.3, and 4485982.3).

Taxonomic distribution analysis

The taxonomic distribution of each read was performed with the MEGAN4 (Huson et al. 2011) software, which compares the given reads against a database of reference sequences by performing a search using the BLASTX algorithm (Altschul et al. 1997) against the NCBI-NR protein database. The MG-RAST server (Meyer et al. 2008) was also used for the taxonomic analysis. The 16S rRNA identification was accomplished using the Meta_RNA software (Huang et al. 2009). The sequences were subsequently classified with RDP Classifier software (Wang et al. 2007) based on the RDP Database (Cole et al. 2009).

Functional metagenomic analysis

The MEGAN4 software was used to perform a functional analysis using the SEED (Overbeek et al. 2005) and KEGG (Kanehisa et al. 2012) databases. Each read was related to its SEED functional role using the best BLAST score to protein sequences without known functional roles. A similar procedure was used to match each read to a KEGG orthology (KO) accession number. To study specific metabolic pathways, the BA and RJ metagenomes were compared using 1e-05 as the maximum e-value cutoff, with a minimum identity of 60% and a minimum alignment length of 15. A functional comparison against the public metagenomes was also performed on the MG-RAST server.

Carbohydrate-active enzyme domains assigned by the CAZy database (Cantarel et al. 2009) were searched for in the predicted protein sequences (Rho et al. 2010; Zhu et al. 2010) using HMMER 3.0 software (Finn et al. 2011) and the Pfam 26.0 database (Punta el al. 2012), with a cutoff of E ≤ 1e-4.

All contig sequences were analyzed and functionally annotated using the System for Automated Bacterial Integrated Annotation (SABIA) (Almeida et al. 2004). According to the automatic annotation criteria, a given ORF was considered “valid” if it had BlastP hits on the KEGG, NCBI-nr or UniProtKB/Swiss-Prot databases, if it had subject and query coverage of ≥ 60%, and if it had positives of ≥ 60%. ORFs that had no BlastP hits on the NCBI-nr, KEGG, UniProtKB/Swiss-Prot, TCDB or Interpro databases or that were excluded by the above criteria were considered “hypothetical”.

Comparative metagenomic analysis

The mangrove samples were compared with other metagenomes using the SEED subsystem content as a comparative metric (Suen et al. 2010 Tringe et al. 2005). Metagenomes from host-associated, water, and soil samples were evaluated using the MG-RAST server. The parameters for inclusion were a maximum e-value cutoff of 1e-05, a minimum identity of 60%, and a minimum alignment length of 15.

Trends in the abundance of the SEED subsystem were examined using Principal Component Analysis (PCA) and hierarchical clustering (Willner et al. 2009). The PCA method is a reduction/ordination technique that clusters the samples based on variations extracted from their normalized abundance profiles (Meyer et al. 2008). This analysis was performed on normalized data, using the bray-curtis dissimilarity.

Statistical analyses

The MG-RAST taxonomic and functional profiles were analyzed using the Statistical Analyses of Metagenomic Profiles (STAMP) software (Parks and Beiko 2010) to detect biologically relevant differences in the relative proportions of the classified sequences. The two-sided Fisher’s exact test with Storey’s FDR method for multiple test correction was employed to analyze the data sets. All unclassified reads were removed from analyses. The most important taxa were filtered according to their q-values (0.01), and only those categories with more than a 2-fold ratio between the proportions or with difference between the proportions of at least 5% were used.

Statistical tests on the taxonomic data were also performed with MEGAN. The BA and RJ counts were normalized to produce data sets of 100,000 reads. Subsequently, MEGAN was used to apply a directed homogeneity test in order to highlight the significant differences in the mangrove comparisons. The highlighting thickness is logarithmically proportional to its significance; that is, the thickness is an integer value of 2log x when P = 1.0e^x (Mitra et al. 2009). Multiple testing correction analysis was not applied, and all unassigned reads were ignored.

Sample description

Two different mangrove locations with distinct features were chosen: one in the Southeast (RJ) and the other in the Northeast (BA) of the country. The RJ mangrove is located in the city of Angra dos Reis, in Rio de Janeiro state, on a natural reserve island called Ilha Grande. Ilha Grande was created by the Brazilian government on August 25, 1978, and is considered to be highly preserved. It is 3,600 ha in size and contains both mangroves and other coastal ecosystems, including the Atlantic rainforest of subtropical southeast Brazil. The BA mangrove is located in the city of Porto Seguro, in the Trancoso district of the Bahia state, in a tropical region of the continent. It also contains both coastal and restinga ecosystems (Additional file 1: Figure S1). Despite its allocation in a hard-to-access site, the area is close to several small rural properties and a few luxury beach hotels and resorts. Consequently, this area is now increasingly threatened by urban development. Therefore, the BA mangrove is considered to be at a medium grade of degradation (Gomes et al. 2008). Although these two mangrove systems differ in terms of their location (island and continent), anthropic action and preservation state, they are similar in terms of their physical-chemical parameters. They both have a humid climate with average temperatures of 24°C and average salinity levels of 1.5%. The RJ mangrove has a neutral pH of 7.0, while the BA mangrove has a slightly acidic pH of 6.5. The samples were collected from the soil surface and the intertidal zone to access different soil layers along with their associated microorganisms.

The isolation and selection of cultivable microorganisms

Samples from the RJ and BA mangroves were used to isolate cultivable microorganisms in four different media. From a total of 296 microorganisms with visual differences in colony morphology and growth (including bacteria, yeast and filamentous fungus), 179 (60.5%) and 117 (39.5%) were isolated from the RJ and BA samples, respectively (Figure 1A).

To determine the ability of these isolates to degrade cellulosic compounds, all microorganisms were grown in a CMC-congo red assay. Here, 77 microorganisms were found to be positive, and 58.4% of these were isolated from the RJ mangrove. In addition, 73% of the CMC-congo red positive microorganisms were initially isolated from mangrove samples through media containing cellulosic compounds (sugarcane bagasse or CMC) (Figure 1B). Twenty-one filamentous fungi with cellulolytic potential were isolated.

The biodiversity and general characteristics of the mangrove metagenome

A large-scale metagenomic analysis was used to profile the microbiota of two Brazilian mangroves and to identify their functional attributes. A total of 503,019 and 494,201 reads from BA and RJ, respectively, were submitted for taxonomic and functional analysis. After applying quality control measures and removing any artificial duplicate reads (ADRs) on the MG-RAST server, the 491,224 (98.7%) reads from the BA library produced a total of 450,087 predicted protein-coding regions. Similarly, a total of 488,630 protein-coding regions were predicted from the RJ mangrove metagenome (Table 1). Of these sequences, 772 (BA) and 457 (RJ) were identified as 16S rDNAs using the MG-RAST server against RDP. This result corresponds to 0.15% and 0.09% of the total sequences obtained from the BA and RJ samples, respectively.

For the BA metagenome, 261,482 (58.1%) of the 450,087 predicted proteins were assigned an annotation, and 188,605 (41.9%) had no significant similarity to any protein in the databases and were therefore considered orphan sequences. In total, 243,812 (93.2%) sequences were functionally categorized. In the RJ sample, 242,868 (49.7%) sequences were assigned an annotation, and 245,762 (50.3%) were found to have no significant similarities to anything in the protein databases. Of the annotated sequences, 224,512 (92.4%) were assigned to functional categories. The average read length for the BA sample was 328 ± 111 bp, with a mean GC content of 54 ± 10%, and the average read length for the RJ sample was 327 ± 110 bp with a mean GC content of 55 ± 10% (Table 1).

The MEGAN results assigned annotations to 340,668 reads from the BA mangrove, which corresponds to 67.7% of the total reads, 162,169 (32.2%) of which are orphan sequences. For the RJ sample, 293,374 (59.3%) reads had significant similarities to hits within the protein databases, whereas 200,458 (40.5%) were unassigned reads. The algorithm implemented by MEGAN assigned more sequences when compared to MG-RAST, and it was able to identify a higher number of sequences related to Bacteria, Archaea, Eukarya, and Viruses (Table 1).

The alpha diversity was calculated for both samples, and the results indicated that the RJ mangrove (alpha-diversity = 974.385 species) is more diverse than the BA mangrove (alpha-diversity = 847.721 species). The index of alpha-diversity summarizes the diversity of organisms in a sample and is estimated according to the distribution of species-level annotations.

The analysis of mangrove-associated microbiota

At the domain level, Bacteria were more abundant than Archaea in the two mangrove samples (Additional file 2: Figure S2, Table 1). The MG-RAST analysis indicated that the percentage of sequences affiliated with each taxa were similar in the BA and RJ mangroves, with an abundance of Proteobacteria (57.8% and 44.6%), Firmicutes (11% and 12.3%), Actinobacteria (8.4% and 7.5%), Bacteroidetes (4.1% and 4.3%), Chloroflexi (3.2% and 6.4%), Cyanobacteria (2.2% and 5.8%), Euryarchaeota (2.2% and 4.5%), and Planctomycetes (1.6% and 1.9%). The statistical test applied by MEGAN revealed that the proportional difference in reads assigned to Chloroflexi, Cyanobacteria, Planctomycetes, Proteobacteria, and Chlamydiae/Verrucomicrobia was highly significant between the BA and RJ mangroves, with the RJ sample having a higher number of reads assigned to those taxa. Moreover, the Actinobacteria, Fibrobacteres/Acidobacteria, and Firmicutes groups were significantly more highly represented in the BA mangrove (Figure 2A).

In terms of the phylogenetic classification within Proteobacteria obtained via MG-RAST, the most frequent class detected was Gammaproteobacteria (13.5% in BA and 13.6% in RJ), followed by Epsilonproteobacteria (11.6%), Alphaproteobacteria (11.5%), Deltaproteobacteria (10.9%), and Betaproteobacteria (9.9%) in the BA sample, and followed by Deltaproteobacteria (12.5%), Betaproteobacteria (8.2%), Alphaproteobacteria (7.9%), and Epsilonproteobacteria (2.1%) in the RJ sample. The statistical analyses from MEGAN indicated that the BA mangrove had significantly more reads in Alphaproteobacteria, Betaproteobacteria, and Epsilonproteobacteria, whereas the RJ mangrove had significantly more reads in Gammaproteobacteria, Deltaproteobacteria, and Zetaproteobacteria (Figure 2B).

At the genus level, Pseudomonas (4.3%), Campylobacter (2.9%), and Geobacter (2.2%) were found to be the main bacterial groups in the BA sample, as measured according to the MG-RAST results. In contrast, Geobacter (2.2%), Roseiflexus (1.9%), and Burkholderia (1.8%) were found to be the main bacterial groups in the RJ sample.

According to the MEGAN results, the majority of the eukaryotic EGTs (40.1 and 37.2% for BA and RJ, respectively) were more similar to the Opisthokonta, Stramenopiles (21.8 and 32.2%), and Viridiplantae (24.3 and 18.7%, respectively). However, there was no significant difference between the mangroves in terms of either their eukaryotic or their viral sequences.

Functional metagenomic analyses

Subsystem-based annotations (SEED) were performed to analyze the metabolic and physiological conditions in the metagenome of the mangrove samples (Figure 3). A wide variety of environmental gene tags (EGTs) from several metabolic routes were found. The mangroves were found to have a similar profile in terms of the percentages of genes from all of the SEED terms identified. Moreover, the MG-RAST (Figure 3) and MEGAN (Additional file 3: Figure S3) analyses gave similar patterns for both groups of mangrove samples.

The RJ sample was found to have a significantly higher number of EGTs related to carbohydrate metabolism (Figure 3). Genes commonly found in other metagenomes related to amino acid metabolism, cellular respiration, and DNA and protein metabolism were also identified. Furthermore, sequences related to virulence, the stress response, membrane transport, cell wall and capsule synthesis, and the metabolism of aromatic compounds were also significantly more abundant in the BA mangrove. A greater number of EGTs related to nitrogen metabolism, potassium metabolism, and iron acquisition and metabolism were also identified in the BA sample. In contrast, genes involved in sulfur metabolism (Figure 3) and CO₂ fixation (Additional file 4: Figure S4) were found to be more highly represented in the RJ mangrove.

Further analyses of the functional composition of mangrove metagenomes using similarity to a non-redundant protein database against the KEGG metabolic pathways produced similar results (Figure 4A). A deep examination of the carbohydrate metabolic pathways (Figure 4B) revealed such subgroups as fructose and mannose, starch and sucrose, and galactose metabolism with a significant number of putative genes potentially involved in the decomplexation of vegetable biomass.

Several genes related to cellulose degradation were found, including endo-1,4-β-D-glucanases (entry 3.2.1.4; BA: 72 and RJ: 90), 1,4-β-cellobiosidases (3.2.1.91; BA: 4 and RJ: 3), and β-glucosidases (3.2.1.21; BA: 279 and RJ: 228). Sequences classified as putative xylan-degrading enzymes (3.2.1.37, 1,4-β-xylosidases; BA: 23 and RJ: 9) were also identified (Figure 5).

The enzymes necessary for the first step of methane metabolism, i.e., the transformation of methane into methanol, were not found. However, several genes related to the conversion of carbon dioxide into carbon monoxide and acetyl-CoA were recovered (Figure 6A). A total of 504 and 562 genes involved in the transformation of formate to CO₂ were identified in the BA and RJ samples, respectively. There were 75 (BA) and 103 (RJ) hydrogen dehydrogenases responsible for the interconversion of H⁺ and H₂ identified in the samples.

In terms of sulfur metabolism, several enzymes in the mangroves (BA: 68 and RJ: 72) convert adenylylsulfate into sulfite (Figure 6B). The majority of enzymes involved in nitrogen metabolism are represented between the two mangroves, with the exception of methylaspartate ammonia-lyase (EC number: 4.3.1.2). This enzyme produces ammonia as a subproduct and is present only in the RJ sample (Figure 6C). The pathways leading to ammonia production are well represented by histidine ammonia-lyase (EC: 4.3.1.3), L-tryptophan indole-lyase (EC: 4.1.99.1), D-amino-acid dehydrogenase (EC: 1.4.99.1), and L-cystathionine L-homocysteine-lyase (EC: 4.4.1.8). These enzymes were found at higher abundance levels in the BA sample.

The BA and RJ samples were found to have 60 and 49 nitrogenases involved in biological nitrogen fixation, respectively. Cyanobacteria, which were highly abundant in both mangroves, as well as Chlorobi (green sulfur bacteria), Azotobacteraceae (Gammaproteobacteria), Frankia (Actinobacteria), and Rhizobia (Alphaproteobacteria) are the diazotrophs involved in this process. Enzymes related to the nitrification were thus identified. However, the denitrification process, which reduces nitrates to nitrogen gas, has an even higher number of sequences in the two mangroves. The bacteria involved in this process are anaerobic and use nitrates as an oxygen alternative to the final electron acceptor in the respiration cycle.

Metabolic pathway comparative analysis with the KEGG database indicated the presence of 202 and 183 alcohol dehydrogenases (EC number: 1.1.1.1) in BA and RJ samples, respectively, which are involved in polycyclic aromatic degradation. Additionally, 27 and 6 salicylate 1-monooxygenases from the same pathway (EC number: 1.14.13.1) were found in the BA and RJ samples, respectively. These enzymes play a role in 1- and 2- methylnaphthalene degradation.

Naphthalene is considered a PAH and is known to cause human disease (IARC, 2002). After evaluating the metabolism of xenobiotics by the cytochrome P450 pathway, two important enzymes for naphthalene metabolism were identified: glutathione dehydrogenase, which had 117 and 40 hits for BA and RJ, respectively, and benzo[a]pyrene-4,5-oxide hydratase, which had 14 and 10 hits for BA and RJ, respectively.

Carbohydrate-active enzyme analysis

Using the CAZy database (http://www.cazy.org), the two mangrove samples were found to have the same metabolic potential to hydrolyze carbohydrates, including cellulosic compounds (cellulose, hemicellulose, pectin, xylan, among others) (Additional file 5: Table S1). This analysis identified more than 2,900 Environmental Gene Tags (EGTs) from 110 different CAZy families, most of which were from Proteobacteria. Some CAZy families were only found in one mangrove: GH 6, 45, 70, 71, 72, and CBM 4 were only found in the RJ sample, while GH 11, 46, 79, 81, CE 5, and CBMs 15, 25, and 33 were only found in the BA sample.

A wide diversity of GH (glycosyl hydrolase) families was also identified: 54 and 56 different GH families were found in the BA and RJ mangroves, respectively. GH families with cellulolytic enzymes involved in the degradation of plant cell walls were found in both metagenomes, and only a few members of GH 6 and 45 were detected exclusively in the RJ sample. In addition, enzymes involved in the hydrolysis of hemicellulose/pectin and xylan side chains, including the β-galactosidases (GH 2, 27, 35), the xylanases (GH 10, 26, 43), the acetylxylan esterases (CE 4), the α-mannosidases (GH 38), the α-xylosidases (GH 31), the pectin methylesterases (CE 8), the α-L-rhamnosidases (GH 78), the α-glucuronidases (GH 67), and the pectin lyases (PL 1), were identified (Additional file 5: Table S1).

The RJ and BA mangroves also possess a wide diversity of carbohydrate-binding modules (CBM), which may promote the interaction between a given enzyme and its target substrate, thereby increasing its catalytic efficiency. Important modules for the degradation of cellulose were found (CBM4/9/16/22). The CBM module with the most identified members was CBM50, known as the LysM domain.

In total, 58 families of glycosyl hydrolases, 23 families of glycosyltransferases, 5 families of carbohydrate esterases, and 3 families of polysaccharide lyases were identified in both mangrove microbiomes.

When comparing mangroves to other host-associated and environmental metagenomes (Additional file 5: Table S1), including those from termites, pandas (Zhu et al. 2011), marine water systems (DeLong et al. 2006) and soil, some GHs involved in the hydrolysis of cellulosic/hemicellulosic compounds were found exclusively in the mangroves (GH 6, 12, 17, 44, 46, 47, 62, and 76). Several CBMs important in the recognition of cellulosic compounds were exclusive to the mangroves (CBM 2, 3 and 15). Families GH 3 and GH 13, which contain a large range of glycosidases able to hydrolyze complex carbohydrates into oligosaccharides, were abundant.

Crucial carbohydrate degradation-related domains, including the dockerin and cohesin domains, were found in both mangroves; the cohesins were highly represented (Additional file 5: Table S1). In other metagenomes, only one dockerin (in the marine metagenome) and few cohesins (in the panda and marine metagenomes) were detected. Even in metagenomes specialized in cellulose degradation, such as the termite metagenome, no dockerin/cohesion sequences were identified, further showing the remarkable potential of the mangrove environment for enzyme identification.

Deep analyses of these results also identifies microorganism sequences specialized for cellulose/biomass degradation. Sequences belonging to Anaerolinea thermophila, which is a filamentous and thermophilic bacterium, were the most abundant in both mangrove samples (30 CAZy sequences) (Additional file 6: Table S2). Interestingly, these results also showed the identification of CAZy sequences from Fibrobacter succinogenes in both microbiomes (RJ: CBM 4, 9, 16, and 22; and BA: GH 30). Sequences from other specialized microorganisms were also found, including Sorangium cellulosum, Solibacter usitatus and Spirochaeta thermophila.

Comparative metagenomic analysis

The functional information provided by SEED was used to compare the Brazilian mangrove metagenome to the published metagenomes of other environments, including soil, water, and host-associated environments. Principal component analysis with only the first two components could explain 77.2% of the data variance. The resulting plot (Figure 7) showed that the BA and RJ mangroves formed a cluster near the metagenomes of water and soil. The metagenomes from the Caribbean Sea (open ocean), the Atlantic Ocean (4,200 m), agricultural soil, tropical forest soil, and rice rhizospheres were the most similar to the mangrove metagenome. The host-associated metagenomes formed two main groups: one formed by the human and termite sequences, and the other formed by the mouse sequences. Additionally, the metagenomes of euphotic, mesopelagic and ocean regions with a minimum oxygen layer were similar to these functional categories and formed a cluster that was separate from the others.