Darwin Harbour is lined by mangroves, with extensive intertidal mudflats, and subject to a tropical savannah climate, with a distinct dry and wet season and average annual rainfall of 1,727 mm (www.bom.gov.au). The harbor is adjacent to the city of Darwin, which has a population growth rate of 1.9%; the second fastest of all Australian capital cities in 2014–2015 (www.abs.gov.au).

Buffalo Creek, a tidal creek in Shoal Bay in the north of the harbor (Figure 1) receives secondary-treated sewage effluent (140 t total N year⁻¹ and 26 t total P year⁻¹ in 2013) from an effluent outfall in the headwaters of the creek, 4.8 km from its mouth. The estimated population of the serviced area of the waste stabilization ponds that discharge to Buffalo Creek was 47,466 in 2013. The ponds consist of two parallel treatment trains with five waste stabilization ponds in each series (one facultative and four maturation) designed to remove organic matter, nutrients and fecal bacteria and viruses. Hypereutrophic Buffalo Creek has a narrow meandering channel through mangroves with multiple barriers to direct water exchange and tidal movement including a large barrier sand bar across the mouth (Figure 1). Suburban stormwater and runoff also flow into the headwaters of Buffalo Creek and parts of it were also artificially filled for mosquito control. Micket and King Creeks are the reference creeks for Shoal Bay (Figure 1) with less human impact although Micket Creek still receives some urban runoff. Micket Creek also has a sandbar at the mouth albeit smaller. There are extensive salt flats behind the fringing mangroves of the Shoal Bay creeks.

In the East Arm area of the harbor (Figure 1), Myrmidon Creek receives secondary-treated sewage effluent (109 t total N year⁻¹ and 18 t total P year⁻¹ in 2013) from an outfall for five waste stabilization ponds in series (one facultative and four maturation), servicing the city of Palmerston with estimated 31,216 households in 2013. The effluent is discharged into a 600 m long tributary which branches off the main creek 900 m upstream from its mouth. Myrmidon Creek has a wide and relatively straight channel surrounded by mangroves and is considered oligotrophic to locally mesotrophic. Reference Creek and Short Creek are the reference creeks for East Arm, as well as two sites off Blaydin Point in the middle of East Arm (Figure 1). Both, Myrmidon and Short Creek also receive urban stormwater.

Water and sediment were collected from 32 sites in the Shoal Bay and East Arm areas in Darwin Harbour (Figure 1). For Shoal Bay, seven sites were chosen along the impacted Buffalo Creek between the site where the effluent entered the creek channel through a pipe and the sandbank at the mouth. A site at the headwaters of Buffalo Creek close to Darwin suburbs was included to represent tidal creek water mixed with urban runoff (Figure 1). Treated sewage effluent and urban runoff were also collected before mixing with the creek water. There were three sites each from two reference creeks, Micket and King Creeks. The distance between the mouth to the effluent outfall or most upstream site for the reference creeks was similar i.e., 4.6–4.8 km for all three creeks. Similarly for East Arm, there were seven sites along the impacted creek (Myrmidon Creek) as well as three sites each from two reference creeks (Short and Reference Creeks). Pure effluent was collected as well as water at the outfall site where the effluent discharged across a bank into the mangroves and a Myrmidon side-tributary (Figure 1). A further site was chosen in this tributary 400 m downstream from the outfall and 200 m upstream from the Myrmidon main channel. Two sites off Blaydin Point captured the water from the center of the East Arm estuary.

There were eight rounds of sampling over 2 years with two rounds each in April 2013, Sept 2013, Feb 2014, and Sept 2014. This covered two consecutive wet and dry seasons and for water samples, a neap and spring tide sampling round for each of the seasons. For access reasons, water samples were collected just after high tide at outgoing tide and thus, represented the best-case scenario in terms of dilution of nutrient inputs from effluent or land runoff. Samples were collected from a depth of 0.5 m and in duplicates from key sites during neap tide. It was only possible to collect sediment during neap tides. Using a corer the top 10 cm were collected in duplicates. All samples were kept on ice and processed within 6 h of collection. Water samples were aliquoted and parts filtered in situ using 0.45 μm inline filters for subsequent nutrients analysis while water for subsequent molecular analysis was filtered in the laboratory within 6 h of collection (see below) and filters were frozen at −20°C. Sediment was aliquoted for molecular and physicochemical analysis and frozen (see below).

On average, 1.5 μg of dried DNA were sent to mrdnalab (www.mrdnalab.com, Shallowater, TX, USA) for 16s rRNA gene amplicon sequencing using the MiSeq platform (Illumina, San Diego, CA, USA). The 16-s rRNA gene was amplified using the V4/V5 primers 16S_F563/16 (5-AYTGGGYDTAAAGNG) and 16S_BSR926/20 (5-CCGTCAATTYYTTTRAGTTT) (Claesson et al., 2010) with barcodes on the forward primer. The HotStarTaq Plus Master Mix Kit (Qiagen, USA) was used with the following conditions: 94°C 3 min, 28 cycles of 94°C 30 s, 53°C 40 s, 72°C 1 min, final elongation step at 72°C for 5 min. Pooled PCR products were purified using calibrated Ampure XP beads and a DNA library was prepared following the Illumina TruSeq DNA library preparation protocol. Sequencing was performed on a MiSeq Illumina sequencing platform following the manufacturer's guidelines using the v2 chemistry for the first two rounds of samples and v3 for the remaining six rounds.

The Usearch-8 fastq (www.drive5.com) and FastQC (www.bioinformatics.babraham.ac.uk) tools were used to assess the quality of the MiSeq R1 and R2 fastq files. Four MiSeq processing pipelines were compared; Mothur (Schloss et al., 2009), UPARSE (Edgar, 2013), QIIME closed reference and QIIME open reference (Caporaso et al., 2010). The QIIME open reference pipeline was chosen (see reasoning Figure S1, Table S1). QIIME 1.9.0 was used on a Linux cluster. In short, paired ends of R1 and R2 files were joined using the QIIME default parameters. Quality control was conducted using the Usearch 8 fastq-filter with a minimum length threshold of 330 bp (median length 372 bp) (Edgar, 2010), an expected error threshold of 1 and no N's allowed. The forward and reverse complement of R1 and R2 sequences were checked for barcodes and forward primers and the sequences were demultiplexed. The maximum unacceptable Phred quality score was set to 20. Chimeras were removed using the Usearch8 uchime_ref command and the QIIME “gold” reference database (Edgar et al., 2011). OTUs were picked using the default parameters incl. the Uclust method at 97% similarity. For taxonomy assignment, the Greengenes database was used (release May 2013) (McDonald et al., 2012). Singleton OTUs and OTUs assigned to Archaea were excluded. The sequencing depth ranged from 10,711 to 588,040 sequences per sample (median 68,842 sequences). To account for differences in filtering volume, DNA extraction and PCR efficiency and sequencing depth, all samples were subsampled to 17,000 sequences. This number was chosen as a trade-off between excluding a minimum number of samples with <17,000 sequences (10 sediment and three water samples were excluded) while still retaining an average Goods coverage of above 90% for all remaining 274 water samples. A flattening of the rarefaction curves was observed for both, water and sediment samples (Figure S2). A second stage analysis (Primer-E 7, Plymouth UK) compared the weighted UniFrac resemblance matrices of all water and sediment samples rarefied to different subsampling depths (i.e. no subsampling, subsampling to 17,000 or 20,000 sequences). The distance matrices showed high Spearman correlations of >0.99 between the different subsampling depths, indicating that the effect of subsampling upon the beta analysis was negligible for both water and sediment samples (Table S1, Figure S1).

Data were analyzed using Primer-E 7 (Plymouth, UK), StataIC 14 (www.stata.com), R studio (R v3.2.2) and QIIME 1.9.1. Alpha diversity (Simpson diversity and Goods coverage) was assessed in QIIME (alpha_diversity.py) and differences tested using Kruskal-Wallis in Stata. Beta diversity was assessed using a weighted UniFrac distance matrix at OTU level based on the abundance weighted fraction of the total phylogenetic branch length not shared between a pair of samples (Lozupone et al., 2011). The weighted UniFrac distance matrices were visualized using unconstrained principal coordinates analyses (PCO).

During eight sampling rounds over 2 consecutive years (2013–2014), two dry and wet seasons and four neap and spring tides, 286 water and 208 sediment samples were collected from 32 sites from two areas in Darwin Harbour, namely Shoal Bay and East Arm creeks (Figure 1). Sixteen sediment (7.7%) and six water (2.0%) samples did not produce 16s rRNA gene amplicon sequences. Samples were rarefied to 17,000 sequences, which resulted in a mean Goods coverage of 90.2% for water samples (n = 277) and 79.2% for sediment samples (n = 192). For sediment samples, 197,507 OTUs were called by the pipeline with an average of 5,420 OTUs per sample [standard deviation (sd) 1,180]. Significantly fewer OTUs per sample were measured for water with an average of 2,391 OTUs (sd 937) and total 162,125 OTUs (Student's t-test t = 30.4, P = 0.001).

The association between the microbiota and seasons, tides, years and location for the East Arm and Shoal Bay areas were analyzed to determine whether the microbiota differed spatially namely between creeks, and/or temporally i.e., between tides, seasons or years.

The dry and wet seasons had the greatest impact on the Darwin Harbour water microbiota. This was the dominant factor and accounted for more variation in the microbiota than any other tested factor, including the unexplained residual variance within groups of samples (Table 1). There was also season-specific clustering in particular for the East Arm water microbiota (Figure S3). Excluding the outfall and urban runoff-related sites, there were more bacteria of the family Chromatiaceae (purple sulfur bacteria) in the dry season in particular in Myrmidon Creek (IndVal 0.92, P < 0.001) while there were more cyanobacteria such as Phormidiaceae in the wet season (IndVal >0.84, P < 0.001). Year of sampling (2013 vs. 2014) was the second most important factor for East Arm and third for Shoal Bay after seasons and tides.

There was a strong effect of neap vs. spring tides and it was the second most important factor for Shoal Bay and third for East Arm (Table 1). There were significantly more Burkholderiales and Phormidiaceae during neap tides in particular for Shoal Bay and the wet season (IndVal >0.81, P < 0.001) while spring tides were associated with more Verrucomicrobia or Planctomycetes (IndVal >0.75, P < 0.001). There was a significant interaction between creeks and tides (Table 1) which indicated that tides also changed the effect of location upon the microbiota. There was a creek specific microbiota signature for neap tides, which was largely absent for spring tides (Figure 2). While the microbiota from the impacted creeks (Myrmidon and Buffalo) significantly differed from their reference creeks for neap and spring tides (t-test, P < 0.01), the reference creeks only differed from each other during neap tides. The creek-specific signature was particularly evident for Shoal Bay where the Buffalo Creek microbiota gradually changed from an impacted signature at the outfall to a Shoal Bay background signature that was also shared by the reference creeks King and Micket (Table 1, Figure 2). The bacteria in the impacted Myrmidon Creek showed a typical disturbance—recovery trajectory with the effluent entering the creek through a side-tributary and the microbiota reverting to the background signature close to where the tributary water entered the main channel (Figure 2). There was no significant difference in the microbiota dispersion between creeks (P > 0.1).

Comparing the microbiota between the two effluent outfalls, the effluent discharging into Buffalo Creek contained significantly more Deinococci, Planctomycetes and Chlamydiales (IndVal = 1 for all, P = 0.038), while the effluent at Myrmidon Creek had more Clostridiales, Cloacamonae and Bacteroidetes (IndVal > 0.9, P = 0.038). The urban runoff into the Buffalo Creek headwaters contained significantly more Enterococcaceae, Aeromonadaceae and Lactobacillales, compared to the effluent at Buffalo Creek (IndVal > 0.8, P < 0.01).

We tested the predictive ability of the microbiota for location i.e., given the microbial signature, was it possible to predict the creek from which the water was collected? In particular for Shoal Bay there was good separation between the impacted and most distant reference creek (Table S2). Buffalo Creek showed the highest correct classification rate with 80% while the water microbiota from Myrmidon Creek showed the lowest predictive ability with only 42% of samples correctly assigned. Control samples had a considerable amount of misclassifications between each other reflecting their similarity.

An eigenvector-based spatial analysis of the neap water microbiota of Shoal Bay showed three independent patterns of which two consisted of linear trends along the x and y axes reflecting the effluent point source at Buffalo Creek (Figure S6). The third non-linear pattern only explained 2.2% of the Shoal Bay microbiota variance and showed a correlation between the urban runoff microbiota into the upper reaches of Buffalo Creek and the most upstream Micket site. Micket Creek also receives urban runoff albeit to a much lesser degree. The last pattern was associated with higher DOC, TN, and TP for these upstream sites (Mann-Whitney U-test, P < 0.01). There were significantly more Neisseriaceae in these upstream samples (IndVal 0.8, P value 0.01). For East Arm, there was only one spatial pattern significantly associated with the neap water microbiota. It explained 14.9% of the microbiota variance and reflected the effluent outfall at Myrmidon (Figure S6). Both harbor areas showed similar variation partitioning patterns. Nutrients accounted for most of the explainable variation in the microbiota with total 34–35%, followed by salinity with 21–22%, and spatial components 15–18% (Figure 4). While 5% of the Shoal Bay microbiota variance was explained by concomitant local changes in salinity and nutrients, there was no such measurable fraction for East Arm. The overall contributions of the spatial fraction were smaller for the East Arm model and only a quarter of the salinity and nutrient fractions were also shared with the spatial fraction compared to a third for Shoal Bay.

In support of the findings above, the Shoal Bay microbiota had a stronger site-specific structure compared to East Arm (Figure 5). Shoal Bay and in particular the Buffalo Creek microbiota were more similar to the treated effluent and urban runoff related microbiota than the East Arm microbiota. A comparison of bacterial phyla between sites showed the distinct higher abundance of Firmicutes in the effluent and runoff microbiota, and green sulfur bacteria of the phylum Chlorobi in the effluent (Figure 5). Marine group A (SAR406) was distinctively more abundant in East Arm water compared to Shoal Bay water.

Temporal factors showed the strongest relationship with the water microbiota. Temporal shifts in the microbiota composition of temperate aquatic habitats have been linked to seasonal variation of physicochemical water properties (Fuhrman et al., 2015) and allochthonous inputs upon increased rainfall (Jeffries et al., 2016). In this study, the dry and wet seasons showed the strongest association with the water bacteria composition. The monsoonal wet season brings high rainfall, which significantly decreases salinity levels while raising TDN and TSS levels. More Phormidiaceae were detected in the wet season, a family of blue-green algae which also contains the genus Planktothrix with members able to form toxic blooms in freshwater habitats (Hossain et al., 2012). In Malaysia and Bangladesh, differences in phytoplankton diversity between monsoonal seasons were found to be dependent on runoff, nutrients, TSS and temperature (Sidik et al., 2008; Hossain et al., 2012). In the dry season, significantly more phototrophic purple sulfur bacteria of the family Chromatiaceae were detected. Members of this family have been previously described from this region (Cornall et al., 2013), and they primarily occur in stagnant water (Pfennig and Trüper, 1992), conditions more likely in the dry season when there is virtually no freshwater input.

The dry season was also associated with lower water temperature and more dissolved oxygen. Intriguingly, while both areas, East Arm and Shoal Bay recorded the same water temperature, only the water microbiota in East Arm was associated with changes in water temperature and dissolved oxygen. Water in East Arm can be trapped for several months with high evaporation in the dry season leading to inverse estuary effects and increased salinity (Williams et al., 2006; Tonyes et al., 2015). While the residence time for the whole of East Arm is long, the creeks feeding East Arm are open channels allowing good mixing of water within East Arm, which weakens spatial patterns and dilutes nutrients from effluent and stormwater. Thus, considering long water residence times and no distinct nutrient and salinity gradients, temporal seasonal patterns might become more pronounced and may explain the clear seasonal clustering of the East Arm microbiota.

In contrast to East Arm, spatial patterns including the association with salinity were decisive for the Shoal Bay water microbiota. Salinity gradients from the upper reaches to the mouth of the creeks were considerably larger compared to East Arm, particularly in the wet season with more freshwater inputs into the headwaters of Shoal Bay. The main reason for the more pronounced salinity gradient across Shoal Bay was likely reduced mixing within the creeks as sand bars at the mouths of the creeks restrict tidal water exchange and wave mixing. These sand bars act as weirs forming barrier bar estuaries—a phenomenon not seen in East Arm. A particularly large intertidal sand bar at the mouth of Buffalo Creek reduces the tidal amplitude by up to 50% (Smith et al., 2012). Flushing of Buffalo Creek is further impeded by the creek's narrow meandering morphology which creates eddy circulations that restrict longitudinal mixing. In the dry season, the upper reaches of Micket and King Creeks turn hypersaline with levels up to 41 ppt (Ridd and Stieglitz, 2002). The Shoal Bay catchment also has extensive areas of salt pans which are regularly inundated at high tides. In relation to these salinity gradients, the Shoal Bay microbiota were more heterogeneous within a creek compared to East Arm, and had more location specific signatures, particularly in the wet season.

Location or creek specific microbial patterns were mainly present at neap tides and virtually absent during spring tides when the water was well mixed. Exceptions were sites close to the effluent outfall and urban runoff. Differences between neap and spring tides were the second most important factor for the Shoal Bay water microbiota after seasons, and ranked third for East Arm after seasons and sampling years. Macro-tides and wet season run-off both potentially represent major challenges to microbiota survival, and we know relatively little about how they respond to shifts in salinity. In a similar setting (Hyun et al., 1999) reported that the microbial community alternated between autochthonous halotolerant estuarine and allochthonous halophobic freshwater populations. We found that Verrucomicrobia, abundant in marine environments (Freitas et al., 2012), were more abundant during spring tides, which suggests increased exchange with oceanic water. In contrast, neap tide conditions during the wet season in Shoal Bay were associated with more Burkholderiales and Phormidiaceae, which reflects the higher freshwater inputs and reduced mixing with marine water. The higher abundance of blue-green algae of the Phormidiaceae family might also reflect the elevated nitrate levels in Shoal Bay (Heath et al., 2016). The finding of clearer spatial patterns of the water microbiota during neap tides suggests that water quality monitoring at neap tides better reflects the local water conditions including anthropogenic nutrient sources. We also found higher nutrient levels at neap tide compared to spring tides. It was of note that the composite of nutrient species showed fewer spatial patterns than the neap water microbiota suggesting the latter to be a superior tool to discern the spatial impact of elevated nutrients.

Nutrient levels were important for explaining the water microbiota for both harbor areas. For the East Arm water microbiota, TDN accounted for most of the microbiota variance in marginal tests, and ammonia was the most important factor in the multivariate distance model. This reflected the high ammonia levels measured in the effluent at the Myrmidon outfall, which were diluted rapidly once reaching the Myrmidon main channel. The Simpson diversity of the effluent microbiota at the Myrmidon outfall was significantly lower than at any other site in this study including the effluent microbiota at the Buffalo Creek outfall. A decrease in microbial diversity has been associated with a decline in ecosystem function (Sun et al., 2012). The water microbiota showed a disturbance-recovery trajectory along Myrmidon Creek with a rapid reversal to the background microbiota after the effluent entered the main channel through the tributary, and the rest of the East Arm water microbiota clustered tightly. This pattern also reflects the increased water mixing and low nutrient levels within East Arm compared to Shoal Bay.

In contrast to East Arm, TDN and NO_x explained most of the microbiota variation in Shoal Bay after salinity. NO_x levels were considerably higher in the effluent discharged into Buffalo Creek compared to the Myrmidon outfall although these levels were still lower than those of ammonia. Deinococci and Planctomycetes bacteria were more abundant at the Buffalo outfall compared to the Myrmidon outfall. Both these taxa contain ammonia-oxidizing bacteria and have also been described in a batch reactor with high NO₂ concentrations (Tan et al., 2008).

Indicative of the reduced water flushing in Buffalo Creek, there was only a gradual change of the water microbiota from the disturbed signature at the outfall back to the Shoal Bay background microbial composition at the mouth of the creek. Similarly, phylogenetic analysis of the water microbiota showed that the urban runoff and outfall microbial signatures from both outfalls were more similar to the microbial signatures from Shoal Bay water than to East Arm water. The urban runoff showed the highest nitrate concentration of all water samples. Nitrate is often the most common soluble nitrogen species in stormwater with various sources including soil nitrification processes, fertilizers, animal waste or abandoned landfills (Wakida and Lerner, 2005). The urban runoff also had the highest bacterial Simpson diversity and compared to the effluent at Buffalo outfall, contained more bacteria of families with known fecal indicators, commensals or human pathogen indicators such as Enterococcaceae, Lactobacillales, or Aeromonadaceae (Cabral, 2010; Roslev and Bukh, 2011).

The water microbiota differed at sites closer to the outfall and urban runoff regardless of seasons and tides. However, it was not clear how much this was due to eutrophication i.e., nutrient load and/or changes in salinity levels due to freshwater inputs. We therefore conducted a variation partitioning analysis on the neap water microbiota for each harbor area to distinguish between nutrient, salinity, and spatial factors. A spatial analysis showed that the microbiota changed with increasing distance from the effluent point source. A non-linear spatial trend also showed an association between the microbiota at the most upstream Buffalo and Micket Creek sites. Both these creeks receive urban runoff and both had significantly more Neisseriaceae than at other sites. Neisseriaceae have been described in groundwater of agricultural land and contain several mammalian commensals and pathogens (Wakelin et al., 2011). For both harbor areas, the composite of nutrient species accounted for the major part of the explainable variation in the water microbiota. The variation partitioning for East Arm showed a distinct lack of a fraction which associated the microbiota variance with salinity and nutrient levels. This indicates that local changes in salinity and nutrients were not well correlated and may reflect the generally low nutrient levels across East Arm with the largest changes in nutrients confined to the Myrmidon outfall and adjacent side-tributary. Compared to Shoal Bay, the spatially structured environmental variation (Borcard et al., 2011) as well as the total spatial fraction explained less of the East Arm microbiota variance. This again reflects the less site-specific bacterial signatures for East Arm.

Mangrove ecosystems are highly productive with a diverse sediment microbial community recycling nutrients and acting as a carbon and nitrogen sink (Holguin et al., 2001). The sediment microbiota in this study also showed a considerably larger diversity than the surface water microbiota. The sediment and water microbiota composition significantly differed which was also due to sediment pore and benthic water typically being anoxic with associated significant changes to redox sensitive reactions. There was a distinct absence of spatial patterns for the sediment microbiota in East Arm, which did not differ beyond the immediate outfall confirming the still healthy condition of the sediment in this part of the harbor. In contrast, the sediment microbiota in Shoal Bay showed a creek-specific signature in particular for Buffalo Creek. The creek has received effluent and urban runoff for over 40 years and sediment pore water is enriched with dissolved inorganic nutrients. Benthic nutrient fluxes are high while denitrification efficiencies are variable including below 10% (Smith and Haese, 2009; Smith et al., 2012). Increased tidal pumping and exchange with nutrient enriched sediment pore water during spring tides can further increase nutrient loads in surface water peaking at the following neap tide (Call et al., 2015). This exchange is further enhanced through numerous crab burrows in the mangrove sediment increasing hydraulic connectivity and the surface area of the sediment water interface. Also pelagic primary production rates were found to be of 1–2 order of magnitude higher in Buffalo Creek than in the reference creeks (Smith et al., 2012). The discharge of effluent, exacerbated by limited creek flushing, leads to increased algal biomass and degraded water quality with hypoxic conditions in particular at night (Smith et al., 2012). We found more anaerobic sulfate reducing Desulfobacteraceae and nitrifying Nitrosomonadales in the sediment of Buffalo Creek, as well as Chloroflexi bacteria which contain green non-sulfur bacteria and Clostridiales, an order of mostly fecal anaerobes with known sewage indicators (McLellan et al., 2013). These bacteria have previously been described in polluted sediment receiving sewage effluent (Zhang et al., 2008; Lu and Lu, 2014).

For both treated sewage outfalls, sediment microbiota were associated with higher levels of copper in sediment which is consistent with other reports for sewage discharge sites in Darwin Harbour (Padovan et al., 2012). The bacteria in upper Buffalo Creek sediment were associated with higher levels of sulfur, which could explain the greater detection of Desulfobacteraceae in this creek. In King Creek, sediment bacteria were linked to a more positive redox potential reflecting the more oxygenated and less nutrient rich environment. It also contained more of the nitrite oxidizing Nitrospirales which have been found to decrease in abundance under anoxic conditions in marine sediment (Devereux et al., 2015). Despite the creek-specific signature of the Shoal Bay sediment microbiota and association with some abiotic factors, they explained a considerably smaller amount of the sediment microbiota compared to the water microbiota for both harbor areas. This is indicative of sediment acting as a long-term sink for nutrients and reflecting historical as well as current water conditions. While for many ecosystems, contemporary environmental conditions are crucial for bacterial species sorting, other studies including on sediment found a slow response of the microbiota to changing conditions and past processes were equally important to shape a bacterial community (Reed and Martiny, 2013; Andersson et al., 2014).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.