Spatial patterns of microbial communities across surface waters of the Great Barrier Reef

Microorganisms are fundamental drivers of biogeochemical cycling, though their contribution to coral reef ecosystem functioning is poorly understood. Here, we infer predictors of bacterioplankton community dynamics across surface-waters of the Great Barrier Reef (GBR) through a meta-analysis, combining microbial with environmental data from the eReefs platform. Nutrient dynamics and temperature explained 41.4% of inter-seasonal and cross-shelf variation in bacterial assemblages. Bacterial families OCS155, Cryomorphaceae, Flavobacteriaceae, Synechococcaceae and Rhodobacteraceae dominated inshore reefs and their relative abundances positively correlated with nutrient loads. In contrast, Prochlorococcaceae negatively correlated with nutrients and became increasingly dominant towards outershelf reefs. Cyanobacteria in Prochlorococcaceae and Synechococcaceae families occupy complementary cross-shelf biogeochemical niches; their abundance ratios representing a potential indicator of GBR nutrient levels. One Flavobacteriaceae-affiliated taxa was putatively identified as diagnostic for ecosystem degradation. Establishing microbial observatories along GBR environmental gradients will facilitate robust assessments of microbial contributions to reef health and inform tipping-points in reef condition.

Correlation variation is more evident for inshore reefs than their offshore counterparts, where conditions seem to be more stable between summer (wet) and winter (dry).
Supplementary Fig. 8. Individual indicator results for GBR benthic categories.    Note that n=3 Riverine samples were excluded due to absence of eReefs data for nonoceanic sites.    Supplementary Fig. 17. Bacterial community composition across all regions, sampling groups and seasons obtained from the case-studies (n=147). a) Phylum-level and b) familylevel community composition averaged across seasons, and c) with seasonal resolution. For simplicity, only the most abundant bacterial phyla across all samples are shown. Note that for this comparison data was rarefied to a depth of 1,000 reads and that 20 samples were consequently removed from the Tully dataset.  Environmental variation was modelled by LDA to predict reef category for the sites with available microbial community data. Salinity, temperature, chlorophyll a, Kd_490, DOC and DIC were non-collinear variables (see Supplementary Fig. 2) and were thus included in the LDA modelling ( Supplementary Fig. 3).

Bacterial response to riverine and seasonal influences within inshore In-Porites reefs
In contrast to the BPA dataset used to compare microbiomes across GBR reef categories, the Tully dataset 1 included only sites assigned by our LDA model to the In-Porites reef category.
The importance of the Tully microbiome dataset is that it encompasses both the spatial dynamics generated by the outflow of the Tully river onto the inshore reef, as well as the superimposed temporal dynamics established between dry and wet seasons, and can be used to understand microbial variation within a single GBR reef category.
Alpha diversity (Richness; Supplementary Fig. 9a) varied significantly with season but not with location (no effect of plume versus marine locations) (see Supplementary Table 4 and Suppl. Fig. 10 for further results). Microbial diversity was higher in the dry season than in the wet season, what is consistent with the BPA dataset. Riverine sites showed a much lower diversity than the plume and marine sites, which did not differ from one another. nMDS ( Supplementary Fig. 9b) shows some overlap between communities based on location, but a clearer structuration of the community based on season, particularly for the plume sites.
Season differences and general heterogeneity seems to be more marked for the plume sites, Environmental variation in the Tully region (4x4 Km eReefs data) was mostly explained by four environmental parameters (DOC, salinity, temperature and chla) after dimension reduction via subtraction of collinear variables (from 16 original variables; Supplementary Fig.   11). The first two components of PCA represented 73.7% of variation in the dataset with n=73 samples included; Supplementary Fig. 12). dbRDA ( Supplementary Fig. 12) proceeded with n=71 microbial samples and model selection showed that all of the four constrains used explained significant variation in the microbial community (see Supplementary  Hence, proximity to river mouths can drive microbial community dynamics which increase taxa in the surrounding waters that have been implicated as causing corals diseases 2 . Marine Group II Euryarchaeota were also more abundant during the wet season 1 . These motile residents of the photic zone that have a photo-heterotrophic lifestyle through which they degrade protein and lipids, are also known to display great seasonal and spatial variation elsewhere 3,4 . Most likely there are also important numbers of Archaea across the abovementioned inshore to offshore gradient, as well as within each of the regions here characterized. Archaea have not been characterized widely (see, for example, Frade et al. 5 ) and their importance for coral reef functioning is far from being understood.
It is noteworthy that reefs in the Tully region show coral cover (46-51%) similar to that of Orpheus Island and higher than that of Magnetic Island, showing that the region seems to be in a good conservation state even after being hit by recent cyclones 6 . Even locations under the effect of a river plume can still sustain reasonable coral (11-32%) and low macroalgal cover (2-8%).

Shelf edge effects on bacterial community profiles.
The dataset derived from the Mackay region 7 , allows investigation of the influence of coastal distance and reef shelf on microbial community structure (Fig. 1). From the inshore and midshelf to the outershelf and shelfbreak there is large increase in relative abundance of Alteromonadaceae and a concomitant decrease in Pelagibacteraceae and the Halomonadaceae ( Supplementary Fig. 15). Major differences in community composition between the different locations studied across the GBR shelf occur during the dry season.
Whereas the seasonal effects for Tully region are likely related to terrestrial run-off and riverine incursion into marine communities along the inshore reefs 1,6 , seasonal drivers in the Mackay region could be of oceanic origin. The region is highly productive due to upwelling along the shelf, bringing nutrients and cold water from the deep and therefore contributing to shaping the microbial communities of outer reefs. However, the upwelling regime in that area is restricted to the summer wet months 8 , but the nutrient data available does not support a hypothesis of upwelling as a driver of changing microbial community structure 7 . This may be related to the very restricted temporal window during which samples were taken in the region.
One important characteristic of the GBR in this region is the extensive distance of the outer reef from shore. This and the prevalent upwelling system that is active during the wet season could contribute to differences seen in community composition. Alteromonadaceae are recognized as copiotrophs that can grow rapidly when organic nutrients are available in the environment 9,10 .

Bacterial community patterns in the southern GBR
The final data set is derived from Heron Island and presents a different microbial community compared to the more northern regions (see Supplementary Fig. 16 likely reflect the oligotrophic conditions in the system. However dominance of Flavobacteraceae, may suggest an increased abundance of opportunistic bacteria that may affect coral health 10,12 .

Methodological considerations
In order to incorporate available microbial community data from current case studies on the GBR with the reef categorization of Mellin et al. 13

Caveats of this meta-analysis
The datasets originating from the Burdekin region, including the Yongala lagoon site, plus the Coral Sea dataset were obtained with the same primer set and processed through the same analysis pipeline of BPA 16,17 . This allowed for a fairly robust comparison across sites and distinct reef categories. For comparisons across all available datasets, reads were further rarefied to a depth of 1,000 reads (See Supplementary Fig. 17  There are also obvious limitations in the extrapolation exercise we present to predict microbial communities for the wide-GBR. Firstly, the spatial resolution of the eReefs data obtained, for example, is still coarse (1x1 Km) if compared to the spatial scales at which microbial data was acquired. This means we are necessarily dealing with environmental conditions averaged over large areas, and therefore excluding any heterogeneity that could in fact better relate to the environmental conditions verified at the exact location where the microbial sampling was performed. However, the known congruence in microbial samples retrieved from seawater across sampling locations 16 counterbalances this limitation. Secondly, the range of environmental conditions covered by the LTMP dataset is broader than that covered in the microbial dataset. Even though these ranges overlap (see Supplementary Fig. 1), the microbial sites only cover a minor part of the environmental variation in the GBR benthic categories, which means that the estimated microbial community is only to be found potentially within a shorter range of the full spectrum of environmental variation. This means our estimations are necessarily too constrained and likely there is room to accommodate a broader microbial community.
With regards to the relative importance of the different reef categories across the GBR area mapped (in Fig. 6a), the undisclosed category Out-Tab only represents 1.4% of all mapped GBR reefs. Other outershelf reefs cover about 30% of the GBR (Out-Digit and Out-Soft accounting for 6.5% and 22.7%, respectively of all mapped reefs). Out-Soft is actually the second largest category, losing only to Mid-Mixed, or midshelf reefs, with 57.5%. Inshore reefs occupy just over 10% of the GBR (In-MA and In-Porites representing 2.3% and 9.7%, respectively, of all mapped GBR reefs). Our microbial predictions thus potentially apply to about 90% of the GBR, but the reality is that much more baseline research is needed until one can have a trustworthy representation of pelagic microbial variation across the wide GBR.
In addition to the longitudinal gradient of environmental and microbial variation, a large latitudinal temperature gradient occurs along the expanse of the GBR. Unfortunately, microbial datasets currently available did not allow robust comparisons to be made across latitudinal scales. The increasing occurrence of the In-Porites category towards the northern sector of the GBR (depicted in Fig. 6a) indicates that a similar bacterial community may be identified at these sites. The positive correlation to temperature also suggests that a slight increase in Synechococcaceae and Rhodobacteraceae may occur. However, there is considerably lower terrestrial run-off in the northern regions of the GBR 20 and so the nutrient levels that drive these changes may not be reached. Brodie et al. 20 reported the absence of a cross-shelf gradient in water quality for the far northern GBR, where mean chlorophyll levels were less than half of those measured for the south and central GBR. This suggests that inshore reef microbial communities could be spatially stable at the northern reef sites and perhaps similar to those found during the wet season in the central GBR (i.e., low chlorophyll levels and high temperature). Establishment of fixed monitoring sites dedicated to sampling and characterizing microbial communities at these spatial and temporal scales are required to test these hypotheses.

Final considerations
The indicator taxa proposed probably consist of specialised lineages that have diversified to occupy a particular niche 21 , allowing prediction of the surrounding environment with fairly high confidence. Theoretically, this same indicator approach can be extended to microbial functions, measured either as abundances (or ratios) of particular genes, gene transcripts or even of the proteins they code for 22 . Potentially, many of these indicators will be constituted as indexes or ratios that are able to integrate multiple microbial responses into a unique detection mechanism. It is even suggested that the resolution at which microbiome composition differs among samples may be informative of how phylogenetically conserved the traits under selection are 23 . This means that in the future one may be able to identify microbial indicators not just for prevailing environmental conditions and short-term fluctuations, but also for selective pressures on the reef and shifts therein that may result from loss of ecosystem resilience, for instance. This could facilitate, for instance, the early determination of inshore reefs that are at the turning point between being classified as communities dominated by hard coral communities (In-Porites) and macroalgae communities (In-MA). While other reef metrics provide this information, there is a recognized inability to accurately measure or predict the role of cumulative impacts across different locations spanning communities with varied susceptibility to runoff-related water quality pressures.
Biological indicators have in the past been incorporated into monitoring programs for the GBR as accepted tools to identify the occurrence of environmental stress over space and time. The application of microbial based monitoring and diagnostics is in its infancy, however once established, these approaches will greatly increase our understanding of the biological response of all trophic levels to impacts affecting coral reefs worldwide.