Effects of substratum and depth on benthic harmful dinoflagellate assemblages

Microhabitats influence the distribution and abundance of benthic harmful dinoflagellate (BHAB) species. Currently, much of the information on the relationships between BHABs and microhabitat preferences is based on non-quantitative anecdotal observations, many of which are contradictory. The goal of this study was to better quantify BHAB and microhabitat relationships using a statistically rigorous approach. Between April 2016 to May 2017, a total of 243 artificial substrate samplers were deployed at five locations in the Perhentian Islands, Malaysia while simultaneous photo-quadrat surveys were performed to characterize the benthic substrates present at each sampling site. The screen samplers were retrieved 24 h later and the abundances of five BHAB genera, Gambierdiscus, Ostreopsis, Coolia, Amphidinium, and Prorocentrum were determined. Substrate data were then analyzed using a Bray–Curtis dissimilarity matrix to statistically identify distinct microhabitat types. Although BHABs were associated with a variety of biotic and abiotic substrates, the results of this study demonstrated differing degrees of microhabitat preference. Analysis of the survey results using canonical correspondence analysis explained 70.5% (horizontal first axis) and 21.6% (vertical second axis) of the constrained variation in the distribution of various genera among microhabitat types. Prorocentrum and Coolia appear to have the greatest range being broadly distributed among a wide variety of microhabitats. Amphidinium was always found in low abundances and was widely distributed among microhabitats dominated by hard coral, turf algae, sand and silt, and fleshy algae and reached the highest abundances there. Gambierdiscus and Ostreopsis had more restricted distributions. Gambierdiscus were found preferentially associated with turf algae, hard coral and, to a lesser extent, fleshy macroalgae microhabitats. Ostreopsis, almost always more abundant than Gambierdiscus, preferred the same microhabitats as Gambierdiscus and were found in microbial mats as well. With similar habitat preferences Ostreopsis may serve as an indicator organism for the presence of Gambierdiscus. This study provides insight into how BHAB-specific microhabitat preferences can affect toxicity risks.


Methods
. The Pulau Rawa and Pulau Serenggeh sampling sites are located on uninhabited islands. Both sites encompassed sheltered, shallow reef flats and gradually sloping from 5 to ~ 20 m. Tokong Laut is a relatively deep pinnacle, dominated by sandy/silty substrate and high currents. It was the deepest site sampled ranging from ~ 12 to ~ 25 m. The D'Lagoon and Batu Nisan sites are located at Perhentian Kecil Island. D'Lagoon is a relatively sheltered, low complexity fringing reef while the Batu Nissan site is a relatively exposed, higher complexity fringing reef.
Physical data collection. The depths at each sampling location where screens were deployed were determined using a dive computer. Data sets for seawater temperatures and light intensity were obtained using a HOBO Pendant temperature/light 64 k data logger (Onset Computer Corporation, MA, USA) at 3 m at Pulau Rawa. A logger also was deployed at 10 m at the same site for part of the study to measure how daily temperature varied with depth. Only the maximum daily water temperatures observed at the 3 and 10 m depths were plotted. In addition, average maximum daily temperatures were calculated for the inter-monsoon, southwest monsoon and northeast monsoon seasons and included as part of the temperature time series graph.
Both light sensors deployed at 3 and 10 m malfunctioned so the time series for light was lost. As an alternative means of approximating the light versus depth relationship, functioning light loggers were deployed at 3, 6, 10, 15 and 18 m at Pulau Rawa and the relationship between photon flux density (PFD, µmol photons m −2 s −1 ) versus depth was determined as detailed in Supplementary Data 1. For subsequent analyses, the PFD versus depth relationship was used to estimate approximate light levels at each sampling point. Because of the way this relationship was determined, the PFD measurements should be regarded as an interchangeable proxy for depth as well as an approximate measure of light availability.
Sample collection, processing and microhabitat mapping and classification. An artificial substrate sampling method utilizing fiberglass window screen mesh 63,70 was employed in this study. The screens were deployed by SCUBA, retrieved 24 h later by carefully placing the screen in a wide-mouth 1-l bottle underwater (Supplementary Data 2). In the laboratory, the screens were shaken vigorously for 5-10 s to dislodge the attached cells. Samples were passed through a 200 µm sieve to remove detritus or particles. The filtrates were then filtered onto a 0.2-μm nylon membrane filter. The membrane filter was transferred into a 50-ml tube, filled with 30 ml of filtered seawater, and preserved with 1% acidic Lugol's iodine solution for cell enumeration.
To characterize the benthic dinoflagellate assemblages in relation to the microhabitat variability, the bottom substratum where the screens were deployed, were characterized simultaneously using a photo-quadrat method. This method utilized a waterproof digital camera mounted perpendicularly 1 m above a 0.25 m 2 quadrat. This assemblage was used to photograph the bottom substratum each time a sample was taken (Supplementary Data 2). Digital underwater images were then analyzed for percent coverage of various bottom substrates using CoralNet 73 (https ://coral net.ucsd.edu). The images were annotated with a total of 100 uniform annotation points www.nature.com/scientificreports/ based on general benthic reef community characterizations that were classified into nine benthic substratum types: invertebrates (Invt); coarse rubble and rocks (Rub); soft corals (SC); hard corals (HC); sponges (Spg); turf algal assemblages (Turf); upright fleshy macroalgae (Fles); fine sand and silt (Sd); microbial mats (MM) (Supplementary Data 2). All photo-quadrat images and annotation data are publicly available via CoralNet (https :// coral net.ucsd.edu/sourc e/503/).

Statistical analysis and data visualization.
The data were first analyzed for normality with the Shapiro-Wilk test using PAST 3.25 74 . As the data were not normally distributed, a non-parametric one-way ANOVA on a Kruskal-Wallis rank with a Dunn's multiple comparison test was used to test for significant differences between benthic harmful dinoflagellate assemblages and locality or microhabitat clusters. The distribution of benthic harmful dinoflagellates at each sampling point, in different benthic microhabitats and depths were conceptualized through bubble plots using ggplot2 75 . To evaluate the degree of benthic microhabitat heterogeneity, a cluster analysis with a Bray-Curtis dissimilarity matrix was performed based on the benthic substrate percent coverage; a dendrogram was constructed by vegan 76 in R (R Core Team 77 ). Non-metric multidimensional scaling (nMDS) was used to visualize the correspondence between distinct major clusters of benthic substrates (Supplementary Data 3). One-way analysis of similarity (ANOSIM 78 ) was performed to test significant differences between the benthic microhabitat clusters. SIMPER analysis was used to assess the average percent contribution of microhabitat characteristics towards dissimilarity between clusters formed in nMDS and to identify probable major contributors of the differences detected in ANOSIM (Supplementary Data 3). These analyses objectively identified distinct microhabitat types based on the various substrates present. A heatmap, where different color intensity represented the percent cover of each substrate type at each quadrate sampled over the course of the study was generated using Heatplus 79 . The heatmap was arranged so samples from different sites falling into each of the microhabitat types were plottted together. This convention made it easy to visualize which of the different substrate types (HC, Invt, SC, Spg, MM, Sd, Rub, Fles, or Turf) defined each microhabitat type. Next, the percent contribution of each of the five genera of benthic dinoflagellates to the total assemblage was determined by dividing the number of cells belonging to each genera by the total number of cells contributed by all five genera in a quadrat and multiplying by 100. These data were plotted in the same 5  www.nature.com/scientificreports/ order within habitat type as used in plotting the substrate heatmap. Plotting the samples in the same order for the microhabitat cluster analysis, the heatmap of substrate type and the generic-specific heatmap allowed a direct comparison of habitat types, substrate types and the relative abundance of the different BHAB genera.
Because not all microhabitat types were distributed equally among sampling sites, the proportional distributions of each microhabitat type at each study site were calculated (in percentage) and presented as a stacked bar chart. It is important to note how the different substrate types were distributed with depth to determine the extent to which depth preferences by any genera were due to a factor such as light or temperature versus unequal distribution of substrate types with depth. To accomplish this, the distributions of the nine benthic substratum types as functions of depth were presented as a violin plot using ggplots.
Published data on the relative cell counts from screen sampling devices for Amphidinium, Coolia, Gambierdiscus, Ostreopsis, and Prorocentrum from various field studies were collated for comparative purposes (Supplementary Data 4). The habitat types and sample locations from each study were included. The goal was to determine if consistent patterns in relative abundance among the different genera measured using the screen method emerged when sites from different geographic locations, including those from this study, were compared.
To illustrate the distributions of the genera at the different depths, the abundance of each genus for each quadrat sample were plotted as a function of habitat type on the x-axis and depth on the y-axis. The abundances were indicated by different sized circles and the circles were colour coded to identify the sampling location. Canonical correspondence analysis (CCA) was used to infer the underlying relationship between the benthic harmful dinoflagellate assemblages and benthic substrate characteristics, depths, and irradiances. CCA is a constrained multivariate ordination technique that extracts major gradients among combinations of explanatory variables in a dataset and requires samples to be both random and independent. Data for cell abundances were Hellinger-transformed prior to CCA to ensure the data met the statistical assumptions of normality and linearity. The analysis was performed using vegan. The significance of variation in benthic harmful dinoflagellates assemblages explained by the explanatory variables was tested using an ANOVA-like Monte Carlo permutation test as implemented in vegan.

Results
Sampling frequency. A total of 234 screens were deployed for 24-h periods and collected from various microhabitats and depths between 1 and 25 m at five different locations between April 2016 and May 2017. Sampling dates are indicated by the vertical lines in Fig. 2. The number of sampling sites at each location are provided in Table 1.

Water temperature and light intensity.
Over the course of the study, the maximum Perhentian Islands  Table 1. This microhabitat classification was further supported by the nMDS plot within a stress fac-  (8) MM. Note the benthic microhabitat dominated by microbial mats (microscopic algae such as diatoms and cyanobacteria) formed a distinct microhabitat different from that dominated by turf algae. The contribution of each BHAB genera, as % of total number of cells counted relative to the total number of all BHAB cell counted in the corresponding samples, is shown in Fig. 3C. The eight distinct benthic microhabitats were patchily distributed across the five sampling sites (Fig. 3D). For example, microhabitat SC was only found on Pulau Rawa and Pulau Serenggeh (Fig. 3D). At the other extreme, HC microhabitat was found at each of the sampling sites and Turf microhabitat was found at all sites except Tokong Laut.

Relative abundance and distribution of BHAB genera across habitats in the Perhentian
Islands. Overall, the benthic microhabitats HC and Turf supported the highest abundances of benthic harmful dinoflagellates as compared to other benthic microhabitat types (Fig. 4). Microhabitats Invt, Rub, SC, and MM supported lower BHAB abundances (Fig. 4).
Relative abundances of the five genera of benthic dinoflagellates differed over the microhabitats examined (Figs. 3, 4). Individual habitats were usually dominated by Prorocentrum (46-71% of total cells counted), except in Turf microhabitat (34%) where Ostreopsis was the most abundant group (51%). Consistent with its numerical dominance, the distribution of Prorocentrum (indicated by the percentage of screen samples in which at least one cell was observed in a habitat type divided by total number of screen samples collected in that habitat × 100) was homogenously distributed across microhabitat types being present in 99% of all screen samples counted (Kruskal-Wallis, p = 0.159; Fig. 4). This genus was most abundant in microhabitats HC, Turf, Fles, and Sd, though the absolute maximum Prorocentrum concentration (1.4 × 10 4 cells 100 cm −2 ) was observed in a sample from the Invt microhabitat (Fig. 4).
Gambierdiscus achieved highest abundances in HC and Turf microhabitats (Fig. 4), with the maximum abundance (255 cells 100 cm 2 ) found in Turf microhabitat (Fig. 4). Lower, but still relatively high cell concentrations were found in Fles microhabitat. Habitat specificity of Gambierdiscus is especially clear, as cells were not found in Invt or SC microhabitats and at low frequencies in samples from Rub, Sd, and MM microhabitats (7-29%) (Figs. 3C, 4).
Coolia did not show significant differences in distribution (Kruskal-Wallis, p = 0.176) in their frequency of occurrence among microhabitats (present in > 70% of samples in all habitat types, except the MM microhabitat with 40% occurrence). The highest cell abundance (368 cells 100 cm 2 ) was observed in microhabitat 2 (Rub) with similarly high concentrations in HC, Turf, Fles, and Sd microhabitats (Fig. 4).
Although abundances of Amphidinium in various habitats were low (0-7.3% of total cells counted), the frequency of occurrences in some microhabitats was relatively high (27-81% of samples counted), except in the SC microhabitat where no cells were detected. The highest occurrence of Amphidinium in samples (81% of samples counted) was in the Fles microhabitat. Maximum abundances occurred in HC microhabitat with slightly lower maximum concentrations found in Turf and Sd microhabitats followed by lower concentrations in the Fles microhabitat. Even lower abundances were observed in the Rub, MM and Invt microhabitats (Fig. 4). Table 1. Categories of benthic microhabitats of Perhentian Islands based on the benthic biological and physical substrates, the eight geomorphic zones were clustered based on percentage covers of substrates (see Fig. 3; Supplementary Data 2 and 3).
Results from the literature survey indicated that BHAB assemblages were most often dominated or codominated by Prorocentrum and Ostreopsis (Supplementary Data 4). Gambierdiscus represented a minor portion of the assemblages present except for two samples taken during local blooms 70,81 . Coolia were not sampled as often so their patterns of abundance were not readily assessed. When Coolia concentrations were measured,  www.nature.com/scientificreports/ data indicated they can numerically dominate benthic microalgal assemblages. Amphidinium was sampled less frequently still and were typically present only at low abundances relative to other species.
Canonical correspondence analysis (CCA). The canonical correspondence analysis (CCA) was carried out to assess the degree to which the various BHAB genera were associated with different benthic substrate characteristics, light (as an inverse proxy for depth) and temperature (Fig. 5). The horizontal first axis (CCA1) explains 70.5% (eigenvalue, 0.1088, p = 0.001***) of this constrained variation, and the vertical second axis (CCA2) explains 21.6% (eigenvalue, 0.333, p = 0.001***). Taken together, both axes of the data set explained 92% of total inertia, which is highly significant at p = 0.001*** (Monte Carlo Permutation test, n = 999; F = 8.94), indicating strong correlations between the BHAB abundances, substrate types, light level and seawater temperature (Supplementary Data 5). Turf algae had the greatest influence on CCA1 in a positive direction (F = 15.77, p = 0.001***), while Rub and Sd had the greatest influence in the negative direction (Rub, F = 4.57, p = 0.002**; Sd, F = 1.88, p = 0.112). Light further influenced CCA1 in a positive direction while temperature influenced CCA1 strongly in the negative direction ( Fig. 5; Supplementary Data 5). These factors clearly separated Gambierdiscus-Ostreopsis (CCA1 > 0) and Prorocentrum-Coolia (CCA1 < 0). Along CCA2, which explains less than a quarter of the total variation, the factors having the most influence in the positive direction (CCA2 > 0) were temperature, fleshly macroalgae, turf algae and to a lesser degree light. Factors influencing CCA2 in the negative direction (CCA2 < 0) were HC and Rub. Amphidinium and Gambierdiscus were positively associated with temperature, compared to Prorocentrum and Coolia which were not strongly influenced by it. Ostreopsis was negatively correlated with temperature indicating abundances were higher during the cooler sampling periods.
Prorocentrum and Coolia were negatively correlated with light (surrogate for depth), but only marginally so. This slight negative relationship may have been influenced by these two genera being the only ones present at  www.nature.com/scientificreports/ the deepest depths (Fig. 4). Gambierdiscus, Amphidinum, and Ostreopsis were positively associated with light as compared to Prorocentrum and Coolia, which were negatively weighted, but only slightly. Amphidinium, Gambierdiscus, and Ostreopsis were positvely associated with Turf. Gambierdisucs and Amphidinium were assocated with Fles and Ostreopsis with HC ( Fig. 5; Supplementary Data 5). Prorocentrum and Coolia were associated with Sd and Rub but not strongly. This is consistent with their wide occurrence across the various habitat types (Fig. 4).
BHAB distribution with depth. With respect to depth, the BHAB assemblages were abundant at the depths of 1-10 m, with the average maximum abundances mostly observed at these depths (Fig. 4). For example, the maximum Gambierdiscus abundance was observed in a Turf microhabitat at 4.9 m. Similarly, maximum Ostreopsis abundances occurred in a HC microhabitat at 1.2 m, Coolia in a Rub microhabitat at 12 m, Prorocentrum in a corallimorph (Invt) dominated microhabitat at 7 m, and Amphidinium in a HC microhabitat at 1 m. Prorocentrum and Coolia were ubiquitous, occurring at all depths down to 25 m (Fig. 4). The results also revealed that both these genera were negatively associated with light (Fig. 5). In contrast, Ostreopsis, Gambierdiscus and Amphidinium are more aggregated at the depths of < 10 m. The greatest depth where Ostreopsis, Gambierdiscus, and Amphidinium were found was 16 m (Fig. 4). Interestingly, Prorocentrum and Coolia were the only genera observed at Tokong Laut and then only at depths > 12 m.

Discussion
Effect of benthic microhabitats on the BHAB assemblages. This study focused on expanding our understanding of the role microhabitat types play in controlling the distribution and abundance of benthic harmful algal bloom species (BHABs) in the genera Amphidinium, Coolia, Gambierdiscus, Ostreopsis, and Prorocentrum. Understanding how different microhabitats foster various BHAB genera is critical for understanding their relative contributions to toxin transfer in marine food webs, identifying hot spots or sentinel sites for monitoring and eventually, modelling efforts. Field efforts to examine the relationship between microhabitat and BHABs were hampered previously by lack of ways to objectively define habitat types and a standardized, uniform BHAB cell sampling method. The current investigation used systematic classification of habitat types in photographs from each sampling site in conjunction with Bray-Curtis dissimilarity cluster analysis to define different microhabitat types (Figs. 3, 4). Cell abundances of the BHAB genera were measured using a method that standardized sampling surface areas. This method enabled normalization of benthic dinoflagellate abundances to a known surface area for comparison among sites and studies 1,63,[70][71][72]80,81 despite the heterogeneity and complexity of benthic habitats 63 .
Most previous studies only sampled macrophyte hosts as the target substrate leading to many contradictory data regarding their association with various BHAB genera (as reviewed in Tester et al. 64 ). Other potential hosts such as hard coral colonies, turf algal assemblages, as well as abiotic substrates like rubble, rocks and sand sediment were less frequently sampled. Field collection of macrophyte substrates in BHAB studies may seem convenient because they are easily accessible, however, in this study there was substantial variability of BHAB www.nature.com/scientificreports/ species occurrence and abundances in various benthic microhabitats, including the Fles habitat type. The high coefficient of variation (CV > 1.0; see Table 2, Supplementary Data 6) is indicative of significantly patchy distributions. The application of an artificial substrate sampling technique coupled with benthic photo-quadrat surveys confirmed other types of benthic substratum, besides macrophytes, support high BHAB abundances. Although BHABs occupied most of the microhabitats examined in the Perhentian Islands, Malaysia, our results demonstrated that some BHAB genera exhibited a degree of preference towards specific microhabitats. Prorocentrum and Coolia were widely distributed among each of the habitat types indicating a broad ecological niche (Fig. 4). The CCA showed the two genera tightly clustered, consistent with their having similar microhabitat preferences (Figs. 4, 5). Both genera were most abundant in the hard coral and turf-dominated microhabitats, followed by sand and fine silt, fleshy macroalgae, invertebrates, course rubble and microbial mats (Fig. 4). Regarding occurrence, neither genus exhibited a strong habitat preference.
Ostreopsis showed a strong preference for microhabitats dominated by hard corals and turf algae where they reached their highest abundances but were also abundant in microhabitats dominated by fleshy macroalgae, sand and fine silt, invertebrates, microbial mats and course rubble and rock (Figs. 4, 5). The result of CCA indicated the genus was negatively associated with temperature but not light, suggesting higher abundance samples were collected when ambient temperatures were lower. The results also indicated that of the diverse microhabitats where Ostreopsis occurs, highest densities are most likely to be found in hard coral, consistent with the cell density estimates shown in Fig. 4. Overall, these observations support Ostreopsis species as opportunists capable of colonizing a wide variety of living and non-living benthic substrates 30,62,[82][83][84][85][86] . The high abundances on turf algae versus relatively low abundances observed on course rubble and rock substrates have implications for research in the Mediterranean, especially where coral reefs are absent. There, Ostreopsis are often noted as being strongly associated with hard substrates, particularly rock and manmade structures 30 , other than the dense macroalgal mats or turfs 82,[87][88][89] . This association may be the algal turfs associated with the hard substrates and represent an area of research that could yield important insights into the population dynamics of toxic Ostreopsis species.
Gambierdiscus exhibited the most restricted microhabitat range of the five genera surveyed. They were found predominantly where substrates were dominated by turf algae, hard coral and to a lesser extent fleshy macroalgae (Fig. 4). They were either absent, or present at only low concentrations, in the other microhabitats (Fig. 4). The genus was also positively associated with increasing light and temperature indicating a preference for conditions occurring at shallower depths (Figs. 4,5). Of the eight habitats defined in this study, the CCA indicated turf algae in association with higher temperature and light would represent the microhabitat most likely to support the highest Gambierdiscus cell densities. This conclusion is consistent with other studies showing an affinity of Gambierdiscus for turf algae 63,90 . Several published arguments have been advanced for why turf-dominated microhabitat are preferable. First is, turf algae provide larger surface area for occupancy as compared to fleshy macrophytes 2,49,53,91 . Secondly, structural architecture of turfs with spatial complexity are more likely to create a microhabitat with low micro-scale flow velocity that provides refugia against flow-related disturbances 53,92 . These results also argue against sampling only macroalgae as a means of estimating overall BHAB cell abundances because they are not the preferred microhabitat for BHAB species, particularly Gambierdiscus, the most toxic of the genera.
The habitat preferences of Gambierdiscus most closely resembled those of Ostreopsis. A major distinction was that Ostreopsis, is more broadly distributed among habitats. The highest Gambierdiscus cell concentrations were associated with turf algae microhabitat followed by the hard coral microhabitat, whereas the relative abundances in these two habitats were reversed for Ostreopsis (Fig. 4). The similarity in microhabitat preferences also suggests Ostreopsis, which are more abundant than Gambierdiscus, may serve as a good indicator of where Gambierdiscus are present ( Fig. 4; Supplementary Data 4 64 ).
Amphidinium was widely distributed among habitat types with greatest abundances observed in turf algae, fleshy macroalgae, hard coral, and sand and fine silt microhabitats (Fig. 4). Its abundances were lower in the other habitats and it was absent from all soft coral samples. CCA indicated that Amphidinium species were positively associated with higher temperature and light, likely to reach maximum densities in microhabitats with substrates dominated by algal turfs or fleshy macroalgae. This microhabitat distribution was similar to the broad range of microhabitats preferred by Prorocentrum. Again, whether this utilization of diverse microhabitats was due to only a few species with broad ecological niches, a larger number of species with specialized requirements or a combination of the two is unknown. A molecular survey of different microhabitats using genus specific rDNA primers and high throughput sequencing represents a promising means of addressing this question.
Prorocentrum were the numerically dominant species ( www.nature.com/scientificreports/ at different depths provides an unbiased means of how the various BHAB genera are distributed with depth, including potential differences in light, temperature (Fig. 2), microhabitat type (Fig. 3) and wave action. Results from this study showed maximum abundances of Gambierdiscus, Ostreopsis, and Amphidinium occurred at depths < 10 m (Fig. 4). All three genera were positively associated with light in the CCA consistent with shallower depth distributions (Fig. 5). These observed distributions may be due to the availability of the preferred microhabitats, in this case, the warm-water coral, turf algae and macrophyte-dominated habitats (Fig. 3E) that were distributed preferentially toward the shallower waters. As shown in this study, Pulau Rawa which sheltered and encompassed the highest complexity of microhabitats as compared to other sites (Fig. 3D), hosted the highest abundances of all five BHAB groups among the sites studied ( Table 2). The combination of relatively lower turbulent environments and greater microhabitat availability, particularly those including turf algae and hard corals, probably contributed the higher cell abundances observed. Many ecological studies also report increased species richness and abundance in more complex habitats 93,94 . Conversely, Prorocentrum and Coolia exhibited a broader depth distribution and can be found in deeper waters (to ~ 25 m), with preferred microhabitat types distributed over all depths (Fig. 3E). Prorocentrum and Coolia were also the only two BHAB genera found in the deeper habitats at Tokong Laut, which is a high-energy pinnacle reef affected by stronger underwater currents and dominated by sand and silt (Fig. 3D). Although the effect of physical disturbance on habitat type was not directly studied, species belonging to these BHAB genera can be found in habitats with a moderately high level of turbulence 53,82 . They may benefit from small-scale turbulence in low nutrient habitats, which would increase nutrient diffusion rates and efficiency of cell nutrient uptake 95 . In the suite of species present, some may exhibit higher growth rates than those in other genera (i.e., Gambierdiscus), allowing them to better survive population losses due to turbulent dispersion such as that caused by the relative high current regime found at Tokong Laut. These genera were also proportionately more abundant on the sand and silty substrate predominating at this deeper site indicating they were adapted to utilize this substrate. However, relatively low sampling efforts at this site could have contributed to these findings. The association with the prevailing substrate type found at the deeper Tokong Laut site may also account for the slight negative association of these species with light.
The greater depth range exhibited by these two genera is not likely due to a greater capacity to cope with lower light levels. Dinoflagellates, in general, are low light adapted, most achieving maximal growth rates at 50-100 µmol photons m −2 s −1 compared to surface irradiance, often > 2,500 µmol photons m −2 s −164 . In clear waters, these irradiances can extend to 50 or 100 m. No systematic differences in photosynthetic capacity have been demonstrated among BHAB genera under light intensities less than 100 µmol photons m −2 s −1 . Some species in each genus exhibit positive growth in light levels as low as 10-50 µmol photons m −2 s −154,57,59,60,64 . The greater problem for these species in shallower waters is photoinhibition. Benthic dinoflagellates cope with high light by taking advantage of shading by substrates and physiological methods such as variations in pigmentation 49,95-100 .

Conclusion
The results of this study revealed that substrate variability in the microhabitats across depth-gradients determined the composition and differentially foster the abundance of BHAB species. This study and Yong et al. 63 represent pioneering efforts to numerically evaluate the influence of benthic microhabitat heterogeneity on the abundance and distributions of BHABs. Both efforts provide a robust sampling and statistical analysis to classify the sites where BHABs were sampled based on the various benthic substrate types and allowed comparison of habitat diversity across all BHAB sampling sites. This approach can be used in designing monitoring programs at sentinel sites and provides insight into site specific differences in BHAB abundances and potential for toxins to enter marine food webs. It also seems clear that disturbances of the bottom substrates, such as coral reef degradation will markedly influence the BHAB assemblages 67-69 .

Data availability
All data generated during this study are included in this published article and its supplementary information files. The primary and secondary datasets are also available via figshare (https ://figsh are.com/proje cts/Effec ts_of_benth ic_subst ratum _chara cteri stics _and_depth _on_benth ic_harmf ul_dinofl agel late_assem blage s/81026 ).