Light and temperature drive the distribution of mesophotic benthic communities in the Central Indian Ocean

Research on mesophotic coral ecosystems (MCEs) has increased exponentially in recent decades, and the significance of this ecosystem has been recognised both in terms of biodiversity and distribution. However, this research has mostly focussed on corals and is globally sporadic, with the Indian Ocean remaining largely unexplored and overall MCEs under protected. Hence, baseline data on MCE benthic communities is lacking, but nonetheless essential for developing adequate management strategies. Here, we assess the variation in diversity and community structure of MCEs along the depth gradient and among sites in the Indian Ocean and the environmental parameters that are potentially driving these differences.


| INTRODUC TI ON
Coral reefs are one of the most iconic and diverse ecosystems in the world.Nevertheless, they are threatened by current anthropogenic activities, including direct human pollution, and human-induced impacts via climate change (Bryant et al., 1998).Slightly deeper than shallow-water coral reefs, reside other highly diverse ecosystems, called Mesophotic Coral Ecosystems (MCEs), broadly found between 30-40 m and 150 m (Hinderstein et al., 2010).MCEs have received increasing attention in recent years, but extensive research on MCEs is still lacking, compared to shallow-water coral reefs (Pyle & Copus, 2019).The paucity of research is mostly due to logistical difficulties and safety issues in accessing MCEs and the expense of using deep-water technologies in these relatively shallower environments (Baker et al., 2016).MCEs are distributed in tropical and subtropical regions and may represent ~80% of all potential coral reef habitat only based on depth ranges (Pyle & Copus, 2019); providing many of the same ecosystem services found in shallow coral reefs but also unique ecosystem services (Baker et al., 2016).Thus, understanding the diversity, distribution and function of these ecosystems is critical.
Research on MCEs has increased exponentially over the past two decades, due in part to enhanced technological capacities, and the formulation of the Deep Reef Refugia Hypothesis (DRRH) (Glynn, 1996), which was further expanded by Hughes and Tanner (2000).The DRRH postulates that deeper reefs could provide a spatial refuge for shallow-water species and serve as a source of larvae for the recovery of shallow-water coral reefs (Glynn, 1996;Hughes & Tanner, 2000).However, it rests on two assumptions: firstly, there is a high overlap in species composition and connectivity throughout the depth range, and secondly, deep reefs are less susceptible to anthropogenic and natural stressors than their shallow-water counterparts (Rocha et al., 2018).While some studies have observed similar scleractinian coral genera throughout the depth gradient displaying less bleaching susceptibility at depth or being less affected by surface disturbances (Pérez-Rosales et al., 2021), there is growing evidence that the DRRH may not hold globally (Godefroid et al., 2022).Indeed, several studies observed distinct communities between shallow and deeper habitats, for both benthic (Hoarau et al., 2021;Lesser et al., 2019;Stefanoudis et al., 2019Stefanoudis et al., , 2022) ) and pelagic (Rocha et al., 2018;Stefanoudis et al., 2022) organisms.In addition, some of the few species with distributions that span from shallow down to mesophotic depths demonstrated lower reproductivity (Shlesinger et al., 2018), slower growth rates (Groves et al., 2018), or were not genetically connected between depth bands (Bongaerts et al., 2017).Other studies have demonstrated that mesophotic reefs are threatened by both natural and anthropogenic sources (Rocha et al., 2018;Smith et al., 2019).
Most of the recent MCE research has focussed on scleractinian corals and their potential role in the DRRH, or qualitative species description along the depth gradient (Pérez-Rosales et al., 2022).While a substantial number of studies have investigated the community structure in deep-sea ecosystems (Bridges et al., 2021;Goode et al., 2021;Howell et al., 2010), only a handful of studies have investigated MCEs at the community level (Lesser et al., 2018;Stefanoudis et al., 2019;Swanborn et al., 2022).In addition, these studies on MCEs are globally sporadic with high discrepancies between ocean regions, with the Indian Ocean considered as highly underexplored and MCEs in general under protected (Pyle & Copus, 2019;Stefanoudis et al., 2023).However, baseline information on MCE community structure and their environmental drivers is missing but is considered essential for developing adequate management strategies specifically for MCEs and to determine their potential role in supporting shallow coral reef systems (Stefanoudis et al., 2019).
In this study, benthic communities of coral reefs from 15 m down to 160 m were investigated using a Remotely Operated Vehicle (ROV) at two sites on one of the five atolls of the Chagos Archipelago, to answer the following research questions: (1) What are the differences in MCE community composition along depth gradient and among sites in the central Indian Ocean, (2) Which benthic species are driving these differences and (3) Which environmental parameters are driving these vertical (throughout the depth gradient) and horizontal (geographically, between the study sites) differences.

| Study area
This study focusses on MCEs of the Chagos Archipelago (6°S, 71°30′E) located in the central Indian Ocean.The Chagos Archipelago can be considered as a hotspot of marine life in the middle of the Indian Ocean, due to its remote location and the creation of a fully no-take marine protected area in 2010, experiencing minimal direct human influence (Hays et al., 2020).Shallow-water reefs have been extensively investigated since the 1980s (Hays et al., 2020), but very little is known about MCEs in the Archipelago, with only a few studies that have reached a maximum depth of 60 m (Andradi-Brown et al., 2019;Sheppard, 1977).This study was located at Egmont Atoll (6°40′S, 71°21′E; Figure 1).Two sites were investigated; Ile Des Rats (IDR), located on the North-West coast and Manta Alley (MA), found along the North-East coast (Figure 1).These two sites were chosen for their contrasting oceanographic regimes, where IDR is open to the wider ocean, benthic community structure, Chagos Archipelago, chlorophyll a, community break, environmental driver, low species overlap, Mesophotic Coral Ecosystem, PAR while MA is close to Great Chagos Bank (Figure 1).All data were collected in November 2019 or March 2020.

| Biological data
A Saab Seaeye Falcon remotely operated vehicle (ROV) was used to collect benthic imagery of MCEs at both sites.It was equipped with four LED daylight lamps (3520 Lumens), with two mounted on a tilt platform either side of a standard definition (SD) camera (720p) with a wide angled lens (91° in water).An information overlay displayed time, depth, heading, pitch and roll.A GoPro Hero 4 high-definition camera (2.7 k, 24 fps, wide field of view) was positioned directly under the SD camera.
Due to the presence of a strong current around Egmont Atoll, it was difficult to maintain a constant heading at an altitude close to the seabed to undertake linear benthic video transects.Thus, still images were sampled along transects.To collect image samples, the ROV descended to the appropriate depth zone, and approached the seabed with the cameras positioned at an oblique angle (altitude <1.5 m).Still frames were captured to collect high-quality images allowing benthic specimen identification and maximise field of view of the seabed.
For each image, the time and depth were recorded in addition to latitude and longitude of the sampling position, monitored by a mounted acoustic ultra-short baseline (USBL) transponder system.The depth gradients were extracted from GEBCO data and define the benthos delimitations.Egmont Atoll is displayed indicating 3D bathymetry data acquired using a multibeam and the two sites studied: Ile Des Rats (North-west coast) and Manta Alley (North-east coast).This map was adapted from Wikipedia, where the original can be found here: https://en.wikipedia.org/wiki/Chagos_Archipelag o#/media/ File:Chagos_map.PNG.
Still images collected along each transect were annotated using BIIGLE 2.0, an online annotation platform (Langenkämper et al., 2017), which enables identification and quantification of organisms.In this study, all benthic organisms (≥1 cm) within each image were identified and quantified, with encrusting algae estimated as percentage cover.As the standard taxonomic approaches are not often applicable to image data, all organisms were identified to the highest taxonomic resolution possible and distinct morphospecies were assigned an Operational Taxonomic Unit (OTU) following the method in Howell et al. (2010).A morphospecies may correspond to species, genus, family, or higher taxonomic level.All specimens were identified using both relevant literature and expert taxonomist knowledge (see Table S1).Alongside the image analysis, a morphospecies catalogue was created for the region (Diaz, Foster, & Howell, 2023) following the global standardised marine taxon reference image database in Howell et al. (2019).The first version of this catalogue is available in Diaz, Foster, & Howell (2023) and can be consulted in the near future at https://smart ar-id.app/(Howell et al., 2019).Quality assessment of the image annotation was undertaken by the same observer, who re-annotated 5% of the total number of images, as suggested by Schoening et al. (2016).Data from the two annotation sessions were assessed for accuracy and reproducibility of annotations: images were compared side by side (original and re-annotated images) with a similarity of percentage (SIMPER) analysis, and visualised with non-metric multidimensional scaling (nMDS) using PRIMER v.6 (based on square root transformed data and Bray-Curtis similarity).The SIMPER analysis revealed a similarity of 76.9-90.1% between the two images within transects, considered as an acceptable consistency in the annotation process.The nMDS plot showed an overlap between the original and re-annotated images (see Figure S1).Furthermore, annotations were verified by a second observer.
The ROV was not equipped with paired lasers during the benthic surveys, thus, calculation of image area was not possible.Instead, the pilots positioned the ROV at a similar altitude from the seabed, with the same angle of view, for each collected image, to ensure where possible that a similar field of view was captured in each image.The mean time taken to complete a single transect was 30 min 3 s (±6 min 58 s), with no significant differences between the transect timings with depth and site (Kruskal-Wallis test, p-value >.05.Transect by depth: chi-squared = 10.244,df = 5, p-value = .06862;transect by site: chi-squared = 0.0022733, df = 1, p-value = .962),indicating that the ROV maintained a consistent speed throughout the study.Thus, it can be assumed that a comparable area of seabed was surveyed within each transect and depth zone.Substratum categories were selected using similar categories as detailed in Stefanoudis et al. (2019Stefanoudis et al. ( , 2022)): boulders (rocks ≥256 mm), bedrock/hard bottom, coral rubble (64 to 256 mm), coral gravel (2 to 64 mm), sand (0.06252 mm), sediment veneer and epibenthic mat/turf.Live coral and live sponge categories were categorised as bedrock, to avoid double counting, as bedrock was the substratum where corals and sponges were attached to Goode et al. (2021).
Categorising live coral or sponge within the substratum categories would be, in effect, double counting, as these individuals had already been quantified within the image annotation.Examples of each category can be viewed in Figure S2.
To determine substratum type per image, the method from Dumas et al. (2009) was followed, adapted for the ecosystem investigated here.Nine points/m 2 ratio was proven reliable to describe shallow reefs and was a trade-off between capturing the finer changes in substratum composition and time efficiency.Our study is focussed on deeper reefs, which have larger-scale changes in substratum than the shallow-water reefs, thus, 20 random points were assigned on each image to identify substratum cover, in accordance with work by Stefanoudis et al. (2022) on MCEs in the Western Indian Ocean.In November 2019, the Indian Ocean was subjected to an unusual thermocline deepening, due to an extreme Indian Ocean Dipole event, leading to high temperatures at depth (Diaz, Foster, Attrill, et al., 2023) (Kirk, 2011).This was done to derive metrics including the 10% light optical depth, considered to be the midpoint of the euphotic zone and the 1% light level, where the euphotic zone is considered to end, as photosynthesis equals respiration, also named the compensation point (Kirk, 2011;Lesser et al., 2018).z1% was initially considered to be the lower boundary of MCEs, but more recently, the maximum depth records of lightdependent corals have extended beyond the 1% light level (Kahng et al., 2010;Pyle & Copus, 2019).The presence of light-dependent corals at these depths suggests that photosynthesis is occurring at levels of 0.1% of surface light, thus, z0.1% could actually be the lower boundary of MCEs (Laverick et al., 2020).Hence, the 0.1% light level was also calculated in this study.

| Environmental data
These metrics were calculated for each of the 22 PAR profiles (upward and downward profiles of the 11 PAR deployments; see Table S2).PAR is highly dependent on location and cloud cover.
Here, the 11 deployments were taken with the same method, at the same time of the day but on different days and at different locations, revealing PAR variations between the deployments.An average of the 22 profiles was considered in the environmental variables for subsequent analyses, based upon the 22 regression equations (see Table S2).PAR average values extracted from the 22 profiles are displayed in Table 1.Attenuation coefficients for PAR (K 0 for the surface & K D(PAR) for a specific depth) are adequate to describe general optical characteristics of the water column (Lesser et al., 2018) and K 0 was used to extrapolate irradiance over depth for each profile.
In clear and homogenous waters with little scattering and a diffuse light environment, the relationship between the vertical attenuation coefficient K 0 and the downwelling irradiance K D(PAR) is nearly one (Kirk, 2011), thus allowing for direct comparison between K 0 and K D(PAR) .The midday vertical profiles showed overall little stratification; however, there were slight differences between profiles, with R 2 ranging from 0.8646 to 0.9883.K 0 was dependent on the date, site and boat shadow of the vertical profiles and ranged from −0.035 to −0.073 m −1 (see Table S2).The midpoint of the euphotic zone (10% of the surface light) is estimated as 2.3/K D(PAR) , the bottom of the euphotic zone (1% of light left) is estimated as 4.6/K D(PAR) and the lower limit of MCEs (z0.1%) is estimated as 6.8/K D(PAR) (Kirk, 2011;Laverick et al., 2020;Lesser et al., 2018;Pyle et al., 2016).the effects of other variables (e.g., temperature and salinity) (Evans et al., 2015).Every environmental variable value mentioned above can be viewed in Table 1 and Table S3.

| Statistical analyses
To assess the variation in community structure of MCEs over the depth gradient and the environmental drivers of their distribution, data were analysed using multivariate routines in the statistical software package PRIMER v.6 (Clarke & Gorley, 2006) with PER-MANOVA+ add-in (Anderson et al., 2008).Morphospecies and associated abundance data were exported from BIIGLE software and imported into PRIMER.Data were square-root transformed, and a resemblance matrix was constructed using Bray-Curtis similarity.Hierarchical clustering with similarity profile (SIMPROF) permutational tests were carried out to determine statistically significant clusters.
A SIMPER test was undertaken using a cumulative cut-off of 70% to identify taxa contributing to differences between the communities previously identified using the SIMPROF test.Taxa that contributed the most to the identified communities were visualised with a ribbon plot created in R software (R Core Team, 2020) (i.e., cumulated contribution of the taxa to the within-group similarity up to 50% of the SIMPER clusters or an individual taxon contribution of a minimum of 1% in a SIMPER cluster).
Environmental variables (latitude, longitude, depth, temperature, salinity, rugosity, slope, BBPI, FBPI, DO 2 , PAR, Chla, substratum categories) were visualised using draftsman's plots, normalised and a resemblance matrix was created using Euclidian distance.A distance-based linear model (DistLM routine in PERMANOVA+) was run in Primer v.6 with all the environmental variables to identify the significant predictors of MCE community structure.A Pearson's correlation was performed to identify variables deemed to be highly correlated (Pearson's r > .90;)(Clarke et al., 2014) and therefore one of each correlated pair was removed prior to further analysis, based on DistLM marginal tests and their ecological relevance.A final DistLM was run using the chosen variables with stepwise selection and Akaike information criterion (AIC) with 9999 permutations.The DistLM results were visualised using distance-based redundancy analysis (dbRDA).In addition, a DistLM with original temperature data collected in November 2019 and the same environmental variables selection was performed to visualise the differences between the two models.Results can be viewed in Tables S4 and S5 and Fifteen statistically significant biological assemblages were identified using a SIMPROF routine (from "a" to "o", Figure 2).At around 20% similarity, the benthic communities were firstly separated by depth and then by site, with wider separations between shallow to upper (15-40 m) and mid to lower mesophotic zones (60-150 m).
Within the deeper bands, the most distinct group was the deepest band (150-160 m), which splits from the other depth zones at around 30% of similarity (Figure 2).Then, the 110-120 m depth band separates out at about 35% similarity, leaving the mid zone from 60 to 90 m grouped together down to 50% similarity.Four communities (c, f, h, l) contained only one transect and were significantly different from the other transects within the same site, at the same depth.
These differences can be largely attributed to the differing proportions of substratum categories between each community.Indeed, outliers had noticeably different proportions of soft (sand, coral rubble and gravel to a lesser extent) and hard substrates (boulders, epibenthic mat, bedrock, sediment veneer) for the same site and depth (Table 1), which is illustrated by differences in morphospecies numbers and abundances (Table 2).The SIMPER analysis revealed that within-community similarity ranged from 62.0 to 76.0% (Table 2).
For the same depth band, we suggest that differences between IDR and MA were mostly due to the frequency of occurrence of morphospecies within a transect, with overall individual morphospecies contribution to within-community structure increasing with depth (Table 2).
A ribbon plot was used to show the distribution and proportional abundances of the morphospecies that contributed most to the differences in biological assemblages, based on SIMPROF/SIMPER results, over the MCE depth range (Figure 3).Communities between 15 and 40 m were largely characterised by scleractinians, shallowspecialist sponges and algae, followed by an increase in octocorals, antipatharians and hydrozoans with depth.Between 60 and 90 m, two zooxanthellate hard corals from the Leptoseris genus were found in abundance (OTU340 and OTU416), with depth-specialist octocorals, hydrozoans and sponges also observed, but in lower abundance (Figure 3).Communities between 110 and 120 m differed greatly compared to other assemblages, with a particular abundance of an octocoral belonging to the Nicella genus (OTU17) and a hydrozoan from the Stylasteridae family (OTU19) (Figure 3, Table 1).At 150-160 m, azooxanthellate corals (OTU42, Tubastraea sp. and OTU40 from the Caryophyllidae family) were observed but depth-generalist sponges and algae characterised the assemblage (Figure 3).In general, cnidarians (comprising scleractinians, octocorals, hydrozoans, and antipatharians) were depth-specialists, being particularly abundant within a specific depth band, while only a few morphospecies extended their distribution across more than one depth band (e.g., Leptoseris sp., Annella sp., Sarcophyton sp., Ellisellidae, among others; Figure 3).Zooxanthellate corals that were observed from shallow depths down to the mesophotic zone mostly belonged to the Leptoseris genus, with a few Pachyseris sp. and Oxypora sp.individuals (Figure 3).A distribution pattern is also observable within sponges, with several depth specialists observed but a relatively high proportion of them being depth-generalists; although the results must be taken with caution as sponge visual identification and distinction from imagery is complicated, particularly at deeper depths.
After undertaking an initial DistLM analysis with all environmental variables (see Table S6), a Pearson's correlation matrix with the 21 environmental variables revealed 4 correlation Cluster dendrogram from the Similarity Profile Analysis (SIMPROF) routine on the assemblage composition of each transect of the two study sites.Ile Des Rats (IDR) and Manta Alley (MA) revealed 15 statistically significant biological assemblages, with SIMPROF groups labelled from "a" to "o" (refer to Table 1 for cluster descriptions).
Distance-based redundancy analysis (dbRDA) plots allow visualisation of the DistLM results in two dimensions, including each significant environmental variable (Figure 4).The first two axes of the dbRDA plot explained 21.9% of the total variation in community structure (Figure 4).PAR was the main driver of the variation in benthic community structure across the depth bands and accounted individually for more than 13% (see Table S8).Decreasing

F I G U R E 3
Morphospecies distribution and abundance with depth by phyla of the most contributing taxa, from SIMPER analysis.Each ribbon shows the proportion of distribution over the depth gradient, in percentage of presence (%).The OTU label number is indicated in front of each morphospecies name.Each colour indicates a phylum to which the morphospecies belongs to.An example image for each taxon can be consulted in Diaz, Foster, & Howell (2023).
decrease can be mainly observed from 60 m, which represents the depth of the thermocline (Figure 4; Table 1).In addition, Chla  S8), while significant in the model, accounted for <4% of the total variation.

| DISCUSS ION
This study represents the first extensive survey of mesophotic coral ecosystems along the entire depth gradient in the Chagos Archipelago.Distinct benthic communities were recorded, from 15 to 160 m (Figure 5), similar to those reported for the benthos in the Western Indian Ocean (Stefanoudis et al., 2022).Broadly, the ben- Of the 582 benthic morphospecies recorded during this study, 110 morphospecies were identified as the main components of the benthic zonation (Figure 3, Table 1).The majority of photosynthetic scleractinian corals were observed shallower than 40 m, along with shallow-specialist octocorals and sponges (Figure 3).The first scientific expedition to the Archipelago in 1905 reported scleractinian corals to a maximum of 90 m via dredging surveys; with octocorals, antipatharians, sponges and a few solitary corals observed at an undisclosed depth, deeper than 90 m (Gardiner, 1907).Between 1972 and 1979, the Joint Expeditions established a more detailed inventory of MCEs down to 60 m, with similar species to those observed in the current study dominating between 33 and 45 m, such as Leptoseris sp. and Pachyseris sp.(Baldwin, 1975;Griffiths, 1981;Sheppard, 1981).Later in 1996, octocorals were recorded down to 50 m, including Annella sp. and Ellisella sp.(Reinicke & Van Ofwegen, 1999), which were also observed in the present study (Figure 3).However, aside from collections of Antipatharia in 1909, this class remains poorly studied in the Archipelago preventing us from making any comparison with our results (Andradi-Brown et al., 2019).
The present study demonstrated distinct benthic communities over the depth gradient, further emphasised by a change in taxa composition (Figures 2 and 3).As mentioned previously, scleractinian corals were dominant above 40 m, with sponges observed mostly from 40 m down to the deepest band of our survey, at 160 m.
Between 60 m and 120 m, octocorals, hydrozoans and to a lesser extent antipatharians, were the dominant taxa.Depth-generalist morphospecies were also present, comprised mainly of algae and sponges.While the great majority of MCE research has focussed on scleractinian corals, a few studies have reported similar changes in the abundance of other taxa with depth.For example, octocorals are more abundant at depths >30 m in the Red Sea compared to shallow reefs (Shoham & Benayahu, 2017); >60 m on the Great Barrier Reef (Bridge et al., 2012) and >70 m, along with antipatharians, around Sesoko Island (Sinniger et al., 2022).In the Au'au Channel (Hawaii), the number of antipatharians and sponges increased from 120 m (Kahng & Kelley, 2007).An increase in sponge cover and biomass along the depth gradient has been previously described in both the Caribbean Sea and the Pacific Ocean (Lesser & Slattery, 2011;Slattery & Lesser, 2019).In addition, two recent studies in Bermuda and Seychelles investigated every benthic taxon encountered along the depth gradient (15-300 m and 9-351 m, respectively), revealing similar benthic taxa distribution with depth compared to the present study (Stefanoudis et al., 2019(Stefanoudis et al., , 2022, respectively), respectively).Indeed, they observed limited species overlap between the depth bands, driven by differences in benthic assemblages: the great majority of scleractinian corals were observed shallower than 30 m, in both studies; antipatharians were not observed above 60 m in Seychelles and 90 m in Bermuda and octocorals were mainly observed from 60 m in Seychelles and with numerous depth-specialists in Bermuda (Stefanoudis et al., 2019(Stefanoudis et al., , 2022)).Only a few species had a wide depth span in Bermuda, which were mainly algae and sponges, and in water above 150 m (Stefanoudis et al., 2019).In addition, a study in Sesoko Island observed scleractinian corals predominantly shallower than 50 m, with a significant decrease in their abundance at deeper depths (Sinniger et al., 2022).The present study provides similar evidence of limited taxa overlap along the depth gradient, revealing the biological uniqueness of MCEs through the presence of statistically different benthic communities with depth.
Our study demonstrates that PAR, temperature and Chlorophyll  S8).Irradiance has been observed worldwide to play a significant role in structuring mesophotic benthic communities with depth (Laverick et al., 2020;Lesser et al., 2019).MCEs are defined as light-dependent ecosystems with several benthic species shaping the mesophotic communities that rely on photosynthesis as a major food source (Hinderstein et al., 2010).In addition, numerous studies have suggested MCE boundaries be defined based on local water clarity (e.g., light) in combination with community data, instead of using a fixed depth of 30-150 m (Baker et al., 2016;Bridge et al., 2012;Kahng et al., 2010;Laverick et al., 2020;Pyle et al., 2016).Indeed, PAR has been observed to play an important role in structuring MCEs, with "community breaks" (defined as distinct change in the composition of multiple species in the same geographical location; Lesser et al., 2019) dependent on water clarity rather than depth (Kahng et al., 2010).
The 10% and 1% surface PAR values (considered respectively as the midpoint of the euphotic zone and endpoint of the euphotic zone, where photosynthesis equals respiration) correspond to the SIMPROF "community breaks" (Figure 2).Ten percent of the surface PAR was recorded at 43 m, where a split was observed at approximately 25% similarity between the shallow-upper mesophotic zone (15-40 m) and the deeper bands, suggesting 40 m as the upper limit of MCEs in the Chagos Archipelago.One percent surface PAR, previously considered as the lower limit of the mesophotic zone (Lesser et al., 2019), was measured at 86 m (see Table S2), within the 80-90 m depth band.In the SIMPROF analysis, the 80-90 m depth band displayed a clear split from the deeper zones, with <40% and <30% similarity with the 110-120 m and 150-160 m depth bands, respectively (Figure 2).While these irradiance values are in accordance with those measured in the Au'Au Channel (Hawai'i) (Kirk, 2011;Pyle et al., 2016), and in the Ryukyu Archipelago (Japan) (Sinniger & Harii, 2018), they can vary substantially geographically (Kahng et al., 2010;Laverick et al., 2020;Lesser et al., 2018;Rouzé et al., 2021).Recently, Laverick et al. (2020) suggested that 0.1% surface PAR be considered as the lower limit of the mesophotic zone, instead of the compensation point at 1%, because of the maximum depth records of light-dependent benthic organisms in some regions.
Here, 0.1% surface PAR was observed at 127 m (see Table S2), where no photosynthetic corals were observed (Figure 3).In addition to PAR, other environmental variables played a significant role in structuring the MCE communities, with temperature being the second most influential factor (explaining ~8% of the variation in community structure; Figure 4, see Table S8).Temperature decreased with depth, in the same manner as irradiance.Temperature is an important ecological factor as it regulates an organisms' metabolism (Bridges et al., 2021).Here, temperature was highly correlated to depth, which is also frequently assessed as a key driver of community structure in MCEs (Pyle et al., 2016;Stefanoudis et al., 2023), as well as in deep-sea ecosystems (Bridges et al., 2021;Goode et al., 2021).Temperature reflects the change in numerous physical factors, including internal waves that can induce a displacement of the thermocline and/or nutrients at large amplitudes (Baker et al., 2016;Hinderstein et al., 2010;Hosegood et al., 2019;Kahng et al., 2010).In addition, temperature has been demonstrated to impact coral reefs in general, by affecting oxygen production for instance, which can in turn be limiting in reef environments (Nelson & Altieri, 2019).In particular, the variability in seasonal temperature with depth observed in the Chagos Archipelago was seen to greatly impact mesophotic scleractinian corals, further potentially influencing the overall structure of benthic mesophotic communities (Diaz, Foster, Attrill, et al., 2023).
The third contributing factor, while only explaining 1.6% of the community variation, was Chlorophyll a, a proxy for food availability and thermocline (see Table S8).In particular, a deep chlorophyll maximum (DCM), that requires stratified conditions, can create two distinct layers in terms of environmental conditions, with nutrientlimited above and light-limited below (Cornec et al., 2021), likely impacting communities residing above and below this layer.In this study, a peak in Chlorophyll a was observed at approximately 60 m (Table 1; see Figure S3A), a widespread and common feature in global ocean regions (Lesser et al., 2009).This DCM impact is suggested in the benthic "community break" (Figure 2), as well as in the clear change of benthic taxa observed at 60 m (Figures 3 and 5).
Other environmental drivers were also significant in shaping mesophotic benthic community composition, but to a limited extent, contributing from 1.4% to only 0.2% of the total variation in the benthic community structure.These include salinity, substratum type (coral gravel, sand, bedrock, sediment veneer and epibenthic mat), benthic terrain variables (BBPI, FBPI), latitude, and longitude (see Tables S6 and S8).Nevertheless, these factors have been shown to play a role in structuring benthic communities worldwide.
Indeed, hard substrates support more biodiverse reef communities by providing settlement areas in deep (Goode et al., 2021), mesophotic (Stefanoudis et al., 2019;Swanborn et al., 2022) and shallow (Francini-Filho et al., 2013) benthic communities.In addition, slope and geomorphology in general direct the transport of sediment, suspended particles, and dissolved nutrients, which controls the distribution of MCEs on a local scale (Locker et al., 2010).These factors can influence mesophotic reef distribution and productivity, including photosynthetic scleractinian corals, which depend on light to grow; or heterotrophic organisms, relying on suspended particulate matter (Swanborn et al., 2022).Longitude and latitude act as proxies for other variables, including correlated factors such as temperature, light, reef/non-reef habitats and boundary currents (Bridges et al., 2021).
Finally, the best DistLM regression only accounted for ~29% of the total variation in benthic MCE community structure at Egmont Atoll.This value was similar, higher or lower to previous mesophotic (Asher et al., 2017;Ponti et al., 2018) and seamount studies (Bridges et al., 2021;Goode et al., 2021), suggesting that MCE community assembly processes are complex and difficult to model (Goode et al., 2021).In this study, oceanographic parameters were not captured, or were captured as proxies.These include oceanographic parameters, such as internal wave activity, which has been observed to vary substantially with depth and site (Pers.Obs.).Internal waves and in-water currents are already known to affect benthic communities worldwide (De Clippele et al., 2019;Leichter et al., 2008) and to influence pelagic animals in the Chagos Archipelago (Harris et al., 2021).Other oceanographic parameters may also influence the benthic community structure, such as water chemistry, or larger processes such as upwelling and the monsoon regime, particularly influencing coral reefs in the Indian Ocean (Gischler et al., 2014).Biological parameters such as competition, predation and herbivory are also known to drive benthic community distributions (Kahng et al., 2010).Hence, incorporation of biotic and oceanographic processes in future studies could help resolve the drivers of mesophotic benthic community structure (Swanborn et al., 2022).

| CON CLUS ION
The entire MCE depth gradient was surveyed, with six depth bands, chosen a priori to the study and based on the literature, covering the shallow reef and the lower mesophotic zones: 15-20 m; 30-40 m; 60-70 m; 80-90 m; 110-120 m; 150-160 m.Three transects were surveyed per depth band parallel to the slope at each of the two study sites.Ninety images were collected per depth with a total of 1080 images from the two sites.All survey transects were collected in November 2019 (IDR: transects 1, 2, 3; MA: transect 1) or March 2020 (MA: transects 2 and 3).

F
Study sites and their locations.Composite figure showing Egmont Atoll (right corner inset) located within the Chagos Archipelago (main map), central Indian Ocean (left corner inset), showing the Marine Protected Area delimitation in orange.

A
calibrated Valeport Modus CTD logger (Conductivity-Temperature-Depth, serial number 31,027) was mounted on the ROV to measure in situ conductivity (from which salinity can be derived), temperature and depth while conducting the video transects in November 2019 (IDR transects 1, 2, 3 and MA transect 1) and March 2020 (MA: transects 1, 2).

.
Anomalous temperature values recorded in November at IDR and MA were not representative of the temperature observed throughout the year.Hence, temperature values surveyed in MA in March 2020, when the thermocline shoaled back to its typical depth (~60 m), replaced temperature values of November 2019 in the model.Chlorophyll a (Chla, in μg/L) and dissolved oxygen (DO 2 , in %) profiles were extracted from a Seapoint Chlorophyll Fluorometer and JFE Advantech RINKO III, respectively, sampling at 12 Hz and deployed off the Atoll on the 02/12/2019 at 15:50 local time, for transects undertaken in November 2019.In March 2020, measurements of Chla and DO 2 were taken from an Argo drifting buoy close to the Archipelago (5°6′28.8″S,81°59′9.6″E) on the 14/03/2020 only and from a combination of a RBR maestro CTD profile deployed on the western side of the Atoll (at Ile Lubine), down to 76 m on the 13/03/2020 and the Argo buoy for deeper measurements (Argo, 2000).Comparisons of Chla and DO 2 profiles from the profiling CTD and Argo data can be viewed in Figure S3.Vertical profiles of photosynthetically active radiation (PAR, also called irradiance and calculated in μmol photons/m 2 /s, between 400 and 700 nm) were conducted using a LI-COR LI-193 spherical underwater quantum sensor.The sensor was attached to a metal lowering frame along with a Star-Oddi Starmon Temperature/Depth recorder and lowered over the side of the vessel to a maximum depth of 180 m.Eleven midday (12:10-13:15) vertical deployments were taken in the Archipelago from the 12th to the 17th of March 2020 and from the 19th to the 29th of March 2022, with a total of 22 profiles recorded (upward and downward profiles of the 11 PAR deployments).The attenuation coefficient of downwelling PAR (K D(PAR) ) was calculated according to the Lambert-Beer Law (I z = I 0 e −Kdz ; where z represents depth, I z the intensity of irradiance at depth z and I 0 the intensity of irradiance just beneath the surface) Topography with bathymetry and derived variables were extracted from multibeam echosounder data collected in November 2019 and generated in ArcMap 10.7 with the Benthic Terrain Modeler extension (Walbridge et al., 2018): broad-scale bathymetric position index (BBPI) with an inner radius of 5 and outer radius of 250 (from a cell size of 9 m), fine-scale bathymetric position index (FBPI, inner radius of 3 and outer radius of 25, cell size of 9 m), slope and ruggedness (neighbourhood size of 3).BPI gives the relative elevation of a point in relation to the overall landscape (Lundblad et al., 2006): positive values indicate features rising above the surrounding terrain, while negative values indicate depressions.In contrast, areas with constant slope or flat areas are represented by near-zero values.BPI acts as a surrogate for various environmental factors that affect species distribution, such as light or current exposure, current speed and sedimentation, without confounding Figure S4.
PAR drives the separation between shallow-water communities (15-20 m), upper mesophotic communities (30-40 m) and deeper mesophotic communities (60-160 m; Figure 4).Secondly, temperature accounted for ~8% of the variation in community structure, with temperature values decreasing with increasing depth.This TA B L E 2 Overview of taxon composition of individual SIMPROF clusters.Pavona sp.(6.13); 377-Atriolum (4.51); 58-Corallinales (4.27); 379-Porites sp.(3.95) was the third significant environmental driver and accounted for almost 2% of the variation in community structure, with the highest values of Chla observed at around 60-90 m.Salinity was the next significant variable, explaining 1.4% of the variation, and with increasing values as the temperature decreased.Coral gravel represented slightly more than 1% of the variation in benthic assemblage.Finally, the remaining eight variables; including substratum categories, latitude and longitude, and the geomorphological variables; see Table thic communities surveyed at Egmont Atoll are separated as follows: shallow to upper-mesophotic communities from 15 to 40 m; midmesophotic communities from 60 to 90 m; and distinct deeper communities of 110-120 m and 150-160 m.A strong benthic community break was observed at approximately 60 m, separating shallow and upper-mesophotic communities from mid-and lower-mesophotic communities (>60 m, Figure2), with a clear change in benthic taxa (Figures3 and 5).This trend has previously been observed across other geographical regions (reviewed inLesser et al., 2019), including the Caribbean Sea(Kahng et al., 2019), the Pacific Ocean(Bridge et al., 2012;Sinniger et al., 2022) and the Western Indian Ocean (Stefanoudis et al., 2022).
However, a selected number of scleractinian coral species were observed below 40 m, with the deepest photosynthetic hard coral (Leptoseris sp.) recorded at 105 m; however, most of the individuals were observed to a maximum depth of 90 m.Below 90 m, octocorals, hydrozoans, antipatharians and sponges dominated communities, aligning with previous studies undertaken in the Chagos Archipelago, (summarised by Andradi-Brown et al., 2019).
tween 50 and 60 m in the Chagos Archipelago.However, Leptoseris spp. was observed as the dominant species below 60 m in this study (identification verified through genetic analysis of collected specimens; Foster et al. in preparation), concurrent with several other F I G U R E 4 dbRDA plot allowing visualisation of the DistLM, with square root data and Bray-Curtis similarity.Each colour and shape represent a different depth band, with cyan closed shapes for IDR (Ile Des Rats) and orange opened shapes for MA (Manta Alley).Chla, chlorophyll a; PAR, photosynthetically active radiation; BBPI and FBPI, broad and fine benthic position index, respectively.locations: Hawaii (Pyle et al., 2016), the Great Barrier Reef (Bridge et al., 2012), the Red Sea (Kramer et al., 2020), Reunion Island (Hoarau et al., 2021), French Polynesia (Pérez-Rosales et al., 2022), a, are significant drivers of the distribution of mesophotic communities at Egmont Atoll, in the Chagos Archipelago, with substratum, F I G U R E 5 Summary figure of the main results, depicting the representing taxa at each depth and their distribution; the shallowmesophotic and rariphotic zones; the depth of the community break and low species overlap and the main environmental drivers of the change in mesophotic community structure with depth.Chla, chlorophyll a; DCM, deep chlorophyll maximum; Light, downward irradiance in percent; T°C, temperature in degree Celsius.latitude, and geomorphology playing minor roles (Figure 4).Light, measured via downward irradiance (and represented by PAR calculations), was the main driver structuring mesophotic benthic communities, accounting for ~13% of the variation in community structure (Figure 4, see Table In the ChagosArchipelago, 1% surface PAR seems more appropriate to define the boundary of the lower mesophotic zone, if the definition is to use maximum zooxanthellate coral distribution as the lower boundary.Indeed, the majority of deeper photosynthetic corals were observed in the 80-90 m depth band, with only one specimen observed at 105 m and no zooxanthellate corals detected in the 110-120 m depth band.However, when considering benthic community breaks, 120 m can be considered as the lower boundary of the lower mesophotic zone (<30% similarity between 110-120 m and 150-160 m depth bands, Figure2).The benthic zone immediately below 120 m in the Archipelago can thus be considered as the rariphotic zone (sensuBaldwin et al., 2018, Figure 5).The relationship between light and zooxanthellate coral depthrange demonstrated in this study concurs with observations made in Japan, finding similar values for PAR(Sinniger & Harii, 2018) and the deepest scleractinian corals at 82 m(Sinniger et al., 2022).However, the photosynthetic coral depth limit observed in the current study must be taken with caution as the sampling area of this study was limited to two locations within a single atoll and PAR values varied greatly between deployments (see Our results highlight the variability in benthic community structure of MCEs, both along the depth gradient and at local geographical scales in the Chagos Archipelago, Indian Ocean.This variability, revealing the uniqueness of MCEs, casts into doubt the possibility of a deep reef refugia in the region.The differences in benthic communities both vertically and horizontally are further emphasised by the numerous environmental processes that are driving them.To date, the vast majority of MCE research in the Chagos Archipelago has focussed on upper mesophotic depths, or specifically scleractinian corals.Hence, this study provides the first baseline data on MCE community structure, and the underlying environmental drivers of this structure, in the Chagos Archipelago.These data will aid further research in predicting MCE distribution, in assessing the potential for deeper reefs to provide a refugia for threatened shallow reefs, and in developing adequate evidence-based management strategies for these unique ecosystems.