A molecular census of early‐life stage scleractinian corals in shallow and mesophotic zones

Abstract The decline of coral reefs has fueled interest in determining whether mesophotic reefs can shield against disturbances and help replenish deteriorated shallower reefs. In this study, we characterized spatial (horizontal and vertical) and seasonal patterns of diversity in coral recruits from Dabaisha and Guiwan reefs at Ludao, Taiwan. Concrete blocks supporting terra‐cotta tiles were placed at shallow (15m) and mesophotic (40m) depths, during 2016–2018. Half of the tiles were retrieved and replaced biannually over three 6‐month surveys (short‐term); the remainder retrieved at the end of the 18‐month (long‐term) survey. 451 recruits were located using fluorescent censusing and identified by DNA barcoding. Barcoding the mitochondrial cytochrome oxidase I (COI) gene resulted in 17 molecular operational taxonomic units (MOTUs). To obtain taxonomic resolution to the generic level, Pocillopora were phylotyped using the mitochondrial open reading frame (ORF), resolving eight MOTUs. Acropora, Isopora, and Montipora recruits were identified by the nuclear PaxC intron, yielding ten MOTUs. Overall, 35 MOTUs were generated and were comprised primarily of Pocillopora and, in fewer numbers, Acropora, Isopora, Pavona, Montipora, Stylophora, among others. 40% of MOTUs recruited solely within mesophotic reefs while 20% were shared by both depth zones. MOTUs recruiting across a broad depth distribution appear consistent with the hypothesis of mesophotic reefs acting as a refuge for shallow‐water coral reefs. In contrast, Acropora and Isopora MOTUs were structured across depth zones representing an exception to this hypothesis. This research provides an imperative assessment of coral recruitment in understudied mesophotic reefs and imparts insight into the refuge hypothesis.


| BACKG ROU N D
Climate change and other human disturbances have propelled an ongoing decline in coral reefs worldwide, and the outlook of reefs remains bleak (Hoegh-Guldberg et al., 2018;Hughes et al., 2017;Van Hooidonk et al., 2016). Mesophotic coral ecosystems (MCEs)-coral reefs between 30 and 150m-may shelter against disturbances which affect shallower reefs (< 30m) and may propagate larvae to recruit on impacted shallow reef ecosystems (Baird et al., 2018;Bongaerts et al., 2010;Bongaerts & Smith, 2019). Reproduction in corals is induced by environmental cues, such as the synergic properties of increasing light and temperature (reviewed in Harrison (2011)).
Artificial units of recruitment (AURs) have been employed by researchers to assess coral recruitment for more than 100 years (Mundy, 2000), and their application to collect and study recruits has seen widespread use (Field et al., 2007;Hill & Wilkinson, 2004). Yet, the availability of research targeting recruitment decreases inversely with depth, constrained by the technical challenges associated with working beyond the recreational scuba depth limit. Historically, few studies have examined patterns of recruitment at depths beyond 30m (Bak & Engel, 1979;Birkeland, 1977;Birkeland et al., 1981;Hughes & Tanner, 2000;Rogers et al., 1984;Vermeij et al., 2011).
As of 2017, only 1% of research targeting MCEs had examined recruitment (Turner et al., 2017); however, improved availability of diving and ROV technology and the development of new methods to study MCEs are accelerating this research. For example , Turner et al. (2018), innovated an approach to examine patterns of recruitment at 40m depth in Western Australia which foregoes diving. Also, Kramer et al. (2019) used technical diving to describe recruitment dynamics at 50m depth in the Red Sea. Albelda et al. (2020) used conventional scuba to compare juvenile and adult assemblages down to 40m depth in the Philippines. Despite recent progress, our knowledge of recruitment in MCEs remains limited and many geographic areas have never been studied, emphasizing the need for more research.
Another challenging aspect of studying the early-life history of corals lies in locating small recruits and identifying them based on few useful morphological characters (Green & Edmunds, 2011).
During settlement, spats metamorphose into their benthic life stage and begin accretion of the corallite matrix (Gilis et al., 2014), on which discrimination of microstructural characters is based (Budd & Stolarski, 2011). Fluorescence has proven useful for locating and identifying corals during this developmental stage (Baird et al., 2006;Eyal et al., 2015;Hsu et al., 2014;Roth et al., 2013); nevertheless, identification of recruits based on morphological traits is mainly limited to the family level (Babcock et al., 2003;Green & Edmunds, 2011;Nozawa et al., 2013). This limitation is particularly important in locations where diversity is high and traits converge between confamiliar taxa (Baird & Babcock, 2000), such as the Indo-Pacific. An inability to identify recruits complicates ecological assessments and hinders our ability to deduce the outlook of threatened coral communities (O'Cain et al., 2019), particularly when confamiliar species fulfill divergent functional roles within an ecosystem (Denis et al., 2017). Therefore, leveraging molecular typing toward improving the taxonomic resolution of early-life stage communities (Hsu et al., 2014;O'Cain et al., 2019;Shearer & Coffroth, 2006) can yield new insights.
This study details a census and comparison of coral recruitment in shallow and mesophotic reef communities at Ludao, an island off southwestern Taiwan. Shallow and mesophotic communities at Ludao have been shown to possess distinctive communities, despite similarities in coral assemblages . In the past, assessments of recruitment in Taiwan were conducted above 15m (Edmunds et al., 2014;Ho & Dai, 2014;Nozawa et al., 2013;Soong et al., 2003) and all have identified recruits morphologically, except Hsu et al. (2014) which identified recruits by barcoding. Here, we employ technical scuba diving and a sampling design in which tiles are fixed to blocks to survey recruitment within shallow and mesophotic zones. We censused recruits with the aid of fluorescent and white light and then barcoded their DNA using three molecular markers to identify recruits and generate molecular operational taxonomic units (MOTUs). which are subjected to high-energy hydrodynamic transport (Lau et al., 2015). These benthic communities are characterized by arborescent, bushy, and tabular hard corals, clustered octocorals, and encrusting actinarians . MCEs at Guiwan and Dabaisha exhibit limited hard substrates interspersed with sediment and rubble . Communities at mesophotic sites are denoted by unattached hard corals, bushy and encrusting octocorals, massive sponges, encrusting ascidians, filamentous cyanobacterians, and bushy hydrozoans .

| MATERIAL S AND ME THODS
Artificial units of recruitment (AURs) were constructed from concrete blocks drilled to accept 12 stainless steel concrete sleeve anchors: six on the superior surface and three on each opposing lateral surface (Figure 1a). A labeled terra-cotta tile measuring 12.5 × 12.5 × 1 cm (total surface area 362.5 cm 2 /tile) was fastened to each sleeve using a stainless steel washer and nut, resulting in an arrangement of 12 plates per block. Plates were fixed at a height of approximately 7 cm from the face of the block. Six seasonal and six long-term tiles were distributed between the superior and lateral faces of the block using a Latin square design, producing an arrangement of three vertical long-term tiles, three vertical seasonal tiles, three horizontal long-term tiles, and three horizontal seasonal tiles.
AURs were deployed during 4-8 April 2017, a few days prior to the mass coral spawning date, expected between 1 and 11 days after the full moon (full moon: 11 April 2017). In total, 20 AURs were deployed during the survey: five AURs were deployed at shallow, and upper mesophotic zones at Guiwan and Dabaisha (Figure 1b An additional survey under white light verified that weakly fluorescent and nonfluorescent spats were collected. All tile surfaces: top, bottom, and four sides, were inspected for recruits. Spats were removed by scraping with a small chisel and preserved in tagged vials containing 99% ethanol. Tiles and remaining assemblages were tagged, bleached, and dried for storage as vouchers. DNA was extracted using DNEasy Blood and Tissue kits (Q iagen) following the manufacturer's instructions but with slight modifications to increase DNA yield. Samples were incubated overnight in lysis buffer/proteinase K solution, and an additional centrifugation was performed (16,000g for 3 min) before DNA elution in order to completely dry the filter column. DNA concentration and purity were verified using a NanoDrop Spectrophotometer (Thermo Scientific). DNA was serially diluted to achieve a concentration approximating 10 ng/ml. PCR was amplified using 15 μl Taq 2X Master Mix (Amplicon) diluted to achieve a 30 μl total reaction volume.
Coral spat DNA was phylotyped using primers which amplify the mitochondrial cytochrome oxidase subunit I (COI) region, commonly used to differentiate metazoan invertebrates (Folmer et al., 1994). Raw forward and reverse sequences were screened by querying the NCBI BLAST database (Megablast) (Altschul et al., 1990), and sequences not matching scleractinian corals were excluded from further analysis. Using UGENE v.1.31 (Okonechnikov et al., 2012), DNA F I G U R E 1 a. Experimental design. Artificial unit of recruitment (AUR): a concrete block filled with cement for ballast and anchored to the seafloor with a rebar. Each block supports 12 plates: 6 in vertical orientation and 6 in horizontal orientation. Plate shading identifies seasonal and long-term plate placement. b. Schematic of blocks on reef (not to scale) sequences were assembled, and consensus sequences were created from forward and reverse reads and aligned using the MUSCLE algorithm (Edgar, 2004). COI sequences were compared to GenBank sequences of the same marker matching the search terms "scleractinia" (Benson et al., 2005). PHYLIP neighbor-joining trees consisting of sample and reference sequences were generated under the F84 distance matrix model in UGENE (Felsenstein, 2004). Molecular operational taxonomic units were generated using a furthest neighbor clustering algorithm defining MOTUs on 5% dissimilarity. Sequences identified with COI as Pocilloporidae underwent additional phylotyping of the ORF region and those identified as Acroporidae were typed with PaxC. PaxC and ORF sequences were aligned with GenBank reference sequences which matched the search terms "acroporidae" and "pocilloporidae," respectively. MOTUs for these markers were assigned following the same protocol used for the COI marker. Finalized phylogenetic trees were rendered using the interactive Tree of Life (iTOL) (Letunic & Bork, 2019). Pearson's chisquared tests were performed using the "stats" package in R version 3.4.2 (R Core Team, 2013).

| RE SULTS
We collected 518 coral-like spats and identified 451 coral spats through the combined barcoding of COI, ORF, and PaxC. Recruitment averaged 19.1 ± 80.0 recruits/m 2 throughout the study period.
55.5% of spats originated from shallow AURs and 44.5% from mesophotic AURs. 65.3% of spats recruited on horizontally oriented tiles and 34.7% recruited on vertically oriented tiles. Sixty-seven samples were excluded from our analysis, representing 7.7% of the overall collection: 37 specimens phylotyped as nonscleractinian invertebrates, 22 spats for which no PCR product was obtained, or which did not return any matches in BLAST database searches, 5 pocilloporid samples which did not amplify ORF, and 3 acroporid samples which did not amplify PaxC. Recruitment was more abundant during S1 (150 recruits) and S3 (126 recruits

| D ISCUSS I ON
We collected and identified 451 recruits, comprising 35 MOTUs and 13 coral families. Many recruits-the smallest of which possessed a diameter of 0.7mm-were too small to reliably identify morphological characters which would enable identification beyond the family tier. Unidentified samples composed 4% of our overall collection and reflect a marked improvement compared to a similar study using a high-salt DNA extraction method (9.7%) (Hsu et al., 2014).
At least 328 scleractinian coral species inhabit the reefs of Taiwan and 8 MOTUs identified in this study have been found in previous diversity surveys (Dai & Horng, 2009a, 2009bDenis et al., 2015Denis et al., , 2019de Palmas et al., 2018;Huang et al., 2015). Past synonymiza-  (Mayfield et al., 2018) and may explain the absence of P. acuta in older species catalogs but its detection by genetic assays. In Ludao, P. verrucosa exhibits a wide bathymetric distribution and constitutes one of the dominant corals structuring shallow and mesophotic seascapes. In addition, P. verrucosa is a functionally competitive species, capable of sustained recruitment (Kayal et al., 2018), explaining its position as the most copious recruiter here.
Acroporidae was the only taxonomic group to exhibit clear bathymetric structuration, and only one MOTU, Acropora sp. (III), was present at both depth zones. At least 90 Acroporidae species are currently documented in Taiwan and this family yielded relatively high richness (11 MOTUs), but low abundance (n = 31). Some acroporids are shallow-dwelling and rapid-growing species and exhibit competitive life-history traits (Darling et al., 2012). In a study of northern Taiwan, Acroporidae were the most abundant family in long-term surveys and formed the largest spats, suggesting superior survivorship compared to other families (Ho & Dai, 2014). Another study in Ludao found vertically structured distribution in acroporid recruitment, recruiting abundantly at 5m, but not at 15m, where pocilloporid and poritid recruits dominated instead (Nozawa et al., 2013). It is therefore plausible that greater diversity of Acroporidae may have been captured by surveying shallower. In addition, acroporids are vulnerable to physical disturbances such as storms (Madin, 2005), which have severely impacted local populations (Chen & Dai, 2004;Kuo et al., 2011Kuo et al., , 2012. It is possible that low acroporid coral cover locally (Ribas-Deulofeu et al., 2016) may have led to scarce recruitment in our study, as recruitment in this group is subject to densitydependent effects (Kayal et al., 2015(Kayal et al., , 2018. Poritidae were significant recruiters in previous surveys around Taiwan (Edmunds et al., 2014;Ho & Dai, 2014;Hsu et al., 2014; F I G U R E 3 Summary of MOTUs generated from barcoded COI, ORF, and PaxC markers, sorted by seasons, sites, and depth zones, in alphabetical order Kuo & Soong, 2010;Nozawa et al., 2013), but were rare in ours.

Number of Individuals
Massive Indo-Pacific poritids and many merulinids are associated with a stress-resistant life-history strategy characterized by slow growth, long generation times, and sustained recruitment (Darling et al., 2012). However, these types of corals can survive long periods in the absence of recruitment (Hughes & Tanner, 2000); therefore, low recruitment in those taxa, while unusual, is not entirely uncharacteristic of this type of life history.
The potential for MCEs to reseed shallow reefs may apply to species with wide depth distributions which inhabit both depths (Bongaerts & Smith, 2019 18-month study period cannot be ruled out, as long-distance dispersal of migrants over ecological timescales (Noreen et al., 2009;van Oppen et al., 2008) or step-wise transgenerational dispersal (Holstein et al., 2015;Vaz et al., 2016) may be sufficient to establish connectivity.
Pocillopora damicornis planulae are zooxanthellate and swim actively, enabling a long PLD of up to 212 days (Harrigan, 1972). In contrast, its sister species, P. acuta, is characterized by brief PLD, which leads to localized recruitment (Bahr et al., 2020). PLD is a critical trait for estimating connectivity and is available for many fish and other commercially important species; however, the PLDs of many coral species are still unknown. We hypothesize that extensive PLDs in Pocillopora may explain the widespread distribution of this genus in our survey. PLD could be a critical characteristic defining species which can disperse broadly and find refuge, and this possibility should be researched further. The prevalence of traits and/ or behaviors which facilitate dispersal could explain the absence or low abundance of coral taxa restricted to one particular habitat in our experiment.
Reproductive modalities may also influence the dispersal of larvae tends to be associated with dispersal ability, although exceptions exist. Brooded larvae often settle soon after they are released which limits their ability to disperse (Nozawa & Harrison, 2005;Sakai, 1997;Warner et al., 2016) and short competency periods in brooded larvae may be responsible for producing localized dispersal patterns observed in high-latitude communities (Tioho et al., 2001).
In spawners, gametes may remain planktonic for several days, enhancing their potential for dispersal (Nozawa & Harrison, 2008). The larvae of spawners A. millepora, A. tenuis, and M. digitata possess a high lipid content which makes them buoyant, enabling dispersal by wind and currents and provides energy storage during periods of extended dispersal (Arai et al., 1993;Richmond, 1987); however, this trait is not present in all spawners. Nevertheless, brooders exhibiting high connectivity across vertical gradients (Hammerman et al., 2018;Serrano et al., 2016) and spawners possessing strong vertical genetic partitioning have been documented (Eckert et al., 2019;Serrano et al., 2014).
Most corals in Taiwan do not spawn during October-April, and observed patterns were consistent with expectations. Pocillopora dominance persisted during S1 and S3, but not during S2, when their abundance was comparable to other taxa. Kuo and Soong (2010) found variable pocilloporid recruitment during wet (May-September) and dry seasons (November to March) and observed similar patterns interannually. Fan et al. (2006)  Recently, P. verrucosa has been found to brood in nearby Philippines (Villanueva et al., 2008) that wintertime P. verrucosa recruits could actually be brooded; however, further study is required. Several other pocilloporids can reproduce asexually, which could provide an alternative explanation for winter recruitment: Pocillopora acuta may generate planula asexually in the absence of sperm (Nakajima et al., 2018;Smith et al., 2019), and P. damicornis may produce clonal larvae, although this behavior may vary along latitudinal gradients (Miller & Ayre, 2004). Additionally, clonal reproduction in this species may be undertaken in response to disturbances (Sherman et al., 2006;Yeoh & Dai, 2010). Corals of the Montipora genus typically spawn between April and June in Taiwan, but may be subject to interannual variability (Lin & Nozawa, 2017); the small spat size (2.1mm) of the Montipora sp. (II) spat collected during S2 suggests it spawned late during this season, potentially explaining its retrieval during our winter survey. Lastly, it is possible that larvae with extended larval development released during the regular spawning season may remain viable to recruit during wintertime.
We emphasize the limitations in making inferences of the distribution of corals based on recruitment patterns; therefore, our conclusions warrant caution. Substrate choice is one of the most important factors determining the survival of coral larvae (Ritson-Williams et al., 2009) and larvae may select locations that maximize their chances of surviving (Martinez & Abelson, 2013). We hypothesize that tile choice may influence results by providing a more favorable habitat to some while detracting others from settling.
Indeed, Harriott and Fisk (1987) found the composition of acroporid and pocilloporid larvae settling on artificial substrates diverged from natural surfaces. Likewise, Burt et al. (2009) found recruitment densities varied among settlement substrate types. In some cases, diverging preferences may apply to close relatives: in azooxanthellate corals, Tubastraea tagusensis settles more densely on concrete substrates, while Tubastraea coccinea exhibits no such preference (Creed & De Paula, 2007). In addition, the date of initiation of the experiment could bias results toward MOTUs with later spawning dates if settlement tiles have not accumulated sufficient biofilm to promote metamorphosis (Webster et al., 2004). However, uncured terra-cotta tiles do not have this effect on Acropora millepora larvae, indicating that some species are less selective than others (Heyward & Negri, 1999).
Fluorescent censusing enhanced our ability to find recruits of small size. The application of fluorescence as an aid for recruit censusing was partly successful, but we observed variable intensity within MOTUs and between depths. It is well documented that not all coral species are fluorescent (Alieva et al., 2008;Gruber et al., 2008;Kenkel et al., 2011;Roth et al., 2015); however, a thorough registry of fluorescent corals does not exist. Additionally, in fluorescent types, intraspecific variation may occur (Eyal et al., 2015;. We were unable to quantitatively measure variation in fluorescence in our recruits, but we hypothesize that variation across shallow and deep light environments, in combination with differences in light exposure due to settlement location (i.e., vertical/horizontally oriented tile and top/bottom of the tile) could induce variation. The intensity of fluorescence may be influenced by light climates, such as depth (Scucchia et al., 2020) and shading surrounding the coral (Lesser & Gorbunov, 2001;Ralph et al., 2002). Eyal et al. (2015) showed that fluorescent signals in some species are completely independent of light exposure, while in others, fluorescence may be lost in dark environments. Alternately, fluorescent signals may be impacted by coral health  and dimmed fluorescent responses may indicate stress (Roth et al., 2015). Further study into interspecific and bathymetric variation in coral fluorescence during early-life stages is warranted. research is required to further expand our knowledge of recruitment and connectivity at depth, while delving into the physiological and environmental processes which affect them. As higher resolution molecular markers are developed, the resolution of molecular taxonomy will improve accordingly. Still, the present work represents a noteworthy improvement over traditional recruit identification.
Future studies should strive to explore understudied geographical areas while developing innovative ways to overcome the challenges of surveying recruitment at depth.

ACK N OWLED G M ENTS
The authors would like to thank Dr. Rodrigo Carballo-Bolaños for assistance with fieldwork, Dr. Lauriane Ribas-Deulofeu for support with coding, and the personnel of the Green Island Marine Research

Station and the Coral Reef Evolutionary Ecology and Genetics
Laboratory at Academia Sinica for logistical assistance. We acknowledge Drs. James Reimer, Yoko Nozawa, and Nina Yasuda for insightful discussions on early versions of this manuscript. We appreciate the constructive commentary provided by Dr. Gal Eyal and an anonymous reviewer. We are grateful to the Taitung County Government for issuance of collection permits.

CO N FLI C T S O F I NTE R E S T
The authors have no conflicts of interest to declare.

PER M ITS
Coral tissue samples were collected under Taitung County Government permit number 1040000285.

DATA ACCE SS I B I LIT Y
Coral recruit metadata and DNA sequences are publicly accessible on Dryad: https://doi.org/10.5061/dryad.msbcc 2fz4.