Alternative functional strategies and altered carbon pathways facilitate broad depth ranges in coralâ•’obligate reef fishes

18 1. Spatial refuges in peripheral habitats will become increasingly important for 19 species persistence as climate change and other disturbances increasingly 20 impact habitat quality and assembalge compositions. However, the capacity for 21 persistence will be determined in part by species-specific abilities to absorb 22 costs related to altered or decreased quantities and quality of resources at range 23 peripheries. 24

2. We compare variations in dietary strategies and energy acquisition tradeoffs 25 along depth gradients in two obligate corallivores that differ in level of diet 26 specialization. We also assess depth-related changes in energy pathways and 27 energy content of their mixotrophic prey. We found no changes in feeding effort 28 or total resource availability (total coral cover) toward deep range margins, but 29 availability of the preferred resource (Acropora coral) decreased. While both 30 species selectively targeted Acropora, the more specialized species (Chaetodon 31 baronessa) exhibited limited feeding plasticity along the depth gradient. The 32 degree of selectivity toward the preferred coral increased rather than decreased 33 with depth, being 40 times greater than expected, given availability, at their 34 range periphery. In contrast, the generalist's diet (C. octofasciatus) varied 35 greatly in response to changes in resource availability with depth. 36 3. Unexpectedly, the energy content of Acropora did not decline with depth, likely 37 due to increased coral heterotrophy in deeper water, indicated by shifts in their 38 molecular isotope geochemistry. This shift was accompanied by a 20 % increase 39 in plankton-sourced carbon in the muscle tissue of deep-resident fish, despite 40 no observations of direct feeding on plankton food sources. with natural reductions in the quantity and quality of resources (Brown 1984, Thomas 59 and Kunin 1999) that often result in costs to consumers (Zammuto and Millar 1985, 60 Badyaev and Ghalambor 2001). Consequently, understanding potential tradeoffs and 61 compensatory mechanisms of energy acquisition at range peripheries will be vital for 62 predicting future trajectories of species vulnerable to extirpation and extinction. 63 64 For energy maximizing species (Hixon 1982), the ability to persist in marginal habitats 65 is likely to rely on flexibility in diets or feeding rates ( To assess patterns in feeding ecology, divers followed focal fish for 3 minutes at a 181 distance of ~ 2-3 m and recorded the total number of bites, and the minimum and 182 maximum depth of the observation period. We quantified overall feeding effort by 183 recording bite rates of individuals of both species (C. baronessa total n. obs. = 344, C. 184 octofasciatus total n. obs. = 107) pooled across all hard coral types on six reefs (see 185 suplemental figure S2). Within a subset of these observations (from random depths 186 between 0 and 30m on three reefs), we also recorded the number of bites targeted on 187 each of 37 coral genera (See supplemental Table S1) (C. baronessa n. obs = 276, C. 188 octofasciatus n. obs = 90). There was some replication among feeding observations 189 within known monogamous and territorial feeding pairs, resulting in possible pseudo-190 replication among this subset of observations, which was accounted for in our analyses 191 (See Data analysis below). There were no temporal patterns in sampling among depths. 192

193
Feeding selectivity 194 The level of selective feeding on Acoropora by both species was calculated across all 195 depths and within each 5m depth bin on the focal reef, Christine's reef (Suplemental 196 figure S2) using Manly resource selection ratios (Manly et al. 2002). Selectivity was 197 tested by comparing observed and expected bite frequencies on each prey genera based 198 on genera abundance at each depth using chi-squared tests. 199

Nutritional quality of corals 201
To identify changes in the nutritional quality of corals across depths, we collected a 202 fragment (~7-10 cm) from replicate colonies of two commonly targeted Acropora 203 morphologies (tabular; n = 30 and hispidose; n = 37) between 0 m and 40 m depths 204 (Suplemental figure S2). Hispidose colony samples were ground in a mortar and pestle, 205 dehydrated then decalcified using 1M HCL prior to freeze-drying. Tissue from tabular 206 colonies (separate collection) was removed from the skeleton within individual 207 collection bags using an air pick and the resultant slurry collected in vials, dehydrated, 208 and frozen. All samples were dehydrated for ~48 hours at ~55°C then stored in a freezer 209 and freeze-dried prior to lipid extraction. Total lipids were extracted from dried tissue 210 samples using a dichloromethane:methanol solvent protocol (see supplemental 211 methods) and were recorded as proportional dry-weight. Again, there were no temporal 212 patterns in sampling among depths. 213 214

Trophic carbon pathways of corals and fishes 215
Depth-related shifts in trophic position and the carbon pathways supporting coral prey 216 were analyzed using bulk stable isotope analyses (SIAB) of tissue samples from six 217 shallow Acropora colonies (0 -5 m) and four deeper colonies (30 -40 m). Decalcified, 218 dried, and homogenized non-lipid extracted samples were combusted and analyzed for 219 d 13 C and d 15 N isotope values on a Costech elemental analyzer coupled to a 220 ThermoFinnigan Delta-V gas source isotope-ratio-monitoring mass spectrometer (EA-221 IRMS). Stable isotope results are reported using standard delta (d) notation in per mil 222 (‰) relative to standards Vienna Pee Dee Belemnite for carbon and atmospheric N2 for 223 nitrogen. Reproducibility of lab standards was ± 0.1 ‰ and ± 0.2 ‰ for d 13 C and d 15 N, 224 respectively. 225

418
The d 13 C values of essential amino acids within C. baronessa muscle tissue were also 419 lower among deeper-reef residents (Fig 3c, Supplimental Table S3). CSIA-AA-based 420 mixing models of relative source end-member carbon contributions to C. baronessa 421 muscle tissue further supported differentiation in the dietary carbon pathways of 422 shallow-reef and deeper-reef butterflyfish populations (Fig 3d). As expected, coral-423 fixed carbon was the dominant carbon source supporting C. baronessa overall (79 ± 424 13%). However, the relative contribution of coral-sourced carbon to the food web 425 supporting C. baronessa decreased by a 25% between depths (Shallow 90 ± 2% (SD); 426 Deep: 67 ± 5%). Concurrently, the relative contribution of water-column derived 427 planktonic carbon increased substantially among deeper-resident fish (Shallow: 7 ± 2%; 428 Deep: 27 ± 4%) (Fig. 3d). In both populations, microbially-reprocessed detritus made 429 up a relatively small contribution of total carbon to C. baronessa (Deeper: 6 ± 1%; 430 Shallow: 3 ± 1%).  The dietary strategies reported for C. baronessa and C. octofasciatus in shallow water 459 were largely consistent along the depth gradient. The specialist remained specialized, 460 while the generalist became more generalized with increasing depth. Unexpectedly, the 461 relative feeding effort (selectivity) targeting Acropora increased with depth for both 462 species. For the specialist, this is likely related to a continued reliance on Acropora, but 463 for the generalist may be related to competitive release of the preferred resource at C. baronessa on non-coral hosts may therefore provide dietary compenstation at depth 492 and greater potential capacity to respond to coral loss than previously assumed. This is 493 surprising because C. baronessa is widley considered an obligate corallivore and 494 supplemental feeding on non-coral diets has not been observed previously (Pratchett et  of shallow-water corals among shelf positions (Sammarco et al. 1999) and broad 530 geographic locations (Heikoop et al. 2000). It is not known whether similar processes 531 operate differentially within tens of meters of depth in Kimbe Bay. Therefore, we 532 cannot rule-out an additional role of depth-related variation in the nutritional-content 533 of heterotrophic food sources as a potential mechanism for the propossed increases in 534 heterotrophic carbon uptake. Both possibilities, however, suggest that multiple 535 mechanisms may act to buffer the marginality of deep reef habitats for specialist species 536 vulnerable to shallow-water habitat loss. 537

538
The results here suggest increased coral heterotrophy and/or substitute feeding on 539 plankton may additionally buffer 'coral-obligate' fish from depth-related declines in 540 the availability and hypothesized declines in nutritional quality of preferred corals.  Table S1: The proportion of bites taken from each coral taxon within 5m depth bins along a gradient from 0 -30 m.

Lipid extraction protocol
Freeze dried coral tissue samples (see main methods), were weighed to the nearest 0.000g. 2ml of a dichloromethane : methanol (2:1) solvent was added to each sample and mixed for 10 minutes in a Sonicator. After cooling, the sample/solvent mix was filtered through solvent extracted cotton-stuffed, glass Pasteur pipettes using pressure from a hand bulb. An additional 1ml of dichloromethane : methanol solvent was passed through the filter to wash all of the lipid solvent solution into collection vials. 3.5ml of sample wash [KCl 0.44% in H20 (3): Methanol (1)] was added and left overnight for lipids to fall out of solution. The top layer of samples (non-lipid) were removed using a Pasteur pipette with bulb. Lipid samples were recovered into pre-weighed and labelled glass vials using a 1.0ml glass syringe and the solvent was evaporated in a nitrogen evaporator. Total lipids were quantified by re-weighing pre-weighed vials to the nearest 0.000g, now containing lipids.

Complete methods and justification for compound specific stable isotope protocol
To examine the relative contribution of carbon source end-members to corals and coral-feeding butterflyfishes, we used an amino acid carbon isotope fingerprinting approach ( 2). We used three data files to parameterize our mixing model: 1) consumer data consisting of d 13 C values for five essential amino acids (threonine, isoleucine, valine, leucine, phenylalanine) for individual coral or butterflyfish (separate models), 2) source end-member essential amino acid d 13 C fingerprints (see description below), and 3) Trophic discrimination factors for the five essential amino acids (0.1 ± 0.1; McMahon et al. 2010). In SIAR, we ran 500,000 iterations with an initial discard of the first 50,000 iterations as burn-in. By using d 13 CEAA values within the Bayesian isotope mixing model, we avoid the major issue that plagues poorly resolved dual isotope approaches in multi-end-member systems (Fry 2013;Brett 2014): underdetermined mixing, and complications of variable and poorly characterized trophic fractionation (Bond and Diamond 2011).
We characterized unique amino acids isotope fingerprints (multi-variate patterns in relative d 13 C among essential amino acids) for three potentially important source endmembers to Chaetodon baronessa: autotrophic coral carbon (zooxanthellae-proxy), herbivorous zooplankton carbon (water column phytoplankton proxy), and detritivorous sea cucumber carbon (microbially-reprocessed detritus proxy). The source end-member data (Table S2)  (2016) collected staghorn coral, Acropora pharaonis, that is targeted by coral-eating butterflyfish (e.g., Berumen and Pratchett 2008) to represent carbon fixed by autotrophic zooxanthellae. The essential amino acid d 13 C fingerprints of these corals aligned with the essential amino acid d 13 C fingerprints of pure cultures of Symbiodinium sp. from Woods Hole Oceanographic Institution, indicating that these corals rely almost exclusively on autotrophically fixed carbon with little to no heterotrophic feeding. As such, we used these corals as proxies for autotrophic coral end-members in our mixing model.  collected pelagic calanoid copepods that feed on water column phytoplankton as 4 proxies for water column phytoplankton carbon. They did not use phytoplankton directly because the fast turnover rate of phytoplankton means that their isotope signatures are just a snapshot of the water column baseline signature. Instead, they analyzed zooplankton, which integrate dietary carbon signals over longer time scales more relevant to the turnover rates of butterflyfish. Furthermore, given that essential amino acids show virtually no isotope discrimination between diet and consumer (McMahon et al. 2010), the essential amino acid d 13 C values of pelagic copepods provided a faithful proxy for pelagic phytoplankton. As expected, the essential amino acid d 13 C fingerprints of these coral reef plankton aligned with the fingerprints of water column phytoplankton from the Larsen et al. Together, these source end-member essential amino acid d 13 C fingerprints provide a robust data set to reconstruct the relative contribution of source end-members to coral and butterflyfish production.
We focused our analyses on only essential amino acids (threonine, isoleucine, valine, leucine, and phenylalanine) for two reasons: 1) The essential amino acid d 13 C fingerprints represent the sum of the isotopic fractionations associated with individual biosynthetic pathways and associated branch points for each essential amino acid (Hayes 2001;Scott et al. 2006), generating phylogenetically diagnostic amino acid fingerprints of different source end-members (Larsen et al. 2009(Larsen et al. , 2013. Because essential amino acids have very long and complex biosynthetic pathways (typically >10 independent enzymatic steps), they provide the best potential for lineage-specific isotope effects (Lehninger 1975;Stephanopoulos et al. 1998). 2) Essential amino acid d 13 C patterns of source end-members are preserved, essentially unchanged, across trophic transfers (14, McMahon et al. 2010). This is because, while plants, algae, and bacteria can synthesize essential amino acids de novo, metazoans have lost the necessary enzymatic capabilities and must acquire essential amino acids directly from their diet with minimal fractionation (Reeds 2000).
In order to compare the essential amino acid fingerprints of our three source endmember groups collected from literature data to the corals and butterflyfish in this study, we examined essential amino acid d 13 C values that were normalized to the mean of all five essential AAs for each sample. As expected, there is strong experimental and field-based evidence that primary producer essential amino acid d 13 C fingerprints are faithful and robust across large environmental gradients in growing conditions and carbon sources that can affect bulk d 13 C values (Larsen et al. 2009(Larsen et al. , 2013(Larsen et al. , 2015. This is because the underlying biochemical mechanisms generating unique internally normalized essential amino acid d 13 C fingerprints are driven by major evolutionary diversity in the central synthesis and metabolism of amino acids. For example, Larsen et al. (2013) examined the extent to which normalized essential amino acid d 13 C fingerprints were affected by environmental conditions by looking at seagrass (Posidonia oceanica) and giant kelp communities (Macrocystis pyrifera) across a variety of oceanographic and growth conditions (see Larsen et al. 2013 Table S1 for details). For both species, the range in bulk d 13 C values was five-to ten-times greater (2.6‰ and 5.2‰, respectively) than it was for normalized essential amino acids d 13 C (0.4‰ to 0.6‰, respectively). By normalizing the individual d 13 CEAA values to the mean, Larsen et al. (2013) showed that natural variability in d 13 C values of individual amino acids is effectively removed, creating diagnostic fingerprints that were independent of environmental conditions. Larsen et al. (2015) further confirmed this concept with the first directly controlled physiological studies of fidelity in normalized essential amino acid d 13 C fingerprints. This study grew the laboratory-cultured marine diatom, Thalassiosira weissflogii, under a wide range of conditions: light, salinity, temperature, and pH. This study showed that normalized essential amino acid d 13 C values remained unmodified despite very large changes in bulk and raw amino acid d 13 C values (>10‰), molar percent abundances of individual amino acids, and total cellular carbon to nitrogen ratios. Together, Larsen et al. (2013Larsen et al. ( , 2015 provide strong evidence that normalized essential amino acid d 13 C fingerprints are diagnostic of the primary producer source rather than the myriad factors affecting bulk d 13 C values, such as carbon availability, growth conditions, and oceanographic conditions. As such, we are confident that the normalized essential amino acid d 13 C fingerprints of literature source end-members are robust, faithful proxies of the identity of major carbon sources relevant in this study, regardless of the exact location and growing conditions of the end-members. Table S2. Mean (‰ ± SD) essential amino acid d 13 C values of three source end-members (n = 24 individuals for each source end-member) characteristic of potential carbon sources fueling coral and butterflyfish (Literature data from McMahon et al. 2016). Each essential amino acid d 13 C value was normalized to the mean of all essential amino acid d 13 C values within each individual to facilitate comparisons of amino acid "fingerprints" across systems and environmental conditions (sensu Larsen et al. 2015).