Prioritizing phylogenetic diversity to protect functional diversity of reef corals

The ecosystem functions and services of coral reefs are critical for coastal communities worldwide. Due to conservation resource limitation, species need to be prioritized to protect desirable properties of biodiversity, such as functional diversity (FD), which has been associated with greater ecosystem functioning but is difficult to quantify directly. Selecting species to maximize phylogenetic diversity (PD) has been shown to indirectly capture FD in certain other taxa but not corals. Here, we test this hypothesis, the “phylogenetic gambit”, on corals within global marine protected areas (MPAs).


| INTRODUC TI ON
Biodiversity loss threatens the functioning of ecosystem processes that depend upon the persistence of species assemblages and the functions they provide (Dıáz & Cabido, 2001;Díaz et al., 2007;McGill et al., 2006;Tilman et al., 1997). Species contribute unequally to ecosystem functioning and drive ecosystem properties more variably than can be predicted by species richness alone (Maureaud et al., 2019;Tilman et al., 1997;Wardle et al., 1997). These contributions are underlain by the diversity of functional traits exhibited by species in a community (Dıáz & Cabido, 2001;Tilman et al., 1997).
The functional diversity (FD) of a community-quantified by the range of trait states and values represented in the community-has been linked to higher ecosystem functioning, and the potential value of the ecosystem services provisioned (Díaz et al., 2007;Funk et al., 2017;Lavorel & Garnier, 2002;Tilman et al., 1997;Violle et al., 2007; but see van der Plas et al., 2020). To preserve the wide-ranging natural benefits of biodiversity (Tucker et al., 2019), conservation research has increasingly been focussed on understanding and measuring species traits with the objective of preserving FD (Barnosky et al., 2011;Cadotte et al., 2011;Isbell et al., 2017;Mouillot et al., 2014;Vane-Wright et al., 1991).
Given that conservation resources are generally limited, it is becoming apparent that even within protected areas, not all species can be effectively conserved, necessitating that we prioritize among them. A wide range of criteria have been proposed and applied for species' prioritization, including threat levels, genetic diversity, cultural importance and potential contributions to ecosystem function (Bottrill et al., 2008;Vane-Wright et al., 1991). FD is often used as a proxy and maximized to retain the greatest diversity of traits within an assemblage (Mazel et al., 2018;Tilman et al., 1997). Higher assemblage FD may also preserve the evolutionary potential of the participating clades, as greater variability of traits would generally result in more stable communities and thus less local extinction (Faith, 1992;Tucker et al., 2019;Walsworth et al., 2019;Winter et al., 2013). However, owing to our limited understanding of the relative importance of traits in various contexts (Cadotte et al., 2011;Díaz et al., 2013), and how FD varies spatially and between clades, directly computing FD might not always be feasible (Etard et al., 2020;Májeková et al., 2016;Mazel et al., 2018;Pakeman, 2014).
These challenges have since prompted some to view phylogenetic diversity (PD)-computed as the sum of all branch lengths from a common root (Faith, 1992), specifically PD F -as a proxy for FD (reviewed in Tucker et al., 2019). Maximizing PD as a conservation strategy has been advocated widely (Cadotte, 2013;Daru et al., 2019;Forest et al., 2007;Mooers et al., 2005;Pollock et al., 2017;Thuiller et al., 2015) and has formed the basis for conservation initiatives such as the EDGE (Evolutionarily Distinct and Globally Endangered) of Existence programme (Isaac et al., 2007). The International Union for the Conservation of Nature (IUCN) has also embraced the use of PD as a metric to inform conservation (IUCN, 2019), particularly for identifying evolutionarily significant assemblages and ecosystems (IUCN, 2016;Keith et al., 2015). The emphasis on PD follows the hypothesis that trait similarities reflect evolutionary relatedness among species. Additionally, PD is favoured as a biodiversity measure because of the relative ease of attaining the necessary data, often in the form of a DNA-based phylogeny, to compute it.
The supposed relationship between PD and total FD has led to the hypothesis that maximizing PD would also maximize the diversity of form and function (Cadotte et al., 2008;Faith, 1992;Mazel et al., 2018;Winter et al., 2013), implying that PD could be an efficient criterion for conserving FD without having to quantify species traits. This approach has been dubbed the "phylogenetic gambit" by Mazel et al. (2018), which refer to the use of PD as a surrogate for trait diversity where information is scant or unavailable (Bottrill et al., 2008;Mouillot et al., 2014;Vane-Wright et al., 1991). To assess the effectiveness of the "phylogenetic gambit", Mazel et al. (2018) quantified the strength of maximizing PD of an assemblage as a surrogate for maximizing FD in mammals, birds and tropical reef fishes with the following metric: where maxPD represents the FD of a PD-maximized assemblage, maxFD is the FD of a FD-maximized assemblage and randomFD is the FD of randomly sampled species, all from the same species pool and of the same richness. While Mazel et al. (2018) found that in the majority (88%) of selection scenarios, PD did capture more known FD than would be expected whether species assemblages were randomly selected and variation in outcomes was high. In the worst-case scenario, maximizing PD performed up to 85% worse than random selection, suggesting that a PD-maximizing strategy could be detrimental toward achieving conservation outcomes in some instances (Mazel et al., 2018). Nevertheless, the conservation of understudied ecosystems containing taxa whose traits are mostly uncharacterized could benefit from a PD-maximizing strategy.
Reef corals of the order Scleractinia constitute one such group that could benefit from a PD-maximizing strategy if the conservation objective is to maximize FD of coral reefs. Amidst elevated threats to reefs and limited resources, it is possible that only a subset of reef corals within an assemblage may be conserved or restored (Carpenter et al., 2008;Huang, 2012;Huang & Roy, 2015). Selecting species in a way that maximizes FD may be the best way to preserve the ecosystem function of reefs (Bellwood et al., 2004;Darling et al., 2012Darling et al., , 2013. Indeed and importantly, biogeography, conservation prioritization, coral reefs, ecosystem functioning, evolutionary diversity, functional traits, marine protected areas, Scleractinia preserving coral FD has been shown to improve the resilience of reefs against climate change (Walsworth et al., 2019). While generating complete data on coral species to directly maximize FD is certainly possible, it would require considerable research effort.
Given the urgency to prioritize species facing increasing levels of threat (Bellwood et al., 2004;Walsworth et al., 2019), were the "phylogenetic gambit" be valid for corals, maximizing PD would be a viable conservation strategy.
The conservation of coastal and marine ecosystems has been identified as one of the targets of the United Nations (UN) Sustainable Development Goal (SDG) 14, "Life below water" (United Nations, 2015). As of 2019, 17% of coastal and marine areas under national jurisdiction has been accorded some degree of protection, exceeding the target of protecting 10% of coastal waters by 2020 (United Nations, 2020). However, it has been suggested that despite the extensive coverage of marine protected areas (MPAs), the global MPA network is not adequately protecting marine biodiversity (Agardy et al., 2011;Daru & le Roux, 2016;Hargreaves-Allen et al., 2011). Relatedly, efficiency is an important factor to consider in conservation planning (Possingham et al., 2006). The use of proxy variables as biodiversity indicators may prove simpler and more cost-effective than collecting extensive data on species and communities (Caro, 2010;Svitok et al., 2018).
If areas with higher PD are found to also have relatively high measured FD, key areas to be chosen as MPAs can be identified by a PD-maximization strategy, i.e. using only data on species identity and phylogeny.
Therefore, our aims in this study are to (1) assess the efficacy of a strategy that prioritizes coral assemblages based on PD, with the goal of maximizing coral FD within reefs bounded by MPAs and (2) investigate the spatial patterns of such a PD-maximizing strategy on conserving FD and how S PD-FD might be associated with the spatial overlap between PD and FD of corals. Specifically, we assessed the relative gains in FD by maximizing PD (S PD-FD ) as opposed to directly optimizing FD for the purposes of marine conservation. To investigate this, the performance of PD as a surrogate for FD was measured against directly maximizing FD as an upper-limit and random species selection as the lower limit for subsets of varying proportions of species richness. We then carried out a spatial prioritization procedure to optimize the global MPA network based on the objectives of maximizing local FD in one set of scenarios and maximizing local PD in another set, subject to other constraints including fishing volume as a proxy of opportunity cost, habitat connectivity, and exposure to anthropogenic threats. We perceive the mismatches between PD-maximized and Database (Veron et al., 2009(Veron et al., , 2011 were also compiled.

| Phylogenetic and trait data
Phylogenetic data used for the computation of PD were based on a published set of 1000 Bayesian supertrees containing 805 reef coral species obtained from Huang and Roy (2015) and Mouillot et al. (2016). All analyses that required PD to be quantified were repeated across all 1000 supertrees to account for phylogenetic uncertainty. PD F (Faith, 1992) was computed using the R package 'picante' (Kembel et al., 2010) for all the species within each MPA and ecoregion with each supertree. The difference between the PD F of an MPA or ecoregion and PD F of a random assemblage of species as a proportion of the PD F within said MPA or ecoregion was computed as PD excess , a richness-independent measure. Because the random assemblages were drawn from the global species pool, PD excess would generally be negative since actual communities tended to be more phylogenetically clustered than random assemblages with the same richness (Huang & Roy, 2015).
Trait data were obtained from the Coral Trait Database compiled by Madin, Anderson, et al. (2016). Eight traits were chosen for analysis representing the major trait categories of morphology, biomechanics, physiology and reproduction, all of which are known to be crucial for explaining phenotypic variation in reef corals (Darling et al., 2012(Darling et al., , 2013Hartmann et al., 2017;Wong et al., 2018) ( Table 1). The trait data were then used to compute the functional distances between species using Gower's dissimilarity index to summarize the trait space in fewer dimensions with principal coordinate analysis (PCoA). Sufficient PCoA axes were retained to capture 70% of total initial variability. Only 379 species with data available for at least four of the eight traits were considered. Functional diversity for each MPA and ecoregion was estimated using the index of functional richness (FRic) (Cornwell et al., 2006;Villéger et al., 2008), which measures the convex hull volume defined by the species at its vertices in the multidimensional trait space based on the PCoA (Cornwell et al., 2006). FRic was computed using the R package 'FD' (Laliberté & Legendre, 2010). Negative PCoA eigenvalues were resolved with the 'Cailliez' correction method (Gower & Legendre, 1986). FD excess , the total FD within each assemblage relative to a random assemblage of species with the same species richness, was also computed for each MPA and ecoregion (analogous to PD excess above; similarly, generally negative).
Linear regressions between PD F and FRic, and between PD excess and FD excess were performed for assemblages in all MPAs and ecoregions. We also fitted a locally estimated scatterplot smoothing (LOESS) curve to these relationships to visualize the degrees to which they deviated from being linear.

| PD as a surrogate for FD
For each MPA, at five different proportions of total species richness (0.1, 0.3, 0.5, 0.7 and 0.9), PD was maximized for a subset of species within the MPA using the approach from Mazel et al. (2018). This was calculated by maximizing the PD for the subset of species within the MPA using the greedyMMD algorithm (see Supporting Information) and then computing the FD (as FRic) of this PD-maximized subset (see Bordewich et al., 2008). It is rare that there is a single unique subset of species that maximizes PD for a given tree (Mazel et al., 2018), and in order to account for this uncertainty, for each pool For each MPA, at the proportions of species selected (0.1, 0.3, 0.5, 0.7 and 0.9), we quantified the strength of PD as a surrogate for FD using S PD-FD (Mazel et al., 2018). PD would be a perfect surrogate for FD whether S PD-FD has a value of one. Positive S PD-FD would indicate that maximizing PD is a better strategy for conserving FD than random selection. Consequently, S PD-FD of 0 would indicate that the PD-maximization scheme does not do better than randomly selecting species, and negative S PD-FD would indicate that it fares more poorly than random. A two-sample t-test was also performed for each MPA at every proportion of species conserved to directly compare maxPD with randomFD.

| Spatial conservation prioritization
We performed spatial conservation planning to obtain six global MPA network scenarios that optimized local diversity-three that were PD-maximized and three that were FD-maximized-with the same area as the existing global MPA network. Our planning units which did not contain either mangrove or seagrass habitat and those which contained the centroids of cities with populations exceeding 300,000. This trimmed the number of PUs to 1858 with a total area of 5,550,160 km 2 . Proximity to mangrove or seagrass areas was included as a selection criterion as these habitats are important contributors of reef ecosystem functioning (Martin et al., 2015;Olds et al., 2013;Unsworth et al., 2008). It has been demonstrated that mangrove and seagrass habitats can enhance the performance of reserves up to 1 km away (Grober-Dunsmore et al., 2007;Martin et al., 2015). Ensuring the connectivity of protected reefs with these other habitats would thus ensure long-term reef resilience while better meeting conservation goals (Magris et al., 2017;Martin et al., 2015). The urban human population was included as a selection criterion to account for the threats of human impact (Agardy et al., 2011;Ban & Klein, 2009), following that human activity has been found to have negative implications on reef health (Mora, 2008). While studies have found that reef degradation is not necessarily correlated with local human population density (Bruno & Valdivia, 2016), urban settlements remain a potential threat to coastal habitats and challenge to MPA management (Heery et al., 2018). Spatial data for mangrove and seagrass habitats were obtained For spatial optimization, the PUs were selected by integer linear programming performed in Gurobi v9.1 (Beyer et al., 2016), constrained by fishing volume and the area of each planning unit (Table 2). For each objective of FD or PD, the selection scheme was run thrice constrained by the 75th, 50th and 25th percentiles of global fishing volume as a proxy for the opportunity cost incurred by designating a particular area as an MPA, representing the lost profits as a result of fishery restrictions (Ban & Klein, 2009;Magris et al., 2017). Global annual fishing volume data were obtained from Sea Around Us (Pauly et al., 2021).

| PD as a surrogate for FD
As expected from their strong correlation with species richness, PD F and FD computed for all 215 MPAs were positively correlated (R 2 = .8764; p < .001) (Figure 1a; Table S1). However, and importantly, the excess of PD and FD for each MPA-the relative PD and FD an MPA has to a random assemblage of the same size, or PD excess and FD excess , respectively-were also positively correlated (R 2 = .7112; p < .001) (Figure 1b). At the ecoregion level, PD F and FD were positively correlated (R 2 = .8348; p < .001) (Figure 1c; Table   S2), and PD excess and FD excess had a significant but much weaker positive relationship (R 2 = .4189; p < .001) (Figure 1d), likely because of their nonlinear relationship at this scale. LOESS curves illustrated that FD and FD excess reached maximum values faster than PD and PD excess , respectively; species in the most diverse regions contributed more PD without concomitantly increasing convex hull volume.
Across five different proportions of species selected, we found with S PD-FD that a PD-maximized assemblage of species contained 18.7% (±SD 26.6%) on average more FD than randomly selected species assemblages selected from the same MPA. While computed surrogacy values were generally positive, there were considerable variations among MPAs and proportions of species selected, with maxPD preserving between 34.9% less and 71.2% more FD (5th and 95th percentiles, respectively) than randomly selecting a pool of species at the same richness for all proportions conserved ( Figure S1).
In fact, the full range of values were between −279.8% and +95.6% relative to random (Table S3). S PD-FD values were mostly positive and increased on average with more species conserved (Figure 2; Figure   S1). There was no consistent relationship between S PD-FD and species richness across all proportions of species richness conserved (R 2 < .2).
On average, maximizing PD (maxPD) consistently preserved more FD than a random selection of species (randomFD) (Figure 3).

Objective Optimization equation
Maximize functional diversity (FD) maximise

Constraint Criterion
Fishing volume (FV) Subject to Note: FV denotes the annual fishing volume within a planning unit (PU), A denotes the area of a PU and TA denotes the total area of reef to be conserved. x i is a binary variable that represents whether a planning unit i is selected. k represents the fishing volume in tons at the z-th percentile for 0.5° × 0.5° grids within exclusive economic zones (EEZs). Note that PUs which did not contain either mangrove or seagrass habitat and those which contained the centroids of cities with populations exceeding 300,000 were excluded to maximize habitat connectivity and minimise exposure to anthropogenic threats, respectively.

| Spatial conservation prioritization
Filtering the original pool of 3780 PUs to meet the criteria of connectivity and distance from cities yielded 1858 PUs with a total area of 5,550,160 km 2 , on which integer linear programming was carried out to maximize PD or FD. A large majority of the selected planning units were common to both the PD-and FD-maximized networks across the three constraints based on global fishing volume (75th, 50th and 25th percentiles). At the 75th percentile, there were 1088 planning units (58.6%) that were common to both networks, with a total area of 3,231,380 km 2 . At the 50th percentile, there were 447 planning units (24.1%) that were common to both networks, with a total area of 1,325,560 km 2 . At the 25th percentile, there were 184 planning units (9.9%%) that were common to both networks, with a total area of 544,683 km 2 , a decrease suggesting fishing pressure was a strong driver of covariance. In both the FD-and PD-maximizing schemes, the PUs selected were largely concentrated around five marine realms-Western Indo-Pacific ( Figures S2 and S3), Central Indo-Pacific MPAs constrained by the 25th percentile of global fishing volume, but they were not statistically significant (p > .05; Figure S10).

| DISCUSS ION
Species are rarely given equal attention for conservation; resource limitation often does not allow all species in a community to be targeted for protection (Vane-Wright et al., 1991). The interest in prioritizing species based on PD is due in part to its potential to preserve trait diversity, which is regarded as a benefit on its own but also underlies other benefits to humanity (Tucker et al., 2019). Beyond the possibility that prioritizing based on PD could result in the conservation of more traits (and thus more FD), there is no known direct ecological benefit for doing so (Winter et al., 2013). Indeed, while there is commonly a strong positive correlation between PD and FD (mediated by species richness) (e.g. Flynn et al., 2011;Lososová et al., 2016;Wong et al., 2018), it has been shown that maximizing PD is not necessarily a good proxy for maximizing FD (Devictor et al., 2010;Gerhold et al., 2015;Mazel et al., 2017Mazel et al., , 2018Pollock et al., 2017;Srivastava et al., 2012). Here, to test the "phylogenetic gambit" for reef corals, we assess the performance of a PDmaximizing strategy to preserve FD relative to arbitrarily selecting species. Critically, we evaluate the risks of this strategy by investigating the mismatches between PD-maximized and FD-maximized MPA networks and highlighting scenarios where this approach could fall short. Deficiency of coral trait data will continue to be a major limitation in FD studies, so it is important to understand the circumstances in which protecting PD could still lead to gains in FD.

| PD as a surrogate for FD
Overall, maximizing PD outperforms random selection and is a viable option for maximizing FD. Indeed, S PD-FD values are positive on average (mean 18.7%) showing that a PD-maximizing strategy generally performs better than random selection. This is comparable with the average S PD-FD of 18% for vertebrates obtained by Mazel et al. (2018). Across the different proportions of species conserved, S PD-FD is mostly positive with values generally increasing with a higher proportion of species conserved (Figure 2), suggesting that a PD-maximizing strategy is more viable when a greater proportion of species to be conserved is considered. In the best-performing case, a PD-maximizing strategy relative to randomly selecting (maxPD) could result in FD gains by as high as 95.6%, again comparable with the maximum S PD-FD of 92% obtained by Mazel et al. (2018).
However, in the worst-case scenario, maxPD performs 279.8% worse than random selection and is more severe than the corresponding scenario in Mazel et al. (2018), where the poorest performance of maxPD was 85% worse than random selection. This situation occurs when randomFD performs nearly as well as directly maximizing FD while a PD-maximizing strategy (maxPD) severely underperforms. It should be noted that the worst case in this study is an isolated one, at Interestingly, maxPD protects significantly more FD than ran-domFD in 863 out of all 1038 selection scenarios (83.1%), much higher than the 64% of cases in Mazel et al. (2018). This suggests that maxPD may be a more reliable conservation strategy for reef corals compared with the mammals, birds and labrid fishes tested by Mazel et al. (2018). Such a strategy would benefit the ecosystem functioning of reefs if these traits do indeed confer ecosystem functions. However, there are also 81 selection scenarios where maxPD protects significantly less FD, and on average 24.8% worse, than randomFD, meaning there are potential risks in adopting a PDmaximizing strategy, as highlighted in the selection scenarios where maxPD protects less FD than randomFD, sometimes considerably so. To manage these risks, there is a need to better understand the factors affecting S PD-FD , so that the suitability of a PD-maximizing strategy can be assessed for specific areas.
There is considerable spatial variation in the strength of PD as a surrogate for FD and variation in S PD-FD values between assemblages. It is important to note that PD and FD are strongly correlated at both the MPA and ecoregion levels (Figure 1a,c), a pattern that is congruent with findings in the literature (see Devictor et al., 2010;Safi et al., 2011;Wong et al., 2018). However, the precision of this relationship is not universal and could be associated with the varying strengths of the phylogenetic signal among the selected traits Flynn et al., 2011). Furthermore, while maxPD may be able to select species assemblages that outperform random selection in protecting FD, it still falls short of directly maximizing FD, evident in the small average S PD-FD of 18.7%, meaning the assemblage of species selected by maxPD likely differ substantially from the species that contribute most to the known FD of the entire assemblage. Mazel et al. (2018) found that the strength of PD as a surrogate for FD decreases with increasing total richness, making maxPD a viable conservation strategy for species-poor clades and regions.
However, reef corals show a distinct result. While species richness is correlated with S PD-FD , the strength and direction of the relationship are highly variable. In the cases of richness conserved at 0.1, 0.3 and 0.9, higher richness appears to predict higher S PD-FD values while the opposite is true for richness conserved at 0.5 and 0.7 ( Figure S1). Furthermore, as conserving 0.1, 0.3 and 0.9 of species richness are associated with a higher likelihood that maxPD would conserve significantly more FD than random selection, adopting a PD-maximizing approach to prioritize the lowest and highest proportions of species in diverse areas may yield greater FD gains.
Overall, however, the effect size of species richness on S PD-FD is relatively small, and given the lack of a consistent relationship between them (even as S PD-FD generally increases with higher proportion of richness conserved), it is unlikely that species richness alone can inform the validity of using an area-specific PD-maximizing strategy.
Many underlying factors such as climate change, herbivory and habitat complexity could affect coral assemblages nonrandomly,  Madin et al., 2008;van Woesik et al., 1999). These factors could select for particular species, removing certain traits over others and thereby decreasing the overall FD of an assemblage. A PD-maximizing strategy would thus vary in its ability to capture FD depending on how each of these factors impacts different geographic regions following prioritization and protection. As illustrated by the nonrandom spatial patterns of S PD-FD (Figure 4), the biogeography of reef corals has strong and semi-independent effects on both trait and species compositions McWilliam et al., 2018) and so plays a major role in how evolutionary history and ecosystem functioning relate to each other (Khalil et al., 2018;McLean et al., 2021;Violle et al., 2014), further explaining the challenge of finding precise relationships between biodiversity indices and S PD-FD . Thus, further studies are required to explore the link between community structure and S PD-FD . While these caveats highlight the potential risks of a PD-maximizing strategy, our findings suggest that in most circumstances, this approach protects greater FD relative to random selection of species.

| Spatial congruence between PD-and FDmaximizing strategies
Given how PD and FD are positively correlated (Figure 1), it is unsurprising that a large majority of PUs have been selected by both the PD-and FD-maximizing schemes across all three constraining levels of fishing ( Figure 5; Figures S2-S7). The large degree of overlap suggests that, in the absence of data for computing FD, a spatial PD-maximizing strategy is a viable proxy for conserving FD.  (Huang & Roy, 2015). A PD-maximizing strategy could therefore result in the over-representation of specific marine provinces in the global MPA network.
Nevertheless, species in the Tropical Western Atlantic represent some of the most phylogenetically unique corals globally, including several placed among the top 20 species based on evolutionary distinctiveness (i.e. Stephanocoenia intersepta, Montastraea cavernosa, Helioseris cucullata, Siderastrea spp.) Redding et al., 2015). Caribbean corals within the region are also at high risk of extinction (e.g. critically endangered Acropora cervicornis and A. palmata, and endangered Orbicella spp.) (Carpenter et al., 2008;Huang, 2012), as the reefs have experienced dramatic declines since the 1980s with a coral-to-macroalgal community phase shift that persists till today (Gardner et al., 2003;Hughes, 1994;Jackson et al., 2014). Therefore, while a PD-maximizing strategy would not capture the most functionally diverse PUs, it could help protect corals in a biogeographically unique and threatened reef region (Briggs, 1974;Huang & Roy, 2015;Jackson et al., 2014). To increase habitat connectivity, the co-occurrence of either mangrove forests or seagrass beds is a precondition for a particular PU to be selected. Therefore, PUs in the resultant optimized networks mostly cover coastal reefs and omit oceanic reefs. We note that many oceanic reefs have high S PD-FD and even outsized FD (Figures 4-5), so they may need to be prioritized separately or with more ecologically relevant conditions of habitat connectivity (Balbar & Metaxas, 2019;McMahon et al., 2012). Our spatial optimization also follows the premise that areas under anthropogenic threats from urban areas should be avoided as protected areas. However, it can be argued that areas facing the greatest anthropogenic impacts should be prioritized instead (Mazor et al., 2021;Nelson & Burnside, 2019). Establishing MPAs in heavily fished regions could also be used as a fisheries management tool (Sala et al., 2021). Regardless, careful planning is an integral part of effective conservation, and similar methodologies that are data-driven and guided by the most relevant metrics should be applied when identifying areas to expand the global MPA network.

| CON CLUS ION
In this study, we show that a prioritization strategy that maximizes PD is viable for reef coral assemblages as it results in 18.7% more known FD (measured as assemblage-level FRic) on average relative to random selection of species. However, as there are instances where this strategy preserves significantly less FD than even random choice, especially when just a few species are to be prioritized, the PD-maximizing strategy is not one that comes without risks. We note that our comparisons with random subsets of species serve primarily as a conceptual test, and gains in FD may be more realistically benchmarked with species selection based on real-world priorities (e.g. Huang & Roy, 2013). This is an important area of work.
In the long-term, research efforts should characterize traits and understand their roles in the natural functioning of ecosystems.
Even for corals, which are relatively well-studied, only about half of the species have data for four or more of the eight traits considered. More complete trait data would enable FD to be computed with greater accuracy, enabling studies on whether FD based on a subset of traits predicts FD overall and potentially negating the need to use PD as a proxy for conservation prioritization. More generally, the spatial optimization approaches based on FD and PD we have developed here could serve as a framework for the future expansion of the MPA network to better meet desired conservation outcomes associated with reef ecosystem functioning.

ACK N OWLED G EM ENTS
This work was initiated as part of the sCAP working group supported by sDiv, the Synthesis Centre of the German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig (DFG FZT 118), and by the Canadian Institute for Ecology and Evolution. Earth Fund. We thank the sCAP working group for discussion, Yuchen Zhang for help with spatial optimization, members of the Reef Ecology Lab for feedback and support, and Qiang Lin and two anonymous reviewers for constructive comments on the manuscript.

CO N FLI C T O F I NTE R E S T
The author declare that there is no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
All datasets and R scripts are available at Zenodo (https://doi. org/10.5281/zenodo.6331588).