Micro-Hotspots for Conservation: An Umbrella Tree Species for the Unique Socotran Reptile Fauna

: Umbrella species are deﬁned as species that can be rare and sensitive to human disturbance, whose protection may confer the protection of other co-occurring species. The dragon’s blood tree Dracaena cinnabari Balf.f. was already considered an umbrella species on Socotra Island (Indic Ocean, Yemen) due to its ecological importance for some native biota. We studied the reptile community living on D. cinnabari from Socotra Island. We sampled reptiles on trees across most D. cinnabari populations and applied co-occurrence and network partition analyses to check if the presence of reptiles on D. cinnabari populations was random or structured. Regardless of its patched and scarce actual distribution, we report the use of this tree as a habitat by more than half of the reptile community (12 endemic reptiles). Co-occurrence and network partition analyses demonstrate that this community is structured across the distribution of dragon’s blood trees, reﬂecting complex allopatric, vicariant, and biotic interaction processes. Hence, these trees act as micro-hotspots for reptiles, that is, as areas where endemic and rare species that are under threat at the landscape scale co-occur. This Socotra endemic tree is currently threatened by overgrazing, overmaturity, and climate change. Its protection and declaration as an umbrella species are expected to beneﬁt the reptile community and to protect evolutionary processes that are partially driven by the ecological links between reptiles and this tree. To our knowledge, no tree species has been proposed as an umbrella species for island vertebrate endemics so far, highlighting the ecological uniqueness of Socotra Island.


Introduction
An umbrella species is defined as a species that can be rare and sensitive to human disturbance, whose protection may confer the protection of other co-occurring species [1,2]. This concept is appealing and offers a simple ecologically-based shortcut for the conservation management of communities [3]. The umbrella species concept was first described to refer to species with wide ranges under the assumption that by protecting them, a suite of species with more modest spatial needs would be protected. Because organisms with large body sizes tend to have large home ranges, large mammals and birds have typically been proposed as umbrella species (see a review in [3]). On a few occasions, umbrella species have been proposed for the protection of island endemics, even though some islands are prominent biodiversity hotspots [4] and are considered to be among the most threatened terrestrial ecosystems [5]. Island ecosystems are characterized by high endemicity, low species richness, short trophic webs, and reduced antagonistic interactions [6,7]. The absence of large vertebrates and the apparent simplicity of the trophic webs and ecological networks [6,7] have likely limited the description of umbrella species on islands.
The Socotra Archipelago is a fragment of the Gondwana Supercontinent, which was isolated from the Indian Ocean around 20 My ago when the Arabian and African plates separated, resulting in the formation of the Gulf of Aden [8]. The Archipelago is composed of a large main island, Socotra, and three islets. It is characterized by highly endemic flora, i.e., 37% of plants [9], and fauna [10]. Within vertebrates, 93.5% of reptiles are endemic, 29 native, and two introduced [11][12][13]. Phylogenetic studies including Socotran reptiles have demonstrated their historical relatedness to Arabian and Madagascan reptiles, therefore suggesting a Gondwanan origin from ancient ancestors [14,15]. Other studies have detected that some Socotran reptiles are sister species to more recent relatives occurring on the mainland [12], with spectacular cases of intra-island speciation [16].
A relevant endemism of the island is the dragon's blood tree Dracaena cinnabari Balf.f. This is a monocotyledonous species considered to be a symbol of the island's wildlife, which is currently considered vulnerable to extinction [17]. Although a significant part of the island was covered extensively with dragon's blood trees, their original range was reduced as a result of overgrazing and overmaturity (lack of natural regeneration and a large proportion of old and dead trees) [18]. Dracaena cinnabari meets the requirements for being considered an umbrella species [19] due to its size, shape, and distribution throughout Socotra, as well as its rarity and sensitivity to human disturbance [2]. Reptiles could be the target beneficiary taxon, as this tree was referred to as being occasionally used by some reptile species and the exclusive habitat of the Critically Endangered Hemidactylus dracaenacolus Rösler & Wranik, 1999 [11]. Only one other tree was referred to as being used by a significant number of reptile species, Phoenix dactylifera L. [11]; however, this tree is extensively cultivated, and it does not constitute a natural habitat to be protected.
In this study, we describe and quantify the reptile community living on D. cinnabari with the aim of determining whether this tree should be classified as an umbrella species for reptiles. In particular, we will answer the following questions: Is this tree a common habitat for the reptile community? Does reptile richness on D. cinnabari habitats depend on tree density or maturity? How many D. cinnabari populations need to be protected to cover most of the reptile community? If this tree serves as habitat for at least one third of the native reptile species and it is important for near-threatened/threatened species, we consider it should be classified as an umbrella species. If species richness is higher in denser or more mature D. cinnabari populations, we should give higher conservation priority to those areas. If the reptile population is structured, the number of D. cinnabari populations to be protected should match the number of sub-communities to guarantee the protection of a higher number of reptile species. Population structure could also be explained by previously known allopatric distributions (disjunct geographic ranges) or unveiled competitive interactions (interactions that result in negative outcome for both parts involved).

Study Sites and Sampling
In February and March 2014, we conducted an extensive sampling of D. cinnabari populations in Socotra aimed at describing co-occurrence patterns and building the ecological network of the reptile community living in this particular habitat. We sampled 11 D. cinnabari populations (280 individual trees) covering the species' entire distribution on the island, including intact forest (two sites, in Diksam and Firmihin forests), sparse woodland (three sites, around Firmihin forest), mountainous areas (three sites, in the Haggeher Mountains), and major isolated D. cinnabari remnants (three sites, in the Eastern side of the island) ( Figure 1A). Currently, only 10% of the island is covered with D. cinnabari forest/woodland ( Figure 1A). The largest populations of D. cinnabari are located in the Haggeher Mountains and the adjacent limestone plateau of Diksam, in the center of the island ( Figure 1A), usually forming a sparse woodland. In the Natural Sanctuary of Firmhin's plateau, D. cinnabari trees form a continuous forest, possibly reflecting the aspect of the island before human occupation ( Figure 1B). Mountains and the adjacent limestone plateau of Diksam, in the center of the island ( Figure 1A), usually forming a sparse woodland. In the Natural Sanctuary of Firmhin's plateau, D. cinnabari trees form a continuous forest, possibly reflecting the aspect of the island before human occupation ( Figure  1B). Sampling sites were selected when at least 20 D. cinnabari trees were less than 15 m apart from each other. Between two to four researchers examined D. cinnabari trees during 13 days and nights, totaling approximately 41 h of effective sampling effort. During the day, trees were selected and georeferenced, and distances among selected trees and the three closest neighbors were measured as estimators of tree density. We measured the girth at 1 m height, stem height, and number of branching events as estimates of maturity of the selected trees. Trunks, visible canopy, and the nearest surrounding ground, including rocks and shrubs, were examined for the presence of diurnal and nocturnal (resting) reptiles. At night, the same trees were examined again with flashlights in order to record the presence of both (resting) diurnal and (active) nocturnal species. The Environmental Protection Authority (EPA) approved this study as it was in the scope of the agreement signed by an EPA representative and Salvador Carranza on March 22, 2010.

Data Analyses
A presence/absence and abundance matrix was compiled treating each D. cinnabari population as a sampling unit. From this matrix, we examined co-occurrence patterns and constructed an ecological network. Co-occurrence patterns were calculated by the C-score [20], which measures the average number of checkerboard units (pairs of sample sites where, when comparing the presence of two species, one is present and the other is not) between all possible pairs of species. In a competitively structured community, the C-score should be significantly larger than expected by chance, so we tested the null hypothesis if C-scores were identical to expectations. The C-score was Sampling sites were selected when at least 20 D. cinnabari trees were less than 15 m apart from each other. Between two to four researchers examined D. cinnabari trees during 13 days and nights, totaling approximately 41 h of effective sampling effort. During the day, trees were selected and georeferenced, and distances among selected trees and the three closest neighbors were measured as estimators of tree density.
We measured the girth at 1 m height, stem height, and number of branching events as estimates of maturity of the selected trees. Trunks, visible canopy, and the nearest surrounding ground, including rocks and shrubs, were examined for the presence of diurnal and nocturnal (resting) reptiles. At night, the same trees were examined again with flashlights in order to record the presence of both (resting) diurnal and (active) nocturnal species. The Environmental Protection Authority (EPA) approved this study as it was in the scope of the agreement signed by an EPA representative and Salvador Carranza on March 22, 2010.

Data Analyses
A presence/absence and abundance matrix was compiled treating each D. cinnabari population as a sampling unit. From this matrix, we examined co-occurrence patterns and constructed an ecological network. Co-occurrence patterns were calculated by the C-score [20], which measures the average number of checkerboard units (pairs of sample sites where, when comparing the presence of two species, one is present and the other is not) between all possible pairs of species. In a competitively structured community, the C-score should be significantly larger than expected by chance, so we tested the null hypothesis if C-scores were identical to expectations. The C-score was measured with EcoSim software [21], with an alpha level = 0.05 and 5000 randomized simulations based on the matrix. As competition among species may result in allopatric distributions, this test is useful to guide conservation planning of the reptiles occurring in D. cinnabari populations.
Based on co-occurrence patterns of reptile species in D. cinnabari populations, we constructed an ecological network. We used the Louvain method [22] implemented by Pajek [23] for identifying reptile sub-communities within the network. This method attempts to optimize the modularity of the network, i.e., the network's strength of division into groups. Firstly, the method detects "small" communities, optimizing modularity locally; secondly, it aggregates nodes (species) belonging to the same sub-community and builds a new network. These steps were repeated iteratively until maximum modularity was attained and a hierarchy of communities was produced. Accordingly, species with similar co-occurrence patterns remained within the same sub-community. The resulting network was visualized with Gephi [24].
If the reptile population is random, it would not matter which D. cinnabari population should be protected in order to protect the reptile community. On the other hand, if it is structured, the number of D. cinnabari populations to be protected should match the number of sub-communities to guarantee the protection of a higher number of reptile species.
We performed statistical analyses to check if the total number of reptile species per site was correlated with any maturity variables or distance between trees calculating Pearson's r and the respective 95% confidence intervals (CI) with the Fisher transformation (F). If the F values were outside of CI, we rejected the null hypothesis of ρ = 0.
We considered D. cinnabari as an umbrella species for the reptile community if it served as habitat for at least one third of the native reptile species and some of those are near-threatened/threatened species, especially if they occurred in areas of higher density or maturity of the trees.
The co-occurrence C-score measured for the reptile community living on D. cinnabari trees was 3.35, which was significantly higher (p = 0.02) than expected (expected C-score = 3.06, variance = 0.01) and confirmed the community's competitive structure. Network partition detected four sub-communities within the reptile community ( Figure 2). The central sub-community was composed of widespread and generalist Socotran species that co-occurred at most of the sites sampled (the widest nodes and a dense set of connections among edges of the network in Figure 2) and were able to thrive in a wide variety of environmental conditions. Interestingly, two of the other three sub-communities likely corresponded to species that only occurred on a reduced portion of the island. One of these sub-communities was composed of two tree/shrub specialists that spatially segregated from the rest of the species, one of them being the Critically Endangered H. dracaenacolus ( Figure 1C), a species that only lives on D. cinnabari trees, and H. trachyrhinus, which is mostly found active in Cissus bushes. Another sub-community included the rock-dwelling and cliff species Pristurus insignis Blanford, 1881, Pristurus insignoides Arnold, 1986, and Hemidactylus granti Boulenger, 1899, the sister species of H. dracaenacolus. Both H. dracaenacolus and H. granti have parapatric distributions possibly linked to a vicariant process that occurred on the island approximately 2.3 My ago [16]. Therefore, at least four D. cinnabari populations should be protected to cover most of the reptile community. cinnabari trees with respect to other habitats, and colors indicate sub-communities according to the partition procedure (green: shrub/tree specialist community; blue: habitat generalists community; pink: rock-dwelling community; red: arboreal/shrub species). Edges join species that share at least one sampling point, and the higher thickness of the edges indicates those species with higher co-occurrence patterns within the 11 sampling points.
Concerning density, D. cinnabari sampled trees had an average minimum distance to the closest tree of 447 ± 505 cm (range = [65, 1500] cm) and an average distance to the three closest D. cinnabari trees of 1301 ± 128 cm (range = [1112, 1500] cm); see Table 1. The sites with the lowest referred average distances to D. cinnabari were Diksam and Haif, respectively (Table 1) Table 1. The site with the selected trees with the largest average girth was Qafshifo and with the highest average stem was Firmihin Protected Area ( Table  1). The site with the selected D. cinnabari with the highest average number of branching events was Qafshifo (Table 1). We counted up to 20% dead or partially damaged crown trees in some remnant woodlands. The average number of total reptile species found per site was 8 ± 5 (range = [2,18]). The site with the highest richness of reptiles was Qafshifo, followed by Killisan (Table 1).

Figure 2.
Ecological network of the Socotran reptiles found in Dracaena cinnabari Balf.f. trees. Nodes represent species; their size indicates the degree of specialization in the use of D. cinnabari trees with respect to other habitats, and colors indicate sub-communities according to the partition procedure (green: shrub/tree specialist community; blue: habitat generalists community; pink: rock-dwelling community; red: arboreal/shrub species). Edges join species that share at least one sampling point, and the higher thickness of the edges indicates those species with higher co-occurrence patterns within the 11 sampling points.
Concerning density, D. cinnabari sampled trees had an average minimum distance to the closest tree of 447 ± 505 cm (range = [65, 1500] cm) and an average distance to the three closest D. cinnabari trees of 1301 ± 128 cm (range = [1112, 1500] cm); see Table 1. The sites with the lowest referred average distances to D. cinnabari were Diksam and Haif, respectively (Table 1) Table 1. The site with the selected trees with the largest average girth was Qafshifo and with the highest average stem was Firmihin Protected Area ( Table 1). The site with the selected D. cinnabari with the highest average number of branching events was Qafshifo (Table 1). We counted up to 20% dead or partially damaged crown trees in some remnant woodlands. The average number of total reptile species found per site was 8 ± 5 (range = [2,18]). The site with the highest richness of reptiles was Qafshifo, followed by Killisan (Table 1). Table 1. Details of the sites and variables measured in Dracanea cinnabari Balf.f. trees with reptiles. The code of the sites (S) where D. cinnabari were sampled, its coordinates (latitude (Lat) and longitude (Long), in decimal degrees, WGS84), and location names are listed. The total number of reptiles species found on trees (N), the average minimum distance from the tree with reptiles to the closest tree (MD1), the average distance to the three closest trees (XD3), the average girth at 1 m height, the average stem height (SH), and the average number of branching events (BE) are detailed per site. We found positive correlations of an estimator of tree maturity with the average number of reptile species present on D. cinnabari populations, but not with estimators of tree density. That is, we found positive correlations between the total number of reptile species found on a site and the average girth

Discussion
This study demonstrated that D. cinnabari was a suitable habitat for half of the native reptile community of Socotra Island, but we could not disregard the occurrence of other unexplored endemic fauna living in the bark and canopy of the tree species. Although few trees have been proposed as umbrella species (however, see [19,25]), we supported the consideration of D. cinnabari as an umbrella species for vertebrates as well. This was due to its ecological, environmental, and evolutionary functions for the Socotra endemic reptiles, in addition to its historical economic and social role for local people that harvest D. cinnabari resin, which is an activity that could be sustainable if carefully regulated.
Network and co-occurrence analyses, together with previously known distribution data and ecological modelling [26,27], showed that reptiles living on Dracaena trees belonged to different structured communities reflecting both allopatric distributions and competitive interactions. Some habitat generalists (e.g., Haemodracon riebeckii and Hemidactylus inintellectus) [11] likely avoided the presence of the Dracaena-specialist H. dracaenacolus, and even in sympatry in the study sites, the former only occupied D. cinnabari trees when the latter were absent, so they were not syntopic. The high intra-island speciation displayed by some reptiles [16] suggested that adaptation to new niche opportunities such as the use of D. cinnabari trees might be one of the evolutionary drivers in Socotra. In this study, we showed higher co-occurrence patterns among some reptile species than others; however, we did not specifically check for species interactions, and co-occurrences and interactions did not necessarily match [28,29]. For these reasons, it would be interesting to check for interactions among reptiles as well in future works using different methodologies.
As our study was performed during a short period of time and reptiles' detectability could vary along the year, it is possible that we have underestimated the importance of D. cinnabari trees for some reptile species. However, the number of species detected was enough to demonstrate that the tree is indeed important for the reptile community, even for species that were not previously described to use D. cinnabari trees (e.g., Ditypophis vivax Günther, 1881) and for the Near Threatened Chamaeleo monachus.
Thus, we recommend that local authorities repeat this survey after the rains to check for undetected reptile species such as the worm snakes and to perform further research to unveil the link between dragon's blood trees and chameleons. It is important to mention that endemic gecko species were recorded with pollen of D. cinnabari on their snouts, suggesting that they may pollinate this highly valued tree [28]. If this is to be proven for chameleons, it may help conservation actions as they are typically persecuted and feared on Socotra [28] (it is said that a person hearing its hiss will lose the ability to speak).
Currently, D. cinnabari woodlands and forests occupy only 10% of its current potential habitat [24], and forests continue to be threatened by a number of local and general processes. The abundance of goats across the island exerts a grazing pressure on seeds and young shoots that precludes D. cinnabari turnover, causing a general reduction of its distribution and density in many areas. Adolt and Pavlis [30] concluded that the lack of regeneration could lead to an irreversible loss of the structure of D. cinnabari forests within 30-77 years. It seems, from our results, that the most mature D. cinnabari populations hold a larger number of reptile species, so those populations should have higher conservation priority, regardless of tree density. This scenario is even more impacting as it especially affects the reduced area where H. dracaenacolus occurs. Moreover, by 2080, a predicted increase in aridity is expected to reduce the potential D. cinnabari distribution by up to 45% [31]. Indeed, fragmentation and degradation of D. cinnabari populations could trigger the extinction of endemic reptile specialists, as well as remove one of the (biotic) drivers that might be participating in the speciation processes of the reptile community on Socotra Island. It is also important to highlight that, in this work, we did not take into account the genetic diversity of the reptiles, most of it hidden cryptic diversity [26], and previous studies have shown its importance for persistent conservation measures [27]. Thus, fragmentation or degradation of some D. cinnabari populations could promote the erosion of particular genetic pools of some reptile species. Measures should be taken to avoid that by protecting the richest and most unique genetically diverse areas where both D. cinnabari and reptiles co-occur.
Conservation of D. cinnabari forests is urgent. Two nurseries funded by international institutions were producing D. cinnabari seedlings in the past, although now, funding is on hold due to the country's political situation. The slow growth of this tree [30] calls for the immediate protection of remnant populations, drastically reducing the grazing pressure by means of necessary agreements with shepherds.

Conclusions
The unique flora and fauna of islands and the evolutionary processes that they hold are fragile treasures that should be integrally protected. On the island of Socotra, the protection of D. cinnabari forests may not only save this relict of the Mio-Pliocene Laurasian subtropical forests, but also a number of endemic species, including reptiles, as well as biotic interactions and ongoing evolutionary processes, such as those involving reptile species as shown by this study. For this reason, we recommend protecting at least four of those Dracaena populations, namely in Haif, Killisan, Skand, and Shibehon, to maximize protection of all the reptile sub-communities detected, specially populations with older and thicker trees, as those seem to hold higher numbers of reptile species.  Table A1.  Table A1.