Extinction risk patterns in a biodiversity hotspot—The case of Thesium (Santalaceae) in the Greater Cape Floristic Region

Ecologically specialized plants are expected to be at greater risk of extinction than generalists due to climate change. Such risk is greatest in biodiversity hotspots such as the Greater Cape Floristic Region (GCFR), which accommodates both ecological specialists and generalists. Thesium L., a genus with the highest number of species in Santalaceae and the most diverse in Africa, offers an appropriate system for evaluating both the correlates of range extent and specialization and the relative extinction risks associated with both. We hypothesized that range size, ecological specialization, and consequent climatically modulated extinction risks are all phylogenetically structured, such that climate change will precipitate a disproportionate loss of phylogenetic diversity in the GCFR Thesium. Past and future species distribution ranges were predicted using MaxEnt models based on present‐day occurrences and environmental conditions. Of the 101 Thesium species modeled, 70% have had large range sizes during the Last Glacial Maximum (LGM), 50% currently have a large range size, and future conditions are predicted to allow 40% to obtain large range sizes. Between the LGM and the present, 17% of species are thought to have undergone a contraction of available range space in the present time whereas 37% are expected to expand their ranges into the future, while 51% of species will experience range contractions. Of the 65 species currently ranked as Least Concern in the South African Red List, 24% will likely shift into higher extinction risk categories. Interestingly, 8.5% of ecological specialists, although having experienced a range reduction from the LGM to the present, are predicted to persist in the face of future climate change. However, the range extent, ecological specialization, and extinction risk are phylogenetically random and therefore should have a negligible impact on the phylogenetic diversity of the GCFR Thesium.


| INTRODUCTION
Ecological specialization concerns how species restrict themselves to a range of environmental niches, with variation in niche breadth distinguishing ecological specialists from ecological generalists (Clavel, Julliard, & Devictor, 2011;Brennan, 2013). Where specialists tend to occupy narrow ecological ranges, generalists tend to occupy broad ranges, with a gradation between these two extremes (Kotze & O'Hara, 2003). Since specialists are restricted by their adaptability to environmental conditions, they tend to occupy narrower ranges relative to generalists (Behroozian, Ejtehadi, Memariani, Pierce, & Mesdaghi, 2020;Botts, Erasmus, & Alexander, 2012). In effect, specialists are locally less abundant, less dispersible, and typically strong competitors within their particular niche space (Cooper, Bielby, Thomas, & Purvis, 2008;Sodhi et al., 2008) but are susceptible to change in environmental conditions (Hammond, Palme, & Lacey, 2018). Conversely, generalists are often wide-ranging, tolerate a variety of conditions (Devictor et al., 2010) and less sensitive to external pressures (Hammond et al., 2018) because they are more successful reproducers and competitors (Botts et al., 2012). Hence, ecological specialists are more vulnerable to range contraction, and the potential risk of extinction that this may bring than the generalists (Clavel et al., 2011;Hammond et al., 2018;Murray, Rosauer, McCallum, & Skerratt, 2011), particularly in the face of accelerated climate change (Lacher & Schwartz, 2016;Thuiller et al., 2011;Urbina-Cardona et al., 2019). Thus, understanding the factors that influence species' distribution ranges is potentially important for predicting species vulnerability to climate change-mediated extinction.
In recent years, species distribution modeling (SDM) has been commonly used to predict species distribution ranges (Luan et al., 2021;Naimi & Araújo, 2016;Urbina-Cardona et al., 2019) and potential range shifts in response to climate change (Araújo, Alagador, Cabeza, Nogues-Bravo, & Thuille, 2011;Rodríguez-Castañeda, Hof, Jansson, & Harding, 2012). Characterization of a species' habitat preference promotes an understanding of how that species is likely to respond to future climate change (Lourenço-de-Moraes et al., 2019;Yang et al., 2017). In addition, it allows an assessment of species range changes in response to historical climate fluctuations (Condamine, Rolland, & Morlon, 2013). Therefore, mapping species distributions on a fine-scale grid using a combination of geological, topographical and climatic data facilitates insights into the extent of species ranges (Elith et al., 2011;Keppel, Robinson, & Wardell-Johnson, 2017), helps predict patterns of local species diversity under current (Guisan & Thuiller, 2005;Slatyer, Rosauer, & Lemckert, 2007) and future climate change scenarios . By comparing the predicted species distribution and phylogenetic diversity patterns, the distribution data provides a basis for assessing associated extinction threats (Pio et al., 2014).
Many ecologically relevant traits show phylogenetic signals, resulting in niche conservatism (Münkemüller, Boucher, Thuiller, & Lavergne, 2015). In addition, there is evidence that ecological specialism and vulnerability to climatically driven extinction are also often phylogenetically structured such that climate change may result in greater-than-expected losses of phylogenetic diversity (Eiserhardt, Borchsenius, Plum, Ordonez, & Svenning, 2015;Willis, Ruhfel, Primack, Miller-Rushing, & Losos, 2010). Presumably, a lack of strong phylogenetic structure in extinction risk indicates that conserved functional traits and the ecological variables that determine species threats are independent of evolutionary history (Davies et al., 2011). Conversely, a strong phylogenetic signal suggests that extinction risks are shared by closely related lineages (Yessoufou et al., 2016). Thus, if extinction risk is phylogenetically non-random, then interrogating the alternative evolutionary processes that best explain the distribution of threat status among tips of the phylogenetic tree might highlight functional traits significant to the extinction risk of species. Therefore, a comparative approach to testing extinction risks in the context of phylogenetic patterns vis-à-vis its impact on phylogenetic diversity can provide bases for conservation planning (Cardillo & Meijaard, 2012;Kelly, Grenyer, & Scotland, 2014;Villero, Pla, Camps, Ruiz-Olmo, & Brotons, 2017). Consequently, the phylogenetic diversity data might provide supporting information for forecasting imminent extinction risks (Kelly et al., 2014).
The Greater Cape Floristic Region (GCFR) is an ideal region for exploring the relationship of species range size and perturbation because its physical environment (climate and geology; Skowno, Poole, et al., 2019) is complex and has facilitated the evolution of both strongly stenotopic and more generalist species. Thus, it provides an opportunity to compare extinction risks between species that are more specialists in nature and those that are more generalists. Existing work suggests that within the GCFR, endemic and narrow-ranged species tend to be concentrated in the Western Cape Floristic Region (CFR; Skowno, Raimonodo, et al., 2019) whose greater climatic stability during the Pleistocene reduced extinction rates and facilitated the persistence of local endemic taxa (Chase & Meadows, 2007;Cowling et al., 2015;Cowling & Lombard, 2002). On the other hand, climate variability can lead to species diversification (Lacher & Schwartz, 2016). Notwithstanding their association with historically stable climatic refugia and narrow ranges, ecological specialists may be at particular risk of extinction, both because their associated cool conditions are expected to shrink disproportionately in the future (Ohlemueller et al., 2008) and because the small size of their ranges inherently increases their sensitivity to change (Hannah, Midgley, Hughes, & Bomhard, 2005;Lacher & Schwartz, 2016;Schwartz, Iverson, Prasad, Matthews, & O'Connor, 2006).
Containing both narrow-ranged ecological specialists and generalists, the genus Thesium L. (Santalaceae) represents an appropriate study group for assessing the correlates of range extent and specialization and the relative extinction risks faced by narrow-ranged, specialists and widespread species due to accelerating climate change (Diamond & Chick, 2018). This study, therefore, focuses on the GCFR Thesium clade (Zhigila, Verboom, & Muasya, 2020), a largely Cape-endemic lineage that originated in the Middle Miocene and subsequently radiated to give rise to the bulk of Thesium species richness (ca. 100 species) found in the contemporary GCFR (Moore, Verboom, & Forest, 2010;Zhigila et al., 2020). Thesium is a lineage of hemiparasitic annual herbs, perennial shrubs to shrublets or small trees ( Figure 1) which typically inhabits sandstone mountain fynbos, quartzsilcrete pebble patches, shaly Renosterveld scrub, coastal sands, and rarely thicket. Thesium species survive fire both by resprouting and reseeding. They possess scale-like, spine, succulent, or terete modifications to their leaves, have autogamous or zoophilous small flowers, and dispersal of their nut-like fruit includes myrmecochory ( Figure 1; Zhigila et al., 2020). These ecophysiological morphological traits are critical in the reproduction, dispersal, and interaction of species in their environment (Liu, Xu, He, Santiago, & Yang, 2019).
Firstly, we assessed the correlation between species range size and environmental variables and tested whether the range extent of ecological specialists and narrow-range species are more susceptible to contraction than those of generalists due to climate change. Secondly, we tested the hypothesis that specialism and climate change sensitivity are phylogenetically structured such that species losses are likely to produce a disproportionate loss of phylogenetic diversity. Finally, species specialism index and projected range extents were used to re-evaluate the conservation status of Thesium species. The approach followed involves (i) generation of geospatial data in a Geographical Information System (GIS) framework to quantify species range extent and ecological specialism; (ii) development of distribution models for each species in the context of contemporary environmental variables, and assessing the impacts of projected climate change on the distribution ranges of these species by using these models to predict the distribution in 2080; and (iii) assessment of the relationship between climate change response to ecological specialism and present-day range extent.

| Species data
A database of specimen locality data for all GCFR Thesium species was compiled from several sources including (i) the public-domain Botanical Database of Southern Africa (BODATSA) managed by the South African National Biodiversity Institute, Pretoria (SANBI, 2020); (ii) species occurrence records deposited at the Global Biodiversity Information Facility websites (GBIF, 2021); (iii) records from South African and international herbaria, not currently captured in BODATSA or GBIF, but having significant holdings of Cape plants (e.g., BOL, FHI, FHO, GRA, K, MO, NBG, OFX, PH, PRE and S; acronyms following Thiers, 2023); and (iv) our own field records. These locality data were converted to coordinate data by georeferencing all specimens. At the geo-referencing stage, the coordinates of each specimen were given a precision code, making it possible to select specimen subsets based on georeference precision. Doubtful records were projected against ArcGIS Desktop version 10.8.1 (2020), and unverifiable species points were excluded accordingly. Since Thesium species are notoriously difficult to identify (Hill, 1925), misidentification of Thesium is a problem in many herbarium collections (Zhigila et al., 2020). All specimens were therefore properly vetted against the type specimens prior to use. Since several studies (e.g., Escobar, Qiao, Cabello, & Peterson, 2018;Pearson et al., 2006) have shown that the accuracy of SDMs is distorted by the quality of data (Fernandes, Scherrer, & Guisan, 2019) and small sample sizes, SDMs were built only for species with >15 occurrence data points following van Proosdij, Sosef, Wieringa, and Raes (2016). For species with <20 locality points, species-specific model settings were used (Radomski et al., 2022). Of the 5510 initial specimen locality data collated, only 3576 specimens representing 101 GCFR species were ultimately used for SDM development (Table S1).

| Selection of environmental variables
For the purpose of niche quantification and SDM development, we downloaded raster layers (30 00 arc spatial resolution) describing 19 biologically relevant climatic variables from Worldclim (2019) and elevation records from the National Elevation Dataset (2002). In addition, we obtained layers describing eight soil variables (Cramer, Wootton, van Mazijk, & Verboom, 2019), giving a total of 28 environmental variables ( Figure S1). Since strong multicollinearity among environmental variables can cause SDM overfitting and so compromise model accuracy (Feng, Park, Ye Liang, Pandey, & Papeş, 2019), a multicollinearity test was performed using Pearson's correlation dimension-reduction methods (Dormann et al., 2013;Pearson et al., 2006) to determine the strongest environmental predictors of species range extents. For this purpose, a threshold of r > 0.7 (Dormann et al., 2013) was used to identify sets of highly correlated variables, with only one variable from each set being retained F I G U R E 1 Diversity in growth form, habitat, and morphology of the South African Thesium (a) a mounded growth form in T. imbricatum; (b) T. fruticosum on sandstone Fynbos; (c) T. pycnanathum tall (up to 2.5 m tall), erect occurring along stream sides; (d) T. rhizomatum, a short (<8 cm tall) plant on alluvial soil; (e) T. carinatum, < 50 cm tall, virgate branching pattern, mostly on sandstonederived soil, (f) T. fragile, occurring on beach sand; (g) T. quartzicolum on quartz-silcrete soil. Photos (a) and (c) by Nick Helme and Charles H. Stirton, respectively, and the remainder by Daniel A. Zhigila. for SDM development ( Figure S1). Ultimately, 14 noncollinear variables, including six soil variables, seven climatic variables, and elevation (Table 1; Figure S1), were retained. Presumably, elevation and soil data will remain relatively constant up to the year 2080 and therefore these contemporary data were used to perform the habitat suitability tests over time (Lourenço-de-Moraes et al., 2019).

| Quantifying species range extent and ecological specialization
To quantify species' range sizes, we first derived a minimum convex hull polygon for each species using the georeferenced specimen data, taking the area enclosed by the polygon as a measure of range size. Although convex hull polygons have been widely used to estimate species' range sizes, the method has been criticized for its failure to accommodate irregularities in the shapes of species distributions and for its sensitivity to sampling gaps (Burgman & Fox, 2003;IUCN, 2022;Preuss, Low, Cassel-Lundhagen, & Berggren, 2014). We therefore employed an additional metric, the maximum pairwise distance between specimen localities (i.e., α-hull), as a measure of range extent. Convex hulls and maximum pairwise distances were determined using GeoRange function in the raster version 2.9-5 (Hijmans, 2019) and dismo version 1.1-4 (Hijmans, 2017) packages for R (R Core Team, 2019). Then, the change in range size for each species was simply quantified as the cell frequencies occupied on the raster layer of the predicted range (probability of presence >0.5) at the LGM, present and future.
To quantify the degree of ecological specialization, we first used the raster package to query the 14 environmental layers using the specimen localities for each species. The degree of specialism in each variable was determined as the standard deviation of the values obtained across the set of occurrence localities (i.e., a small standard deviation implies strong specialism). Then, following Júnior and Nobrega (2018), the z-standardized, variable-specific standard deviations were subjected to principal component analysis (PCA), as implemented in R, to derive an overall index of specialism (i.e., the first principal component). Linear regression was used to test the relationship of ecological specialism to range extent, the latter expressed both as convex hull area and maximum pairwise distance.

| Species distribution modeling
To test whether range sizes of narrow-range and/or ecological-specialist species are more susceptible to contraction under predicted climate change scenarios than widespread and/or generalist species, we generated SDMs for each Thesium species. These SDMs were in the context of the present-day environmental layers. We used these to predict the distribution ranges: (i) in the present day; (ii) at the Last Glacial Maximum (LGM); and (iii) under the emissions scenario for the year 2080 (RCP 8.5;IPCC, 2015). For this purpose, model development was done using MaxEnt (Elith et al., 2011;Phillips, Anderson, & Schapire, 2006) as implemented in dismo package version 1.1-4 for R (Hijmans, 2017(Hijmans, , 2019. For the purpose of testing model performance, 75% of the occurrence data for each species were used to train the algorithm, the remaining 25% was reserved for model evaluation. Model quality was evaluated using the area under the receiver operating characteristic curve (AUC) values, with AUC < 0.8 being considered random, AUC >0.8 and <0.9 being considered good, and AUC > 0.9 excellent (Fielding & Bell, 1997). There are several debates regarding the limitations of the use of AUC to assess model performance (e.g., Escobar et al., 2018). However, many authors argued that the use of MaxEnt model for the SDM takes such limitations into consideration (e.g., Galante et al., 2018;Radomski et al., 2022;Sheth et al., 2014) and AUC can be used if a threshold of >0.5 via jack-knifing is adopted (Escobar, Qiao, Cabello, & Peterson, 2018;Hijmans, 2019). Further, for improved model performance, we employed a null model test (van Proosdij et al., 2016).
To compare the contemporary ranges of species with their projected ranges at the LGM and in 2080, we quantified for each species the absolute change in an area identified as environmentally suitable at the probability threshold of p > 0.5 (i.e., species with ≤0.5 not suitable while ≥0.5 suitable), as well as the percentage of range loss or expansion.

| Phylogeny and phylogenetic signal
To assess the phylogenetic signal in ecological specialism and extinction risk, a dated phylogenetic hypothesis was F I G U R E 2 Histograms illustrating the cross-species distributions of standard deviations of (a) Bio16 precipitation of the wettest quarter, (as an example of rainfall), (b) Bio 03 isothermality (as an example of temperature variable), (c) soil pH (as an example of an edaphic variable), (d) elevation, (e) species geographic range size (km 2 ), and (f) values for the first principal component as an index of specialization. generated for Thesium using the molecular data matrix of Zhigila et al. (2020). Using the Bayesian log-normal uncorrelated relaxed clock model in BEAST v1.10.4 (Suchard et al., 2018), a dated molecular phylogeny was obtained. The programme BEAUti v1.10.4  was used to obtain the input file for BEAST. The (GTR + I + G) sequence evolutionary and the Yule speciation models (Gernhard, 2008) were used as the tree priors. A single age prior estimation for the divergence between Buckleya and Thesium was used for the analysis. A normally distributed calibration mean age of 73.7 million years ago (Mya) with a standard deviation of 7.00 million years was used based on the secondary calibration of Der and Nickrent (2008) and Moore et al. (2010). The resulting tree was visualized in Tracer v1.10.4 , with 20% burnin tree discarded prior to generating the maximum clade credibility tree. The resulting chronogram was imported into R to trim outgroups, the non-GCFR species and duplicate accessions to match the species names of the spatial data matrix.
To test the hypothesis that range size, ecological specialization, and extinction risk are phylogenetically structured (e.g., species at greatest risk are closely related) and, therefore, the extent to which future climate change might influence the overall phylogenetic diversity, we quantified phylogenetic signal in these traits. The phylosignal function in the picante package (Harmon et al., 2019;Keck, Rimet, Bouchez, & Franc, 2016) was used in R to calculate these statistics.

| Niche breadth and range size
The majority of the GCFR Thesium species show intermediate specialism in most environmental variables, resulting in most variables showing bell-shaped frequency distribution across the species set (Figure 2a-d).
Precipitation of the wettest quarter representing rainfall variables displayed a bell-shaped spread (Figure 2a). This indicates reduced geographical range size as precipitation increases above a certain threshold value. A similar curve was true for temperature (represented by isothermality, Figure 2b) and soil (represented by pH, Figure 2c). In terms of elevational niche, the range was normally F I G U R E 3 Principal component analysis (PCA) plot applied to the GCFR Thesium species-specific deviations from the original means of environmental correlates. The ordination scores for each species are represented by black dots. Post hoc variable vectors as determined by the PCA axis loadings with bioclimatic, elevation, and soil data are indicated in red, green and magenta, respectively. distributed across taxa, with most species having a standard deviation of <400 m above sea level (Figure 2d). Most species lay between being strongly specialist (<100 m) and highly indiscriminate (>800 m). The temperature and rainfall variables corresponded with the elevation range in the GCFR mountainous areas at which there was relatively high rainfall and reduced temperature. The derived specialism index showed a skewed right distribution with few species showing strong specialization. Values for the first principal component as an index of specialism, show an increasing degree of specialization from left to right (Figure 2e). The range size values indicated that the standard deviations of the geographic distribution coverage of the GCFR Thesium vary from 2.5 to 1200 km 2 with ca. 80 species having ranges <700 km 2 (Figure 2f). The curve became stable at >700 km 2 indicating an increase in geographical range size for about 20 species.
Plots of individual Thesium species against the first two axes extracted by a PCA based on the species-specific standard deviations of the 14 environmental variables are presented in Figure 3 and Table S2. Like most of the baseline variables, the PCA-derived specialism index shows a bell-shaped distribution albeit with a slight negative skew. Respectively, PC1 and PC2 capture 31.6% and 16.2% of the total variance. Since all variables associate positively with PC1, we interpret the latter as an index of specialism (niche breadth). Species with low PC1 score indicates strong ecological specialism and high PC1 scores weak ecological specialism. In contrast to PC1, PC2 had positive loadings for soil variables and negative loadings for elevation and climatic variables (Figure 3; Table S3). Hence, PC2 describes the relative importance of soil as opposed to climatic variables underpinning specialism. For details of PCA loadings and scores, descriptions of environmental correlates, standardized data matrix, and species names, see Tables S2-S5).
Our data show that there is a strong positive relationship between specialism and range extent, but there is invariably a high unexplained residual variance (Figure 4). Consequently, the PCA-derived specialism index is also associated positively with range size. The same is true for the environmental variables which show positive correlations with the species' range size ( Figure S2). However, three variables (maximum temperature of warmest month, soil K, soil total N), were unrelated to range size ( Figure S2C,K,L).

| Species distribution modeling and prediction
The AUC values for the predicted species distribution were high, having an average value of 0.9 (Table S6).
Such high values demonstrate a reasonable predictive performance of the SDM models. In addition, the omission rate increased linearly with occurrence probability ( Figure S3). Further, the relatively low values between the omission rate and the predictive probability of occurrence support the preference and suitability of the model.
On the evidence of SDMs, Thesium species showed a diversity of range size responses to climate change, this being apparent in comparisons of predicted range size both between the LGM and the present, and between the present and 2080 ( Figure 5; Table S6). There is no correlation between the predicted range size change of the past and the present-day range size or ecological specialism (Figure 5a,b). However, there is a negative relationship between the nature of range change between the LGM and the present, and of that between the present and 2080. This implies a tendency for species whose ranges expanded since the LGM to contract in the future, and vice versa (Figure 5c,d). Of the 101 species modeled, F I G U R E 4 Assessment of correlation between the average range extent of individual species (maximum distance [m]) and specialism. Values obtained from PC1 were used as the index for specialization to evaluate the relationship between species range extent (m) and ecological specialization. The blue line indicates the best linear fit for each species and the gray band is the 95% confidence level. The square of the correlation coefficient (r 2 ) and the corresponding p value is included in the top left corner of the panel. 71 (70%) are inferred to have undergone range contractions since the LGM, the balance showing range expansions (Table S6). Comparing species ranges between the present and 2080, 37 species (ca 37%) are predicted to undergo range expansions, 51 (50%) range contractions, and 10 (8.5%) are predicted to maintain their geographical range sizes.
For most species of GCFR Thesium, distribution models identified climatic variables as better predictors of habitat suitability than soil and elevation variables (Table S6). Although there is a continuum in the relative influence of climatic vs edaphic variables as determinants of distribution, we identified 84 species as having primarily climatically driven distributions with climatic variables having a 45%-91% estimated contribution. For instance, T. capitellatum had Bio 2 (mean diurnal range) and Bio 16 (precipitation of wettest quarter) as the main correlates of habitat suitability. Eighteen species had primarily edaphically driven distributions with strong edaphic influence as observed in for example T. gnidiaceum (mostly pH and N), with 36%-40% soil variable contributions. However, the remaining soil variables contributed to only one species each. The distributions of 15 species (with 42%-60% relative contributions) were predominantly driven by elevation. Of the 15, 11 were mostly influenced by low elevation (mean elevation <200 m a.s.l), for example, T. rufescens and four species had high elevation as the main distribution driver (mean elevation >1000 m a.s.l), for example, T. annulatum.
The three climate change scenarios yielded different distribution ranges for Thesium species. The SDMs predicted several changes in range extent between the three-time slices, as illustrated using three exemplar species (Figure 6a-c). The predicted habitats of the GCFR Thesium species with climatically determined ranges attained their maximum range extents in the LGM, (ca. 22,000 years ago) e.g., T. strictum as opposed to specialists (Table S6). The model predicted a range reduction of 51 species from the LGM to the present and will continue in range reduction into 2080. In contrast, the present and future scenarios predicted range expansions for 37 species. In some species such as T. namaquense (Figure 6b), the model predicted a marked increase of suitable geographical ranges toward the northern regions of the GCFR. This increase in range size indicates F I G U R E 5 Correlation between (a) log-transformed changes in range extent (km 2 ) and present to future range change (%), (b) specialism index (PC1) and range change from the present to future (%). Predicted range expansion or contraction difference from (c) the past to future in relation to the present to future (2080) expressed in log-transformed percentages and (d) absolute range change values from the past to present and the present to future. Each dot represents a species. The square of the correlation coefficient (r 2 ) and the corresponding p values are included at the bottom right corner of the panel.
resilience to environmental change. Note that, 24 species are expected to shift their geographical ranges. For example, T. spinosum is predicted to have shifted and will potentially shift further from the eastern coast to the western and toward the northern coasts under climate change (Figure 6c).

| Phylogenetic signal
Mapping species range extents, ecological specialization, and extinction risk on the MCC tree reveal a lack of phylogenetic structure (Figure 7a-d). These indicate that range size, specialism and extinction rate are evolutionarily labile and highly homoplasious in the GCFR Thesium.

| DISCUSSION
Based on the contemporary locality points and environmental conditions, range extents of GCFR Thesium species were assessed (Figure 2). The species range extent was normally distributed, with most GCFR Thesium species having a narrow range size of less than 400 km 2 . Only a few species are wide-ranging, with range sizes up to 900 km 2 . Species with a range extent of less than 400 km 2 were also specialists and predictably confined to areas of high endemism. Meanwhile, an increase in species range size indicates generalization and such species occupy a wide range of habitats (Barret, 2013). Such range size distributions have been observed in several plant and animal taxa globally (Davies et al., 2011;Rocha-Ortega, Rodriguez, Bried, Abbott, & Cordoba-Aguilar, 2020). Consistent with the F I G U R E 6 Predicted distribution ranges for exemplar species (i) backward projections to the Last Glacial Maxima (LGM, past), (ii) under contemporary conditions (present) and (iii) in the future. The predictions were based on species-specific MaxEnt models. The green color indicates areas with the highest suitability probability grading to lowest grayish white areas. The threshold probability of occurrence being <0.5 equals 0 (absent), while >0.5 to 1 equals 1 (present). Circles are the actual species occurrence localities plotted based on the present-day model predictions.
F I G U R E 7 Phylogenetic signal mapped on a Maximum Clade Credibility (MCC) phylogenetic tree adopted from Zhigila et al. (2020). (a) range extent (maximum distance [km]); (b) first principal component values as specialization index; and (c) estimate of cell frequency of species range change from past (LGM) to present (contemporary to 2080); (d) estimate of cell frequency of species range change from present to the future (year 2080). The phylogenetic tree was obtained from a concatenated nuclear (ITS) and three plastids (trnL-F+matK +rpL32-trnF) markers. Phylogenetic signals for ecological specialists (in red) grading to generalists (in blue) are indicated by clades and tip colors (note the color ramp for interpretation of patterns). specialization-disturbance theory, which is largely supported in ecology, conservation and evolution literature (e.g., Vazquez & Simberloff, 2002), the model demonstrated a strong positive linear correlation between the range extents and inferred ecological specialism of species, with generalists being generally more widespread than specialists (Figure 4; Figure S2). However, Vamosi, Armbruster, and Renner (2014) suggested that a species might be a generalist under one perspective, but a specialist under another, depending on the influence of the evolutionary and environmental trade-offs of a species (Chichorro, Juslen, & Cardoso, 2019;Davies et al., 2011). Yessoufou et al. (2016) reported that species range sizes correlate with their suitable environmental conditions. Therefore, it follows that environmental heterogeneity becomes a key trait in the evolution of ecological specialization (Vamosi et al., 2014).
The main environmental determinants of species range size in Thesium are mean diurnal range, maximum temperature of warmest month, precipitation of wettest quarter, elevation, soil pH and soil nitrogen (Figure 2) with variations among species. Consistent with previous studies, these variables are considered the most important biological drivers of habitat suitability across GCFR species (Skowno, Raimonodo, et al., 2019) such as Protea L. (Midgley, Hannah, Millara, Thuiller, & Booth, 2003;Skowno, Poole, et al., 2019) and Rooibos tea (Lotter & le Maitre, 2014). Range contractions are more likely to occur in narrow-ranged species (Skowno, Poole, et al., 2019), especially the climate specialists (Reed & Tosh, 2019), while wide-ranging species would probably experience a range expansion (Willis et al., 2010 ;  Table S6) or large-scale death of species, for example, Widdringtonia cedarbergensis J.A.Marsh (Skowno, Poole, et al., 2019). Similar range expansion and contraction in 28 species of GCFR Protea were reported in the GCFR (Midgley et al., 2003;Skowno, Poole, et al., 2019). Empirically, our data show that there is a relationship between inferred specialism and range extent, but that there is invariably a high unexplained residual variance (see Figure 4; r 2 -values). The mismatch between range extent and specialism in the Cape, where environmental gradients are invariably steep (Cramer et al., 2019;Harris et al., 2019) is as expected. It is unsurprising for species to be very local in distribution while at the same time spanning a range of conditions.
According to the SDM predictions, Thesium suitable habitats, particularly around the southwest up to southeast, have been present since the LGM (Figure 6). The exception was the Karoo Mountains and its fringes, which lacked suitable habitat during the LGM. However, the habitats of some species, for example, T. namaquense, persists and are projected to expand their distribution range into the hinterlands in future (Figure 6c). There was a strong negative relationship between the nature of range change between the LGM and the present, and that between the present and 2080; implying a tendency for some species whose ranges have contracted since the LGM to expand in the future, and vice versa. With these fluctuations in range size over time, some species will potentially respond negatively to change scenarios, while others benefit or are not affected.
Our analyses forecast varying responses of Thesium species to the increase in climate change ( Figure 6). The modeled range sizes of 51 species are expected to contract (Table S6). As such, climate change is a significant risk to the persistence of these species. At the global scale, several studies predict a substantial species range change due to the rapidity of global warming (e.g., Alaniz, Pérez-Quezada, Gallegiollos, V asquez, & Keith, 2019;Weiskopf et al., 2020;Williams & Blois, 2018). At a local scale (the GCFR), the range extent of the Fynbos Biome is predicted to decline between 51% and 65% under different climate change scenarios (Lotter & le Maitre, 2014). With the GCFR regional drying and warming (Altwegg, West, Gillson, & Midgley, 2014), particularly in the western area (Midgley et al., 2003), rainfall is projected to decline (Altwegg et al., 2014) and 23% of its species would probably be at risk of extinction (Lotter & Maitre 2014). Conversely, the 50 species with projected range maintenance or expansion are likely climate change-tolerant species. The range expansion and contraction projected under different climate scenarios in this study corroborate patterns demonstrated for other Cape plant lineages, including Diastella, Leucadendron, Leucospermum, Protea and Serruria (Midgley et al., 2003), Aspalathus (Lotter & le Maitre, 2014) and the CFR endemic plants (Hoveka et al., 2022;Lenoir, Gegout, Marquet, de Ruffray, & Brisse, 2008). However, these results should be interpreted cautiously given that the predicted range contractions do not necessarily translate into extinction (Midgley et al., 2003;Zhang et al., 2017). Also, modeled range reduction, even from accurate data of environmental variables of species, is probabilistic (Midgley et al., 2003;Naimi & Araújo, 2016). Moreover, our models did not incorporate other factors that influence fynbos systems such as fire regimes (typical of Fynbos dynamics; Bradshaw & Cowling, 2014). Also, resprouters persist and regenerate after fire, but given the drier, hotter conditions that foster higher fire frequencies, climate change may prove problematic particularly to reseeders (Cowling et al., 2015). In addition, land use has had a profound negative impact on habitat suitability in the GCFR flora (Raimondo et al., 2009). The GCFR Thesium species seldom occur on cultivated farmlands, residential areas or industrial sites (Zhigila, pers. obs). We hypothesize that the lack of occurrence in such disturbed sites could be linked to disruption of community dynamics, including the disruption of host-hemiparasite networks at critical stages of establishments, on geographical distribution and ecology of Thesium species. In scenarios such as altered sites, species could shift ranges to habitats at higher altitudes or rugged terrains where the impacts of land use are limited.
In the context of molecular phylogeny, we assessed whether range extent, ecological specialization, and extinction risk show phylogenetic signal and it was asked whether extinction risk will possibly prune Thesium phylogenetic diversity. Low phylogenetic signal was observed (Figure 7). Suggesting that climate change will not result in a disproportionate loss of Thesium phylogenetic diversity. This contradicts the previous studies, for example, Willis, Moat, and Paton (2003) Eiserhardt et al. (2015), and Suhonen, Ilvonen, Korkeamäki, Nokkala, and Salmela (2022) who postulate that climate change will result to loss of phylogenetic diversity. However, the labile phylogenetic structure observed in the GCFR Thesium might be due to evolutionary convergence in the deep nodes of the phylogenetic tree compared to individual lineages (Felsenstein, 1985).
What are the implications of the realized species geographical extents for conservation of the GCFR Thesium? Of the 65 species currently ranked as Least Concern on the South African Red List (Raimondo et al., 2009), more than 24 species will likely shift into higher extinction risk categories. These findings reiterate the South African National strategy call for Thesium as a priority plant taxon for taxonomic and conservation study (Victor, Smith, & van Wyk, 2015), as nearly 30% of the species have uncertain conservation status and are likely to be threatened (Von Staden, 2015). For effective conservation planning, it is necessary to understand the distribution ranges of species and their relationship with abiotic and other biotic variables (Soultan, Wikelski, & Safi, 2019;Villero et al., 2017). Our findings present several important outcomes for the local and regional conservation planning of Thesium and as a proxy to other GCFR biota.
Range-restricted species have been shown to be more vulnerable to range loss than widespread species (Davies et al., 2011;Reed & Tosh, 2019;Yessoufou et al., 2012). Since ecological specialists are mostly endemics, any extinction drivers associated with these species conveys a significant probability in the loss of important local and global biodiversity (Wiens, 2016). Therefore, a high concentration of specialists and by implication endemics in each biome implies that many species will be affected in the event of any single threat (Raimondo et al., 2009). Although labile on the diversity of Thesium, the extirpation of range-restricted ecological specialists from a biome may impact phylogenetic diversity, pruning the tree of life (Eiserhardt et al., 2015;Reed & Tosh, 2019;Vamosi et al., 2014). This will ultimately affect the stability of the species operational ecosystem (Clavel et al., 2011;Zhang et al., 2017). Therefore, we reiterate that narrow-ranged species are worth conservation priority. Also, identifying threatened habitats with a high concentration of ecological specialists and making decisive efforts to protect them is an important conservation action that should be taken (Bland et al., 2019).

| CONCLUSION
The GCFR is the center of Thesium diversity, but the driving factors behind the rapid diversification and distribution of the genus are poorly understood. In the context of the GIS data used in this study, it was revealed that the GCFR Thesium species show intermediate levels of specialism for most environmental variables and the multivariate specialism index. The number of extreme generalists and extreme specialists is relatively small with most species being moderately range restricted. In addition, although there was a positive relationship between ecological specialism and range extent, there was invariably a high unexplained residual variance. The observed mismatch between range extent and specialism particularly in the Cape can be linked to the invariably steep environmental gradient. Hence, it is possible for species to be highly localized in distribution while at the same time spanning a range of conditions.
Range shifts between the LGM and the present and between the present and 2080 vary widely between species. where several species that underwent contractions over the last period are predicted to increase under climate change, and those that have been expanding are now predicted to contract. Absolute changes in range size between both the LGM and the present, or between the present and future (2080) are not related to the current range extent or ecological specialism. Absolute and percent range change between the LGM and present, and between the present and 2080, are negatively related. In other words, ca. 50% of species whose ranges have increased since the LGM are predicted to suffer range decline in future, implying a conservation concern. As inferred on the basis of range contractions and range shifts, there is a relationship between specialism and extinction risk, posing a conservation concern. Hence, more conservation efforts are recommended for the coastal regions of the northwest, southwest, and southeast of the GCFR as several Thesium species are predicted to lose their ranges in these regions due to climate change.
Further, because habitats of many Thesium species are threatened by both accelerating climate change and anthropogenic activities such as excessive farming, a multifaceted approach is required for conserving the species both in situ as well as beyond their natural habitats. Further, the predicted reduction in range size of 51 Thesium species implies probable shifts of the conservation status of these species into higher extinction risk categories. South Africa has conserved extensive areas of nature reserves and contributed to environmental education and awareness of both local farmers and conservationists. However, ex-situ conservation efforts to conserve the Cape flora, particularly Thesium species remain elusive. Conservation efforts to protect the GCFR being one of the world's Biodiversity Hotspots are needed.

ACKNOWLEDGMENTS
We thank the curators of BOL, FHI, FHO, K, NBG (including SAM and STE), OXF, and PRE for allowing access and loan of materials. The field work was supported by the South African National Research Foundation (NRF Grant 93559-Quartz fields of Southern Africa) and the Science Faculty, University of Cape Town. We thank the Cape Nature for permission to collect plants in the GCFR (permit number CN35-28-5831; CHS and CN35-28-17379; DAZ). Support from the African-German Network of Excellence in Science is acknowledged. We appreciate the anonymous reviewers and the Conservation Science and Practice Editors for their tireless comments and suggestions which have greatly improved the manuscript.

CONFLICT OF INTEREST STATEMENT
The authors declare no conflict.

DATA AVAILABILITY STATEMENT
Herbarium specimens collected for this study are currently available at Bolus Herbarium, University of Cape Town, South Africa, and duplicates of such specimens deposited at NBG, PRE, and K. DNA sequences related to the phylogenetic reconstruction are available on GenBank: https://www.ncbi.nlm.nih.gov/genbank/ Thesium.

SUPPORTING INFORMATION
Additional supporting information can be found online in the Supporting Information section at the end of this article.