Climatic and edaphic niche shifts during plant radiation in the Mediterranean biodiversity hotspot

Abstract

Background and Aims

Ecological speciation is frequently invoked as a driver of plant radiation, but the behaviour of environmental niches during radiation is contentious, with patterns ranging from niche conservatism to niche divergence. Here, we investigated climatic and edaphic niche shifts during radiation in a western Mediterranean lineage of the genus Linaria (Plantaginaceae).

Methods

Detailed distributional, phylogenomic and environmental data were integrated to analyse changes in climatic and edaphic niches in a spatiotemporal context, including calculation of niche overlap, niche equivalency and similarity tests, maximum entropy modelling, phylogenetic comparative methods and biogeographical analyses.

Key Results

Active divergence of climatic and edaphic niches within a limited subset of available conditions was detected among the eight study species and particularly between sister species. Speciation and niche divergence are estimated to have happened in the southern Iberian Peninsula in Mediterranean conditions, followed by waxing and waning of distribution ranges resulting from the Quaternary climatic cycles.

Conclusions

The results support the idea that the prevalence of niche conservatism or niche divergence patterns is a matter of phylogenetic scale. Habitat isolation pertaining to both climatic and soil conditions appears to have played a role in plant speciation in the western Mediterranean biodiversity hotspot, most probably in combination with pollinator isolation and some degree of geographical isolation. These findings are in agreement with an adaptive radiation scenario incorporating certain non-adaptive features.

INTRODUCTION

Evolutionary radiations involving rapid diversification from a common ancestor are excellent systems for the investigation of evolution from multidisciplinary perspectives (Naciri and Linder, 2020; Schluter, 2000). These include phylogenomics, phylogenetic comparative approaches, evolutionary developmental biology and ecological (biotic and abiotic) interactions, among others. In particular, ecological differentiation is frequently invoked as a driver of radiations by means of ecological speciation, i.e. the process by which barriers to gene flow between populations arise as a result of ecologically based divergent selection (Rundle and Nosil, 2005; Rundell and Price, 2009; Hernández‐Hernández et al., 2021; Matsubayashi and Yamaguchi, 2022). However, the behaviour of abiotic environmental niches (or Grinnellian niches, i.e. the set of ‘scenopoetic’ variables that are not affected by the focal species; Soberón, 2007) during radiation is still a contentious topic in ecology and evolution.

The movement of lineages through environmental niche space in the course of diversification is thought to be governed broadly by the process of phylogenetic niche conservatism, i.e. the tendency of populations to maintain their current niche through time (Pyron et al., 2015). According to the conceptual framework of Pyron et al. (2015), the phylogenetic niche conservatism process can lead to a diverse range of evolutionary patterns depending on the geographical and environmental setting of speciation, from a pattern of niche conservatism, in which closely related species display highly overlapping niches, to one of niche divergence, in which closely related species have non-overlapping niches (Rundle et al., 2000; Wiens, 2004; Wiens and Graham, 2005; Holt, 2009; Ahmadzadeh et al., 2013; Anacker and Strauss, 2014; Hiller et al., 2019). To achieve a more complete understanding of evolutionary radiations, it is necessary to characterize niche conservatism and divergence patterns, and the factors underlying them, by using robust methodological approaches and high-quality data. This endeavour has become increasingly possible thanks to the growing availability of high-throughput sequencing and phylogenomic techniques (capable of resolving evolutionary relationships in rapidly evolving clades; McCormack et al., 2013), detailed occurrence datasets (derived from biodiversity collections and databases; Ball-Damerow et al., 2019), worldwide, standardized environmental data (Hengl et al., 2017; Karger et al., 2017) and robust statistical approaches (Warren et al., 2008; Broennimann et al., 2012).

In plants, two major determinants of environmental niches are climate and soil, and they can both act as pre-pollination barriers to gene flow driving ecological speciation, although their relative roles can vary across biomes and biogeographical regions (Rajakaruna, 2004, 2018; Givnish, 2010; Anacker and Strauss, 2014). For example, adaptation to particular soil conditions is thought to play a key role as a driver of plant diversification in mediterranean-type regions, in combination with summer-dry climatic conditions (Rundel et al., 2016). In particular, a recent study of the whole endemic flora of the Iberian Peninsula (western Mediterranean Basin) indicated that soil conditions significantly explain the spatial distribution of neoendemism, with recently evolved endemic species arising primarily in areas with high soil pH (Buira et al., 2020). To gain a deeper understanding of the mechanisms behind this large-scale association between soil conditions and recent diversification, and of the additional role of climatic conditions, detailed analyses of particular plant clades are necessary (e.g. de Luis et al., 2018; Castro et al., 2019, 2020; Alonso et al., 2022; Otero et al., 2022). Although a role of climatic and edaphic conditions as drivers of recent diversification of Mediterranean plants has been widely hypothesized (Rundel et al., 2016; Thompson, 2020; Kadereit and Abbott, 2021), there is a scarcity of case studies investigating both climatic and edaphic niche shifts during diversification of Mediterranean plants using robust phylogenetic and distributional data, in combination with cutting-edge analytical approaches (e.g. López‐Jurado et al., 2019; Oberprieler et al., 2023).

Linaria sect. Versicolores (Antirrhineae, Plantaginaceae) is a clade of ~30 species and subspecies of toadflaxes distributed mainly in the western Mediterranean Region (Sutton, 1988; Fernández-Mazuecos et al., 2013). They are characterized by their specialized flowers with personate, spurred corollas; wingless seeds with no obvious capabilities for long-distance dispersal; and bifid styles or emarginate stigmas (Sutton, 1988). Within this section, the Iberian clade of Linaria subsect. Versicolores includes eight well-defined toadflax species endemic to the Iberian Peninsula (except for the subendemic Linaria spartea, also found in southern France) that radiated between the late Pliocene and the Pleistocene (Fig. 1A) (Fernández-Mazuecos and Vargas, 2011, 2015; Fernández-Mazuecos et al., 2013, 2018a, 2018b, 2019). Because of its remarkable characteristics, speciation in this group is being investigated using a multidisciplinary approach integrating phylogenomics, ecology and evolutionary developmental biology. To date, the focus has been on pollinator-driven floral evolution and, in particular, on the role of the nectar spur, considered a key innovation that promotes diversification in flowering plants (Fernández-Mazuecos and Glover, 2017; Cullen et al., 2018; Fernández-Mazuecos et al., 2018b, 2019). Indeed, Fernández-Mazuecos et al. (2018b) found that pairs of sister species in this clade display divergent floral morphologies, which supports a critical role of pollinator-driven speciation (van der Niet and Johnson, 2012; Phillips et al., 2020). In addition, the clade is also an ideal system in which to investigate the role of climatic and edaphic niche shifts during radiation by taking advantage of the well-grounded taxonomy, robust phylogenetic framework (based on genome-wide sequence data), availability of reliable occurrence data, and wide variation in ecological requirements and geographical range size (i.e. from narrow endemics to a species widely distributed in the Iberian Peninsula) (Sáez, 2009; Fernández-Mazuecos et al., 2018a, 2018b). In fact, a role of soil conditions in lineage divergence has been hypothesized for this clade (Fernández-Mazuecos and Vargas, 2015), but has not been tested explicitly to date. Likewise, changing climatic conditions during the Quaternary are thought to have played an important role in speciation of Iberian plant lineages, including Linaria (Abellán and Svenning, 2014; Blanco-Pastor et al., 2015). Placing ecological shifts and speciation events within the biogeographical context of Iberian areas of plant endemism (Buira et al., 2017) can provide a more comprehensive view of evolutionary patterns and processes during radiation.

Fig. 1.

Phylogenetic relationships, environmental niches and geographical distributions of the eight species of the Iberian clade of Linaria subsect. Versicolores. (A) Phylogenetic relationships based on coalescent analysis of genome-wide genotyping-by-sequencing data (Fernández-Mazuecos et al., 2018b), shown alongside floral morphologies. (B) Climatic niches along the first two axes of principal component analysis (PCA-env) based on six CHELSA variables. (C) Edaphic niches along the first two axes of PCA-env analysis based on six SoilGrids variables. (D) Filtered species occurrences used in the analyses, obtained from Fernández-Mazuecos et al. (2018a). In each PCA-env plot, grey shading indicates the density of occurrence of the species, and solid and dashed contour lines represent 100 % and 50 %, respectively, of the background environment (Iberian Peninsula and southern France). See Table 2 for variable descriptions and Fig. 2 for correlation circles and percentages of variation explained by PCA-env axes. Species acronyms are indicated.

Here, we investigated the effect of climatic and edaphic niche shifts on the rapid diversification of the Iberian clade of Linaria subsect. Versicolores as a case study to understand the environmental drivers of recent plant diversification in the Mediterranean Region. Our objectives were as follows: (1) to test niche conservatism vs. niche divergence scenarios in relationship to the biogeographical and environmental setting of speciation; (2) to evaluate the relative incidence of climatic vs. edaphic niche shifts in the course of plant radiation; and (3) to assess the effect of Quaternary climatic changes on isolation among species. Our hypothesis was that both climatic and edaphic niche divergence have occurred during speciation of recently diverged pairs of sister species, in association with shifts in floral morphology described in previous studies.

MATERIALS AND METHODS

Study species and occurrence data

For this study, we followed the taxonomic treatment of Fernández-Mazuecos et al. (2018a), which recognized eight species in the Iberian clade of Linaria subsect. Versicolores: L. algarviana (ALG), L. becerrae (BEC), L. clementei (CLE), L. incarnata (INC), L. onubensis (ONU), L. salzmannii (SAL), L. spartea (SPA) and L. viscosa (VIS) (see Table 1). An additional species of this group, L. bimaculata, has been proposed recently (Farminhão and Carapeto, 2024), but we considered it a synonym of L. spartea following the taxonomic criterion of Fernández-Mazuecos et al. (2018a). Occurrences of the eight study species covering their entire distribution ranges were obtained. For six species (ALG, INC, ONU, SAL, SPA and VIS), occurrences were taken from Fernández-Mazuecos et al. (2018a). To reduce spatial clustering of occurrences, and thus sampling bias, we used the ReduceSpatialClustering function from the R package sdmflow (Benito, 2021). A different minimum distance between occurrences was used for each species depending on distribution range size: 0.2° for SPA, 0.1° for INC and VIS, 0.05° for SAL and 0.008333° (the resolution of environmental layers; see below) for ALG and ONU. Occurrence data for CLE and BEC were newly collected in the field in spring 2014–2017 and filtered manually to leave a maximum of one occurrence per pixel of environmental layers. Although a small number of populations of SPA and VIS in eastern Spain are thought to be introduced (Sáez et al., 2000; Fernández-Mazuecos et al., 2018a), they were not excluded, because of the lack of clear evidence to distinguish introduced from native populations and because this small number of occurrences was unlikely to alter the results substantially.

Table 1.

Study species forming the Iberian clade of Linaria subsect. Versicolores, species acronyms and numbers of filtered occurrences (n) used in analyses of environmental niches.

Species	Acronym	n
L. algarviana Chav.	ALG	17
L. becerrae Blanca, Cueto & J.Fuentes	BEC	8
L. clementei Haens.	CLE	18
L. incarnata (Vent.) Spreng.	INC	51
L. onubensis Pau	ONU	5
L. salzmannii Boiss.	SAL	35
L. spartea (L.) Chaz.	SPA	146
L. viscosa (L.) Chaz.	VIS	34

Phylogenomic data

The most probable phylogenetic relationships among the eight species of the Iberian clade of Linaria subsect. Versicolores, considering available datasets, were obtained from Fernández-Mazuecos et al. (2018b) (Fig. 1A). In particular, we used a phylogeny obtained using coalescent-based (NJst) analysis of genome-wide genotyping-by-sequencing data. To obtain an ultrametric, time-calibrated tree suitable for model-based evolutionary reconstructions, we used the penalized likelihood method implemented in treePL (Smith and O’Meara, 2012). Three calibration points were used: (1) the crown node of Linaria sect. Versicolores; (2) the crown node of Linaria subsect. Versicolores; and (3) the crown node of the Iberian clade of Linaria subsect. Versicolores. Minimum and maximum ages for all three nodes were taken from each of three previous dating analyses in wider phylogenetic contexts, leading to three calibration schemes implemented in three separate dating analyses: cal1 (old divergence times; Fernández-Mazuecos et al., 2019); cal2 (intermediate divergence times; Fernández-Mazuecos and Vargas, 2011); and cal3 (recent divergence times; Fernández-Mazuecos et al., 2013) (see Supplementary Data Table S1). For each analysis, we first ran a ‘prime’ analysis to determine the best optimization parameters (opt, optad and optcvad). After setting these parameters, we ran a cross-validation analysis to determine the optimal smoothing parameter value. This value was then used in the final analysis. The ‘thorough’ option was used in all runs, and the number of iterations was 200 000 for penalized likelihood and 5000 for cross-validation.

Environmental data

Geographical information system (GIS) layers of present values for 19 bioclimatic variables at 30 arc-second resolution (Supplementary Data Table S2) were obtained from the CHELSA 1.2 database (https://chelsa-climate.org/; Karger et al., 2017) and clipped to the extent of the study region (36°–45°N; 10°–4°E) using ArcGIS 10 (Esri, Redlands CA, USA). To avoid collinearity, we followed a two-step procedure using the R packages raster (Hijmans and van Etten, 2020) and HH (Heiberger, 2020). First, we excluded variables displaying a high correlation coefficient with other variables (|r| > 0.7) in the study area (Dormann et al., 2013). Then, we excluded additional variables with a high variance inflation factor until all remaining variables had a variance inflation factor of less than ten. Layers of the same bioclimatic variables for past periods (see below) at 2.5 arc-minute resolution were obtained from the PaleoClim database (Brown et al., 2018). Likewise, GIS layers of ten edaphic variables (Supplementary Data Table S2) were obtained from the SoilGrids 2017 database (https://www.isric.org/explore/soilgrids/faq-soilgrids-2017; Hengl et al., 2017), downscaled to 30 arc-second resolution and clipped to the extent of the study region. To avoid collinearity, variables were filtered following the methods described above for bioclimatic variables.

An additional GIS layer mapping major lithological classes at 30 arc-second resolution (IberLit) was built using information from lithological maps of continental Spain (IGME, 2006), continental Portugal, Andorra and southern France (http://portal.onegeology.org/; Laxton et al., 2010). Nine lithological classes potentially influencing plant distributions were defined. Correspondence between lithological classes and dominant materials in the original maps is shown in Supplementary Data Table S3.

Analyses of environmental niches

Climatic and edaphic niches of the eight species of the study clade were evaluated using a PCA-env approach, i.e. a principal component analysis calibrated on the entire environmental space of the study area (Broennimann et al., 2012). Given that we were interested in evaluating the relative roles of climatic and edaphic niches in diversification and their potentially different evolutionary patterns, separate analyses were conducted using the selected CHELSA and SoilGrids variables, respectively, in combination with filtered occurrences of the eight species (see Tables 1 and 2). PCA-env analyses were implemented in the R package ecospat (Di Cola et al., 2017), and niches were visualized along the first two axes. Niche position and breadth (Thuiller et al., 2005) were calculated for each species. To that end, random pixels weighted by density were sampled within the niche of each species, and niche position and breadth were calculated as the median and variance, respectively, of the scores of these random pixels along the first two axes of the PCA-env. The number of random pixels sampled for each species depended on niche size (see Supplementary Data Table S4).

Table 2.

Variables used in analyses of environmental niches of the eight species of the Iberian clade of Linaria subsect. Versicolores, including climatic variables from the CHELSA database, edaphic variables from the SoilGrids database, and a lithological variable. Climatic and edaphic variables were selected after applying procedures to avoid collinearity.

Type	Acronym	Description	Source
Climate	ts	Temperature seasonality	CHELSA 1.2
	mtwq	Mean temperature of wettest quarter	CHELSA 1.2
	mtdq	Mean temperature of driest quarter	CHELSA 1.2
	mtcq	Mean temperature of coldest quarter	CHELSA 1.2
	pwq	Precipitation of wettest quarter	CHELSA 1.2
	pdq	Precipitation of driest quarter	CHELSA 1.2
Soil	vpcf	Volumetric percentage of coarse fragments (>2 mm)	SoilGrids 2017
	wpcp	Weight percentage of clay particles (<0.0002 mm)	SoilGrids 2017
	wpsp	Weight percentage of sand particles (0.05–2 mm)	SoilGrids 2017
	cec	Cation exchange capacity of soil	SoilGrids 2017
	soc	Soil organic carbon content	SoilGrids 2017
	ph	pH index measured in water solution	SoilGrids 2017
Lithology	IberLit	Lithological classes (see Supplementary Data Table S3)	This article

We also used ecospat to calculate pairwise values of niche overlap (Schoener’s D; Schoener, 1968) and to conduct tests of niche equivalency and niche similarity for all possible species pairs (Warren et al., 2008; Broennimann et al., 2012). These analyses were also conducted for climatic and edaphic niches separately. The niche equivalency test was used to evaluate niche divergence, i.e. whether the observed niche overlap between two species is significantly lower than a null simulated by randomly reallocating the occurrences of both species between their ranges (alternative=‘lower’ option). Therefore, significance of the niche equivalency test would support niche divergence. The niche similarity test was used to evaluate niche conservatism, i.e. whether the observed niche overlap between two species is significantly greater than that obtained if random shifts within the environmental space are allowed (alternative=‘greater’ option). Consequently, significance of the niche similarity test would support niche conservatism. All tests were based on 100 iterations. In similarity tests, both ranges were randomly shifted (rand.type = 1). Given that 28 tests of each type were conducted, adjusted P-values were calculated using the Benjamini–Hochberg (BH) correction (Benjamini and Hochberg, 1995).

We evaluated the lithological preferences of the eight species of the study clade by calculating the frequency of their filtered occurrences on substrates of each of the nine lithological classes of the IberLit dataset. Statistical differences were explored using Fisher’s exact test (Fisher, 1925) in R, with P-values calculated by Monte Carlo simulation and adjusted using the BH correction.

Divergence in climatic and edaphic niches was analysed further for the three pairs of sister species (BEC/CLE, VIS/ONU and SPA/INC) using the R package humboldt (Brown and Carnaval, 2019). Analyses in ecospat described above, considering the total environmental space of the study area, assume that species distribution ranges are at equilibrium and that the entire study area is accessible to them. These assumptions are relaxed in humboldt by incorporating the accessibility of environments into statistical tests. In this framework, niche divergence for a given pair of species is confirmed by statistical significance of an equivalency test considering only the portion of the accessible environmental space that is shared by the two species, termed ‘analogous accessible environmental space’ (Brown and Carnaval, 2019). For each pair of sister species, we used this approach to evaluate climatic and edaphic niche divergence in humboldt. Accessibility was incorporated by using the input localities to calculate a buffered minimum convex polygon with a buffer distance of 100 km, considered reasonable given the range sizes and dispersal mechanism of the species. The buffer distance was increased to 150 km for ONU in the analysis of climatic niche divergence with VIS because using 100 km resulted in an insufficient number of localities for ONU. Other parameters followed the humboldt documentation (https://jasonleebrown.github.io/humboldt/). Given that three tests of each type were conducted, adjusted P-values were calculated using the BH correction.

Maximum entropy modelling

We adopted a maximum entropy approach to model the distributions of the four more widely distributed species (INC, SAL, SPA and VIS). The four narrow endemics were not analysed because of the issues associated with small numbers of occurrences (Pearson et al., 2007). Thirteen variables were used, including six continuous climate variables, six continuous edaphic variables and one categorical lithological variable (see above; Table 2). Maximum entropy models for the four species in present conditions were implemented in Maxent v.3.4.1 (Phillips et al., 2006). For each species, 100 subsample replicates were run by randomly partitioning the occurrences into a training set (80 %) and a test set (20 %), and an average model was calculated. The area under the receiver operating characteristic curve (AUC) for test data was used to evaluate the predictive power of the models. Logistic outputs were converted to presence/absence data using the maximum training sensitivity plus specificity logistic threshold (Jiménez-Valverde and Lobo, 2007). Importance of variables was assessed using jackknife.

Models were then projected to nine time slices using PaleoClim data: Marine Isotope Stage 19 (MIS19, ~787 ka); Last Interglacial (LIG, ~130 ka); Last Glacial Maximum (LGM, CCSM4 model, ~21 ka); Heinrich Stadial 1 (HS, 17.0–14.7 ka); Bølling-Allerød Interstadial (BA, 14.7–12.9 ka); Younger Dryas Stadial (YDS, 12.9–11.7 ka); early Holocene (EH, 11.7–8.326 ka); mid-Holocene (MH, 8.326–4.2 ka); and late Holocene (LH, 4.2–0.3 ka) (Otto-Bliesner et al., 2006; Fordham et al., 2017; Karger et al., 2017; Brown et al., 2018). These time slices postdate divergence times among species according to dated phylogenies under all three calibration schemes (see Results), except for MIS19 under the cal3 scheme (recent divergence times). Additionally, projections to past time slices rested on two assumptions: (1) constancy of ecological niches through time within species; and (2) constancy of edaphic and lithological conditions (present edaphic and lithological layers, downscaled to 2.5 arc-minute resolution, were used for all past time slices). These assumptions are less likely to hold as models are projected further back in time. Consequently, projections to the earliest time slices should be treated with caution.

Phylogenetic comparative analyses

Phylogenetic comparative analyses were conducted using the three alternative ultrametric phylogenetic trees obtained under the cal1, cal2 and cal3 calibration schemes.

To visualize the magnitude and direction of climatic and edaphic niche shifts during diversification of the study clade, we adopted a ‘phyloecospace’ approach (Baldo et al., 2017). This is equivalent to the ‘phylomorphospace’ approach, which is used to map the history of morphological diversification (Sidlauskas, 2008), but applied to ecological variables. The trees were projected onto the ecospaces defined by the first two axes of the climatic and edaphic PCA-env analyses. Each species was represented by niche position values (see above). The analyses were implemented in the R package phytools (Revell, 2012).

Correlation between pairwise niche overlap and phylogenetic (patristic) distance was assessed using Mantel tests (Mantel, 1967). Patristic distance values were calculated from the three ultrametric phylogenies in Mesquite (Maddison and Maddison, 2011). Mantel tests were implemented in the R package ade4 (Thioulouse et al., 2018) with 9999 permutations, and scatterplots were generated using ggplot2 (Wickham, 2016).

Ancestral range estimation was conducted in RASP v.4.2 (Yu et al., 2015) under the dispersal–extinction–cladogenesis (DEC) model (Ree et al., 2005; Ree and Smith, 2008). Six discrete areas were defined following a previous biogeographical regionalization of the Iberian Peninsula based on endemic vascular plants (Buira et al., 2017): a, southwestern Atlantic coast; b, Baetic Mountains; c, southwestern inner quadrant; d, northwestern inner quadrant; e, southeastern Mediterranean coast; and f, rest of the Iberian Peninsula and southern France. Area f encompasses five of Buira et al.’s (2017) areas and southern France, all of which contain a single species of the study clade (SPA). Separate DEC analyses were run using maximum values of two and six areas for ancestral ranges. Area connectivity was accounted for by scaling dispersal probability values by a factor of 1 for adjacent areas and 0.5 for non-adjacent areas.

RESULTS

Species occurrence, phylogenomic and environmental data

Resulting numbers of species occurrences after filtering ranged from 5 for the narrow endemic ONU to 146 for the widely distributed SPA (Table 1; Fig. 1D; Supplementary Data Dataset 1).

Time-calibrated phylogenetic trees displayed a wide range of divergence time estimates depending on the calibration scheme (Supplementary Data Fig. S1). The time to the most recent common ancestor of the study clade ranged from 3.5 Ma (cal1 tree) to 740 ka (cal3 tree), and the age of the most recent speciation event (involving SPA and INC) ranged from 2.4 Ma (cal1 tree) to 513 ka (cal3 tree).

Filtering of variables to avoid collinearity led to the selection of six bioclimatic variables and six edaphic variables (Table 2; Fig. 2). The distribution of lithological classes across the study area is shown in Fig. 3A.

Fig. 2.

Correlation circles from PCA-env analyses of environmental niches and phyloecospaces of the eight species of the Iberian clade of Linaria subsect. Versicolores based on climatic (A) and edaphic (B) variables. Correlation circles (left) represent the contributions of variables to the two axes of each PCA-env analysis; percentages of variation explained by axes are also indicated. Phyloecospace plots (right) represent patterns of niche shift along the phylogeny based on the first two axes of PCA-env analyses; each species is represented by median niche position values (coloured circles; Supplementary Data, Table S4); black dots indicate the estimated niche positions of ancestors, and lines represent phylogenetic branches. The phyloecospace analysis uses the time-calibrated phylogenetic tree obtained with the cal2 calibration scheme (see Supplementary Data Fig. S3 for similar results obtained from the cal1 and cal3 trees). See Table 1 for the meaning of species acronyms. See Table 2 for the meaning of climatic and edaphic variables.

Fig. 3.

Differences in the frequency of occurrence of the eight species of the Iberian clade of Linaria subsect. Versicolores on substrates of nine lithological classes mapped at 30 arc-second resolution. (A) Geographical distribution of the nine lithological classes (1–9) across the Iberian Peninsula and southern France (IberLit dataset). (B) Histograms representing the number of observations of each species (from the filtered dataset, see Fig. 1D) occurring on each lithological class; phylogenetic relationships are shown on the left for reference. See Table 1 for the meaning of species acronyms and Supplementary Data Table S3 for further explanation of lithological classes.

Analyses of environmental niches

Climatic niches of the eight study species along the first two PCA-env axes (explaining 69.13 % of variation) are shown in Fig. 1B, and contributions of variables to these axes are shown in Fig. 2A (left). Climatic niche position and breadth values are shown in Supplementary Data Table S4. In the context of the study region, climatic niches of all study species had low values of PC1, characterized by low precipitation of the driest quarter, high mean temperature of the driest quarter and high mean temperature of the wettest quarter. Along PC2, ALG displayed the lowest values (characterized by low temperature seasonality, high mean temperature of the coldest quarter and high precipitation of the wettest quarter), and SAL had the highest values (characterized by the opposite conditions). Values of climatic niche overlap (Schoener’s D; Table 3) obtained in ecospat considering the entire environmental space of the study area ranged from 0 for 10 of the 28 species pairs to 0.360 for the pair INC/CLE. The three pairs of sister species (BEC/CLE, VIS/ONU and SPA/INC) displayed low but non-zero values. Tests of niche equivalency (Table 3) produced highly significant results (P < 0.05 before and after BH adjustment) for nearly all species pairs, including the three pairs of sister species, thus supporting pervasive divergence of climatic niches. Tests of niche similarity (Table 3) produced highly significant results (P < 0.05) for only four species pairs, but all tests were non-significant (P > 0.1) after BH adjustment, indicating limited support for niche conservatism across species. Among pairs of sister species, a weakly significant result (P < 0.1) was obtained for BEC/CLE before adjustment, and non-significant results were obtained for the other two pairs.

Table 3.

Analyses of climatic and edaphic niches among the eight species of the Iberian clade of Linaria subsect. Versicolores considering the entire environmental space of the study area. The table shows pairwise values of niche overlap (Schoener’s D) and P-values of equivalency tests and similarity tests obtained in ecospat. In niche equivalency tests, statistical significance indicates niche divergence (D is lower than expected under niche equivalency). In niche similarity tests, statistical significance indicates niche similarity (D is greater than expected at random). For all statistical tests, original P-values and BH-adjusted P-values are shown. Values for pairs of sister species are highlighted in bold. See Table 1 for the meaning of species acronyms. Significance codes: *0.05 ≤ P < 0.1; **0.01 ≤ P < 0.05; ***P < 0.01.

Species pair	Climatic niche					Edaphic niche
	Schoener’s D	Equivalency test		Similarity test		Schoener’s D	Equivalency test		Similarity test
		P-value	BH P-value	P-value	BH P-value		P-value	BH P-value	P-value	BH P-value
BEC/ALG	0.000	0.0099***	0.0111**	1.0000	1.0000	0.318	0.0693*	0.1021	0.0693*	0.2156
CLE/ALG	0.000	0.0099***	0.0111**	1.0000	1.0000	0.548	0.1386	0.1687	0.0495**	0.1733
INC/ALG	0.000	0.0099***	0.0111**	1.0000	1.0000	0.009	0.0297**	0.0693*	0.3861	0.4324
ONU/ALG	0.000	0.0099***	0.0111**	1.0000	1.0000	0.359	0.3861	0.4158	0.0297**	0.1733
SAL/ALG	0.000	0.0099***	0.0111**	1.0000	1.0000	0.495	0.0891*	0.1188	0.0495**	0.1733
SPA/ALG	0.007	0.0099***	0.0111**	0.0297**	0.3465	0.174	0.0396**	0.0853*	0.0495**	0.1733
VIS/ALG	0.002	0.0099***	0.0111**	0.2970	0.4892	0.146	0.0297**	0.0693*	0.1683	0.2626
BEC/CLE	0.008	0.0099***	0.0111**	0.0792*	0.3564	0.338	0.0495**	0.0866*	0.0198**	0.1733
INC/BEC	0.039	0.0099***	0.0111**	0.0396**	0.3465	0.001	0.0099***	0.0693*	0.4555	0.4654
ONU/BEC	0.000	0.0099***	0.0111**	1.0000	1.0000	0.063	0.0198**	0.0693*	0.2277	0.3036
SAL/BEC	0.002	0.0099***	0.0111**	0.1782	0.4620	0.575	0.6733	0.6733	0.0099***	0.1733
SPA/BEC	0.000	0.0099***	0.0111**	0.3168	0.4928	0.044	0.0198**	0.0693*	0.1584	0.2626
VIS/BEC	0.000	0.0099***	0.0111**	0.2475	0.4620	0.061	0.0198**	0.0693*	0.1287	0.2626
INC/CLE	0.360	0.0099***	0.0111**	0.0099***	0.2772	0.009	0.0297**	0.0693*	0.2772	0.3375
ONU/CLE	0.078	0.0099***	0.0111**	0.0693*	0.3564	0.219	0.1386	0.1687	0.0495**	0.1733
SAL/CLE	0.002	0.0099***	0.0111**	0.3663	0.5398	0.539	0.0495**	0.0866*	0.0396**	0.1733
SPA/CLE	0.002	0.0099***	0.0111**	0.2475	0.4620	0.168	0.0297**	0.0693*	0.1782	0.2626
VIS/CLE	0.004	0.0099***	0.0111**	0.2376	0.4620	0.153	0.0792*	0.1109	0.0792*	0.2218
ONU/INC	0.042	0.0099***	0.0111**	0.0495**	0.3465	0.006	0.0495**	0.0866*	0.3762	0.4324
SAL/INC	0.049	0.0495**	0.0513*	0.2079	0.4620	0.003	0.0099***	0.0693*	0.4654	0.4654
SPA/INC	0.006	0.0099***	0.0111**	0.2772	0.4851	0.148	0.2277	0.2657	0.1386	0.2626
VIS/INC	0.008	0.0099***	0.0111**	0.4059	0.5683	0.007	0.0099***	0.0693*	0.4654	0.4654
SAL/ONU	0.000	0.0099***	0.0111**	1.0000	1.0000	0.157	0.0594*	0.0978*	0.2574	0.3276
SPA/ONU	0.000	0.0099***	0.0111**	0.1980	0.4620	0.106	0.2970	0.3326	0.1287	0.2626
VIS/ONU	0.002	0.0198**	0.0213**	0.1485	0.4620	0.040	0.0693*	0.1021	0.2079	0.2911
SPA/SAL	0.042	0.0099***	0.0111**	0.1584	0.4620	0.121	0.0297**	0.0693*	0.1782	0.2626
VIS/SAL	0.001	0.0099***	0.0111**	0.5149	0.6865	0.157	0.0099***	0.0693*	0.1782	0.2626
VIS/SPA	0.097	0.0693*	0.0693*	0.0891*	0.3564	0.294	0.5545	0.5750	0.0990*	0.2520

Edaphic niches of the eight study species along the first two PCA-env axes (explaining 74.63 % of variation) are shown in Fig. 1C, and contributions of variables to these axes are shown in Fig. 2B (left). Edaphic niche position and breadth values are shown in Supplementary Data Table S4. In the context of the study region, edaphic niches of all study species had high values of PC2, characterized by low cation exchange capacity and low organic carbon content. Along PC1, VIS displayed the lowest values (characterized by high soil pH, high clay content and low sand content), and INC had the highest values (characterized by the opposite conditions). Values of edaphic niche overlap (Schoener’s D; Table 3) obtained in ecospat considering the entire environmental space of the study area ranged from 0.001 for the pair BEC/INC to 0.575 for BEC/SAL. The three pairs of sister species displayed a range of values from 0.040 for VIS/ONU to 0.338 for BEC/CLE. Tests of niche equivalency (Table 3) produced highly significant results (P < 0.05) for 16 of 28 species pairs before BH adjustment, and all these tests were still weakly significant (P < 0.1) after adjustment, thus suggesting widespread divergence of edaphic niche. Among pairs of sister species, the result was significant for BEC/CLE. Tests of niche similarity (Table 3) produced highly significant results (P < 0.05) for eight species pairs before BH adjustment, but all tests were non-significant (P > 0.1) after BH adjustment. Among pairs of sister species, a highly significant result was obtained for BEC/CLE before adjustment, and non-significant results were obtained for the other two pairs. Therefore, support for conservatism of edaphic niches across species was limited.

Regarding lithology, species occurrence frequencies suggested diverse preferences (Fig. 3B). Fisher’s exact tests produced highly significant results (P < 0.05 before and after BH adjustment) for 17 of the 28 species pairs, including two of the three pairs of sister species (BEC/CLE and SPA/INC), thus indicating widespread differences in lithological preferences (Supplementary Data Table S5). Among pairs of sister species, we found a diversity of situations. In BEC/CLE, the former occurs on lithologies classified as coarse-grained clastic, whereas the latter occurs mostly on calcareous lithologies. In SPA/INC, both species occur mainly on siliceous (both igneous and metamorphic) and coarse-grained clastic lithologies with different frequencies. In VIS/ONU, both species occur chiefly on coarse-grained clastic lithologies, but the former also has a relevant presence on calcareous lithologies.

In the humboldt analyses (Table 4), the strongest evidence for niche divergence was found for BEC/CLE, with highly significant results (P < 0.05 before and after BH correction) in equivalency tests of both climatic and edaphic niches considering the portion of the accessible environmental space that is shared by the two species. For SPA/INC, a highly significant result was found when analysing climatic niche. The remaining equivalency tests produced non-significant results.

Table 4.

Analyses of climatic and edaphic niches for the three pairs of sister species of the Iberian clade of Linaria subsect. Versicolores considering the portion of the accessible environmental space that is shared by the two species of each pair. The table shows pairwise values of niche overlap (Schoener’s D) in analogous environmental space and P-values of equivalency tests obtained in humboldt. See Table 1 for the meaning of species acronyms. Significance codes: *0.05 ≤ P < 0.1; **0.01 ≤ P < 0.05; **P < 0.01.

Species pair	Climatic niche			Edaphic niche
	Schoener’s D	Equivalency test		Schoener’s D	Equivalency test
		P-value	BH P-value		P-values	BH P-values
BEC/CLE	0.001	0.00332***	0.00996***	0.063	0.00664***	0.01992**
VIS/ONU	0.116	0.22259	0.22259	0.039	0.24917	0.37376
SPA/INC	0.144	0.01329**	0.01994**	0.416	0.69767	0.69767

Maximum entropy modelling

The average test AUC was close to 1 for SAL (0.969), VIS (0.949) and INC (0.949), but lower for SPA (0.785). In all four cases, inferred potential distribution ranges in present conditions encompassed actual distribution ranges plus some additional areas where the species are presently absent. Jackknife analyses of variable importance are shown in Supplementary Data Fig. S2. Overall, edaphic and lithological variables were more important for SPA and INC, whereas climatic variables were more important for VIS. Both groups of variables were of similar importance for SAL.

For SAL (Fig. 4A), model projections estimated a stable range in the SE Iberian mountains (where the species currently occurs) since the MIS19 (~787 ka), with variable suitability through time outside of this region. For VIS (Fig. 4B), stable suitable conditions were inferred for the SW Iberian coast, where the species is most abundant at present, and a much wider distribution during the LGM (~21 ka) was estimated. For SPA (Fig. 4C), a relatively stable range in central and western Iberia, with some waxing and waning, was inferred. In contrast, the potential range of INC (Fig. 4D) was remarkably unstable, sometimes being confined to small areas in the western Iberian coast (e.g. during the Heinrich Stadial 1, 17.0–14.7 ka) and sometimes extending to wide areas of central Iberia (e.g. during the LGM).

Fig. 4.

Potential distributions of the four most widely distributed species of the Iberian clade of Linaria subsect. Versicolores throughout the late Quaternary: L. salzmannii (A), L. viscosa (B), L. spartea (C) and L. incarnata (D). Results are based on maximum entropy modelling using 12 continuous variables (six climatic and six edaphic; see Fig. 2 and Table 2) and one categorical lithological variable (see Fig. 3). Models were built for the present and projected to nine time slices: MIS19, Marine Isotope Stage 19 (~787 ka); LIG, Last Interglacial (~130 ka); LGM, Last Glacial Maximum (~21 ka); HS, Heinrich Stadial 1 (17.0–14.7 ka); BA, Bølling-Allerød Interstadial (14.7–12.9 ka); YDS, Younger Dryas Stadial (12.9–11.7 ka); EH, early Holocene (11.7–8.326 ka); MH, mid-Holocene (8.326–4.2 ka); and LH, late Holocene (4.2–0.3 ka). Potential distributions according to the maximum training sensitivity plus specificity logistic threshold are shown.

Phylogenetic comparative analyses

Phylogenetic comparative analyses produced nearly identical results when using each of the three time-calibrated phylogenetic trees generated with different calibration schemes. Results for the cal2 tree are described here and shown in Figs 2, 5 and 6, and those for the cal1 and cal3 trees are shown in Supplementary Data Fig. S3.

Fig. 5.

Relationship between pairwise phylogenetic (patristic) distances and niche overlap (D) values for the eight species of the Iberian clade of Linaria subsect. Versicolores based on climatic (A) and edaphic (B) variables. Regression lines and results from Mantel tests (correlation coefficients r, and P-values based on 9999 permutations) are shown. Results are based on the time-calibrated phylogenetic tree obtained using the cal2 calibration scheme (see Supplementary Data Fig. S3 for similar results obtained from the cal1 and cal3 trees).

Fig. 6.

Biogeographical reconstructions of the Iberian clade of Linaria subsect. Versicolores based on dispersal–extinction–cladogenesis (DEC) modelling. (A) Regionalization of the Iberian Peninsula and southern France adapted from endemism units inferred by Buira et al. (2017): a, southwestern Atlantic coast; b, Baetic System; c, southwestern inner quadrant; d, northwestern inner quadrant; e, southeastern Mediterranean coast; f, rest of the Iberian Peninsula and southern France. (B) DEC reconstructions using maximum values of two (left) and six (right) areas for ancestral ranges; pie charts at nodes indicate probabilities of ancestral ranges, represented as colours (see A); each node is additionally labelled with the most likely ancestral range. See Table 1 for the meaning of species acronyms. Results are based on the time-calibrated phylogenetic tree obtained using the cal2 calibration scheme (see Supplementary Data Fig. S3 for similar results obtained from the cal1 and cal3 trees).

In the climatic phyloecospace (Fig. 2A, right; see also Supplementary Data Fig. S3A, C), the estimated niche position of the root node was closest to the position of CLE. Shifts towards lower values of principal component (PC)1 and PC2 were detected for ALG and the VIS/ONU pair, whereas shifts towards higher values of PC1 and PC2 were estimated for SAL and the SPA/INC pair. Limited divergence in climate niche position was inferred for two of the three sister species pairs (VIS/ONU and SPA/INC). Somewhat higher divergence, mostly along PC1, was detected for the BEC/CLE pair.

In the edaphic phyloecospace (Fig. 2B, right; see also Supplementary Data Fig. S3B, D), the estimated niche position of the root node was closest to the position of ALG. A shift towards lower values of PC2 was estimated for the clade formed by SAL and the pair BEC/CLE. Meanwhile, the SPA/INC pair experienced a shift towards higher values of both PC1 and PC2 (to a higher degree for INC). In the VIS/ONU pair, the former shifted towards lower values of PC1, and the latter shifted towards higher values of PC2.

Correlations between pairwise niche overlap and phylogenetic distance were non-significant for both climatic and edaphic niche (Fig. 5; see also Supplementary Data Fig. S3E–H).

Ancestral range estimations using a total of six areas (Fig. 6A) and maximum values of two and six areas for ancestral ranges produced similar results (Fig. 6B; see also Supplementary Data Fig. S3I–L). Although uncertainty was higher for the analysis with a maximum of six, the most likely ranges were the same at all nodes in both analyses. A southern Iberian range (including areas a and b) was estimated for the most recent common ancestor of the study clade. Divergence between the two major subclades led to parallel diversification in the Baetic Mountains (area b, producing SAL, BEC and CLE) and the southwestern Atlantic coast (area a, producing the five remaining species). Diversification in the southwestern Atlantic coast was followed by dispersal to other areas by four of the five species (VIS, ONU, INC and SPA).

DISCUSSION

This study provides evidence for the climatic and edaphic niche shifts undergone by plant lineages during radiation in the Mediterranean biodiversity hotspot. This evidence has unprecedented detail thanks to the use of resources that have been made available only in recent times, including high-resolution phylogenomic, distributional and environmental data, in combination with robust statistical analyses. Our findings add a relevant piece of knowledge to ongoing multidisciplinary research on speciation patterns and processes in the study clade by incorporating the environmental niche dimension, which had not been studied thus far (Fernández-Mazuecos et al., 2013, 2018a, 2018b; Fernández-Mazuecos and Vargas, 2015; Cullen et al., 2018). More generally, these results add to the body of evidence on the drivers of recent plant speciation in the Mediterranean biodiversity hotspot (Rundel et al., 2016; Thompson, 2020; Kadereit and Abbott, 2021) by pointing to phylogenetic-scale-dependent patterns of niche conservatism and niche divergence (Pyron et al., 2015) and to a widespread occurrence of climate and soil niche shifts that appear to have driven diversification in combination with pollinator interactions and geographical isolation.

Allopatric differentiation of clades in southern Iberia

The Iberian clade of Linaria subsect. Versicolores is one of the multiple clades of the tribe Antirrhineae (including toadflaxes, snapdragons and relatives) that have radiated in the Mediterranean Region since the establishment of a mediterranean-type climate in this region during the Pliocene (Fernández-Mazuecos et al., 2019; Gorospe et al., 2020) (Supplementary Data Fig. S1), a pattern also found in many other plant lineages (Guzmán et al., 2009; Valente et al., 2010; Fiz-Palacios and Valcárcel, 2013; Moazzeni et al., 2014). Ancestral range estimation indicates a most recent common ancestor of the group in southern Iberia, followed by allopatric differentiation of two subclades, one in southeastern Iberia that produced three species and one in southwestern Iberian that produced five species and subsequently colonized other areas in central–western Iberia (Fernández-Mazuecos et al., 2013, 2018b; Fernández-Mazuecos and Vargas, 2015). Geographically similar splits have been described for several organisms, including plants and animals (García-París et al., 1998; Fitze et al., 2011; Benítez‐Benítez et al., 2018), and some of these cases could be explained by the marine barrier that isolated the Baetic Mountains from the rest of the Iberian Peninsula before the sedimentary filling of the Guadalquivir Basin in the Pliocene (Meléndez Hevia, 2004). For our study group, the Quaternary split times estimated from plastid DNA sequences appeared to rule out this possibility (Fernández-Mazuecos and Vargas, 2015), but other analyses have obtained uncertainty intervals overlapping the Pliocene (Fernández-Mazuecos et al., 2019). A Pliocene divergence would be compatible with a role of geographical isolation resulting from a marine barrier in the divergence between the two subclades.

Additionally, given that calcareous and dolomitic substrates are frequent in southeastern Iberian mountains, whereas siliceous, acid substrates predominate in central–western Iberia, it has been suggested that edaphic specialization (Rajakaruna, 2004, 2018) drove the differentiation of the two subclades (Fernández-Mazuecos and Vargas, 2015). Such a pattern has been found, for example, in Carex sect. Spirostachyae (Escudero et al., 2008). Our new detailed analyses support this hypothesis to some degree, as shown by the basal split of ancestral edaphic niches leading to contrasting niche positions for the two subclades. Thus, the three southeastern species (L. salzmannii, L. becerrae and L. clementei) are placed close together in the edaphic ecospace owing to their similarly low values of PC2 and intermediate values of PC1, whereas the other five species occupy a wider space, with higher values of PC2 and a wide range of PC1 values. Nevertheless, a closer look at edaphic niches, including their position, breadth and overlap, indicates a complex pattern of edaphic niche differentiation that goes beyond a simple split between a basophilous subclade in southeastern Iberia and an acidophilous subclade in central–western Iberia. This is shown, for example, by L. viscosa, which frequently occurs on high-pH soils but belongs to the central–western subclade, and by the southeastern Iberian L. salzmannii, which occurs on both calcareous–dolomitic and siliceous soils (Figs 1–3) (Sáez, 2009; cf. Fernández-Mazuecos et al., 2018a). A pattern of repeated shifts between acidophilous and basophilous preferences associated with recent diversification has been found in plant genera such as Adenostyles (Dillenberger and Kadereit, 2013) and Saxifraga (Gerschwitz‐Eidt et al., 2023).

In contrast to edaphic niche, no overall differentiation in climatic niche at the split between the two subclades of our study group is observed. This is congruent with the similarity in climatic niche between species belonging to different subclades, such as L. clementei and L. incarnata. Similar to climatic niches, previous data indicated no overall pattern of differentiation in floral morphology at the basal split of the Iberian clade (Fernández-Mazuecos et al., 2013, 2018b). Altogether, these results indicate that the split between southeastern and southwestern lineages was driven by geographical isolation accompanied by some degree of edaphic specialization.

Environmental niche shifts associated with recent speciation

Patterns of environmental niche conservatism or divergence during speciation have been reported variously in plant lineages and particularly in mediterranean-type floras. Anacker and Strauss (2014) examined 71 pairs of putative sister species from the Californian flora and found a predominance of sympatry associated with shifts in ecological traits, particularly habitat and soil, i.e. niche divergence. Similar results were found by van der Niet and Johnson (2009) after examining 188 sister species pairs from the Cape flora, also mediterranean type. However, Anacker and Strauss (2014) also found a greater ecological similarity of sisters to each other than to non-sister congeners, in agreement with niche conservatism. As far as we know, no comparable large-scale studies have been conducted for the flora of the Mediterranean Basin, but several studies have addressed particular plant lineages. For example, niche divergence has been proposed for Festuca (Marques et al., 2016), Stauracanthus (Chozas et al., 2017), Gypsophila (de Luis et al., 2018) and Iberodes (Otero et al., 2022), whereas niche conservatism has been suggested for Brachypodium (López‐Alvarez et al., 2015) and Helianthemum (Martín-Hernanz et al., 2021). Interestingly, a combination of both patterns has been reported for several genera, including Anthemis (Lo Presti and Oberprieler, 2009), Aquilegia (Jaime et al., 2015), Carex (Benítez‐Benítez et al., 2018) and Lythrum (Gazaix et al., 2021). In our study clade, the basal split between the two subclades of Iberian Linaria subsect. Versicolores was followed by complex patterns of niche shift during speciation. In agreement with the results of Anacker and Strauss (2014), our research supports the idea that the prevalence of niche conservatism or niche divergence patterns is a matter of phylogenetic scale (see Pyron et al., 2015). On a broad phylogenetic scale, it is clear that diversification of the study clade has occurred within some environmental boundaries characterized by mediterranean-type climates and poor soils, whereas Iberian areas without these conditions have remained essentially unoccupied by the clade. Therefore, a niche conservatism pattern predominates at this scale, because closely related species tend to retain ancestral ecological characteristics (Wiens and Graham, 2005). On a more detailed phylogenetic scale, however, pairs of sister species tend to display patterns of niche divergence pertaining to climatic niche, edaphic niche, or both (cf. Anacker and Strauss, 2014). In other words, niches are constrained, i.e. they are divergent within a limited subset of available niches (Pyron et al., 2015). Additionally, the association between species divergence and niche overlap is non-significant for both climatic and edaphic niches, suggesting a high evolutionary lability of niches within that subset (cf. Losos et al., 2003).

The role of niche shifts in particular speciation events can be evaluated by looking at pairs of sister species, as done by van der Niet and Johnson (2009) and Anacker and Strauss (2014). However, these studies relied on phylogenies obtained from a small number of loci (which are known to produce spurious results when dealing with recent diversification) and on relatively simple metrics of ecological divergence. In our study, ecological divergence between sister pairs could be examined more thoroughly (although in a smaller number of species) thanks to the use of phylogenomic data and more advanced metrics. For each of the three pairs of recently diverged sister species, the two species display geographically close or overlapping distribution ranges (Fernández-Mazuecos et al., 2018a, 2018b), suggesting that the same range of environmental conditions is available to them. In fact, there are no obvious geographical barriers explaining either of these speciation events. Under this scenario, ecological speciation is likely to have played a more important role than geographical speciation (see Hiller et al., 2019; Hernández‐Hernández et al., 2021). Ecological speciation might have involved divergence in environmental (climatic and/or edaphic) niches, possibly in combination with evolutionary changes in biotic interactions (Rundle and Nosil, 2005). These include pollinator interactions (van der Niet and Johnson, 2012; Phillips et al., 2020), as indicated for the study group by divergence in floral traits characterizing the three recent speciation events (Fernández-Mazuecos et al., 2013, 2018b).

The first pair of sister species is formed by L. becerrae and L. clementei, two narrow endemics in southern Spain whose closest populations are ≤5 km apart. They have non-equivalent environmental niches (both climatic and edaphic), and strong evidence of divergence in both climatic and edaphic niche was found even when considering the shared portion of the accessible environmental space. They also occur on distinct lithological classes. In fact, field observations confirm that L. becerrae is restricted to calcarenite substrates, whereas L. clementei occurs on marble substrates (Fernández-Mazuecos et al., 2018a). Additionally, their flowers are similar in shape and colour, but they display a striking difference in the length of the nectar spur, possibly associated with differences in pollination mechanisms (Fernández-Mazuecos et al., 2013; Cullen et al., 2018).

The second pair is formed by L. viscosa and L. onubensis, whose distribution ranges overlap in southwestern Spain. The former is distributed mostly in southwestern Spain, with a few populations in eastern Spain (some of them possibly introduced; Fernández-Mazuecos et al., 2018a), whereas the latter is a narrow endemic from southwestern Spain. Their climatic niches are non-equivalent, although climatic niche divergence could not be confirmed when considering the shared portion of the accessible environmental space, possibly as a result of the limited number of L. onubensis localities. The two species occur chiefly on coarse-grained clastic lithologies, but L. viscosa additionally occurs on calcareous substrates. In addition to their environmental niche divergence, L. viscosa and L. onubensis display divergent floral morphologies in terms of both corolla colour (yellow in L. viscosa, purple in L. onubensis) and shape (broad tube in L. viscosa, narrow tube in L. onubensis) (Fernández-Mazuecos et al., 2013, 2018b).

The third pair is formed by L. spartea and L. incarnata, whose distribution ranges overlap extensively in western Iberia. Linaria spartea is widely distributed in central–western Iberia, whereas L. incarnata has a relatively wide distribution in western Iberia. Their climatic niches are non-equivalent, and climatic niche divergence is supported when considering the shared portion of the accessible environmental space. In contrast, equivalence of edaphic niches could not be rejected. The widespread distribution of L. spartea is clearly associated with an expansion of climatic niche, but not so much of edaphic niche. Linaria spartea and L. incarnata occur on similar lithologies, including siliceous and coarse-grained clastic ones, but in different frequencies. Their divergence in floral morphology closely parallels that between L. viscosa and L. onubensis (Fernández-Mazuecos et al., 2013, 2018b).

Adaptive radiation with some non-adaptive features

Many evolutionary radiations contain elements of both adaptive and non-adaptive diversification (Rundell and Price, 2009; Matsubayashi and Yamaguchi, 2022). Under non-adaptive radiation, allopatric, ecologically similar sister species would be expected (Gittenberger, 1991; Albaladejo et al., 2020). In contrast, in the Iberian clade of Linaria subsect. Versicolores, adaptive radiation (defined as the evolution of ecological and phenotypic diversity within a rapidly multiplying lineage; Schluter, 2000) is supported by the combination of environmental (climate and/or soil) niche differentiation and floral differentiation in sister species pairs occurring in sympatry, parapatry or close allopatry (with no clear geographical barriers). The question then is the nature of reproductive barriers isolating these pairs of sister species. Although intrinsic post-pollination barriers appear to be weak in this group based on crossing experiments (Valdés, 1970; Viano, 1978), hybrids are rare in the wild. Therefore, reproductive isolation appears to be based on pre-pollination mechanisms, specifically a combination of habitat isolation and pollinator isolation (Rundle and Nosil, 2005), and temporal isolation is unlikely considering the overlapping flowering times of these sister species pairs (Sáez, 2009). Given that barriers to gene flow important to speciation are those that arise before reproductive isolation is complete (Rundle and Nosil, 2005), the presence of potential ecological barriers (both habitat and pollinator isolation) among our study species in the absence of strong post-pollination barriers indicates that ecology has been crucial to their speciation.

It should be noted that phylogenetic relationships considered here represent the predominant vertical inheritance pattern as estimated from genotyping-by-sequencing data, but homoploid hybridization appears to have played a role in radiation of the group (Fernández-Mazuecos et al., 2018b) and, more generally, in diversification of the genus Linaria (Blanco-Pastor et al., 2012). For example, L. clementei is considered sister to L. becerrae, but it might also contain some genetic contribution from an earlier-diverging lineage. As part of adaptive radiation, hybridization might facilitate shifts to novel niches outside the ranges of parental species (Seehausen, 2004). The way in which homoploid hybrid speciation (Abbott et al., 2010; Nieto Feliner et al., 2017) might have shaped radiation and niche shifts in the study clade remains to be studied using robust phylogenetic network approaches and network-based comparative methods (Solís-Lemus and Ané, 2016; Bastide et al., 2018).

Finally, in addition to adaptive processes, non-adaptive allopatric speciation might also have contributed to this radiation, particularly in the early divergence of the two subclades, as discussed above. These results add to the growing evidence of intertwined adaptive and non-adaptive processes contributing to rapid plant diversification in the Mediterranean Region as a result of a great environmental heterogeneity and a complex geological and climatic history during the Cenozoic (Comes et al., 2008; Guzmán et al., 2009; Valente et al., 2010; Nieto Feliner, 2014; Thompson, 2020; Martín-Hernanz et al., 2021).

Effect of Quaternary climatic cycles on species distributions

Maximum entropy models of the four widespread species, in combination with biogeographical reconstructions, provide additional insights into the evolutionary history of the clade. Projections of L. viscosa, L. spartea and L. incarnata models to the oldest analysed time slice (MIS19, ~787 ka), potentially close to their speciation times according to the cal2 tree (Supplementary Data Fig. S1), support the presence of suitable habitats for all three species in southwestern Iberia at that time (Fig. 4B–D). This is consistent with a southwestern range for the ancestor of the central–western subclade, as estimated by the biogeographical analysis (Fig. 6). The current presence of early-diverging intraspecific lineages of L. spartea (Fernández-Mazuecos et al., 2018b) and of the narrowly distributed L. algarviana and L. onubensis in this area are also consistent with the role of the southwestern Atlantic coast as the cradle of diversification of the central–western subclade. Likewise, projection of the L. salzmannii model to the MIS19 supports the early presence of this species in the Baetic Mountains of southeastern Iberia. In combination with the biogeographical reconstruction and the current presence of L. clementei and L. becerrae in this area, these results clearly support the Baetic Mountains as the cradle of diversification of the southeastern subclade.

Although the Baetic Mountains are well known as one of the most important areas of endemism for the Iberian flora, the role of the southwestern Atlantic coast as an area of endemism and diversification is not so frequently highlighted (Médail and Quézel, 1997; Fernández-Mazuecos et al., 2016; Buira et al., 2017, 2020; Vila-Viçosa et al., 2020; Ramos‐Gutiérrez et al., 2021). According to model projections, these two areas were not only cradles of diversification, but also areas of long-term persistence throughout the late Quaternary climatic changes for species of the central–western and southeastern subclades, respectively (Fig. 4). Remarkably, the two areas include several putative Quaternary refugia described from numerous phylogeographical studies of Mediterranean plants (Médail and Diadema, 2009). Apart from these areas of long-term persistence, model projections support significant waxing and waning of distribution ranges during the late Quaternary, with no clear patterns of isolation between species resulting from these range shifts. Therefore, the evidence for a role of Quaternary climatic cycles on speciation of the study clade appears to be limited (cf. Comes and Kadereit, 1998; Abellán and Svenning, 2014). The degree to which these cycles shaped the distribution of intraspecific genetic diversity, as found for other species of Linaria (Blanco-Pastor et al., 2013; Fernández-Mazuecos and Vargas, 2013), remains to be studied.

Conclusions

Our analysis of high-resolution distributional, phylogenomic and environmental data has enabled a detailed reconstruction of environmental niche shifts during radiation of a western Mediterranean plant clade following allopatric differentiation of two subclades in southern Iberia. Niches of closely related species are divergent within a conserved subset of available niches, supporting the idea that the prevalence of niche conservatism or niche divergence patterns is a matter of phylogenetic scale. Habitat isolation pertaining to both climatic and soil conditions appears to have played a role in speciation in the western Mediterranean biodiversity hotspot, most probably in combination with pollinator isolation, in agreement with an adaptive radiation scenario with some non-adaptive features. Quaternary climatic cycles might have had a limited role in speciation of the study clade, but they caused significant geographical range shifts through time. More detailed analysis of recently diverged species pairs within this radiation, including population genomics, pollination ecology and environmental requirements, will provide deeper insights into the relative roles of key factors driving their speciation.

SUPPLEMENTARY DATA

Supplementary data are available at Annals of Botany online and consist of the following.

Table S1: minimum and maximum ages for calibrated nodes used to obtain ultrametric, time-calibrated phylogenetic trees. Table S2: full list of environmental variables before applying procedures to reduce collinearity. Table S3: correspondence between lithological classes in the IberLit layer and dominant materials according to the source lithological maps. Table S4: position and breadth of climatic and edaphic niches of the eight study species. Table S5: analysis of lithological preferences among the eight study species. Figure S1: time-calibrated phylogenetic trees obtained using three alternative calibration schemes. Figure S2: jackknife analyses assessing the importance of variables for maximum entropy models. Figure S3: phylogenetic comparative analyses based on the time-calibrated trees obtained using the cal1 and cal3 calibration schemes. Dataset S1: filtered occurrence data for the eight study species.

FUNDING

This work was supported by the European Commission through Marie Curie Intra-European Fellowship LINARIA-SPECIATION (FP7-PEOPLE-2013-IEF, reference 624396), the Isaac Newton Trust (Trinity College, Cambridge, UK) through an Isaac Newton Trust Research Grant, the Spanish Ministry of Economy and Competitiveness through a Juan de la Cierva fellowship (reference IJCI-2015-23459) and the Spanish Ministry of Science, Innovation and Universities through a Ramón y Cajal fellowship (reference RYC2022-036418-I) to MF-M.

CONFLICT OF INTEREST

The authors declare no conflicts of interest.

DATA AVAILABILITY

Filtered occurrence data on which this study is based are available as Supplementary Data Dataset 1. The IberLit dataset is available at https://doi.org/10.5281/zenodo.14337975.