Supplementary materials and methods Study species Linaria glacialis Boiss. (Fam. Plantaginaceae, tribe Antirrhineae) is an annual or perennial narrow endemic toadflax inhabiting schistose screes of the alpine vegetation belt of Sierra Nevada (2700-3400 m). It has an occluded corolla similar to that of snapdragons (Antirrhinum), but with a long spur. The corolla is similar in size and shape to that of closely related congeneric species (subsect. Supinae, sect. Supinae; Blanco-Pastor et al. 2012). L. glacialis is mainly differentiated from other Linaria species by its large leaf-like calyx lobes and bracts (see Fig. 1B). The conservation status of this species is Vulnerable –VU D2; UICN category– (Blanca et al. 1998) and the number of individuals has been estimated to be less than 10,000 (Blanca et al. 2001). The scarcity of adequate habitat is considered the main threat for this species (Blanca et al. 1998; Blanca et al. 2001). Study area Sierra Nevada is located in the southern part of Spain nearby the Mediterranean Sea (see Fig. 1A) and has the highest altitude of the Iberian Peninsula (Mulhacen, 3481 m a.s.l.). This mountain range was covered with glaciers only in areas above ~2,500 m a.s.l. during Quaternary glaciations (Gómez-Ortiz et al. 1996), while large areas remained free of permanent ice. Sierra Nevada was a refuge area for many European species during glacial ages (Blanca et al. 1998; González-Sampériz et al. 2010). Indeed, location and altitude of Sierra Nevada has interested scientist to test speciation patterns in plants (Gutiérrez Larena et al. 2002; Kropf et al. 2006). The upper vegetation belt (alpine or cryoromediterranean) of Sierra Nevada has an extension of 3875,7 ha and its lower boundaries span between 2750 m in northern and western zones and 3290 m in southern and eastern zones (Fernández Calzado & Molero Mesa 2011). It shelters an abrupt landscape with steep slopes, screes and rocky areas and has a metamorphic substrate composed by feldspar, micaschist, slates and quartzites. The effect of glacier phenomena is still present in the northern part of the range, although at the beginning of 20th century the last ice mass (Veleta glacier) was strongly reduced and since 1995 there are no traces left (Gómez-Ortiz et al. 2009). In this area the vegetation is exposed to stressful climate conditions with large daily temperature oscillations and a pronounced summer drought. Sampling and DNA sequencing strategy Apart from the L. glacialis dataset, an additional dataset (species-level dataset) was generated for the present study. The species-level dataset included DNA sequences of two L. glacialis individuals and 15 individuals of other closely-related species, which were chosen based on previous phylogenetic studies (Blanco-Pastor et al. 2012; Fernández-Mazuecos et al. 2013): L. aeruginea (two individuals), L. almijarensis, L. anticaria, L. lilacina, L. platycalyx, L. tristis, L. amoi, L. verticillata, L. depauperata, L. polygalifolia (3 individuals), L. supina, L. arvensis (two individuals) and L. saxatilis (see above). The species-level dataset comprised both previously published and newly generated sequences of the ITS region (Blanco-Pastor et al. 2012; Fernández-Mazuecos et al. 2013) and the low copy nuclear gene AGT1 (partial intron) (Blanco-Pastor et al. 2012) (see Supplementary table 1 and 2). Deciphering of haplotypes from unphased genotypes and test for recombination In addition to AGT1 sequences of the L. glacialis dataset, AGT1 and ITS sequences of the species-level dataset were also analyzed with PHASE. In the species-level dataset, when haplotypes inferred were not statistically supported, we obtained haplotype data by cloning the PCR products as done in Blanco-Pastor et al. (2012). In the cloning process several paralogous copies were obtained for the ITS region from one species (L. aeruginea) and for the AGT1 region from another species (L. verticillata). We then assessed the orthology of amplification products following Whittall et al. (2006). Recombination was also tested within the AGT1 and ITS alignments of the specieslevel dataset using the software RDP 3.44 (Martin et al. 2010). Methods and parameter settings were as described in Blanco-Pastor et al. (2012). Species tree estimation In order to investigate the evolutionary framework for L. glacialis we performed a *BEAST species tree analysis (Heled & Drummond 2010) as implemented in the BEAST software v.1.7.2 (Drummond & Rambaut 2007; Drummond et al. 2012) using ITS and AGT1 intron haplotype sequences of the species-level dataset after excluding two AGT1 haplotypes with hard incongruence (L. amoi, L. lilacina) subject to be caused by hybridization/introgression (as the *BEAST model does not account for such processes). The crown age of section Supinae was calibrated with a normal distribution prior with mean 1.95 Ma and standard deviation 0.66. This date was previously obtained in a three-locus species tree phylogeny of Linaria sect. Supinae with detection and exclusion of plausible hybrids (Blanco-Pastor et al. 2012). The remaining prior settings of the analysis were equal to those used in that study. Population size history Extended Bayesian skyline plot Extended Bayesian skyline plot (EBSP) analysis allows the joint analysis of multiple loci and uses Bayesian stochastic variable selection to select the appropriate smoothness of the demographic function (number of groups of coalescent intervals). The analysis was carried out combining information of the four regions of the L. glacialis dataset two linked plastid loci (ptDNA) and two linked regions of the AGT1 gene. We included the four regions as independent partitions for the analysis with distinct substitution models as obtained with jModeltest 0.1.1 (Posada 2008) (see Supplementary table 1). The two plastid regions were linked for the partition tree and the clock model priors. AGT1 exon and intron regions were also linked for the partition tree but unlinked for the clock model priors. Upper limits for the mean.rate parameters were set to 5 s/s/Ma for the AGT1 regions and 1 s/s/Ma for the ptDNA loci. Operators of the analysis were also modified following the recommendations of the EBSP tutorial available in http://beast.bio.ed.ac.uk/Tutorials. Supplementary results Haplotype data gathering and recombination test The species-level dataset included 36 ITS and 33 AGT1 sequences. Recombination was neither detected in the ITS or AGT1 alignments of the species-level dataset by any of the five methods used. This allowed us to perform the species tree analysis without discarding any recombinant region. Species tree Despite the low resolution and the requirement of additional loci to unravel the species relationships within this group (sect. Supinae subsect. Supinae), the higher number of species and sequences used here compared to the analysis of Blanco-Pastor et al. (2012) extends the previous phylogenetic information obtained of this recently diversified group of Linaria. The species tree obtained from dataset A (Supplementary Fig. 1) showed L. glacialis within subsect. Supinae sensu Blanco-Pastor et al. (2012) and forming a monophyletic group together with eight southern Iberian endemics (0.99 PP). Additionally it also showed L. glacialis as sister to the sub-endemic low-land species of Sierra Nevada L. verticillata (with moderate support; 0.84 PP) (see Supplementary Fig. 1), with a divergence time 0 – 230 ka 95% HPD (60 ka median). References Blanca G, et al. (1998) Threatened vascular flora of Sierra Nevada (southern Spain). Biological Conservation 85, 269-285. Blanca G, et al. (2001) Flora Amenazada y Endémica de Sierra Nevada Consejería de Medio Ambiente. Junta de Andalucía, Granada. Blanco-Pastor JL, et al. (2012) Coalescent Simulations Reveal Hybridization and Incomplete Lineage Sorting in Mediterranean Linaria. Plos One 7, e39089. Drummond A, Rambaut A (2007) BEAST: Bayesian evolutionary analysis by sampling trees. BMC Evolutionary Biology 7, 214. Drummond AJ, et al. (2012) Bayesian phylogenetics with BEAUti and the BEAST 1.7. Molecular Biology and Evolution 29, 1969-1973. Fernández-Mazuecos M, et al. (2013) A Phylogeny of Toadflaxes (Linaria Mill.) Based on Nuclear Internal Transcribed Spacer Sequences: Systematic and Evolutionary Consequences. International Journal of Plant Sciences 174, 234-249. Fernández Calzado MR, Molero Mesa J (2011) The cartography of vegetation in the cryoromediterranean belt of Sierra Nevada: a tool for biodiversity conservation. LAZAROA 32, 101-115. Gómez-Ortiz A, et al. (2009) Evidences from historical documents of landscape evolution after Little Ice Age of a Mediterranean high mountain area, Sierra Nevada, Spain (eighteenth to twentieth centuries). Geografiska Annaler: Series A, Physical Geography 91, 279-289. Gómez-Ortiz A, et al. (1996) Contribución al conocimiento de la deglaciación reciente y morfología asociada del Corral del Veleta (Sierra Nevada). Cadernos do Laboratorio Xeolóxico de Laxe 21, 543-558. González-Sampériz P, et al. (2010) Steppes, savannahs, forests and phytodiversity reservoirs during the Pleistocene in the Iberian Peninsula. Review of Palaeobotany and Palynology 162, 427-457. Gutiérrez Larena B, et al. (2002) Glacial-induced altitudinal migrations in Armeria (Plumbaginaceae) inferred from patterns of chloroplast DNA haplotype sharing. Molecular Ecology 11, 1965-1974. Heled J, Drummond AJ (2010) Bayesian Inference of Species Trees from Multilocus Data. Molecular Biology and Evolution 27, 570-580. Kropf M, et al. (2006) Long-distance dispersal vs vicariance: the origin and genetic diversity of alpine plants in the Spanish Sierra Nevada. New Phytologist 172, 169-184. Martin DP, et al. (2010) RDP3: a flexible and fast computer program for analyzing recombination. Bioinformatics 26, 2462-2463. Posada D (2008) jModelTest: Phylogenetic model averaging. Molecular Biology and Evolution 25, 1253-1256. Whittall JB, et al. (2006) Generating single-copy nuclear gene data for a recent adaptive radiation. Molecular Phylogenetics and Evolution 39, 124-134.