The sampling comprises a total of 41 populations of 18 species of genus Carduncellus, including three subspecies, two species of Phonus and one population of Femeniasia balearica (J. J. Rodr.) Susanna. Origin, collectors and dates are shown in Table 1. Voucher specimens for each population are deposited in the herbarium BC (Botanical Institute of Barcelona).

Fresh young leaves of the plants studied were co-chopped using a razor blade with an internal standard in the proportions 2:1 in 1200 μl of LB01 buffer (Doležel et al., 1989) with 0.5% of Triton X-100 and supplemented with 100 µg/ml ribonuclease A (RNase A, Boehringer, Meylan, France) in a plastic Petri dish. Pisum sativum L. ‘Express Long’ (2C = 8.37 pg) and Petunia hybrida Vilm. ‘PxPc6’ (2C = 2.85 pg) were used as internal standards and were first analysed separately in 600 μl of LB01 buffer to locate their peak positions. Nuclei were filtered through a 70-μm nylon filter in order to eliminate cell debris before the addition of 36 μl of propidium iodide (1 mg/ml, solution in water; Invitrogen Eugene, Oregon, USA). Samples were kept on ice before measurement. For each population (Table 1), two samples of each individual were prepared and measured independently. Fluorescence analysis was carried out using an Epics XL flow cytometer (Coulter Corporation, Hialeah, Florida, USA) at the Centres Científics i Tecnològics de la Universitat de Barcelona with the standard configuration as described in Garnatje et al. (2006). Acquisition was stopped at 8000 nuclei. The DNA content was calculated for 10 of the aforementioned runs, assuming a linear correlation between the fluorescence signals (of the stained nuclei) and DNA amount. Mean and standard deviations were calculated for 2C values of each population based on five individuals.

The nuclear ribosomal dataset includes ITS1 and ITS2 regions for 23 species, including two outgroups. ITS sequences of 17 species were available from GenBank and four taxa were sequenced for this study following the same protocol as described in Barres et al. (2013). Carthamus glaucus M. Bieb. and C. oxyacantha M. Bieb. were chosen as outgroup based on published phylogeny by Vilatersana et al. (2000a ). DNA sequences were edited with Chromas v2.6.4 (Technelysium PTy, Tewantin, Queensland, Australia) and Bioedit v7.0.9 (Hall, 1999) and aligned visually. The aligned matrix is available from the corresponding author. The General Time Reversible model (GTR + G) was chosen for ITS nrDNA dataset based on AIC criterion implemented in jModeltest v2.1.2 (Darriba et al., 2012). Markov Chain Monte Carlo (MCMC) analysis was carried out in MrBayes v3.2.6 (Ronquist et al., 2012) for 2,000,000 generation sampling every 100 generations. The first 25% of the trees were discarded as the ‘burn-in’ period, after confirming that the average standard deviation of the split frequencies was <0.01, and the potential scale reduction factor approached 1.0 for all parameters. The remaining samples were pooled to construct a 50% majority rule consensus tree. The resulting summary trees were visualised in Figtree v1.4.2 (http://tree.bio.ed.ac.uk/software/figtree).

Bar plots showing the distribution of genome size values across the taxa of the phylogenetic inference were generated with the package phytools (Revell, 2012; implemented in R v3.2.2; R Core Team, 2016). A graph illustrating the distribution of mean genome size values for the species of the Carthamus-Carduncellus complex at different ploidy levels was generated with the package ggplot2 (Wickham, 2016; implemented in R) using the new genome size assessments together with the previously published data of Garnatje et al. (2006).

One-way ANOVA was carried out in order to test the 1Cx differences between ploidy levels and between Carthamus and Carduncellus lineages. The analyses were performed with XLSTAT 2018.7 (Addinsoft Inc.).

Genome size values (2C) range from 3.24 pg in Carduncellus calvus Boiss. & Reut., a diploid species endemic to Maghreb region (Morocco and Algeria), to 11.16 pg in a hexaploid accession of C. eriocephalus Boiss., a species showing several ploidy levels (Vilatersana et al., 2000b ). This species occurs in the Maghreb region, reaching Egypt. Femeniasia balearica, endemic to Menorca (Balearic Islands), presents a 2C value of 3.84 pg, and Phonus, 2C values from 4.54 to 4.60 pg. Monoploid genome size (1Cx) ranges from 1.29 pg in a tetraploid accession of C. duvauxii Batt. et Trab. (2n = 48; Vilatersana et al., 2000b ), which has also been reported as a diploid by López González (1990), to 2.28 in C. mareoticus (Del.) Hanelt, a diploid species endemic from the northern part of Egypt and Libya. Femeniasia balearica displays a 1Cx value of 1.92 and the 1Cx values oscillate between 2.27 and 2.30 in the two studied species of the genus Phonus, all of them being diploid.

Three ploidy levels have been found (2x, 4x and 6x) in genus Carduncellus, some of which have been inferred from genome size values. Fifteen taxa are diploid, seven tetraploid and only one (C. eriocephalus) is a hexaploid. It is to note that a triploid chromosome count was also reported in this same Moroccan accession of C. eriocephalus (Vilatersana et al., 2000b ), indicating within-population cytotype diversity. In Carduncellus, where diploids and tetraploids predominate, the only other reports of triploidy was for C. calvus and of hexaploidy, for C. caeruleus (L.) C. Presl. and C. pinnatus (Desf.) DC. (López González, 1990; Rice et al., 2015). B chromosomes have been found in several species of Carduncellus and this could be one of the reasons for the wide variability in genome size found within the same ploidy level in phylogenetically closely related species, since dysploidy is not frequent in this genus (Vilatersana et al., 2000b ; Vilatersana, 2002). Our results suggest a loss of DNA per basic genome in polyploids (mean 1Cx = 1.66 pg) with respect to diploids (mean 1Cx = 1.94 pg), a phenomenon known as genome downsizing, largely observed in plants (Leitch & Bennett, 2004), present in the family Asteraceae (e.g. Pires et al., 2004; Chrtek et al., 2009; Pellicer et al., 2010; Vallès et al., 2013). The decrease of 1Cx values in the polyploid species has been found statistically significant by ANOVA test (F = 7.1914, p = 0.0143).

The insular Femeniasia has a lower nuclear DNA content than Phonus, a continental sister genus with which it shares clade and chromosome number (Fig. 1, Table 2). The insularity may have played a role in the reduced genome size of a genus endemic of a Mediterranean island (Menorca) but further studies will be needed to confirm this hypothesis. Similar cases were reported in Carthamus (Garnatje et al., 2006) and Cheirolophus (Garnatje et al., 2007; Hidalgo et al., 2017) species, from the same tribe, as well as in other Asteraceae (Zahradníček et al., 2018). This phenomenon has been attributed to island colonisation pressure (Suda et al., 2003) and to the higher facility of naturalisation of plants with smaller genomes (Kapralov & Filatov, 2011).

Considering only the diploid taxa, the 1Cx average was 1.33 pg for Carthamus and 1.85 pg for Carduncellus-Femeniasia-Phonus clade. Statistically significant differences in the 1Cx values between Carthamus and Carduncellus (F = 99.8583, p = 0.0000) support the previously stated independent genomic evolution of these two lineages, although phylogenetic inferences have not yet fully resolved the relationships between Carthamus and the western group (Carduncellus, Femeniasia and Phonus; Vilatersana et al., 2000a ). For each ploidy level, genome size values in the Carduncellus-Femeniasia-Phonus clade are consistently higher than those of Carthamus; indeed, their range do not even overlap (Fig. 1B). This trend of higher genome sizes in perennial taxa compared to relatives with short life cycle is frequently observed in plants (for Asteraceae, see e.g. Hidalgo et al., 2008; Siljak-Yakovlev et al., 2017; Qiu et al., 2019; but see Pellicer et al., 2014). However, as suggested by Vitales et al. (2019), the observed associations between genome size and life cycle in Asteraceae could be better explained by phylogenetic relatedness between taxa. Yet, more data is needed, and also analysed in an evolutionary context, to establish such an association.