We performed a phylogenetic reconstruction to infer the phylogenetic placement of the Balearic and Tuscan individuals and to assess the degree of genetic differentiation between C. sylvatica subsp. sylvatica and C. sylvatica subsp. paui. Our sampling (Appendix 1) included: 18 samples of C. sylvatica subsp. sylvatica, selected to representatively cover its distribution area; 5 samples previously classified as C. sylvatica subsp. paui from northwestern Africa and northern Spain; 3 samples of the C. sylvatica subsp. paui-like plants from the Balearic and Tuscan Archipelagos; 4 samples of two other species of C. sect. Sylvaticae —C. rainbowii and C. cretica—; and 6 samples representing two species of each of the three sections phylogenetically closely related to C. sect. Sylvaticae (Waterway & Starr, 2007; Martín-Bravo & al., 2013): C. demissa Hornem. and C. flava L. —C. sect. Ceratocystis—; C. distans L. and C. punctata Gaudin —C. sect. Spirostachyae—; C. pendula Huds. and C. bequaertii De Wild. —C. sect. Rhynchocystis—. DNA was extracted from the specimens using a DNeasy Plant Mini Kit —Qiagen—. Materials were PCR-amplified following PCR conditions from Escudero & Luceño (2009). Sequence chromatograms were visualized and edited using the program Geneious v. 6.1.7 —Biomatters—. Two matrices were built, one containing the ITS sequences —nrDNA matrix—, and the second containing the 5’trnK sequences —ptDNA matrix—. Informative indels were coded as binary characters. We performed Maximum Likelihood —M L— and Bayesian Inference —BI— phylogenetic analyses on each matrix as explained in Escudero & al. (2008), Martín-Bravo & al. (2013), and Villaverde & al. (2015) for the ITS and 5’trnK datasets individually. ML analyses were run with RAxML v. 7.2.6 (Stamatakis, 2010), using a GTR+GAMMA model of sequence evolution, as implemented in a Phylocluster —California Academy of Sciences—. Bootstrap support for branches was calculated through 1,000 replicates. BI analyses were run with MrBayes v. 3.2.5 (Ronquist & Huelsenbeck, 2003). Four simultaneous Markov Chain Monte Carlo —MCMC— chains were run for 5 million generations, sampling trees every 100 generations. The simplest models of nucleotide evolution that best fit the data for each studied DNA region were HKY for 5’trnK, HKY+I for ITS1 and ITS2, and JC for 5.8S region. Characters corresponding to coded indels were analysed with a F81 model. Congruence between the two resulting topologies was checked by eye and using Hompart test as implemented in PAUP* v. 4.0b10 (Swofford, 2002) in the same Phylocluster —California Academy of Sciences—. As no significant incongruences were found, the two data sets were combined into a total evidence matrix —combined matrix—, which was analysed again with ML and BI using the same parameters stated above. In order to assess and compare the degree of genetic differentiation between the studied taxa of C. sect. Sylvaticae, we calculated pairwise Kimura-2-parameter genetic distances between pairs of samples with MEGA v. 5.2 (Tamura & al., 2011), using the ITS, 5’trnK, and concatenated sequences.

22 herbarium specimens of typical C. sylvatica subsp. sylvatica (Appendix 1) and 11 specimens previously classified as C. sylvatica subsp. paui (Appendix 1; including the 3 vouchers from the Balearic and Tuscan Archipelagos, and the type of C. laxula from Sicily) were included in our study. For the morphological characterization of the materials we measured the diagnostic characters reported in previous taxonomic studies of C. sylvatica and allied taxa (Chater, 1980; Nilsson, 1985; Luceño & al., 2008; Egorova, 1999; Martín-Bravo & al., 2013), as well as additional characters derived from our observations, making a total of 32 quantitative and one qualitative traits (Table 2). Measurements were taken using a binocular micrometer —Nikon SMZ645—, with the exception of the largest macromorphological characters, which were measured using a standard 30-cm rule.

All statistical analyses of morphometric data were performed using the software SPSS Statistics 20 —IBM Corp., New York, Armonk—. First, we chose those variables with a higher correlation level —> 0.8—. Secondly, we removed those variables that contributed less to the first significant principal components. When those characters were removed, a clear morphological discontinuity was found between the 2 subspecies as analyzed through Principal Component Analysis —PCA—. A total of 7 variables were kept for the final analysis. We performed boxplot analysis for these variables in order to show the degree of overlapping between the two studied taxa. In the boxplot analysis the Navarran samples previously classified as C. sylvatica subsp. paui were treated as C. sylvatica subsp. sylvatica —see Results.

ITS and 5’trnK sequences from the 3 problematic C. sylvatica subsp. paui-like specimens from the Balearic and Tuscan archipelagos clustered with those from typical specimens from northwestern Africa and northeastern Spain —Catalonia—, both in separate —nuclear vs. plastid data sets— and combined analyses (Figs. 1, 6, 7). The northern Spanish Navarran samples, previously classified as C. sylvatica subsp. paui, however grouped with C. sylvatica subsp. sylvatica. The phylogenetic relationships revealed by the separate nrDNA and ptDNA trees and the combined tree were slightly different (Figs. 1, 6, 7), though they were not significantly incongruent after the Hompart test —p > 0.05—. The topology of the combined tree mostly agreed with the one yielded by the nrDNA data due to the higher number of informative characters from this data set in comparison with those from the ptDNA data set (Table 3). Carex sect. Rhynchocystis was strongly supported as sister group to C. sect. Sylvaticae in the combined and nrDNA tree —1.0 PP, 100% BS; 1.0 PP, 98% BS; Fig. 1 and 7, respectively—, whereas in the ptDNA tree C. sect. Rhynchocystis and C. sect. Ceratocystis were resolved as sister groups and in turn sister to C. sect. Sylvaticae, but without high support (Fig. 7). ITS and combined analyses yielded a strongly supported monophyletic C. sect. Sylvaticae —1.0 PP, >90% BS; Figs. 1, 6—, whereas ptDNA analysis showed a moderate support for the monophyly of C. sect. Sylvaticae —0.91 PP, < 50% BS (Fig. 7)—. The South African C. rainbowii was resolved as sister to the remaining lineages of the section —C. sylvatica and C. cretica— in the combined and nrDNA trees (Figs. 1, 6), with C. cretica being resolved as sister to C. sylvatica only in the combined analysis (Fig. 1). The 3 species collapsed in a basal polytomy in the ptDNA tree (Fig. 7). Carex sylvatica subsp. sylvatica and C. sylvatica subsp. paui were monophyletic but showed low support in the combined analysis —PP < 0.9, BS 65% (Fig. 1)—, while their sequences collapsed in a polytomy in the nrDNA and ptDNA phylogenies (Figs. 6, 7). Interestingly, while C. sylvatica subsp. sylvatica was only recovered as monophyletic in the ptDNA tree —0.99 PP, 81% BS (Fig. 7)—, C. sylvatica subsp. paui was strongly supported as a monophyletic group by all markers and analyses —1.0 PP, > 90% BS (Figs. 1, 6, 7).

The calculated genetic distances (Table 4) revealed that C. sylvatica subsp. sylvatica —including the Navarran samples mentioned above— was more distantly related to C. sylvatica subp. paui —ITS 0.016; 5’trnK 0.007; concatenated sequences 0.009— than to C. cretica —ITS 0.004; 5’trnK 0.005; concatenated sequences 0.006— and to C. rainbowii for 5’trnK sequences —0.003.

For the sake of simplicity, and according to our molecular results, we considered the Navarran samples previously classified as C. sylvatica subsp. paui to belong to C. sylvatica subsp. sylvatica.

The PCA including only seven variables —SL, INFL, UL, UBL, SPKMN, SPKAN, and PAP (Table 2)— revealed a clear separation between C. sylvatica subsp. sylvatica and C. sylvatica subsp. paui (Fig. 2). The first three principal components —PCs— accounted for 73.51% of the total variance —31.74%, 23.88% and 17.89% respectively—. The characters that contributed the most to the first 3 components were related to plant and utricle sizes, as well as the androgynous spike number (Table 6).

Boxplots showed that at least UL, UBL, SPKAN, and PAP displayed less than 25% overlap between the two studied taxa (Fig.3).

Table 5 summarizes those diagnostic characters that distinguish between both taxa according to our own results and previous studies (Luceño & al., 2008; Martín-Bravo & al., 2013).

Our molecular phylogenies revealed that the samples identified as C. sylvatica subsp. paui formed a well-supported clade, sister to the C. sylvatica subsp. sylvatica clade in the combined tree (Fig. 1). Moreover, the genetic distance between these 2 taxa was larger than that found between C. sylvatica subsp. sylvatica and C. cretica (Table 4), being this latter taxon usually considered a separate species, morphologically distinct from any other member of C. sect. Sylvaticae (Escudero & Luceño, 2009; Martín-Bravo & al., 2013). In addition, the results retrieved by the morphometric study revealed a good degree of differentiation and a number of diagnostic characters separating the 2 taxa, despite the small number of specimens examined —including the type of C. laxula (Fig. 2)—. At least 4 characters displayed no or few —< 25%— overlap (Fig 3, Table 2). The number of male spikes has been considered the most important and clear diagnostic character in previous studies (Table 5). Our study has revealed the existence of other important distinctive characters between both taxa: the number of androgynous spikes, the length of the stem, the inflorescence, the utricle, and the beak utricle, all of which are larger in C. sylvatica subsp. paui than in C. sylvatica subsp sylvatica (Fig.4). Also, the adaxial leaf surface is strongly rough in C. sylvatica subsp. paui, but smooth or slightly rough in C. sylvatica subsp sylvatica (Fig. 4, Table 5). Our data indicate that the taxon should be recognized at the species level. Based on the nomenclatural priority of the C. laxula name over C. paui Sennen, C. sylvatica subsp. paui is subsumed under C. laxula.

Our study clearly shows that the studied specimens from the northeastern Spain —Catalonia— and the Balearic and Tuscan archipelagos, as well as the type specimen of C. laxula formed indeed a well-characterized morphological taxon (Fig. 2). It is also in agreement with the phylogenetic nesting of the samples included in the molecular study (Figs. 1, 6, 7). By contrast, the Navarran samples previously classified as C. sylvatica subsp. paui fall within the variation of C. sylvatica subsp. sylvatica in both molecular and morphometric studies (Figs. 1, 2) and should therefore be classified within the latter taxon.

Carex sylvatica subsp. paui was first described at the specific rank —C. paui Sennen (Sennen, 1925)—, but the taxon was later recognized at the subspecific rank —C. sylvatica subsp. paui (Sennen) Bolòs & Bolòs— by Bolòs & Bolòs (1950) and later authors (Luceño & al., 2008; Jiménez-Mejías & Luceño, 2011; Govaerts & al., 2016). Early works indicated that the taxon was only known from Spain (Luceño, 1994; Luceño & al., 2008). Later, the populations of C. algeriensis Nelmes from Algeria and Tunisia (Maire, 1957) were synonymized to C. sylvatica subsp. paui based on morphological data (Jiménez-Mejías & Luceño, 2011; Martín-Bravo & al., 2013). This treatment has been supported by our phylogenetic study —v.gr., the nesting of the Tunisian sample in the phylogenetic trees (Figs. 1, 7)—. The finding of C. sylvatica subsp. paui to be conspecific to C. laxula, from Sicily, and its new records from the islands of Mallorca —Balearic Islands— and Elba —Tuscan archipelago—, greatly expands the presence of this taxon in the western Mediterranean (Fig. 5). There were no previous reports of this species for the Balearic and Elba Islands (Pignatti, 1982; Innamorati, 1991; Bolòs & Vigo, 2001; Conti & al., 2005). It implies an important range expansion of C. laxula —C. sylvatica subsp. paui— in Spain and a new taxon for the Italian flora.

Misidentifications of C. laxula and C. sylvatica have been common as a consequence of the very subtle morphological differences between both taxa (Table 5). Thus, the finding of more populations of C. laxula in other adjacent areas of the western Mediterranean could be expected. The misidentification of the Navarran populations of C. sylvatica subsp. sylvatica as C. laxula —C. sylvatica subsp. paui (Luceño, 1994; Luceño & al., 2008)— depicts a classical taxonomic problem. The Navarran individuals of C. sylvatica subsp. sylvatica display larger morphological variation than expected for the taxon (Global Carex Group, 2016). The detailed examination of these samples reveals that they show characters, such as a smooth upper side and margins of leaves, and the presence of only a few sparse prickles at the utricle beak, that match those detected in other studied samples of C. sylvatica subsp. sylvatica (Fig. 3). Nevertheless, these specimens also have a higher number of male and androgynous spikes and longer inflorescences than those usually found in C. sylvatica subsp. sylvatica, explaining previous misidentification of these plants.

The ecology of C. laxula contrasts with that of C. sylvatica subsp. sylvatica. The latter taxon inhabits Eurosiberian and Atlantic forests, being part of the understory in broad-leaf deciduous woods of beech —Fagus sylvatica L.—, oaks —Quercus spp.—, and riparian forests, mostly on moist to wet soils on sandy or stony-clay substrates (Hegi, 1969; Luceño, 2008; Nilsson, 1985; Pignatti, 1982). In contrast, C. laxula mostly grows in shady humid Mediterranean forests dominated by evergreen oak —Quercus ilex L.— and Corylus avellana L. woods, as well as in riparian forests with Alnus glutinosa (L.) Gaertn. and Populus spp., mainly on siliceous substrates such as sandstones, at relatively low altitudes —150-300 m s.n.m.— (Maire, 1957; Luceño, 1994; Luceño & al., 2008). The previous reports of C. laxula in beech forests —v.gr., C. sylvatica subsp. paui (Luceño, 1994; Luceño & al., 2008)— correspond to the Navarran populations here identified as C. sylvatica subsp. sylvatica, in agreement with the newly circumscribed niches of both taxa.

We provide an updated description for C. laxula and an identification key to separate it from C. sylvatica subsp. sylvatica:

Caespitose plant. Flowering stems up to 200 cm long, sharply trigonous, smooth. Inflorescences length 25-35(70) cm. Leaves 10-14 mm wide, shorter than stems, ± carinate, ± rigid, with the adaxial surface scabrid; ligule 1.5-3(5) mm long, longer than wide, apex obtuse; basal sheaths pale brown, entire, rarely fibrous. Lowermost bract shorter than the inflorescence. Male spikes (1)2-4(7), 14-45 mm long, fusiform, sometimes with a few utricles at the base, very rarely with utricles also at the top. Female spikes 3-4, (22)30-53 mm long, occasionally shortly branched at the base, at least the lower ones separated from the upper ones, with long filiform and pendant peduncles, the upper ones with shorter peduncles, sometimes arising very close to each other. Androgynous spikes (0)1-2(4). Male glumes oblong to obovate, light brown, acute, subacute or obtuse, rarely mucronate; female glumes elliptic, shorter than the utricles, hyaline or, exceptionally, pale brown with a wide scarious margin. Utricles 4-4.5 × 1-1.2(1.5) mm, suberect, ovoid, trigonous, greenish or brownish, with only 2 prominent veins, more or less abruptly contracted into a narrow, slender, bifid beak, 1.2-2(2.3) mm long, conspicuously scabrid, with prickles towards the top. Achenes (2)2.2-2.5 × 0.9-1.4 mm, ovoid, trigonous, greenish to pale brown.

It is a western Mediterranean endemic: northeastern Iberian Peninsula, Balearic Islands, Elba Island, Sicily, northern Algeria, and northern Tunisia (Fig. 5).

One of the isolectotype specimens we listed for C. algeriensis housed at BM —BM 000922724— displays a label that says it was collected in May —5— instead of April, as indicated in the protologue and other specimens. However, the collector number —37614—, and also the day of the month —27— and year —1937—, are the same. We consider that the difference in the label is just a typo when transcribing the new label, thus the material should be considered an isolectotype.