Field collections and herbarium material

Mature seeds of 11 Melocactus species were collected between June 2016 and June 2018. Mature fruits of 10 individuals in each one of the 14 studied localities were randomly selected (Fig. 1, Table 1). We chose seeds without deformations and that were neither apparently sterile or broken; vouchers are listed in Appendix 1. In cases of the seeds obtained from private collections, it was confirmed that individuals had been collected in wild localities. Each locality was analysed as a taxon, the identity of which was determined according to the types and protologues of the names of the taxa. In M. actinacanthus and M. acunae subsp. acunae, it was possible to study two localities per taxon, which allowed us to evaluate possible intraspecific variation.

Table 1.

Taxa and locations included in seed micromorphology study of Melocactus in Cuba. Species groups indicated according to Mészáros (1976) and Taylor (1991) criteria consensus. * indicates fruits obtained from ex situ collections with parents from natural populations. Herbarium specimens deposited in Herbarium “Prof. Dr. Johannes Bisse” (HAJB), Jardín Botánico Nacional, Universidad de La Habana, Cuba (see Appendix 1).

Stereomicroscopy and scanning electron microscopy

Seeds were examined by a stereoscopic light microscope (LM) Motic SMZ-168 and a scanning electron microscope (SEM) Hitachi Stereoscan Model Su1510 (10 kv voltage and secondary electron detector). For SEM, three seeds per taxon were randomly selected and washed in distilled water, TWENG detergent (10 drops), 4% sodium hypochlorite and five minutes of ultrasound. Later, they were fixed in an aluminium porta-sample with adhesive tape and coated with gold. This technique was carried out at the Electron Microscopy Laboratory of the Instituto de Biología de la Universidad Nacional Autónoma de México. For each seed, SEM micrographs were taken in several views: lateral, frontal, apical, and cells of several regions: lateral, hilum-micropylar border, and apical. In the case of seeds of Melocactus perezassoi, it was not possible to take SEM micrographs, and their morphology was evaluated according to the photographs presented by Guiggi (2010) and observations made with LM; the seeds photographed by Guiggi (2010) and those used in this study came from the same population (the only known population for this taxon). Seed morphological characters were analysed using the micrographs obtained and observations in LM of 50 seeds per taxon (or per locality, when a taxon was sampled from more than one locality). Observations in LM allowed checking of all characters analysed in SEM.

Qualitative analysis

Seed morphological and microrelief characters were evaluated for each of the seed regions according to Barthlott & Hunt (2000). For general morphology, size based on length and seed shape according to length/breadth ratio was described. The colour, lustre, multicellular sculpture, relief, periphery differentiation degree, keel extension, cell size gradient near keel, and hilum-micropylar region were evaluated for the testa. For the seed regions (Fig. 2), we evaluated seed shape, periclinal wall relief, and relief and curvature of anticlinal cell boundaries. In some regions, specific characters were evaluated, such as microrelief in the lateral region and testa border expansion around the hilum-micropylar region border. In the hilum-micropylar region, the characters evaluated were seed shape, position relative to border, as well as hilum and micropyle configuration.

Fig. 2.

Melocactus seed general scheme, regions and quantitative characters evaluated in lateral view (A) and frontal view (B).

Most of the characters and character states evaluated were assumed according to Barthlott & Hunt (2000). For periclinal wall relief, modified states of Barthlott & Hunt (2000) were used: flat, low-convex and high-convex. Seed relief was evaluated according to the states proposed by Stearn (1983). Keel extension and cell size gradient near keel are described for the first time for Melocactus in the present study. Keel extension was evaluated in correspondence to the regions occupied by this structure (Fig. 2A). Cell size gradient toward keel was evaluated in the same way as cell size gradient toward hilum-micropylar region, as in Barthlott & Hunt (2000).

Quantitative analysis

The study of the dimensions was performed in 50 seeds randomly selected per accession. Photos were made in LM in lateral (Fig. 2A) and frontal (Fig. 2B) views, on millimetre paper that served as a scale. The quantitative variables measured in the AxioVision Rel. 4.8 program were: seed length and breadth in lateral view (Fig. 2A), as well as hilum-micropylar region length and breadth and hilum-micropyle distance in frontal view (Fig. 2B). From the respective length and breadth values, the length/breadth ratio of both, seed and hilum-micropylar region, were determined. The hilum-micropylar region length/seed breadth ratio was also calculated.

The study of seed mass and testa mass was carried out from 50 seeds randomly selected per accession. The seed mass was evaluated in an analytical balance Sartorius (± 0.00001 g). Due to the similarity between the seed mass and the precision value of the balance, five seeds per replicate were randomly selected, for a total of ten replicates per taxon (or locality). In this way, the mass obtained for one seed was determined as the division between five of the mass of each replicate. Prior to the weighing, the seeds were placed in an electric drying chamber for 17 hours at 105 °C (± 2 °C) according to the ISTA (2007) standards to eliminate the water content. The dry mass of seed (DMS) was evaluated, and we then extracted the testa to determine the dry mass of testa (DMT). With these variables, the relative embryo mass (REM) was calculated according to Sánchez & al. (2002): REM = (DMS – DMT) / DMS.

Seed morphological similarity

A cluster analysis was performed based on quantitative non-correlated characters, and qualitative characters with discontinuity states and variation between taxa (Table 2). Quantitative characters were coded according to the proposal of Crisci & López-Armengol (1983) using average and standard deviation as two independent characters. In this way, a basic data matrix was made for 12 taxa (from 14 localities) and 20 characters, of them 14 multistate continuous quantitative, 4 bistate exclusive qualitative and 2 multistate qualitative characters with a logical sequence.

Statistical analysis

For quantitative characters, normality and variance homogeneity was checked by a Kolmogorov-Smirnov and Levene test. We performed three principal component analyses (PCA) using Past 2.17 (Hammer & al. 2001) to explore the multivariate relationships and identify which characters were the ones that most contribute to distinguish among taxa or inter-locality variation in each Melocactus group. Differences between taxa were analysed by a non-parametric multivariate variance analysis (PERMANOVA) with pairwise comparisons. PERMANOVA was carried out in Past 2.17 and was considered statistically significant for p < 0.05. For each variable, a graph with mean and standard deviation was made. Among quantitative characters, we carried out a Spearman correlation test, which were considered statistically and biologically significant for p < 0.05 and r > 0.7, respectively. The basic data matrix was analysed by cluster analysis, as suggested by Crisci & López-Armengol (1983). The similarity matrix was generated with the Euclidean distance coefficient and a phenogram was produced with the method of unweighted pair-group average. All analyses were carried out in Statistica 8.0. The interpretation of the results focused at the level of the species group defined according to the consensus of the criteria of Mészaros (1976) and Taylor (1991).

Table 2.

Qualitative characters and encoded states to evaluate affinities among taxa of Melocactus in Cuba.

Variation in qualitative features

General morphology — Seeds of Melocactus in Cuba vary from small (0.81–1.19 mm) in M. radoczii, to medium-sized (1.20–1.63 mm) in M. actinacanthus (Sierra Alta de Agabama), M. guitartii, M. matanzanus and M. perezassoi. Variation in seed shape occurs within most species between circular to broadly oval, but broadly oval were exclusively found in M. guitartii, M. harlowii and the three localities of M. acunae.

Testa features — Testa was black in all species, glossy in Melocactus acunae subsp. acunae (Punta de Maisí), M. borhidii, M. evae, M. harlowii, M. holguinensis and M. perezassoi, but matte occurs in the remaining eight taxa evaluated. Multicellular sculptures are absent in Cuban Melocactus. Relief shows different patterns: flat in the M. curvispinus group (Fig. 3A, B), and almost half of the M. harlowii group: M. acunae subsp. acunae (Punta de Maisí; Fig. 4A), M. borhidii (Fig. 4C), M. evae (Fig. 3C) and M. harlowii (Fig. 4B). On the other hand, the relief is coliculate in the M. matanzanus group (Fig. 5A–C), M. perezassoi and M. radoczii (Fig. 3D) and tuberculate in M. acunae subsp. acunae (Cajobabo; Fig. 6A), M. acunae subsp. lagunaensis (Fig. 6B) and M. nagyi (Fig. 6C).

All taxa evaluated have a keel, which is exclusively ventral in the Melocactus harlowii and M. matanzanus groups, but ventro-apical in the M. curvispinus group. Cell size gradient near the hilum-micropylar region is abrupt, except in M. actinacanthus (Sierra Alta de Agabama; Fig. 5L), M. acunae subsp. lagunaensis (Fig. 6N), M. evae (Fig. 3S) and M. harlowii (Fig. 4N), in which it is gradual. Cell size gradient near the keel is gradual, except in M. acunae subsp. acunae (Punta de Maisí), M. acunae subsp. lagunaensis, M. matanzanus, M. nagyi and M. radoczii.

Micromorphology of apical region — There is no variation in cell shape, all are elongate (Fig. 3U–X, 4P–R, 5O–Q, 6P–R). Periclinal wall relief is flat in Melocactus acunae subsp. acunae (Punta de Maisí), M. borhidii (Fig. 4R), the M. curvispinus group (Fig. 3U, V) and M. evae (Fig. 3W). Low-convex relief is present in the M. matanzanus group (Fig. 5O, P), although in M. matanzanus there are zones with high-convex periclinal walls (Fig. 5Q). Low-convex relief appears in M. acunae subsp. acunae (Punta de Maisí), M. acunae subsp. lagunaensis in which there are high-convex relief cells (Fig. 6Q), M. harlowii in which there are flat relief zones (Fig. 4Q), M. perezassoi and M. radoczii (Fig. 3X). Finally, high-convex relief appears in M. acunae subsp. acunae (Cajobabo; Fig. 6P) and M. nagyi (Fig. 6R). Anticlinal cell boundaries are generally straight with some irregularly curved areas.

Micromorphology of lateral region — Cells of the lateral region are elongate (Fig. 3N–P, 4J, K, 5I–K, 6J–L), isodiametric near the centre. In Melocactus perezassoi, toward the centre of the seed, there are elongated cells, and the isodiametric cells appear near the hilum-micropylar region border. Periclinal walls are flat, except in M. acunae subsp. acunae (Cajobabo; Fig. 6J), M. acunae subsp. lagunaensis (Fig. 6K), M. nagyi (Fig. 6L), M. perezassoi and M. radoczii (Fig. 3P), in which two states appear, flat and low-convex. Anticlinal cell boundaries are channelled and generally straight with some irregularly curved areas (Fig. 3M–P, 4J–L, 5I–K, 6J–L). The microrelief is striate, except in M. evae, which has no microrelief (Fig. 3O).

Micromorphology of ventral region — Cells in the ventral region are elongate with flat periclinal walls, except in Melocactus acunae subsp. acunae (Cajobabo), M. acunae subsp. lagunaensis and M. nagyi, which have low-convex periclinal walls. Anticlinal cell boundaries are channelled and straight.

Micromorphology of hilum-micropylar border — Hilum-micropylar border cells (Fig. 3Q–T, 4M–O, 5L–N, 6M–O) are isodiametric with flat periclinal walls. Anticlinal cell boundaries are channelled and straight. The hilum-micropylar region border is constricted in Melocactus holguinensis (Fig. 3B), expanded in M. evae (Fig. 3C), M. perezassoi and M. radoczii (Fig. 3D) and not expanded in the rest of the taxa studied.

Micromorphology of hilum-micropylar region — The hilum-micropylar region is oval except in Melocactus guitartii (Fig. 3E), which is a keyhole type. This region is superficial with respect to its border except in M. acunae subsp. acunae (both localities), M. borhidii and the M. curvispinus group, in which it is impressed. The hilum and micropyle form a group but are separated by sclerified tissue.

Fig. 3.

Seeds of Melocactus in Cuba, scanning electron microscopy in lateral view (A – D), frontal view (E – H) and apical view (I – L), and cells of lateral region (M – P), hilum-micropylar border region (Q – T) and apical region (U – X). – Columns 1 – 4 correspond to taxa: M. guitartii (1); M. holguinensis (2); M. evae (3); M. radoczii (4). – Scale bars: A – L = 500 µm; M – X = 100 µm.

Table 3.

Eigenvectors of principal component analysis (PCA) based on correlation matrix of seed dimensions of three Melocactus groups.

Variation in quantitative features

The PCA for the Melocactus matanzanus group showed overlap between taxa except for one point for the dimensions of both localities of M. actinacanthus that did not show overlap (Fig. 7A). The seed dimensions of M. actinacanthus from Sierra Alta de Agabama almost completely overlapped those of M. matanzanus (c. 80% of the smallest convex polygon), while the seeds of M. actinacanthus from La Coca were 38% of the smallest convex polygon. In this PCA, the first two components explained 71.84% of the variability. The characters that contributed the most to the first component were seed length, seed breadth and hilum-micropylar region length, while the second component was determined by hilum-micropylar region (HMR) length/seed breadth ratio, seed length/breadth ratio and hilum-micropyle distance (Table 3).

The PCA in the Melocactus curvispinus group showed both taxa well defined without interposition in the point cloud (Fig. 7B). In this PCA, the first two components explained 87.31% of the variability. The characters that contributed the most to the first component were HMR length, seed length and hilum-micropyle distance, while the second component was determined by seed length/breadth ratio, seed breadth and hilum-micropylar region breadth (Table 3).

Table 4.

Quantitative characters measured in Cuban Melocactus with F values of non-parametric multivariate analyses of variances with pairwise comparisons test and associated probability (p). – HMR = hilum-micropylar region.

The PCA in the Melocactus harlowii group showed interposition between taxa, except in M. perezassoi, where all its points were well separated from the rest of the group (Fig. 7C). For both localities of M. acunae subsp. acunae, the seeds from Punta de Maisí coincide in 86% with respect to the smallest convex polygon of the points of the seeds from Cajobabo. In this PCA the first two components explained 76.72% of the variability. The characters that contributed the most to the first component were hilum-micropylar region length, seed length and hilum-micropyle distance, while the second component was determined by HMR length/seed breadth ratio, seed length/breadth ratio and HMR length/breadth ratio (Table 3).

Significant differences were found for all quantitative characters evaluated in Melocactus seeds (Table 4). In the M. matanzanus group, the seeds of three localities differ from each other with respect to seed length (Fig. 8A) and seed breadth (Fig. 8B), as well as the dry mass of seed (Fig. 9C), although were similar with respect to seed length/breadth ratio (Fig. 8C) and HMR length/breadth ratio (Fig. 9A), and relative embryo mass (Fig. 9D). Melocactus actinacanthus (Sierra Alta de Agabama) and M. matanzanus were similar to each other and different from M. actinacanthus (La Coca) with respect to hilum-micropyle distance (Fig. 8F) and to HMR length/seed breadth ratio (Fig. 9B). The two localities of M. actinacanthus were similar in terms of HMR length (Fig. 8D) and HMR breadth (Fig. 8E). On the other hand, M. guitartii and M. holguinensis differed in all quantitative variables evaluated (Fig. 8, 9). In the M. harlowii group, M. perezassoi had the greatest seed dimensions of the genus in Cuba (Fig. 8). On the other hand, M. radoczii had the smallest seeds for most of the analysed variables. The remaining taxa of the M. harlowii group had seeds with intermediate values for analysed variables (Fig. 8, 9).

Fig. 4.

Seeds of Melocactus in Cuba, scanning electron microscopy in lateral view (A–C), frontal view (D–F) and apical view (G–I), and cells of lateral region (J–L), hilum-micropylar border region (M–O) and apical region (P–R). – Columns 1–3 correspond to taxa: M. acunae subsp. acunae (Punta de Maisí) (1); M. harlowii (2); M. borhidii (3). – Scale bars: A–I = 500 µm; J–R = 100 µm.

Fig. 5.

Seeds of Melocactus in Cuba, scanning electron microscopy in lateral view (A–C), frontal view (D–F) and apical view (G–H), and cells of lateral region (I–K), hilum-micropylar border region (L–N) and apical region (O–Q). – Columns 1–3 correspond to taxa (or localities): M. actinacanthus (Sierra Alta de Agabama) (1); 2: M. actinacanthus (La Coca) (2); 3: M. matanzanus (3). – Scale bars: A–H = 500 µm; I–Q = 100 µm.

Fig. 6.

Seeds of Melocactus in Cuba, scanning electron microscopy in lateral view (A–C), frontal view (D–F) and apical view (G–I), and cells of lateral region (J–L), hilum-micropylar border region (M–O) and apical region (P–R). – Columns 1–3 correspond to taxa: M. acunae subsp. acunae (Cajobabo) (1); M. acunae subsp. lagunaensis (2); M. nagyi (3). – Scale bars: A–I = 500 µm; J–R = 100 µm.

Fig. 7.

Principal component analysis (PCA) based on seed dimensions of Melocactus matanzanus group (A), M. curvispinus group (B) and M. harlowii group (C). – Convex hull is smallest convex polygon containing all points of each taxon or locality. – Seed L = seed length; Seed B = seed breadth; Seed L/B = seed length/breadth ratio; HMR L = hilum-micropylar region length; HMR B = hilum-micropylar region breadth; H-M Dist. = hilum-micropyle distance; HMR L/B = HMR length/breadth ratio; Lhmr/Bs = HMR length/seed breadth ratio; localities: AGA = Sierra Alta de Agabama; CAJ = Cajobabo; COC = La Coca; MAI = Punta de Maisí.

Fig. 8.

Variation of length (A) and breadth (B) seeds, seed length/breadth ratio (C), length (D) and breadth (E) of HMR and hilum-micropyle distance (F) of Melocactus seeds in Cuba. – Black circle (•) indicates mean; lines extending above and below mean indicate standard deviation. Different letters represent statistically significant differences among taxa (or localities), according to non-parametric multivariate analyses of variances and pairwise comparisons for p < 0.05. – HMR = hilum-micropylar region; localities: AGA = Sierra Alta de Agabama; CAJ = Cajobabo; COC = La Coca; MAI = Punta de Maisí. – Shading refers to three groups: M. matanzanus, M. curvispinus and M. harlowii in this order.

Seed morphological similarity

The basic data matrix (from quantitative and qualitative characters) was formed by 20 characters (Table 5). The clustering pattern showed at the 2.1 level, two large, well-defined groups (Fig. 10). The separation at this level is given by keel extension. Species of the C group are separated from 1.9, because their seeds differ in lustre, hilum-micropylar region shape, testa border expansion and all quantitative variables. Also, from 1.9, the H3, H4 and H5 groups differ from the M, H1 and H2 groups by lustre of seeds.

At a cut-off of 1.5, Melocactus perezassoi is separated from the H4 and H5 groups (Fig. 10) by seed relief, testa border expansion and seed dimensions. Melocactus evae (H4 group) is differentiated from the H5 group at a distance of 1.45 by testa border expansion, microrelief, seed length/breadth ratio and HMR length/seed breadth ratio. On the other hand, M. radoczii (H2 group) is separated from the group formed by H1 and M at a distance of 1.3 by testa border expansion, seed length and HMR length/breadth ratio. At a distance of 1.1, H1 differs from the M group by seed relief, seed and HMR length, hilum-micropyle distance, HMR length/seed breadth ratio and dry mass of seed.

There were three subgroups with high levels of seed morphological similarity, in which the Euclidean distance is less than 0.5. In the case of the H5 group, the taxa with the highest similarity are Melocactus acunae subsp. acunae (Punta de Maisí) and M. borhidii, which differ by a cut-off of 0.3 for seed length and its seed length/breadth ratio. Taxa of the M group are similar to each other—only M. actinacanthus (La Coca) differs by a cut-off of 0.25 due to seed length, hilum-micropylar region length/seed breadth ratio and relative embryo mass. In the H1 group, M. acunae subsp. lagunaensis and M. nagyi (Fig. 10) differ by a distance of 0.25 due to HMR length/breadth ratio.

Fig. 9.

Variation of HMR length/breadth ratio (A), HMR length/seed breadth ratio (B), dry mass (C) and relative embryo mass (D) of Melocactus seeds in Cuba. – Black circle (•) indicates mean; lines extending above and below mean indicate standard deviation. Different letters represent statistically significant differences among taxa (or localities), according to non-parametric multivariate analyses of variances and pairwise comparisons for p<0.05. – HMR = hilum-micropylar region; localities: AGA = Sierra Alta de Agabama; CAJ = Cajobabo; COC = La Coca; MAI = Punta de Maisí. – Shading refers to three groups: M. matanzanus, M. curvispinus and M. harlowii in this order.

Fig. 10.

Position of 12 taxa (from 14 localities) of Melocactus in Cuba, based on Euclidean distance matrix through unweighted pair-group average. – M = M. matanzanus group; H = M. harlowii group; H1–H5 = M. harlowii group subgroups; C = M. curvispinus group; localities: AGA = Sierra Alta de Agabama; CAJ = Cajobabo; COC = La Coca; MAI = Punta de Maisí.

Taxonomic implications of seed variability in Cuban Melocactus

Taylor (1991) argued that there is limited variation in seed size and morphology within the Melocactus taxa that he recognized for Central and South America. Our results allow us to recognize that several characters of the seed have discontinuous variation. Based on the taxonomic value of these characters, we argue for their importance in recognizing the species in Melocactus groups native to Cuba.

Melocactus matanzanus group

The three populations of the Melocactus matanzanus group in Cuba have seeds of similar morphology. The seed similarity in this case agrees with the taxonomic proposal of Taylor (1991), and followed by Hunt & al. (2006), Acevedo-Rodríguez & Strong (2012), Hunt (2016) and Greuter & Rankin (2017), who consider M. actinacanthus as synonymous with M. matanzanus. Vegetative, carpological and floral characters that sustain the similarity between these taxa are simple stems from 8 to 9 ribs, with flowers less than 2 cm and pink fruits (León 1934; Areces-Mallea 1976a). Guiggi (2010) recognized one species with two subspecies in this group. The main differences between M. actinacanthus and M. matanzanus according to Areces-Mallea (1976a) and Guiggi (2010) are diameter and height of the plant without the cephalium, cephalium diameter, number of seeds per fruit, flower length, radial spine number and the presence of a central spine only in M. matanzanus.

Seeds of two populations of Melocactus actinacanthus showed differences among seed dimensions (Fig. 7–9). However, point distributions of seed dimensions of M. matanzanus overlapped with both populations of M. actinacanthus. With respect to plant morphology, M. actinacanthus (La Coca) has intermediate characteristics between M. matanzanus and M. actinacanthus (Sierra Alta de Agabama) (Hernández & al. 2014). In addition, the common mentioned presence of a central spine in the individuals of M. matanzanus in its locus classicus is not a constant character (Barrios pers. obs.), because individuals without a central spine can be found. According to our results, we consider that there is insufficient evidence to separate the two entities into different species or subspecies. Based on our seed morphological comparisons, we consider M. actinacanthus as synonymous with M. matanzanus.

Melocactus matanzanus seeds are similar to those of the M. violaceus group formed by M. concinnus Buining & Brederoo, M. neryi K. Schum., M. salvadorensis Werdermann and M. violaceus Pfeiff., as shown in the micrographs presented by Taylor (1991). Keel extension, testa border expansion, seed relief, microrelief and HMR shape are among the similar seed features of these species. Regarding seed length and breadth, there is overlap among intervals of M. matanzanus and similar species of the M. violaceus group. On the other hand, there are vegetative, carpological and floral characters that lend support for seed similarity among these species. The common characters among M. matanzanus and the related species of the M. violaceus group are the number of ribs, which varies to around 8 (León 1934; Taylor 1991), the same number of radial spines (León 1934; Taylor 1991) and cephalium length, which varies between 4–5 cm in M. matanzanus (León 1934), M. cocinnus, M. neryi and M. violaceus; however, this differs in M. salvadorensis, in which it exceeds 10 cm and can reach 15 cm (Taylor 1991). According to Taylor (1991), some species of the M. violaceus group crossed the Amazon and colonized part of northeastern South America and Cuba. Disjunct distributions among M. matanzanus and the M. violaceus group have been considered by Taylor (1991) as an example of long-distance dispersal, and birds are known to disperse the seeds of Melocactus (Casado & Soriano 2010).

Melocactus curvispinus group

Melocactus guitartii and M. holguinensis share only the presence of a ventro-apical keel in the seeds. Therefore, the seed characters sustain the recognition of two species, which agrees with Mészaros (1976), Areces-Mallea (1993), Rodríguez (2005) and González-Torres & al. (2016). When comparing the results of this study with the micrographs presented by Taylor (1991), seeds of M. guitartii are similar to those of M. curvispinus subsp. curvispinus, but not like those of M. holguinensis. Hilum-micropylar region shape, keel extension, seed relief, testa border expansion and seed length show a great similarity between the M. curvispinus and M. guitartii seeds referred to by Taylor (1991).

On the other hand, seed characters concur with the morphological similarities of stems, fruits and flowers found between Melocactus curvispinus subsp. curvispinus and M. guitartii. Therefore, this study supports the criterion of Taylor (1991), who considers that M. guitartii is synonymous with M. curvispinus. On the other hand, the seed morphology of M. holguinensis does not correspond to that of M. curvispinus subsp. curvispinus or M. guitartii. Taylor (1991) suggested that seeds of this group are characterized by their great length, which is not the case in M. holguinensis. Therefore, in the M. curvispinus group, two taxa are recognized in Cuba, M. curvispinus subsp. curvispinus (= M. guitartii) and M. holguinensis.

Melocactus harlowii group

Morphology of seeds in the Melocactus harlowii group allows differentiation of three of its taxa, M. evae, M. perezassoi and M. radoczii. In the remaining localities, according to seed morphology, two groups are distinguished (H1 and H5; Fig. 10). The first group (H1) is formed by M. acunae subsp. acunae (Cajobabo), M. acunae subsp. lagunaensis and M. nagyi. The second group (H5) is formed by M. acunae subsp. acunae (Punta de Maisí), M. borhidii and M. harlowii. The definition of the taxonomic identity of the taxa formed in these two groups will need subsequent studies that integrate other aspects of the biology of these species, such as floral, carpological, karyological morphology and phylogenetics. Although Taylor (1991), Hunt & al. (2006), Acevedo-Rodríguez & Strong (2012), Hunt (2016) and Greuter & Rankin (2017) considered that the M. harlowii group represents a single species, our results show that within this group seed morphology is useful to recognize more than a single species.

Melocactus evae, M. perezassoi and M. radoczii differ from each other by seed relief, microrelief and seed dimensions. The recognition of these three taxa does not correspond to the delimitation proposed by Guiggi (2010), because he suggested M. perezassoi as a subspecies of M. harlowii, and M. evae and M. radoczii as synonymous with M. harlowii subsp. harlowii. According to Areces-Mallea (1993), the population of M. perezassoi seems to have been left as a remnant of an original population on coastal cliffs, after the regression of the sea. Seeds of M. evae and M. radoczii are more similar to M. perezassoi than to M. harlowii, due to their expanded testa border. The reasons why M. perezassoi can be recognized as an independent species within the M. harlowii group are geographic isolation and marked differences in terms of vegetative, carpological and floral characters (Areces-Mallea 1993), as well as the seed morphology analysed here. This species presents more separated areoles with a greater number of spines that are less robust, longer flowers with stigmatic lobes not exserted above the stamens, and long, clavate fruits (Areces-Mallea 1993). On the other hand, characters such as size and time of flower opening, as well as the habit of forming very tight clusters, distinguish M. evae from the rest of the M. harlowii group. Melocactus radoczii, despite possessing similar characteristics to M. harlowii, is among the species with the greatest number of spines per areole (up to 22 spines) (Barrios pers. obs.).

The H1 group formed by Melocactus acunae subsp. acunae (Cajobabo), M. acunae subsp. lagunaensis and M. nagyi is characterized by tuberculate seeds. The seeds with tuberculate relief are present only in this group, which allows differentiating these populations from the rest of M. harlowii group. According to seed morphology, taxa belonging in this group show a marked difference with M. harlowii s.s., which contrasts with the taxonomic proposals of Taylor (1991), Areces-Mallea (1993), Rodríguez (2005), Hunt & al. (2006), Guigii (2010), Hunt (2016) and Greuter & Rankin (2017), which consider these taxa as synonymous with M. harlowii. Seeds that Taylor (1991) considered as M. harlowii are from a locality that is part of the distribution of M. acunae subsp. acunae (Mészarós 1976; Howard 1988); therefore, seed morphology of M. harlowii referred to by Taylor (1991) coincides with M. acunae within the M. harlowii group. Another species putatively related to M. acunae is M. lemairei (Monv. ex Lem.) Miq. ex Lem. from Hispaniola (Taylor 1991). Melocactus lemairei has seeds similar to those of M. acunae subsp. acunae (Cajobabo) and M. acunae subsp. lagunaensis, due to its tuberculate appearance, ventral keel, oval hilum-micropylar region and non-expanded border.

The H5 group formed by Melocactus acunae subsp. acunae (Punta de Maisí), M. borhidii and M. harlowii has in common lustrous seeds with flat relief, an unexpanded testa border and oval hilum-micropylar region. The seed morphology of these three taxa favours considering them as a single species, which does not correspond to the criterion of Mészarós (1976), which considered them as three species. However, other authors have included M. borhidii under the synonymy of M. harlowii (see Taylor 1991; Areces-Mallea 1993; Rodríguez 2005; Hunt & al. 2006; Guigii 2010; Hunt 2016; Greuter & Rankin 2017). The seeds in this group are similar to those of M. intortus (Mill.) Urb., because they have flat relief, striate microrelief, an unexpanded testa border and oval HMR. However, they differ by keel extension; due to the seeds of M. intortus having a ventro-apical keel (Taylor 1991).

One peculiarity in this study is the existence of different seed patterns for two evaluated localities of Melocactus acunae subsp. acunae (Fig. 10) according to seed dimensions (not distinguishable in PCA; Fig. 7C), which places them in different groups (H1 and H5). This result could be explained by the presence of inter-locality seed variation in M. acunae subsp. acunae, and this variation has been widely documented in plants (see Imbert & al. 1997; Zhang 1998; Tieu & al. 2001; Sánchez & al. 2002; Pérez-García & al. 2003), even in Cactaceae (see Taylor 1991; Cota-Sánchez & Bomfim-Patricio 2010; García-Beltrán & al. 2017). Morphological discontinuities are the basis of the phenetic species concept (Judd 1981; 2007), which allows the identification of populations with discontinuous morphological patterns as different species. In taxa such as M. acunae subsp. acunae, in which we observed inter-locality variation, it is not possible to apply such a concept, and the evaluation of other characters that complement this study will be necessary.

Based on our results, we propose the hypothesis that the distribution of Melocactus in eastern Cuba seems to be in response to several speciation processes. Parapatric speciation could be occurring in seven of the localities evaluated, because the localities are close to each other (Fig. 1), while allopatric speciation processes could have been responsible for the differentiation between M. nagyi and M. perezassoi (i.e. the rest of the M. harlowii group). An essential element in speciation in this group is dispersal and the fact that the fruit are readily consumed, and seeds could be easily dispersed. Melocactus seeds are characterized by being small to medium-sized and according to Matilla & al. (2005), dispersal probability increases in smaller seeds. In this sense, as hypothesized by Borhidi (1996) for succulents in the Cuban flora, it is possible that the ancestral populations of the M. harlowii group in eastern Cuba dispersed their seeds over long distances into different habitats, where geographic isolation caused by changes in vegetation distribution during the Pleistocene caused by climatic fluctuations favoured differences among localities through various speciation processes. In the future, an extensive phylogeographic study is required, to explain the historical processes responsible for the current geographic distributions, the possible gene flow between populations, but also as a comprehensive evaluation of the current circumscription of species.