In situ morphometric survey elucidates the evolutionary systematics of the Eurasian Himantoglossum clade (Orchidaceae: Orchidinae)

Individual plants, all taxa present

The first two coordinates based on analysis of individuals of all taxa for all usable characters (Fig. 7A) together account for 43% of the total variance, and work together to organise the plants in a diagonal array. All members of the hircinum–caprinum clade other than H. montis-tauri form a near-linear arrangement from the top-left to the mid-bottom of the plot, whereas the morphologically distinctive subgenus Barlia is isolated in the top-right quadrant. Placed between these two groups as separate clusters are H. montis-tauri, H. formosum and H. comperianum. Both coordinates are dominated by markings found on the sepals and/or perianths (Table 2). All individuals located below the solid line superimposed onto Fig. 7A lack any internal markings on the sepals and all butH. comperianum lack discrete labellum markings. Larger gynostemia and broader ‘abdomens’ also help to separate the hircinum–caprinum clade (left) from the remainder on the first coordinate. The considerably less informative third and fourth coordinates are not depicted here. The third coordinate partially separates H. hircinum, H. adriaticum and H. caprinum from the remaining taxa on the basis of the larger sepals and labella of the latter, whereas the fourth coordinate uses primarily the diffuse background colours of the sepals and labella to wholly separate the purple-flowered H. formosum from the paler, greenish-flowered H. caprinum and H. montis-tauri.

The main consequence of omitting the 12 pigmentation characters from the full matrix was to collapse H. montis-tauri and H. comperianum into the main group of plants (Fig. 7B), demonstrating that their apparent morphological distinctiveness relies heavily on anthocyanin-based characters. Their downward displacement on the second coordinate leaves only H. formosum as morphologically intermediate between the main group and subgenus Barlia. Predictably, the dimensionality of the variation is reduced, such that the first two coordinates account for an increased 47% of the total variance. In compensation, H. adriaticum becomes separated (just) from H. hircinum on the first coordinate, and a narrow discontinuity opens between them and H. jankae—morphologically the closest member to section hircinum of the remainder of the hircinum–caprinum group. This separation of H. hircinum and especially H. adriaticum from the remaining taxa reflects several characters, including their comparatively small columns, narrow shoulders and torsos, and narrow limbs (Table 3). The much weaker third coordinate (not shown) widely separates H. comperianum and, to a lesser degree, H. formosum from the remainder on the basis of their long, curved spurs and, in the case of H. comperianum, their few-flowered inflorescences. The fourth coordinate separates H. formosum from H. comperianum, due primarily to its longer stem and correspondingly longer inflorescence (Table 3).

Individual plants, taxa restricted to hircinum–caprinum clade

Himantoglossum comperianum, H. robertianum, H. metlesicsianum and H. formosum were then removed from the analysis in order to better explore the more subtle variation evident within the hircinum–caprinum clade (Fig. 8).

Despite these additional constraints, the first two coordinates utilising all characters account for a similar proportion (42%) of the total variance (Fig. 8A). Although a wider range of dimensions of floral organs now dominates the first coordinate (Table 4), the relative positions of the taxa on the first coordinate resemble those evident in the all-taxon analysis (Fig. 7A): hircinum plus adriaticum occupy one end of the coordinate and caprinum plus jankae ‘robust’ occupy the other. The second coordinate summarises a wide, heterogeneous range of characters, including several that represent anthocyanin markings. It largely separates hircinum, caprinum and ×samariense from the remainder, including narrowly distinguishing H. adriaticum from H. hircinum. The much weaker third coordinate separates the anthocyanin-deficient, vegetatively comparatively weak H. montis-tauri from the remainder.

Removing pigmentation characters (Fig. 8B) increases the amount of variance accommodated by the first coordinate, which now dictates a narrow discontinuity that separates hircinum plus adriaticum from the remaining taxa, primarily on the basis of their small sepals, though many other characters also contribute to the coordinate (Table 5). The second coordinate is almost entirely determined by characters that represent vegetative vigour and consequently has limited taxonomic relevance, serving primarily to distinguish the comparatively small-bodied H. montis-tauri. The third coordinate (not shown) succeeds only in partially separating adriaticum, calcaratum and jankae s.s. from the remainder.

The most striking feature of both ordinations is that the positions of plants across the plot broadly reflect their relative longitudes, western European plants being confined to the left-hand region of the plot and plants of Asia Minor being confined to the right (Figs. 8A and 8B).

Population means, all taxa present

Ordinations of population means also have superimposed upon them the corresponding minimum spanning trees, which are useful for indicating the relative strengths of the links connecting populations. Theory predicts that populations of the same species should most closely resemble each other rather than populations of other species. Ideally, to optimise this similarity test, more populations of each species would have been measured by us (indeed, H. formosum is represented in our matrix by only one population and so is effectively untestable in this way). Also, within several populations, sample sizes are undesirably small, epitomised by the three populations of H. ×samariense—the single measurable plant found in each population risks incurring serious sampling error when we are obliged to view that plant as representing the entire source population.

The plot using all characters for all populations (Fig. 9A) encompasses a similar amount of variation as the plots for individual plants (45%). It greatly separates from the main cluster both H. comperianum (on both coordinates) and subgenus Barlia (on the first coordinate only); they are linked to the main cluster of populations only weakly, as is H. formosum, which is distanced from all other populations on the plot of the third and fourth coordinates (not shown). Himantoglossum montis-tauri is also somewhat distanced from the main cluster. Most conspecific populations link to each other strongly, the exception being the single plant representing the population of H. ×samariense from its type locality in the Samaria Gorge of western Crete; this instead links weakly to H. caprinum.

Omitting pigmentation characters from the analysis (Fig. 9B) increased the variance accounted for to 52% but yielded broadly similar positioning of most populations. The most sigificant changes were that H. montis-tauri was pulled deeper into the main cluster of populations, whereas H. hircinum and H. adriaticum became attached to each other rather than to H. jankae, and were further distanced from section caprinum. In addition, the Bukovki population of H. jankae became interpolated between the two populations of H. caprinum.

Population means, taxa restricted to hircinum–caprinum clade

Restricting the population-level analysis to the H. hircinum–caprinum clade (Fig. 10A) considerably reduced the degree of disparity among maximum similarities—in other words, the taxonomic relationships appear more egalitarian. Conspecific populations are reliably connected with strong links, H. jankae seemingly occupying a central position within the clade. But as in the all-taxon analysis, the Samaria population of H. ×samariense is linked to H. caprinum. And in this case, the two remaining H. ×samariense ‘populations’ (strictly, plants) are linked, albeit weakly, to the Sandwich population of H. hircinum. The third coordinate (not shown) primarily separated H. montis-tauri from the remainder.

Omitting pigmentation characters from the analysis (Fig. 10B) once again unified the three populations of H. ×samariense (this time weakly attached to H. montis-tauri), and as in the all-taxa analysis, the two populations of H. caprinum became separated. More surprisingly, the single population of H. jankae ‘robust’ became strongly attached to the Mehmetali population of H. caprinum. The third and fourth coordinates offered no taxonomic separation.

Pigmentation

The summary of frequencies of characters representing discrete floral markings (Fig. 11A; see also Fig. 7A) makes clear how the presence or absence of each of the four categories of floral marking (discrete spots on the labellum, discrete spots on the interior of the sepals, discrete dashes on the interior of the sepals, peripheral stripes on the exterior of the sepals) interact to diagnose four groups of species. The only species to possess labellum markings but no sepal markings is H. comperianum (labellum markings were especially numerous in the Sanctuario population of comperianum), whereas H. robertianum and H. metlesicsianum also possess sepal spots. The only marking type possessed by most individuals of H. formosum and H. caprinum is the peripheral stripe on the sepals. Most plants of the remaining species possess all four kinds of marking, except that the majority of H. montis-tauri lack peripheral stripes. However, the presence of each kind of marking in each taxon cannot be wholly relied upon; only H. adriaticum, together with H. metlesicsianum and H. comperianum, appeared to be fixed for presence or absence of all four categories of marking (Fig. 11A). And given our small sample sizes for these species (only four and five plants, respectively), it is likely that we may simply failed to detect such variants.

Flower colour proved to be challenging to summarise when considering Himantoglossum species, as the perianth segments typically had a base colour of yellow-green that then appeared to be ‘overwashed’ with brown, purple or red pigments (Figs. 1–3). Figure 12 shows mean values for two of the three quantified CIE parameters (x and y) that together represent the background colour of the marginal regions of the labellum; Fig. 13 presents comparable data for the exterior surfaces of the sepals.

Most study populations of the same taxon averaged similar labellum colours (Fig. 12), though the three plants of H. ×samariense are spread especially widely on the colour grid and the Spanish (Torcal) population of H. robertianum exhibited unusually purplish hues that approached those more typical of H. metlesicsianum. Himantoglossum formosum and H. comperianum tended to have labellar margins that are purplish rather than greenish-brown, whereas H. hircinum, H. adriaticum, H. montis-tauri and especially the mainland Greek (Taygeti) H. ×samariense leaned toward brownish-yellow.

A wider range of mean values is evident in the equivalent plot for sepal colour (Fig. 13). Although most populations have broadly similar colours in the labellum and sepals, there are exceptions. Most notably, the sepals of H. comperianum are unusually red, whereas their labella are purple. The converse is true of two of the three populations of H. robertianum, which have purplish sepals associated with browner labella. In addition, sepal colour usefully distinguishes the yellow-green-sepalled H. hircinum from the mauve-sepalled Köszeg population of H. adriaticum (Fig. 13).

In addition to the precise hue, we can also usefully consider the depth of coloration of the labellum margin (Fig. 14A). Clearly, the majority of taxa are reliably dark flowered (defined here as a reflectivity of incident light of less than 20%). However, H. caprinum, H. jankae s.s. and H. calcaratum show wider spreads of pigmentation density; a minority of individuals of these taxa, together with some plants of H. comperianum, have moderately reflective labella (20–40% reflectivity). The remainder of the H. comperianum plants, together with all of the H. montis-tauri and H. metlesicsianum plants measured here and approximately one tenth of the H. robertianum, have comparatively pale flowers that are characterised by reflectivities that exceed 40%.

Sepal, petal and gynostemium

Gynostemia of H. adriaticum, H. jankae and H. montis-tauri are narrower relative to their length than are those of the remaining taxa (Fig. 14A). Those of H. adriaticum and H. hircinum are shorter than those of the remaining taxa, though there exists partial overlap in size with the gynostemia of H. jankae s.s.

Dimensions of sepals and lateral petals readily distinguish the long-hooded H. comperianum from the remaining species (Table 1, Fig. 14B). Although overlapping in length with H. robertianum, section hircinum has the shortest sepals (excepting the Moroccan population of H. hircinum: Bateman et al., 2013), and possesses lateral petals that are both the shortest and narrowest. Also, the sepals of H. ×samariense are unusually narrow relative to their length.

Lateral sepal orientation, as perceived relative to the vertical when the flower is viewed from the front, readily distinguishes subgenus Barlia plus H. formosum (Fig. 11B). They show a mixture of partially and wholly spreading sepals, whereas the remaining taxa reliably incorporate the lateral sepals into the hood (galea) that is consistently formed by the median sepal and lateral petals. The hood in turn completely overhangs the gynostemium.

Long, filiform lateral teeth proved to be ubiquitous on the lateral petals of H. comperianum. Shorter, sturdier teeth projected from the petals of the majority of plants of H. formosum, as well as from a small minority of plants of each of H. jankae s.s., H. calcaratum and H. hircinum (Table 1). Such teeth are less frequent across the genus as a whole than was implied by some previous authors (e.g., Delforge, 1999; Sramkó et al., 2014).

Labellum, including spur

Several characters were needed to represent with acceptable accuracy the unusually complex labellum shape of Himantoglossum s.l. species (Figs. 15A, 16–18).

The plot of maximum length versus width of the labellum (Fig. 16A) alone is sufficient to distinguish several of the study species. Comparatively short, broad labella characterise subgenus Barlia, which has an especially high width : length ratio and incurs a greater coefficient of variation for width than for length (mean width is greater for H. metlesicsianum than for H. robertianum). In contrast, the remaining taxa have labella that are much longer than wide and greater coefficients of variation for length than for width. Furthermore, section caprinum can achieve greater mean and maximum lengths than can section hircinum. The width : length ratio is greatest for H. montis-tauri and least for H. adriaticum. Individuals of H. formosum form a fairly compact, intermediate cluster.

A broadly similar pattern is evident in labellum torso dimensions (Fig. 16B); species of subgenus Barlia and subgenus Himantoglossum differ radically in length : width ratios, H. formosum occupying a position intermediate between them. The main exception is H. comperianum, which is long overall (Fig. 16A) but has a comparatively short torso, of similar length of H. formosum and similar width to H. robertianum (Figs. 16B and 18). Shoulders are reliably narrow in section hircinum.

The length of the ‘thorax’—the proximal portion of the labellum that stretches from the spur entrance (‘neck’) to the ‘armpit’ (Fig. 6)—is greatly expanded in H. comperianum relative to the other species (Figs. 15A and 18), which may be the reason that it lacks the three-dimensional ‘crenulae’ that characterise the shoulders of all other species in the genus. Although present, these marginal irregularities are larger—and therefore fewer in number—in subgenus robertianum. The ‘thorax’ is also comparatively long in H. formosum and H. metlesicsianum (which exceeds H. robertianum in this character), whereas it is short in H. adriaticum and especially in H. hircinum. The spread of values for this character is especially wide within H. montis-tauri.

The main plot of labellum arm versus leg length (Fig. 17A) shows that arms are shortest in H. formosum and, to a lesser degree, in the robertianum group. The smaller inset diagram shows clearly that the legs of H. comperianum are on average longer, and the arms much longer, than those of any other Himantoglossum species. Variation in arm length is great within most species, incurring remarkably large coefficients of variation. Himantoglossum hircinum and H. calcaratum are capable of generating longer arms than the remaining species. Even more striking variation surrounds leg length in most taxa of the hircinum–caprinum clade, though not in H. hircinum itself (Figs. 17A and 18). Indeed, 11% of the H. hircinum plants studied lacked legs altogether, the central labellar lobe being entire rather than apically notched into the characteristic leg-like ‘lobules’. Legs exceeding 5 mm in length form a great majority of most other taxa in the aggregate, notably in the largest-flowered taxa—H. calcaratum and H. jankae ‘robust’ (Fig. 18). The two populations of H. adriaticum differ significantly in this character; those from Köszeg have legs that are considerably shorter than those from Nyirád (mean values 1.8 versus 7.5 mm), thus being more comparable in size with those of H. hircinum. Arms are considerably longer than legs in most individuals of section hircinum (Fig. 18).

Limb widths (Figs. 17B and 18) readily separate the filiform elongations of H. comperianum labella from the wide-armed and especially wide-legged subgenus Barlia. In between these two extremes, section hircinum plus H. ×samariense tend to have narrower limbs than do either section caprinum or H. formosum. Only 32% of subgenus Barlia, together with a single plant of H. hircinum, possessed small fifth lobes (‘tails’) located between the legs in the ‘crotch’ of the labellum.

Spur dimensions (Fig. 15B) are also highly diagnostic. In particular, H. formosum and H. calcaratum have long spurs (those of H. calcaratum being broader than those of H. formosum, comparable in width with spurs of H. montis-tauri and subgenus Barlia) and H. comperianum has even longer spurs; those of both H. comperianum and H. formosum typically exceed 75% of the length of the corresponding ovary. Spurs of section hircinum are shortest, and within that section, those of H. adriaticum tend to be even narrower than those of H. hircinum. Greater length appears to permit greater downward curvature of the spur, most notably in H. comperianum (Table 2).

Recording the approximate angle subtended by the labellum relative to the stem showed that H. ×samariense possesses the most outwardly projecting flowers, whereas in contrast, those of H. metlesicsianum are held even closer to the vertical than are those of H. robertianum. Also, relative to the attitude of the torso, the minute arms of H. formosum project forward, the more substantial arms of H. metlesicsianum are borne in approximately the same plane as the torso, but those of all other species (including H. robertianum) usually recurve—most strongly so in H. montis-tauri (Table 2).

Vegetative organs

Unusually for a morphometric study of European orchids, vegetative vigour here plays a comparatively minor role in providing highly variable, and thus potentially taxonomically diagnostic, characters. Moreover, only occasionally do highs and lows in the number and sizes of various vegetative organs strongly co-vary. Hence, vegetative characters are not explored in detail in the present text.

In summary, subgenus Barlia have the most robust stems, though they are matched in this character by the Ifrane population of H. hircinum. Himantoglossum caprinum, H. ×samariense, H. montis-tauri and especially H. comperianum tend to have fewer flowers, whereas inflorescences are densest in subgenus Barlia and in H. hircinum.

Total leaf numbers are comparatively low in subgenera Comperia and Barlia, and also in H. montis-tauri and H. ×samariense. Both subgenus Barlia and the relevant members of subgenus Himantoglossum appear to compensate in other ways for this potential deficiency in photosynthetic surface area; subgenus Barlia produces comparatively large leaves, whereas both H. montis-tauri and H. ×samariense produce comparatively large bracts.

Geographical separation: overview

The Himantoglossum clade brings into sharp relief arguably the most serious general problem that besets systematic biology—that of distinguishing between (1) clinal change across contiguous geographic regions that is best viewed as infraspecific variation versus (2) hybrid zones separating two bona fide species that are distinguished by substantial, but nonetheless incomplete, reproductive isolation.

As they are currently conceived, all species other than the Canary Island endemic H. metlesicsianum abut geographically at least one other species of Himantoglossum—only the High Alps wholly lack Himanoglossum populations. However, substantial overlap of species distributions is also uncommon within each subgenus, most notably within the hircinum–caprinum clade that is comparatively rich in formal taxonomic epithets (exceptions are the two supposed regional endemics of subgenus Himantoglossum: H. calcaratum within the broader territory of H. jankae s.l., and H. montis-tauri within the broader territory of H. caprinum: Fig. 5). This largely jigsaw-like biogeographic arrangement of putative species is consistent with taxonomic partitioning of a morphological (and thus potentially a genetic) continuum. The fact that the morphometric ordinations substantially reconstruct the west–east distribution of the sampled populations (Figs. 7 and 8)—most notably within subgenus Himantoglossum—across ca 65° of latitude (Fig. 5) obliges us to consider this conundrum particularly seriously. Especially when this pattern is also viewed in the context of the rarity of reports of natural hybrids among the taxa within subgenus Himantoglossum.

In his study of the dimensions of flower parts of Greek populations of the jankae–caprinum clade, Tsiftsis (2016) detected a south-to-north increase in the lengths of labellum, spur and ovary (but not lateral labellum lobes) potentially driven by differences in regional climates. Although populations from Lesvos and the Peloponnese attributed by some authors to H. caprinum (or even occasionally to H. montis-tauri) possessed on average smaller flowers than those studied further north, the apparently clinal nature of these size differences caused Tsiftsis (2016) to argue that, among populations of the H. jankae group in Greece, there was no clear taxonomic structure. He therefore concluded that Greece supports only a single (albeit highly morphologically variable) species within section caprinum. This would be a defensible position to adopt if taxonomic decisions were to be based entirely on morphological data rather than involve reciprocal illumination with genetic data.

Cytology and ploidy change

Chromosomal data for members of the genus Himantoglossum are patchy and, for some species (notably H. comperianum), appear contradictory (reviewed by Bateman et al., 2003). The best-characterised karyotypes (D’Emerico et al., 1990; D’Emerico, Bianco & Medagli, 1993; D’Emerico, Pignone & Scrugli, 2001) show H. adriaticum and arguably also H. hircinum to be typical members of the 2n = 36 clade of subtribe Orchidinae—a monophyletic group that was first recognised by Pridgeon et al. (1997) and whose integrity was most recently reinforced by the study of Tang et al. (2015).

Unfortunately, data on karyotypes and especially on ploidy levels remain weak across the Himantoglossum clade as a whole (Bateman et al., 2003). Polyploidy (both allo- and auto-) has long been known to be rife in the digitate-tubered clade of subtribe Orchidinae that includes Dactylorhiza (e.g., Hedrén, Nordström & Bateman, 2011) and Gymnadenia (e.g., Trávníček et al., 2012). Although the 2n = 36 clade that includes Himantoglossum and Steveniella presently appears less prone to ploidy change, two ploidy levels have been reported in H. hircinum via two unconnected studies, one employing chromosome counts (Bernardos & Amich, 2002) and the other employing flow cytometry (Leitch et al., 2009). Even given these very limited data, it is tempting to speculate whether polyploidy could, for example, be responsible for generating the parapatric populations of H. hircinum in Germany that reportedly flower on average three weeks later than the nominate race and were recently formally described as var. aestivalis by Kreutz & Steinfeld (2013). Population-level application of flow cytometry would be the simplest and easiest method of surveying Himantoglossum s.l. for potential ploidy-change events.

Artificial versus natural hybridisation

Artificial crosses produced by Scopece et al. (2010) between H. hircinum and members of other genera of subtribe Orchidinae showed near-complete postzygotic isolation, yielding little if any putatively fertile seed (1.7% with Serapias cordigera, none with Ophrys fuciflora, and 0.3% with the more phylogenetically distant Dactylorhiza saccifera). Artificial hybrids raised by Malmgren & Nyström (2016) between two highly divergent species within Himantoglosum s.l.—H. jankae and H. robertianum—similarly showed reduced degrees of fertility. However, it would be unwise to over-interpret this observation, as the major phenological divergence between these two species meant that it was necessary to freeze the pollinaria of H. robertianum for approximately three months until they could be defrosted and applied to the stigma of the captive H. jankae ’mother’ plant. The fact that the resulting F1 plants both grew and flowered vigorously (Fig. 22) suggests that intrinsic sterility barriers are at best weak within the group, even when the parents of the primary hybrids are sampled from within different subgenera.

However, if sterility barriers are indeed weak, remarkably few putative examples of recent natural hybridisation within the Himantoglossum clade have been reported with confidence. The best-known case—also combining two subgenera—is a few plants on the Aegean island of Lesvos that have been monitored for several years (e.g., Van Lent, 2015) and provide an example of ’wide hybridisation’ between H. comperianum and members of section Caprinum. The morphology of the plants left little doubt regarding their hybrid origin (Karatzas, 2004). A subsequent molecular investigation of the Lesvos plants, supported by landmark analysis of the labellar outline, not only confirmed their identity but also showed that either H. caprinum or its hybrid H. ×samariense was their seed parent and H. comperianum was their pollen parent (K Hürkan, A Molnár, R Bateman & G Sramkó, unpublished data). This polarity of gene transfer was also demonstrated for two similar hybrid plants found by Molnár and colleagues at Kücükcuker (Samsun) in northeast Turkey.

Phenological divergence

Subgenus Barlia has undoubtedly acquired strong phenological isolation, its flowering period rarely overlapping with those of other Himantoglossum species. Flowering of H. metlesicsianum typically extends from December to February, and that of the far more widespread H. robertianum from December (in Morocco) to April, depending on latitude and altitude.

In contrast, the remaining species of Himantoglossum flower from May to July, the season ending with high-altitude populations of H. jankae. Juxtaposed populations of H. adriaticum and H. jankae in Hungary have flowering periods that overlap but detailed studies have shown that the latter peaks a fortnight later than the former (Molnár, 2011; Biró & Bódis, 2015). Steveniella, sister-group of Himantoglossum, typically flowers in May. Together, these observations suggest that it is the robertianum group that has diverged from the remaining taxa rather than vice versa, its phenology having shifted radically during evolution to achieve a flowering period that is evidently much earlier, even when latitude and altitude are taken into account.

Hybrid origin of H. ×samariense

We noted previously (Sramkó et al., 2014) that our molecular data strongly indicated a relatively recent hybrid origin for H. ×samariense, reinforcing a suggestion made on morphological evidence by some previous authors (Alibertis & Alibertis, 1989; Delforge, 1999). The three plants analysed, sampled from populations in S Greece, W Crete and E Crete respectively, diverged from each other in both plastid and especially LEAFY sequences. In the case of LEAFY data, the E Crete sample of H. ×samariense clustered with plants of H. caprinum from the Crimea and S Turkey, whereas the W Crete and S Greece samples clustered with the H. jankae–calcaratum group (Fig. 20A). In the case of plastid data, all three samples of H. ×samariense are clustered with all samples of H. caprinum and some samples of H. jankae s.s. (but not calcaratum); in trees derived from this organellar genome it is the mainland Greek plant that differs slightly from the two remaining ×samariense plants (Fig. 20B).

When morphology is considered, it is the W Crete plant of ×samariense that differs from the other two, primarily in lacking labellum markings—a characteristic that this plant shared with most plants of H. caprinum, though unlike caprinum it did possess interior dashes on its sepals (Fig. 11A). The three ×samariense plants are placed comparatively close together on the multivariate ordinations (Figs. 7–10), typically occupying the space between the eastern European H. jankae and the Turkish H. caprinum, as they do in the plots of paired floral dimensions (Figs. 14–17). However, these populations of ×samariense are not reliably shown as being most similar to each other; the W Crete sample understandably tends to associate with H. caprinum in analyses that include the pigmentation characters lacked by all of these plants (Figs. 9 and 10A). The three ×samariense plants are also especially divergent in both labellum colour and sepal colour (Figs. 12 and 13).

Taken together, these observations suggest that each of the three H. ×samariense populations that we studied had a separate, comparatively recent origin through hybridisation between H. jankae s.l. and H. caprinum. Unfortunately, the similarity of their plastid genomes to at least some populations of each putative parental species means that we cannot presently distinguish the seed-parent from the pollen-parent in any of the three cases (Fig. 20B). Also, ×samariense plants are on average smaller-bodied and fewer-flowered than other members of the jankae–caprinum clade; they certainly do not show evidence of hybrid vigour.

Gene flow between H. hircinum and H. adriaticum

Within subgenus Himantoglossum, H. hircinum has the most westerly distribution, which centres on France and extends further longitudinally than it does latitudinally. Moreover, its eastern contact zone with H. adriaticum is severely constrained by the presence of the Alps (Fig. 5). However, the opportunity for gene flow between these species is increased by the presence of disjunct populations of H. hircinum reputed to occur in Sicily and the southernmost quarter of mainland Italy, where morphologically intermediate populations have been reported (J Bódis, pers. comm., 2016).

These disjunct populations attributed to H. hircinum merit more detailed study. The single population from southern Italy that was analysed by Pfeifer et al. (2009) deviated substantially in AFLP spectra from all other populations, while Bateman et al. (2013) reported equally deviant ITS ribotypes in multiple samples of H. hircinum analysed from Sicily (admittedly, a further ribotype similarly characteristic of both Sicily and southern Italy was also found in northern France and southeast England). Unfortunately, we do not yet possess morphometric data from Italy to allow comparison with those datasets obtained by us from the core distribution of H. hircinum further west.

Gene flow between H. adriaticum and H. jankae

Also in need of explanation is the fact that all H. adriaticum plants are shown as sister-group to all H. hircinum plants in the plastid tree but in the LEAFY tree they cluster with two Hungarian plants of H. jankae. Within Hungary (and elsewhere in eastern Europe), a sharp divide has been documented between H. adriaticum to the west and H. jankae to the east, the two species being separated west of Budapest by as little as 20 km (Molnár, 2011). There are two possible interpretations of these observations: either (1) H. adriaticum may currently occasionally hybridise with H. jankae along a north–south oriented hybrid zone, or (2) H. adriaticum itself may be an older, stabilised hybrid that formed between H. jankae and H. hircinum, though in this case, theory would have predicted that plastid sequences of H. adriaticum would nest within a paraphyletic H. hircinum rather than the arrangement seen in Fig. 20B where the plastids of both species are represented as mutually monophyletic.

The present morphological data reliably place H. adriaticum as close to H. hircinum but with a marginally more extreme phenotype; it shows little potential evidence of any morphological features that are likely to have been inherited from H. jankae or its relatives; the two species exhibit noteworthy similarities only in the small width-to-length ratio of their gynostemia (Fig. 14A) and in each possessing only a minority of plants with labellar legs exceeding 5 mm (Fig. 17A). Thus, recent gene flow from H. adriaticum into H. jankae appears to be the more likely hypothesis to explain the incongruence observed between the LEAFY tree and the plastid tree.

Gene flow within the H. jankae–caprinum clade

Moving eastward, it is section caprinum that presents the most serious challenge, both to circumscription of specific/infraspecific taxa (cf. Gölz & Reinhard, 1986; Delforge, 1999; Baumann & Baumann, 2005; Baumann & Lorenz, 2005a; Baumann & Lorenz, 2005b; Shifman, 2008; Ponert, 2014; Sramkó et al., 2014; Tsiftsis, 2016) and to confident recognition of hybrids. When two taxa are distinguished by only subtle phenotypic differences and/or character states that are not fixed in all individuals of each taxon, it becomes impossible to reliably identify hybrids between them using morphology. Any identification based on phenotypic characters requires the presence of a clear morphological discontinuity into which any primary hybrids will fall as a result of combining numerous characteristics of both parents (e.g., Bateman & Denholm, 1983; Bateman, Smith & Fay, 2008).

Admittedly, by definition, hybrids between species that are genuinely morphologically cryptic will be identifiable only via appropriate genetic analyses. In this context, application of a combination of morphometric landmark analysis of labellum shape and LEAFY sequencing to taxonomically controversial populations of section caprinum on the Aegean island of Lesvos strongly indicate that they reflect hybridogenic origin between members of the H. jankae and H. caprinum groups (K Hürkan et al., unpublished data)—an explanation that might also apply to morphologically similar (and equally controversial) populations located on the western side of the Aegean in the Taygetos Mountains of the Peloponnese (cf. Petrou, Petrou & Giannakoulias, 2011; Tsiftsis, 2016). Ironically, these potentially hybridogenic plants in turn provided one parent of the hybrids between subgenus Himantoglossum and subgenus Comperia found on Lesvos.

Considering first the DNA-based datasets, both the plastid and LEAFY trees suggest (albeit with limited statistical support) that the other named taxa collectively forming section caprinum all emerged from within the comparatively widespread H. jankae—a hypothesis that is neither supported nor refuted by the comparatively undiscriminating ITS tree (Fig. 20). The LEAFY tree shows both of the putative species that are endemic to Turkey and adjacent regions of Asia Minor—H. montis-tauri and H. caprinum—to be both monophyletic and derived relative to H. jankae. Neither statement applies to the plastid tree, which instead shows H. calcaratum (supposedly exclusively Balkan) as being both monophyletic and derived. And lastly, the ITS tree shows H. montis-tauri as monophyletic and derived. Thus, within the context of our datasets, the only named taxon within section caprinum that wholly lacks molecular autapomorphies relative to H. jankae s.s. is H. jankae subsp. robustissimum from Turkey, a localised taxon originally described by Kreutz (2006).

From a morphological perspective, derivation of the other section caprinum taxa from within H. jankae s.s. appears feasible; their often subtle differences are multi-dimensional and hence the populations are difficult to resolve into credibly divisible units. This inference is tentatively reinforced by the fact that H. jankae has a median overall morphology for subgenus Himantoglossum as a whole (Figs. 7–10).

Pollination mode and fruit set

Despite occasional assertions to the contrary, no credible evidence has accumulated to suggest that any member of the genus Himantoglossum produces nectar (reviewed by Teschner, 1976; Teschner, 1980; Claessens & Kleynen, 2011; Bateman et al., 2013). Although dense papillae evidently occur throughout the interiors of the spurs of both H. hircinum and H. robertianum (illustrated on pages 262 and 266 of Claessens & Kleynen, 2011)—papillae that are theoretically capable of secreting and/or resorbing nectar—in practice there is a poor positive correlation between the presence in spurs of papillae and production of nectar when they are analysed across subtribe Orchidinae (Bell et al., 2009).

We are not aware of any reports of pseudocopulatory behaviour shown by insects visiting Himantoglossum, and fruit-set figures rarely exceed the 80% threshold that constitutes the lower limit typical of autogamous orchids (e.g., Bateman, Sramkó & Rudall, 2015). In fact, fruit-set averages 34.3 ± 20.9% in H. hircinum, 28.7 ± 16.1% in the closely related H. adriaticum, and 39.6 ± 16.8% in the earlier flowering H. robertianum (figures derived by us from 21 individual studies summarised in Appendix 2 of Claessens & Kleynen, 2011). Degradation of the rostellum and/or loss of coherence of pollinia have been reported in H. robertianum (Teschner, 1976), but although these phenomena increasingly appear to be widespread among European orchids, they rarely effect self-pollination if they occur only in flowers that have already become senescent. It is therefore reasonable to assume that all members of the genus Himantoglossum are pollinated through food-deceit and are dominantly allogamous (cf. Cozzolino & Widmer, 2005; Scopece et al., 2007). The reward-less nature of the flowers is likely to encourage a significant proportion of geitonogamous pollinations, effected while increasingly frustrated visiting insects explore the reliably large, many-flowered inflorescences produced by Himantoglossum (Summerhayes, 1951; Kropf & Renner, 2008).

Pollinator attractants

Pollinator attraction by orchid flowers typically occurs first through air-disseminated scent, then through vision as the pollinator approaches its target inflorescence, and finally through tactile cues as it lands on its chosen flower.

In the case of H. hircinum, the goat-like floral scent has been shown using mass spectroscopy to be a cocktail of alkanoic acids, specifically decenoic acid plus dodecanoic (lauric) acid (Kaiser, 1993). Despite being the closest relative of H. hircinum, H. adriaticum is said to have a sweeter scent (Vöth, 1990), though it has yet to be analysed biochemically. In contrast, volatile organic compounds released by H. robertianum (occasionally said to collectively resemble the scent of hyacinths) are reputedly dominated by monoterpenes, notably pinenes and limonene (Gallego et al., 2012). As already noted, in members of subgenus Himantoglossum—at least, in those individuals that bear labellum markings—the epidermal cells that contain the anthocyanins and so delimit the markings occur within a central region of the labellar epidermis that expands outward to form a continuous mat of prominent, densely packed papillae (Fig. 22B). This papillose mat may offer footholds to visiting insects, but more significantly, it may also be an osmophore that is responsible for much of the volatile organic compounds emitted by the plant (see also Vöth, 1990).

Similarly, despite the great variation that we have documented above in floral background colours and discrete markings, exploration of the biochemistry of floral pigments of Himantoglossum s.l. has been disappointingly limited. Strack, Busch & Klein (1989) found the flowers of both H. adriaticum and H. robertianum to be dominated (at least, among those compounds that could be confidently identified) by the evolutionarily labile, cyanin-based compounds serapianin and seranin—compounds that were also shown to be dominant in some other members of the 2n = 36 clade of subtribe Orchidinae such as Anacamptis papilionacea and Serapias vomeracea laxiflora. Surprisingly, H. metlesicsianum apparently differed from H. robertianum in containing greater relative proportions of orchicyanins (an observation that may actually reflect lower absolute concentrations of serapianin and seranin).

The majority of Himantoglossum species have labella where at least the marginal zone, and in some cases all but a small central area, are dominated by a range of brownish hues that are infrequently encountered among European orchids. We suspect that the brownish colour represents the dual presence of green pigment(s) underlying purple anthocyanins, both categories of pigment varying among plants in both relative and absolute densities. If so, the difference in colour between the brownish labellar margins of H. robertianum and the pink-purple labellar margins of H. metlesicsianum (Fig. 1) would simply reflect the absence from the latter of underlying green pigments. By this logic, the uniformly olive-green labella of H. montis-tauri would mean that, unlike other members of the hircinum–caprinum clade, it does not express pink-purple pigments diffusely across the labellum, but rather confines them to discrete labellar markings (cf. Fig. 2 vs Fig. 3).

Pollinator identity

As summarised by Claessens & Kleynen (2011), most of the pollinator observations among Himantoglossum species have been obtained from the usual triumvirate of the three widespread western European species (robertianum, hircinum, adriaticum) and almost exclusively concern bees. Eight species of the solitary bee Andrena have been reported as visiting H. hircinum, together with representatives of five further bee genera plus the beetle genus Oedermera (e.g., Teschner, 1980; Vöth, 1990; Kropf & Renner, 2008). A similar but slightly narrower range of bees are known to visit H. adriaticum (Claessens & Kleynen, 2011; Biró et al., 2015), including the social honey bee Apis mellifera—a species that has also been witnessed visiting H. caprinum and H. jankae. Unsurprisingly, compared with visitors to subgenus Himantoglossum, bees observed visiting subgenus Barlia flowers are on average larger bodied; they encompass at least six species of four genera, most commonly Bombus (Teschner, 1977; Claessens & Kleynen, 2011). Only Bombus canariensis has so far been seen visiting H. metlesicsianum (Teschner, 1993), though it seems to us unlikely that this bumble bee is the sole pollinator of this orchid. The two species of subgenus Barlia have similar average frequencies of fruit set (Claessens & Kleynen, 2011).

In summary, there is little evidence in Himantoglossum s.l. of the strong pollinator specificity that is all too frequently invoked (though often with negligible evidence) for other groups of European orchids (e.g., Sun, Gross & Schiestl, 2014). Any genuine differences between these orchids in pollinator spectra are more likely to reflect the geographic distributions of the potential pollinators than the functional morphology of the orchid flowers. In contrast, Steveniella satyrioides is suspected to operate a contrasting mode of food deceit, wherein social wasps transfer pollinaria as they reputedly seek insect prey within the proportionately substantial spur (Nazarov, 1995; Claessens & Kleynen, 2011).

Conceptual prologue

Cladistic representation of relationships between a pair of species as sisters rather than ancestor and descendant is a logical necessity if a hypothesis of relationship is to be derived from a data-matrix with maximum objectivity. Unfortunately, this approach also severely handicaps any attempt to reconstruct the morphologies of the ancestors that would have occupied the internal nodes of a cladogram (e.g., Bateman, Hilton & Rudall, 2006). Early attempts to address this problem often relied on ‘Brownian motion’ models. These models assumed that the (hypothetical) common ancestor of species A and species B possessed a phenotype that was in all parameters the precise average of the two descendant sister species. Thus, if species A possesses six white petals and species B possesses four red petals, their common ancestor is assumed to have possessed five pink petals! Unfortunately, logic tells us that their common ancestor almost certainly possessed either four red petals or six white petals, but we cannot be certain which of these two conditions actually pertained; the one fact of which we can be confident is that the ancestral species did not possess five pink petals!

Hence, more complex models of nodal reconstruction have since become predominant, based on various kinds of probability estimate. These models all depend to varying degrees on weight of numbers; if several successive nodes on the cladogram subtend species possessing six white petals, there is a high statistical probability that those intervening ancestors also possessed six white petals. It is important to realise that this logic does not necessarily mirror evolutionary reality; consider, for example, a single long-lived, widespread, evolutionarily conservative species that repeatedly gives rise to more localised species (e.g., Fig. 2 of Bateman, 2011), each of which possesses the alternative character state to that shown by the long-lived ancestor. This is no mere theoretical consideration; one incontrovertible example is provided by the numerous local autogamous species of the Eurasian orchid Epipactis that evolved independently from within a single geographically widespread allogam, E. helleborine (Squirrell et al., 2002). At best, reconstruction of internal nodes on a cladogram remains a case of building somewhat subjective scenarios, albeit within a framework of (hopefully) less subjective data.

Seeking the most credible outgroup for the Himantoglossum clade

In addition, the chosen outgroup(s) play a particularly important role in influencing the perceived set of character states hypothesised to have been possessed by the common ancestor of the entire ingroup, which by definition is located at the so-called root node. This issue is problematic in the case of the Himantoglossum clade (reviewed by Bateman, 2012a). Topological congruence at the genus level among well-sampled molecular phylogenetic studies of subtribe Orchidinae is limited to two relevant nodes, specifically agreement that: (1) Neotinea s.l. (2n = 42) is sister to a 2n = 36 clade consisting of the genera Himantoglossum s.l., Ophrys, Serapias and Anacamptis s.l., and (2) Serapias and Anacamptis s.l. are sisters. Irrespective of whether they constructed trees using maximum parsimony, likelihood or Bayesian algorithms, all three published ITS-only studies yielded uncertain relationships among the five well-supported groups in the 2n = 36 clade: Serapias plus Anacamptis s.l., Ophrys, Himantoglossum s.l., and Steveniella (Bateman et al., 2003; Sramkó et al., 2014; Tang et al., 2015). By adding to a subset of Bateman et al.’s ITS data a modest number of base-pairs derived from the plastid rpl16 intron, Inda, Pimentel & Chase (2012) strengthened statistical support for a topology in which Himantoglossum s.l. was sister to Serapias plus Anacamptis s.l. and Ophrys was sister to all three genera. Unfortunately, this topology was obtained in the absence of Steveniella—a genus whose presence in a phylogenetic analysis of Orchidinae is crucial, as it was tentatively placed as sister to Himantoglossum s.l. in the molecular studies of Bateman et al. (2003) and Sramkó et al. (2014).

When seeking close relatives of Himantoglossum s.l. that might usefully inform speculation regarding the initial phenotype of this lineage, clearly both Ophrys and Serapias are too divergent (molecularly and morphologically) and too reproductively specialised to provide useful guidance. Anacamptis s.l. also appears unsuitable, given its reliably derived phylogenetic position as sister to Serapias. This understanding leaves Steveniella as the only credible extant candidate for the role of archetype of the Himantoglossum clade.

Eventual acquisition of next-generation sequencing data (e.g., Olson, Hughes & Cotton, 2016) across the 2n = 36 clade of subtribe Orchidinae clade should, in theory at least, help to resolve several outstanding issues in phylogeny reconstruction and taxon circumscription. They would provide a further test regarding whether the improbably early divergence of H. formosum suggested by our LEAFY tree (Fig. 20A) is, as we suspect, a misleading aberration (Sramkó et al., 2014), and could also provide a much-needed additional test of whether Steveniella is viewed correctly by us as sister genus to Himantoglossum s.l.

Having emphasised the assumptions that we felt obliged to make in pursuing this line of thought, we can now proceed to consider character evolution in Himantoglossum s.l. As described by Bateman et al. (2003, p. 16), “the smaller, 1–2-leaved, small-flowered Steveniella superficially resembles members of the Neotinea s.l. and the [dominantly East Asian: Tang et al., 2015] Neottianthe∼Hemipilia clade, though its gynostemium structure, purplish-brown galea, strongly three-lobed labellum and short, robust spur do—as Sprengel (1826) perceived—suggest similarities to the more derived himantoglossids”. Here, we further extend that earlier morphological comparison, placing Steveniella at the root of the evolutionary scenario that we have built upon our present morphometric matrix (Figs. 18 and 23).

Labellum shape

We will begin this part of the discussion by considering the ‘floral skeletons’ that we abstracted from our raw data in order to facilitate systematic comparison of labellum size and shape (Fig. 18).

Steveniella is, in shape and especially size, more typical of earlier-divergent groups of subtribe Orchidinae than of any member of Himantoglossum s.l. Admittedly, the labellum of H. formosum could in theory be generated simply by considerably expanding that of Steveniella and incising a small distal notch into its central lobe. This is an especially interesting observation in the light of the fact that these two species have the eastern-most distributions of any of the taxa under scrutiny (they actually co-exist in the Caucasus), and might therefore be predicted to be closely related. However, both ITS and plastid topologies refute any such closeness of relationship (Sramkó et al., 2014) (Fig. 20).

Labella of subgenus Comperia and subgenus Barlia were most likely generated from a Steveniella-like ancestor by lengthening all structures except the abdomen, which consequently became proportionately shorter, and substantially broadening the labellum (most notably in H. metlesicsianum). It is equally parsimonious for the comparatively short abdomen to have arisen in parallel in both lineages, or to have been acquired only once, before becoming greatly elongated during the origin of subgenus Himantoglossum. In subgenus Barlia the limbs and spur became short and robust, whereas in subgenus Comperia they became slender but greatly elongated (Fig. 18). This developmental trend for extreme elongation even affected the lateral teeth on the lateral petals, causing the teeth to become filiform. Following divergence of subgenus Barlia, the arms became very short and the abdomen became considerably narrower in H. formosum, thereby permitting development of the sinistral spiralling that characterises the remainder of subgenus Himantoglossum. The abdomen then became greatly elongated (and often more tightly spiralled) to generate the distinctive labellar bauplan of the hircinum–caprinum clade. Separation between section caprinum and section hircinum involved substantially increasing the length of the legs and slightly broadening the shoulders of the former (Fig. 18). In section hircinum, spur size was reduced back to that characteristic of Steveniella and legs became even narrower.

Although it is not clear from the molecular phylogenies which member of section caprinum originated first, H. jankae is most geographically widespread and most similar in labellar ratios to H. hircinum and H. adriaticum. Taking into account all the available evidence, it appears most likely that H. jankae gave rise to H. calcaratum, H. montis-tauri and H. caprinum comparatively recently; each most likely arose independently as a parapatric variant. The concomitant changes in flower dimensions were subtle; simple enlargement of the labellum in the case of H. calcaratum (abdomen and spur became especially large) but slight shortening of the labellum of H. caprinum and H. montis-tauri plus, in the case of the latter, widening of the labellum (Fig. 18).

Shapes of other flower parts

It is tempting to view changes in the length and/or width of the two lateral petals and especially of the three sepals as direct functional consequences of the changes documented in the shape and/or size of the third, median petal—the labellum—which must be wholly enclosed throughout the development of the bud (Fig. 22A). These perianth segments became elongated in H. comperianum to accommodate the filiform limbs of its distinctive labellum and broadened in the remaining species to encompass their more robust labella. The entire flower became more compact during the origin of section hircinum, the column also being shortened. Both the column and labellum narrowed further in H. adriaticum relative to H. hircinum.

The sepals of Steveniella and H. comperianum are laterally fused to about one-third of their length from their base (a feature also characteristic of early-divergent members of the related orchid genus Anacamptis s.l.: Bateman & Hollingsworth, 2004), but this feature was then lost prior to the origins of the remaining taxa of Himantoglossum. The lateral sepals were thus free to spread partially outward in subgenus Barlia and even more strongly outward in H. formosum (Fig. 11B), leaving only the median sepal and lateral petals to form the galea overarching the gynostemium. It is possible that spreading sepals had a single origin before the divergence of subgenus Barlia, followed by re-integration of the lateral sepals into the galea in the hircinum–caprinum clade. However, it seems more likely that spreading lateral sepals arose independently in the two lineages. Surprisingly, lateral fusion of the sepals was lost at the same point in the evolutionary history of the group as the viscidia became laterally fused (Fig. 23). Consequently, in the case of subgenera Barlia and Himantoglossum, both pollinaria must be removed by pollinators as a single physical unit.

Flower markings and colour

The Himantoglossum clade possesses a remarkable range of floral markings, but it is even more remarkable that each different category of markings appears to be inherited independently of the others. Labellar markings certainly originated at the evolutionary origin of Himantoglossum (and may have been inherited from its direct ancestor—although labellum markings are absent from Steveniella, species of the more closely related of the genera in the 2n = 42 clade, Neotinea s.l. and Orchis s.s., all routinely possess them). With regard to markings on the sepals of Himantoglossum, interior spots arose only after the divergence of H. comperianum, the marginal stripe only after the divergence of the robertianum group, and the interior dashes only after the separation of H. formosum, when the labellum markings also became raised on a central papillate region (Figs. 22C and 23) (Bateman et al., 2013). All markings other than sepal marginal stripes were subsequently lost independently from H. formosum and from H. caprinum (which also lost the papillae that are reliably correlated with labellum markings in the hircinum and jankae groups: Sramkó et al., 2012).

The base colour of the labellum appears to have become on average paler independently in H. comperianum, H. metlesicsianum and H. montis-tauri, whereas the reverse (abaxial) surface of the sepals darkened independently in H. comperianum and H. formosum. Himantoglossum offers particularly graphic examples of how combinations of contrasting pigments can generate unusual flower colours; the way in which these plants achieve such results merits more detailed investigation of both the nature of the pigments involved and the precise location within the anatomy of the flower where those pigments are concentrated.

Vegetative characters

It is striking how small a role vegetative features play in the evolutionary scenario summarised in Fig. 23. By far their most critical involvement was the remarkable increase in overall body size (and associated increase in leaf number) that must have occurred during the origin of the genus Himantoglossum, as no close relative of the Himantoglossum clade contains species capable of showing an equivalent degree of vegetive vigour (among subtribe Orchidinae, only the Anacamptis palustris group can rival Himantoglossum for average body size). Otherwise, we detected only a further, and more modest, increase in body size in H. calcaratum, and a further, and equally modest, increase in leaf number at the origin of subgenus Himantoglossum. This increase in average leaf number subsequently reversed during the origin of H. montis-tauri (and that of H. ×samariense), where it was apparently compensated for photosynthetically by increased size of the already foliose bracts. Otherwise, mention need be made only of the somewhat fleshier stems of subgenus Barlia, and of the comparatively condensed inflorescence of H. hircinum (Fig. 23).