Introduction

Over the past ∼ 150 years, our planet’s biodiversity has declined at such a staggering rate that some are talking of sixth mass extinction (Wake and Vredenburg 2008; McCallum 2015). The ultimate goal of nature conservation is to preserve the natural biodiversity on Earth. There are many approaches employed to reach this goal, from protecting ecosystems and particular habitats to focusing on endangered taxa. Identifying operational conservation units is the key to effective conservation management in taxon-based conservation. The basic concept is to identify evolutionary significant units (ESU; populations representing genetically unique entities, Ryder 1986; Moritz 1994). While there were some controversies in the past on defining ESUs, the concept’s fundamental is clear (Fraser and Bernatchez 2001). In practice, the idea is to find populations that should be prioritized based on their unique heritable characteristics (e.g. Zink 2004).

The field is heavily DNA-driven, but purely molecular approaches can be seen as “reductionistically mistaken” (Casacci et al. 2014). Further, even though studying ‘historical’ or ‘ancient’ DNA from old specimen from various collections is more and more promising recently, such approaches are not always straightforward and easy to adopt for every potential study object (Anderung et al. 2008; Ellis 2008; Lis et al. 2011; Call et al. 2021; Raxworthy and Smith 2021). Morphology represents the primary data we use to infer species boundaries and to describe biodiversity (Schlick-Steiner et al. 2007; Steiner et al. 2009; Yazdi et al. 2012), and morphological trait variation shows the highest heritability among various trait types (e.g. Mousseau and Roff 1987). Hence, methodology of morphometric taxonomy can be used as a surrogate for genetic analyses, whenever the latter is not available or attainable fast. From the conservation aspect, the distinction and management of many subspecies with a poor taxonomic basis can be detrimental to the efficient allocation of resources (Haig et al. 2006). Therefore, a clear taxonomic overview of a species not only serves the purpose of classifying individuals, but also benefits as a guide for conservation biology, suggesting intraspecific ESUs requiring special treatment.

The general biodiversity decline can be detected in butterflies (Papilionoidea). The Apollo butterfly, Parnassius apollo, is a strongly declining Palearctic butterfly species (van Swaay and Warren 1999; Nakonieczny et al. 2007). It is classified as Least Concern in the IUCN Red List of Threatened Species considering its wide distribution in Europe and Asia with an estimated decline no more than 15% globally, but with a 30% regional decline in Europe (Nadler et al. 2021). However, it was assessed regionally as Vulnerable in the first Red Data Book of European butterflies (van Swaay and Warren 1999), and as Near Threatened in the European red list of butterflies (van Swaay et al. 2010). Maes et al. (2019) ranked it as 36th out of 496 European butterfly species according to a weighted Red List value calculated for each species and classified as regionally. It is a large, single-generation species with a wide, but disperse area, occurring in Europe mainly in the 500–2500 m above sea level mountain zone, inhabiting lowland areas only in the northern part of its range (Scandinavia, East European Plain and Siberia) (Bryk 1935; Glassl 2005; Möhn 2005). Its decline also draws attention to the threat to other butterfly species in open mountain habitats, and P. apollo is considered a flagship species (Nakonieczny et al. 2007). Until the middle of the 20th century, there was a strong trend among lepidopterists to introduce new subspecies names for numerous populations inhabiting spatially more or less separate parts of the species’ range (Braby et al. 2012; Domagała and Lis 2022). In the case of P. apollo, 290 subspecies have been described altogether (Möhn 2005), including 26 subspecies rank taxa from the Carpatho-Pannonian region (for a map of the region, see Fig. 1; for detailed compilation of traits used in original descriptions and subsequent taxonomical works, see Electronic Supplement Material [hereafter ESM] Table 1). Interestingly, several recent works recognize (almost) all of the described subspecies in the Carpatho-Pannonian region (Table 1), defining the subspecific areas in a good agreement (Möhn 2005; Glassl 2005; Kříž 2011). However, a common analyses based on modern methodology of the dozens of subspecies is lacking, even though it would be crucial not only for scientific interest, but also for effective distribution of conservation management efforts.

Fig. 1
figure 1

Map depicting the former (early 20th century) distribution of the putative Parnassus apollo subspecies (area boundaries drawn by brown lines, brown dashed lines indicate areas or parts of areas that were not included in our analysis) and the geo-regions (black dashed lines) that were inhabited by the species in the Carpatho-Pannonian region. NCARP: Northern Carpathians including 15 examined subspecies, ECARP: Eastern Carpathians including two subspecies, WCARP: Romanian Western Carpathians including one subspecies, BURGL: Burgenland (foothills of Eastern Alps) including one subspecies, SERB: Serbian Carpathians including one subspecies, ALPS: Swiss Alps (our outgroup) including one subspecies. Different shapes (circles or triangles, no specific meaning) with darker colours indicate localities (10 × 10 km UTM quadrats) for which photographs of specimens have been collected, lighter colours indicate additional literature data of occurrence (Jakšić 1988; Ruşti and Dragomirescu 1991; Konvička and Fric 2002; Höttinger 2003; Popov and Plushtch 2004; Dabrowski 2008; Kalivoda 2008; Kříž 2011). Dark red lines mark the boundaries of Pannonian Basin and the Carpathians (including its foothills areas) according to Kocsis et al. (2018) and the thick blue lines indicate the country borders (AUT – Austria, CZE – Czech Republic, HRV – Croatia, HUN – Hungary, POL – Poland, ROM – Romania, SRB – Serbia, UKR – Ukraine). Some 100 × 100 km UTM quadrats are marked with two-letter abbreviations (capital black letters), and we provide the scheme for 10 × 10 km UTM resolution in the WL quadrat for reference (the colour version is available online as open access)

Full size image
Table 1 The studied subspecies (Name), their abbreviations (Abbr.), the relevant geo-region (GEO), sample sizes (male: N m; female: N f; Total: N t) and their recent status (RS; extant: +; extinct: -; doubtful: ?). The following geo-regions are included within the Carpatho-Pannonian region: BURGL: Burgenland (foothills of Eastern Alps) including one subspecies, ECARP: Eastern Carpathians including two subspecies, NCARP: Northern Carpathians including 15 subspecies, SERB: Serbian Carpathians including one subspecies, WCARP: Romanian Western Carpathians including one subspecies. ALPS: Swiss Alps including one subspecies was added as our outgroup
Full size table

In the present study, our goal is to provide an extensive wing-morphology-based analysis (the original descriptions are typically based on wing characteristics; ESM Table 1) of the described subspecies occurring in the Carpatho-Pannonian region based on available specimens from museum and university collections. We aim (i) to provide taxonomically relevant information and (ii) to equip conservation management with potential ESUs in the region instead of having only the old, one-by-one subspecies descriptions available.

Methods

Sampling

Sampling from natural populations was not realistic due to the vulnerability of the species, especially with the sample sizes that are relevant for us (N > 10 for each sex from every subspecies), therefore we relied on collection material. About 2/3 of the Carpatho-Pannonian specimens used for our analyses can be found in the collection of the Hungarian Natural History Museum [= HNHM] (Bálint et al. 2016). This sample was supplemented with specimens from other museums and university collections (Table 1; for individual-level details, see ESM Table 2). Furthermore, we selected a Swiss outgroup (P. a. rhaeticus) with 40 males and 27 females, which, based on mtCOI sequences, belongs to a separate lineage (Todisco et al. 2010). The starting material consisted of 995 males and 487 females. As the differential characters in the original descriptions of subspecies (ESM Table 1) were in many cases not present on the specimens examined, the identification of specimens on subspecific level was based on geographic location. Individuals from the Northern Carpathians (Slovakia) were identified by overlaying Kříž’s (2011, see p. 77.) map illustrating subspecific ranges on the surface of Google Earth. In the other cases, we used the available literature (Issekutz 1952; Zečević and Radovanović 1974; Jakšić 1988; Ruşti and Dragomirescu 1991; Glassl 2005; Möhn 2005).

Table 2 Geo-region level results of the Cross-Validated Linear Discriminant Analyses. For the geo-regions’ abbreviations, see Table 1. N: sample size; RC: Ratio of correctly discriminated individuals. The lower limit of acceptable RC is 0.95. Successful discrimination is indicated by bold font
Full size table

We could not include subspecies represented by a low number of specimens to our analyses (P. a. ruthenicus and P. a. vistulicus) or specimens of uncertain provenance (P. a. cominius and P. a. artemidor, and doubtful specimens from the Western Romanian Carpathians and South Carpathians (Romania)). We were also unable to deal with specimens with too vague locality designations, if the subspecific assignment was impossible. Note that the majority of the distribution area of P. a. albus is outside of the Carpatho-Pannonian region, and thus it was not considered. Regarding the subspecific affiliation of the 170 Carpatho-Pannonian localities, only four modifications have been applied compared to the previous studies (ESM Table 2). Altogether, we analysed 20 Carpatho-Pannonian subspecies and one outgroup from the Swiss Alps (Table 1).

The subspecies included in our study belongs to six geo-regions (following Móczár 1967), each of which is well delimited in space and corresponds well to certain previously determined zoogeographical units within the Carpatho-Pannonian area (Fig. 1; ESM Table 2). The geo-regions are: (i) Northern Carpathians including 15 subspecies, (ii) Eastern Carpathians including two subspecies, (iii) Romanian Western Carpathians including one subspecies, (iv) Burgenland (foothills of Eastern Alps) including one subspecies, (v) Serbian Carpathians including one subspecies and (vi) Swiss Alps as our outgroup, with the P. a. rhaeticus subspecies.

Photographing

Due to the high museum value of the specimens, removal of the wings from the body was out of question and thus we used whole specimen digital photographs. In the case of specimens from HNHM, the images were taken using an OLYMPUS Camedia C 7070 camera attached to a stand. The butterflies were pinned to a plastic foam sheet, with a ruler placed into the plane of the wings as reference. The butterflies are usually mounted on a spreading board with wings positioned at a 5° inward slope to compensate for the wing drop over time. In the majority of the specimens, the wings were indeed in horizontal plane. In the non-horizontal cases, we positioned the individuals so that one wing was horizontal. Since butterfly forewings are more important for flight than hindwings (Jantzen and Eisner 2008), forewing shape variation is more likely to describe adaptive variation. Therefore, we focussed on the forewing shape. We measured the right forewing in most cases and used the left only if the right was damaged or had some anomalies. Left wing digital photographs were flipped horizontally for the shape analyses (see below). Specimens with both forewings damaged were discarded. Individuals from other than HNHM collections were photographed by collaborators using similar setups, but with different cameras. Considering the size of P. apollo and the photographing setups, camera-dependent distortion issues are negligible. Finally, 949 males and 477 females were included in our investigations. We note that even though forewings might have higher relevance in local adaptations than hindwings, we can not exclude the possible taxonomic value of the latter. However, we also note that reliable measurement of hindwing shape would have been hampered by the overlap between fore- and hindwings in the museum specimen (see Fig. 2).

Fig. 2
figure 2

Landmark and semi-landmark positions on the right forewing of Parnassus apollo. The abbreviations for venation follows Bryk (1935): ax1 and ax2: axillaris 1 and 2 veins, cu1 and cu2: cubitus 1 and 2 veins, m1, m2 and m3: the three branches of media veins, r1, r2, r3, r4 and r5: the radius (radial) veins, of which 2 and 3 are fused in the case of P. apollo, this is marked by “r2 + 3”, sc: subcosta. For the explanation of semi-landmark positioning, see ‘Landmarks’ in Material and methods (the colour version is available online as open access)

Full size image

Landmarks

Landmark-based geometric morphometrics (Bookstein 1991; Rohlf and Marcus 1993; Zelditch 2012) were used to explore the differences in wing shape and wing size between subspecies and geo-regions. The method turned out to be helpful in butterflies also in an intraspecific context (Jones et al. 2013; Cespedes et al. 2015; Paučulová et al. 2018). The landmarks (N = 24) were placed in tpsDig v.2.31 software (Rohlf 2017) by the first author throughout the whole workflow. 19 standard landmarks were defined mostly by starting and termination points of veins (Fig. 2). Additional semilandmarks were placed to characterize the curvature of the upper (3 semilandmarks) and inner (2 semilandmarks) margins (Fig. 2), via the ‘Append tps curve to landmarks’ function of tpsUtil v.1.81 (Rohlf 2021). To generate the sliders file, which gives the sequence of the semilandmarks, identifying the adjacent points between which each semilandmark slides, we used the ‘Make sliders file’ function in tpsUtil. We used the tpsRelw v.1.53 software (Rohlf 2013) for the Procrustes superposition (where semilandmarks were slid to minimize the Procrustes distance) and gathering the scores from partial warps and uniform components, which we used as the raw shape data for the subsequent analyses (see below). We also estimated centroid size (the square root of the sum of the squared distances between each landmark and the centroid) as a wing size proxy independent of shape (Bookstein 1991), using landmarks from the wing outline only.

Statistical analyses

We compared wing size (centroid size) between subspecies and geo-regions by running Linear Models (LM) separately for the sexes using “lme4” and “lmerTest” packages (Bates et al. 2015; Kuznetsova et al. 2017). Pairwise differences were evaluated by the presence/absence of overlaps between 95% Confidence Interval ranges.

All shape analyses below were run separately for males and females. As a first step, we disclosed patterns in morphometric data without a priori groupings via two Exploratory Data Analysis (EDA) algorithms. First, we ran the ordinating Principal Component Analysis (PCA) by tpsRelw v.1.53 (under the name of ‘relative warp analysis’) on the raw data. The PCA searches for discontinuities in continuous morphometric data and displays patterns graphically, aiming to cover the maximum variation in the data, but has no estimation of the ideal number of clusters, and the “classification” of objects has to be subjectively interpreted. Second, the ideal number of clusters was determined via a Gap statistics partitioning algorithm (Tibshirani et al. 2001) ran on the PC scores, using the package “clusterGenomics” (Nilsen et al. 2013) with the function “gap”. Our setup was: Kmax = 5, (maximum number of clusters) B = 1000 (bootstrap iterations), ref.gen = “PC” (reference data are generated uniformly over a box aligned with the principal components of the data). This method estimates the optimal number of clusters based on statistic thresholds and automatically assigns objects into partitions.

Since the above EDA analyses based on covering maximum variation in the data revealed no pattern (see Results), we switched to an approach that aims for maximum group separation. Here, we employed biologically relevant grouping hypotheses based on the original subspecific classifiers. We first ran a supervised cumulative Linear Discriminant Analysis (LDA) complemented with a cross-validated LDA (CV-LDA), using package “MASS” (Venables and Ripley 2002) on the raw data. Note that sample sizes between subspecies differed markedly, which can seriously hamper the estimates. To overcome this problem, we ran our LDA and CV-LDA with contrast parameter (prior) sets giving equal weight to the subspecies, so they contribute to the model equally. Again, this approach did not support any of the subspecies. Therefore, we first ran CV-LDA on geo-region level, and second, we employed agglomerative hierarchical clustering with multiscale bootstrap resampling (N bootstrap = 10,000; distance: Euclidean, method: average) via “pvclust” (Suzuki and Shimodaira 2006) on the LDA scores, keeping the subspecies categories. The latter analysis results in a dendrogram providing approximately unbiased P value for the clusters. Significant clusters were defined by approximately unbiased P value > 95. Validity of the clusters were checked by CV-LDA. Statistical analyses have been done in R (R Core Team 2020), version 4.0.2. Finally, we generated wireframe graphs (Siegel and Benson 1982; Klingenberg 2013) to compare the shape for the different groups supported by our analyses.

Results

Our LMs on centroid size revealed significant differentiation between subspecies (males: F20,928 = 18.583; P < 0.001; females: F20,456 = 11.92; P < 0.001). We found that apart from a few pairwise differences, the Swiss population was the only one systematically differing (being smaller) from the others in both sexes: in females it was significantly smaller than the 20 Carpatho-Pannonian populations, while in males it was significantly smaller than 19 and only tended to be smaller than ssp. antiquus (Fig. 3). This finding was strengthened by geo-region level comparisons (males: F5,943 = 42.035; P < 0.001; females: F5,471 = 24.144; P < 0.001), where only Swiss Alps differed (being smaller) from the rest (Fig. 3).

Fig. 3
figure 3

Centroid size variation among the studied Parnassus apollo subspecies. Means + 95% Confidence Intervals are shown. Males: A, B; Females: C, D. For the abbreviations of subspecies and geo-regions, see Table 1

Full size image

Our PCA analyses on wing shape revealed three PCs for males and three PCs for females with proportion of variation explained over 10% (proportion of variation explained, males / females: PC1 = 21.17% / 21.91%; PC2 = 19.78% / 17.80%; PC3 = 12.23% / 11.79%). The patterns revealed no recognizably separated entities in the extensive set of males and females (Fig. 4a, c). Gap statistics estimated a single cluster in both sexes (Fig. 4b, d) based on the PC scores, supporting the (lack of) pattern returned by the PCA. Supervised cumulative LDA suggested that the cluster of the two Eastern Carpathian subspecies is different from the rest in both females and males (Fig. 5a, c), but the CV-LDA rejected the separation of the subspecies in both sex from the remaining subspecies or from each other (ESM Table 3). However, we note that classification success approached the minimum acceptable value of 95% for the two Eastern Carpathian subspecies in females.

Fig. 4
figure 4

Graphical display of our Principal Component Analyses (A, C) and Gap statistics (B, D). Males: A, B; Females: C, D. For the abbreviations of subspecies and geo-regions, see Table 1. At Gap statistics, Number of clusters (k) (X axis), the total within-cluster dispersion for each evaluated partition (Wk) and the number of clusters returned by the Gap statistic (Gap [k]) is illustrated. The outgroup is marked with an asterisk (the colour version is available online as open access)

Full size image
Fig. 5
figure 5

Graphical display of our Linear Discriminant Analyses (A, C) and hierarchical clustering (B, D). Males: A, B; Females: C, D. For the abbreviations of subspecies and geo-regions, see Table 1. On B and D the approximately unbiased P values (au) are shown. The outgroup is marked with an asterisk (the colour version is available online as open access)

Full size image

CV-LDA on geo-region level separated Eastern Carpathians from the rest with a 100.00% classification success in females and 95.24% in males, other geo-regions being indistinguishable (Table 2). The agglomerative hierarchical clustering on subspecies level supported two clusters in both sexes: an Eastern Carpathian cluster and another cluster including the rest of Carpatho-Pannonian populations (Fig. 5b, d). The two clusters were fully supported by CV-LDA in both sexes (classification success ranging from 96.4 to 100%). We note that the position of the subspecies from the Serbian Carpathians was somewhat questionable in the dendrograms (Fig. 5b, d), and thus we reran the CV-LDAs with adding this geo-region (containing one subspecies) as a third group. However, Serbian Carpathians as a separate group was not supported (classification success: 90.0% for males, 72.7% for females), while the other two groups (Eastern Carpathians vs. the rest) remained supported in both sexes (classification success > 95%). The visualization of shape differences between the supported clusters suggest that Eastern Carpathian populations have slightly narrower forewings than the other studied populations in both sexes (ESM Fig. 1).

Discussion

The original descriptions of the studied 20 Carpatho-Pannonian P. apollo subspecies are based on the ‘triad’ of size (characterized in 85% of the original descriptions and mentioned as a separating trait in 40% of them), wing shape (characterized in 65% of the original descriptions and mentioned as a separating trait in 40% of them) and wing pattern (characterized in 100% of the original descriptions and at least one of its elements mentioned as a separating trait in 60% of them) (ESM Table 1). Here, we focussed on size and wing shape variation, and we found only two groups: (i) the cluster of two subspecies from the Eastern Carpathians and (ii) the rest, including Northern Carpathians, Romanian Western Carpathians, Serbian Carpathians and Burgenland (and the Swiss outgroup). We discuss our results from the taxonomist’s and the conservation biologist’s point of view below.

In descriptions of P. apollo subspecies (as in other butterflies), “body size” or “size” is actually estimated through wing size, typically via forewing length. This approach has a drawback: a single linear measure inherently mixes size and shape. This problem was covered in our study by using centroid size. In our forewing size analyses, we found that our Swiss outgroup, P. a. rhaeticus, belonging to a genetically distinct lineage (Todisco et al. 2010), was clearly smaller than the Carpatho-Pannonian subspecies, with only male P. a. antiquus showing an overlap with it. The forewing size variation within the Carpatho-Pannonian subspecies held almost negligible information. In males, P. a. strambergensis and P. a. antiquus had smaller forewings than the other Carpatho-Pannonian subspecies, while they did not differ from each other. In the case of females, even these two subspecies did not differ clearly from the others. Considering the high end of the size range, none of the samples could be separated. Our results suggest that size relationships between the subspecies were oversimplified and incorrectly used in a discriminatory context in some of the original descriptions. Hence, in the light of our results, size labels as “medium” (e.g. P. a. oravensis Eisner and Zelný 1969 and P. a. antiquus Eisner and Zelný 1974) or “large” (e.g. P. a. nitriensis Issekutz 1952 and P. a rosnaviensis Issekutz 1952) are unlikely to be trustworthy. When looking at the pattern on the biologically relevant geo-region level, the previous between-group variation disappeared, only the Swiss outgroup remaining significantly smaller than the rest. However, we have general doubts about the taxonomic relevance of body size on the intraspecific level, especially when the studied populations are geographically close to each other and the size divergence is not considerable. This is because without targeted experiments, the contributions of local adaptation vs. phenotypic plasticity (see. e.g. West-Eberhard 2003) in this highly plastic life history trait cannot be separated.

Our forewing shape analyses were based on landmark-based geometric morphometrics instead of analysing an a priori set of linear measurements. This approach is superior to the latter, because any set of a priori selected variables is (i) biased, potentially leaving out relevant information, (ii) comprises a set of nonindependent, often directly confounding variables and (iii) mixing information of size and shape. Further, a set of linear measures contains no information about the spatial relationships of the endpoints used for the measurements, making the visualisation of the total shape changes difficult. Landmark-based geometric morphometry using Procrustes superposition, on the other hand, covers all shape information independently from size and any a priori expectations (Webster and Sheets 2010; Owens et al. 2020; Viacava et al. 2023). Our PCA (+ GAP statistics) found no divergence in forewing shape, providing only a single cluster in both sex, which also included the Swiss outgroup. Therefore, the taxonomy of 20 (+ 1) subspecies received zero support with this approach. Our LDA – CV-LDA detected a suggestive trend of the two Eastern Carpathian subspecies (P. a. transsylvanicus; P. a. rosenius) being somewhat different from the rest, but not from each other. Again, the taxonomy of 20 (+ 1) subspecies received zero support and thus they may not meet subspecies criteria (Braby et al. 2012; Seifert 2020). These results have two alternative explanations. First, one might consider that forewing shape is not a useful character in P. apollo taxonomy. However, it is highly unlikely, considering the taxonomic relevance of wing shape in other insect taxa (e.g. Reis et al. 2021; Viertler et al. 2022), including lepidopterans (e.g. Zhong et al. 2016; Wang and Li 2020) and P. apollo (ESM Table 1). Second, it is possible that the current taxonomy of Carpatho-Pannonian P. apollo, based on the often old, one-by-one verbal descriptions of the different populations as subspecies (Glassl 2005; Möhn 2005; Kříž 2011) needs reconsideration.

The same LDA – CV-LDA analyses conducted among the biologically relevant geo-regions revealed a clear divergence between the Eastern Carpathian populations (ECARP) and the rest of the Carpatho-Pannonian populations (ROCP), which groups could theoretically have valid taxonomic value. Agglomerative hierarchical clustering also supported the same two clusters. The distinction of ECARP is strengthened by the fact that our Swiss outgroup from a separate genetic lineage (Todisco et al. 2010) is clustering with ROCP, while being divergent from ECARP. Recently, the high heritability of the aspect ratio of the forewing (the ratio of wing length to wing width) in Heliconius species has been revealed (Montejo-Kovachevich et al. 2021). As the main difference in wing shape between ECARP and ROCP is revolving around the aspect ratio of the forewing, it is reasonable to assume that this change in shape is also heritable. Thus, ECARP appears to be congruent with the requirements of Braby et al. (2012) that would justify its treatment as a distinct subspecies. We note that the COI sequence (NCBI GenBank accession number: MW501566.1) from a Northern Carpathian population (Štramberk, Czech Republic, introduced population from Strážovské vrchy, Slovakia: Konvička and Fric 2002) and the sequence from an Eastern Carpathian population (Harghita, Romania; accession number: HQ004900.1) are identical. However, there are several cases where complete or near-complete COI barcode sharing between even different species have been described (Hausmann et al. 2011; Takáts and Mølgaard 2016; Dincă et al. 2021). From a functional aspect, we can say that ECARP individuals had relatively narrower forewings than their ROCP conspecifics in both sexes, but the adaptive significance of this shape shift is yet unknown in our case. We also note that male (but not female) individuals from the Serbian Carpathians (putative P. a. timacus) showed a trend of separation from ROCP and obviously, from ECARP, in the LDA and the agglomerative clustering. However, CV-LDA did not support this pattern. It might be a statistical power problem stemming from the relatively low sample size.

From the conservation aspect, our results suggest the presence of two ESUs, instead of the 20 + potential ESUs one could consider based on the recent taxonomic treatments of Carpatho-Pannonian P. apollo (Glassl 2005; Möhn 2005): (i) ECARP (two subspecies) and (ii) ROCP (18 subspecies). The questionable validity of the subspecies / ESUs has been already foreseen in the national conservation strategies of each country, for instance in Slovakia, emphasising that further studies are needed to clarify the question (Žlkovanová and Havranová 2017). It is worrying that there have been no published observations of P. apollo in the last two decades from the Eastern Carpathian populations (Vizauer 2010; Moldovan 2016), apart from a single observation by Lajos Németh from the valley of Bistrița Aurie in 2003 (Lajos Németh, pers. comm.). Eastern Carpathian populations warrant intensive monitoring to see whether the morphologically unique P. apollo populations have indeed gone extinct.

Since the presence of existing ECARP populations is unfortunately highly questionable, we can only suggest one ESU being unequivocally present currently in the Carpatho-Pannonian region. Given that proven (in the sense that individuals can be found in known locations) populations are currently restricted to the Northern Carpathians, the actual conservational consequences of our results might be the most prominent in this geo-region. The 15 putative subspecies (not counting P. a. cominius with uncertain origin, and P. a. vistulicus and P. a. ruthenicus, which are known only by a few specimens) from this geo-region, out of which 11 are existing today (Karol Kříž, pers. comm.), seem to consist a single unit for conservation, or with other words, they might be considered as populations of one ESU. This conclusion is strengthened by Todisco et al. (2010) reporting the lack of genetic differentiation between four Northern Carpathian putative subspecies (P. a. antiquus, P. a. interversus, P. a. niesiolowskii and P. a. liptauensis). The areas of the Northern Carpathian putative subspecies are often less than 30 km apart, so genetic isolation between these subspecies is highly unlikely. Further, a few centuries ago (most of) these areas might have been interconnected and the recent fragmentation might be a result of recent habitat loss and climate change. However, considering the different foodplant use of the Western and Eastern populations of the Northern Carpathians (Nakonieczny et al. 2007; Kříž 2011) and the potential for consequent genetic adaptations, one might consider two possible operational units for conservation within the ESU based on the ecological divergence. To avoid any misunderstandings, we must emphasize that we do not imply that conservation efforts on P. apollo populations should be lowered in the region by any means, but we hope that our results can lead to a more efficient allocation of resources, simpler administrative and professional training mechanisms, and legislation (Thomson 1997).

Taken together, despite the numerous disperse populations, we found only two statistically supported clusters and minimal wing shape variation in a comparison of 20 P. apollo subspecies in the Carpatho-Pannonian region. Further, the only divergent group, consisting of two subspecies from the Eastern Carpathians, has high probability of being already extinct. Our results bring attention to the potential problems of “old” taxonomy, based on one-by-one subspecies descriptions lacking the methodology and rigour of modern taxonomy. We must note that some putative subspecies were represented by relatively low sample size. This means that some differences could have remained hidden. However, this is unlikely to affect the general picture. We also note that in the case of statistically significant, but biologically minor morphological shifts, phenotypic plasticity alone as a driver cannot be unequivocally excluded, especially in an intraspecific context. For a final taxonomical conclusion, integrative taxonomy should be employed, supplementing our wing shape analyses with colour and fine scale genetic analyses.