Are the Yellow and Red Marked Club-Tail Losaria coon the Same Species?

Losaria coon (Fabricius, 1793) is currently comprised of ten subspecies, which were originally described under two names, Papilio coon and P. doubledayi before 1909, when they were combined as one species. The main difference between them is the colour of abdomen and hindwing subterminal spots—yellow in coon and red in doubledayi. Wing morphology, male and female genitalia, and molecular evidence (DNA barcodes) were analysed for multiple subspecies of L. coon and three other Losaria species—rhodifer, neptunus, and palu. Our molecular data support the separation of L. coon and L. doubledayi stat. rev. as two distinct species, with L. rhodifer positioned between them in phylogenetic analyses. Wing morphology and genitalic structures also confirm the molecular conclusions. Our findings divide L. coon into two species occupying different geographic ranges: with L. coon restricted to southern Sumatra, Java, and Bawean Island, while L. doubledayi occurs widely in regions from North India to northern Sumatra, including Hainan and Nicobar Islands. Hence, future conservation efforts must reassess the status and threat factors of the two species to form updated strategies.

The species currently recognised as Losaria coon contains a number of taxa, some of which were originally described as species or as subspecies of the yellow or red marked species, coon and doubledayi, respectively, which were originally described in the genus Papilio (Linnaeus, 1758) [3,4]. All 19th century publications (e.g., Rothschild [5]) treated P. coon and P. doubledayi as separate species, as did Moore [6]. Jordan [7], without explanation united them as a single species, Papilio coon, which has been followed by all subsequent publications (e.g., Evans [8] and Talbot [9]). Modern works have placed  (Fabricius, 1793), with currently accepted subspecies delimitation; the blue dash line encircles the tentative distribution range of L. coon.
The yellow and red marked coon are separated from each other geographically, even in Sumatra, where the yellow marked ssp. palembanganus and the red marked ssp. delianus are not sympatric, according to available records, as shown in Figure 1. The constant morphological differences coupled with this distribution pattern leave an unanswered question-whether the yellow and red marked coon really belong to the same species.
The aim of the present study is to revise the taxonomic identities of the yellow and red marked coon using mitochondrial DNA barcode sequences combined with morphological comparisons. Our findings will first clarify taxonomic confusions, draw updated geographic ranges for relevant taxa, and finally benefit from the formulation of better conservation strategies [12].

Taxon Sampling
Specimens of L. coon (including four subspecies), L. rhodifer (Butler, 1876), L. neptunus (Guérin-Méneville, 1840) (including two subspecies), and L. palu (Martin, 1912) in this study were mainly sampled from the authors' private collections. Subspecies identification for Losaria species mainly followed Tsukada and Nishiyama [2]. Specimens were collected and dried in paper triangles at room temperature until use. For each individual used in molecular analysis, two legs from the same side were taken for DNA extraction before the specimen was rehydrated for spreading. An individual of Pharmacophagus antenor (Drury, 1773) and an individual of Pachliopta polyphontes (Boisduval, 1836) were chosen as outgroups for phylogenetic analyses, similar to those used by Hancock [1]. In addition, three sequences, including two individuals of L. coon doubledayi and an individual of L. neptunus neptunus, were mined from the Barcode of Life Database v.4 (BOLD) (http://www.boldsystems.org) for the molecular analyses. The collected data and GenBank accession numbers of all samples and mined sequences used in our molecular analyses are listed in Table 1.

DNA Extraction and Amplification
The phenol-chloroform protocol was used to extract genomic DNA from two legs pulled from the same side of a specimen of all sampled taxa. The legs were homogenised in protease buffer containing 450 µL STE (10mmol/L Tris-HCl, 1 mmol/L EDTA, 100 mmol/L NaCl, pH = 8.0), 25 µL Proteinase K (20 mg/mL) and 75 µL SDS (10%) and incubated at 55 • C for 12 h to rehydrate and lyse the tissue. The subsequent extraction protocol followed that reported by Hu et al. [23]. The resultant genomic DNA was preserved at −40 • C.

Phylogenetic Analyses
We proofread and aligned the raw sequences with Clustal W [25] in BioEdit 7.0.9 [26] by examining the chromatograms to ascertain polymorphic sites, and problematic sequences were excluded when double peaks were present in the chromatograms. MEGABLAST was applied to check the identities of all sequences against the genomic references and nucleotide collections in GenBank, and amino acid translation was realised with the invertebrate mitochondrial criterion in MEGA 7.0 [27] to detect possible Numts (nuclear copies of mtDNA fragments). A search for non-synonymous mutations in-frame stop codons and indels was carried out to further minimise the existence of cryptic Numts [28,29]. The Kimura two-parameter distances [30] between taxa were calculated in MEGA 7.0.
All sequences were used in phylogenetic reconstructions without pruning identical haplotypes, since the monophyly of samples identified as species by morphological characters [2,13,14] needs to be tested. The phylogeny was reconstructed using a Bayesian Inference (BI) as implemented in MrBayes 3.2.6 [31], with the most appropriate partition scheme recovered by PartitionFinder 2.1.1 [32] using the unlinked branch lengths and the greedy algorithm. We used the partitioning scheme among site rate variation, suggested by PartitionFinder, but instead of selecting one substitution model a priori, we used reversible-jump Markov Chain Monte Carlo (rj-MCMC) to allow sampling across the entire substitution rate model space [33]. BI analyses consisted of two independent runs, each with eight rj-MCMC running for five million generations (sampled every 1000th generation) to calculate the clade posterior probabilities (PP). The marginal likelihood estimate was performed with the stepping-stone sampling [34], implemented in MrBayes with 100 steps, each with 10 million generations, and a diagnostic frequency of 1000. Through the computation of Bayes factors, marginal likelihood estimates were used to compare the model fit of an unconstrained topology with the fit of a constrained topology, in which a group of specimens is forced to be monophyletic.

Molecular Species Delimitation
In order to test the delimitation of Losaria species, we relied on a tree-based approach with the Bayesian-Poisson tree process model (bPTP; https://species.h-its.org/ptp/) [35], and on a DNA-based approach with the Automatic Barcode Gap Discovery (ABGD; https://bioinfo.mnhn.fr/abi/public/ abgd) [36]. The bPTP model estimates the probability of each clade being a putative species based on the branch lengths of a tree. The bPTP analyses were performed with the guide tree, reconstructed from the BI in MrBayes with the following parameters: 100,000 MCMC generations, thinning every 100 generations, 0.1% of generation discarded as burn-in. In comparison, the ABGD model estimates the barcode gap separated the intraspecific from the interspecific molecular divergence. Thus, the prior maximum divergence of intraspecific diversity (P) is an important component of ABGD as it provides approximate indications on the barcode gap. If P is set too high, the whole dataset will be considered as a single species, while if P is set too low, only identical sequences will be considered as part of the same species. Previous results showed that the number of species ranges from 1 (generally when P = 0.1) to a large number of species that corresponds to groups of identical sequences (generally when P = 0.001). We followed this practice and set the range of P from 0.001 to 0.1 to explore a lumped versus split delimitation. In addition, the Kimura two-parameter distance was selected but the analyses made with the Jukes-Cantor model provided identical results. The sensitivity of the method to gap width was left by default (1.5), but analyses performed with higher gap width (2 and 3) yielded identical results. The remaining parameters were left by default. Finally, the Monophylizer (http://monophylizer.naturalis.nl/) [37] was performed on the obtained BI tree to test the monophyly of each identified taxon.

Morphological Comparison
Specimens were spread for examination, with the anal scent scales exposed. Species identification was performed prior to molecular work. All spread specimens were photographed using a digital camera, and Adobe Photoshop CS (Adobe, San Jose, CA, USA) was used to adjust the exposure of these photos.
The methods for preparing male and female genitalia followed Hu, Cotton, Condamine, Duan, Wang, Hsu, Zhang, and Cao [16]. The abdomen was removed from the specimen and placed into a 1.5 mL microcentrifuge tube, and 1 mL water was added to rehydrate the tissue at 50 • C for 30 min, then 1 mL 10% sodium hydroxide solution was used to digest soft tissue at 70 • C for 1h. The treated abdomen was neutralised with 2% acetic acid and then dissected in a water-filled Petri dish under the stereoscope to remove residual tissues, scales, and hair. The genitalia were then transferred to 80% glycerol for 12 h to render them transparent. Photographs were taken with a Nikon DMX1200 digital camera (Nikon, Japan), mounted on a Nikon SMZ1500 stereoscope (Nikon, Japan) and automatically stacked using Helicon Focus 7 (Helicon Software, Kharkiv, Ukraine). All parts of the genitalia were fixed on a glue card with water-soluble white glue and pinned with the specimen after observation and photography.

Molecular Phylogenetic Relationships
Bayesian phylogenetic analyses converged well, as indicated by the average standard deviation of split frequencies close to 0 (0.003864), potential scale reduction factors equal to 1 (maximum = 1.003), and effective sample size >> 200 for all parameters. The phylogeny of the Losaria species was recovered as two major clades with maximal PP value, as shown in Figure 2. The clade (PP = 1) containing both the yellow and red marked coon and L. rhodifer is younger than the other clade including L. neptunus and L. palu (PP = 1). In the first major clade, the yellow and red marked coon is divided into two monophyletic subclades by L. rhodifer (PP = 0.83), indicating a specific level divergence between the two morphological types. The subclade of yellow marked coon contains L. coon coon and L. coon sangkapurae, and that of red marked coon contains L. coon insperatus and L. coon doubledayi, which were further split into two paraphyletic branches, as shown in Figure 2. The basal branch includes samples collected from the northern Malay Peninsula (not far from the type locality of doubledayi), and the sister branch includes samples from Indochina, implying a certain degree of divergence between the two geographical populations, as shown in Figure 2.
The estimations of marginal likelihoods with stepping-stone sampling in MrBayes confirmed the non-monophyly of Losaria coon when all specimens (coon coon, coon sangkapurae, coon doubledayi, and coon insperatus) are constrained to form a monophyletic group (logL = −1955.68 for the unconstrained analysis, logL = −1960.01 for the constrained analysis and Bayes factors = 8.66).

Molecular Species Delimitation
Both bPTP and ABGD approaches produced similar results of taxon delimitation. They identified five Losaria species and four subspecific taxa from the Bayesian tree, excluding two outgroups. The bPTP model identified the yellow and red marked coon as separate species with a probability of 0.81 for L. coon doubledayi and of 0.89 for L. coon coon, and with probabilities above 0.65 for L. rhodifer (probability = 0.98) L. neptunus (probability = 0.76) and L. palu (probability = 0.87), as shown in Figure 2. For subspecies, the bPTP model identified subspecies of the red marked coon with supporting values ranging from 0.66 to 0.81, whereas a combination of L. coon sangkapurae with L. coon coon produced a very low supporting value of 0.03, as shown in Figure 2. The ABGD approach also identified the yellow and red marked L. coon as distinct species in all the analyses, whatever the settings for the nucleotide model, the gap width, and the range of P. At a prior intraspecific divergence P > 0.008, we inferred five groups of sequences corresponding to the five Losaria species. Four subspecific taxa are recovered at P > 0.002. ABGD lumps all coon coon, coon sangkapurae, coon doubledayi, coon insperatus, and rhodifer specimens into a single species for a prior intraspecific divergence P > 0.02.
The monophylizer analysis also identified the yellow and red marked coon as two monophyletic clades in the BI tree, along with L. rhodifer, L. neptunus, and L. palu. Paraphyletic clades were found in subspecies, such as L. coon coon (one of the yellow marked coon) and the two doubledayi races (red marked coon) from southern Thailand to the Malay Peninsula and Indochina, as shown in Table 2. The Kimura two-parameter (K2P) distances between the yellow and red marked coon reached 2.95%. The K2P distances between all taxa within both groups, and the species of L. rhodifer, L. neptunus, and L. palu ranged from 0.19% to 9.61%, with that between L. coon coon and L. coon sangkapurae being the smallest, and that between L. rhodifer and L. palu being the greatest, as shown in Table 3. For those distances between the yellow and red marked coon, the overall range fell from 2.44% to 3.17%, which exceeds the proposed barcoding gap of 2% [38], while the K2P distances within the yellow or red marked coon, respectively, remained at infraspecific level (0.19-1.08%), as shown in Table 3. It is noteworthy that the K2P distances between the yellow marked coon and L. coon insperatus (one of the red marked subspecies) are greater (3.57-3.77%) than those between the yellow marked coon and L. rhodifer (3.17-3.35%), and the distances between the yellow marked coon and the Malay Peninsular doubledayi are very close to those between the yellow marked coon and L. rhodifer (3.08-3.28% vs. 3.17-3.35%), as shown in Table 3. Table 3. The Kimura two-parameter distances (in percentages) between all taxa of genus Losaria (Moore, 1902), with species and subspecies identified as in the Bayesian phylogenetic tree, as shown in Figure 2. Given the aforementioned molecular evidence, we propose that the yellow and red marked coon belong to two distinct species, namely Losaria coon (Fabricius, 1793) and Losaria doubledayi (Wallace, 1865) stat. rev., respectively. The subsequent morphological analyses and discussion will be based on the separated names.

Morphological Differences
Comparison of the wing morphology between L. coon and L. doubledayi showed constant differences, as illustrated in Figure 3. Beside the evident yellow and red colouration on the abdomens and hindwings of both species, the following characters are also important in separating the two species: (1) the hindwing white discal cell patch reaches the wing base in L. coon, but only reaches about half way to the base in L. doubledayi (a); (2) the forewing ground colour in L. coon is buffish brown, whereas in L. doubledayi it is blackish (b); the colour of the male scent scales is light buffish in L. coon, but greyish black in L. doubledayi (f); (3) the subterminal spot in the cell M 2 of the hindwing is usually larger in L. coon than in L. doubledayi (c); (4) the junction between the termen and the lobe at the end of vein CuA 1 on the hindwing is evidently angled in L. coon, but much straighter in L. doubledayi (d); (5) in the hindwing tail, the stalk is broader in L. coon than that in L. doubledayi, and the club is narrower in L. coon than that in L. doubledayi (e).
In overall appearance, among the remaining three Losaria species, L. rhodifer is more similar to L. coon and L. doubledayi. However, the more fragmented hindwing white patch, much larger subterminal crimson spots, and the iconic crimson-tipped tail are ready separation characters, as shown in Figure 4. Morphological similarity and differences of L. rhodifer agrees with its phylogenetic position obtained by our molecular analyses, as shown in Figure 2. The other two species, L. neptunus and L. palu, are less closely related to the previously mentioned taxa, judging from the distinctive yellow-tipped abdomens and lack of hindwing discal patches, as shown in Figure 4.  Male genitalia of available subspecies were dissected, constant differences were found between L. coon and L. doubledayi, while the characters remain consistent within each group, as shown in Figure 5. The main differences in male genitalia are: (1) the dorsal view of the superuncus is broader in L. coon (0.29 ± 0.03 mm, n = 12), while it is rather slender (0.16 ± 0.02 mm, n = 15) in L. doubledayi; (2) the tip of the valve is only truncated in L. coon but strongly indented in L. doubledayi, forming a bifid tip; (3) the apical half of the juxta is 1.3 to 1.5 times broader in L. coon than in L. doubledayi. Female genitalia of available subspecies were dissected, constant differences were found between L. coon and L. doubledayi, while the characters remain consistent within each group, as shown in Figure 6. The main differences in female genitalia are: (1) the ostium is smaller in L. coon but broader in L. doubledayi; (2) the lamella antevaginalis sunk inwardly, forming a crescent edge, in L. coon, while forming an even edge in L. doubledayi; (3) the lamella postvaginalis is 0.5 to 0.7 times shorter and 1.8 to 2.0 times broader in L. coon than that in L. doubledayi; (4) the lateral area surrounding the ostium is less developed in L. coon than in L. doubledayi; (5) the signum is leaf-shaped with one end longer than the other in L. coon, while nearly round in L. doubledayi, and the length of the signum in L. coon is 1.7 to 2.0 times longer than that in L. doubledayi. Among the male and female genitalia of the other three species of Losaria, those of L. rhodifer are similar to the previously mentioned two species, but the shape of valve tip of the male genitalia and the ostium, lamella antevaginalis, and signum are still different, as shown in Figures 7A and  8A,D. Such resemblance indicates that L. rhodifer is closely related to both L. coon and L. doubledayi, as previously mentioned. The genitalia of both sexes of L. neptunus and L. palu are all significantly different, as shown in Figure 7B,C and Figure 8B-F, supporting the phylogenetic relationships with L. coon and L. rhodifer, as inferred by the molecular data.

L. neptunus neptunus (Guérin-Méneville, 1840)
shown in Figure 2. Morphological analyses of genitalic structures and wing patterns are consistent with the molecular evidence, as shown in  therefore, we confirm the species status of L. doubledayi. Furthermore, our morphological analysis, based on the literature [2] and type photos of all recognised subspecies of L. coon and L. doubledayi, allowed us to update the subspecies list of each species, with L. coon, containing four subspecies, and L. doubledayi, containing six subspecies. As stated under Materials and Methods (Taxon sampling), the present study was unable to include all subspecies due to difficulties in obtaining specimens from India, northern Myanmar and some Indonesian islands, and consequently partial findings can be presented here for subspecies. Future research must address this point to form a better understanding of the subspecies divergence, diversity, and distribution pattern when those taxa become accessible.
The divergence between L. coon coon and L. coon sangkapurae is limited, as shown in Tables 2 and 3 and Figure 2, despite the two subspecies being isolated on different islands. Geographic proximity and recent separation between Sumatra, Java, and those offshore small islands might explain such limited divergence [40][41][42][43]. Based on the distribution pattern and morphological characters of the other two subspecies [2], namely ssp. palembanganus and ssp. patianus, it is likely that the divergence between them would also be limited.
In comparison, the divergence within L. doubledayi is greater. The phylogenetic tree, genetic distances and molecular species delimitation results showed that even the populations from the Malay Peninsula and upper Indochina are different, let alone the insular ssp. insperatus from Hainan, as shown in Table 3 and Figure 2. Our examination of a large series of specimens, type photos, and literature records also showed morphological differences among these populations/subspecies. For instance, populations from upper Indochina are usually smaller in size and possess larger hindwing white discal patches compared to populations from the Malay Peninsula, and the population from South Vietnam (samples LCN018 and LCN019 in our dataset, other examined specimens, and also in the cited literature) possesses unique ochreous abdomens, rather than crimson ones [44]. Similarly, photos of live specimens of ssp. cacharensis, plus type photos and the description of ssp. putaoa also showed constant morphological differences compared to other populations [14,45]. Future research should address the question of the population divergence of L. doubledayi across its distribution range.
The molecular and morphological divergence within L. doubledayi across its range might be attributed to two factors. The first factor lies in the more heterogeneous natural environment (e.g., terrain topology, climate, and landscape). The valley habitat in North Myanmar and Northeast India, separated by mountains, where ssp. putaoa and ssp. cacharensis, respectively, occur, is a typical example of divergence. Such extreme altitude shifts restricted populations within narrow ranges (mostly river valleys with broadleaf forest patches), and they subsequently became distinct evolutionary entities (subspecies). Similarly, the southern-central savannah plain in Thailand and Cambodia formed an effective barrier due to the lack of suitable forests for the larval foodplants. The second factor is the larval foodplants for Losaria, which constitute a handful of species in the genus Thottea (family Aristolochiaceae) [46]. Thottea is a genus of lowland forest-dwelling shrub species with high endemism; habitats without dense forests would be unsuitable for Losaria [47,48]. The superimposition of both factors resulted in spatially fragmented suitable habitats, which eventually lead to divergence within L. doubledayi.
Given the separation of L. coon and L. doubledayi, conservation strategy and threat evaluation of the original L. coon should be reassessed. According to Fernando,Jangid,Chowdhury,Kehimar,Lo,and Moonen [12], L. coon and L. doubledayi were considered as a single wide-ranging species, with an IUCN rank of Least Concern (LC). After its work separating these two species, L. coon now becomes a narrow-ranged species that only occurs in several Indonesian islands, including southern Sumatra, Java, and Bawean, as shown in Figure 1. Hence, conservation and assessment strategies must be updated accordingly to ascertain the current population status. Furthermore, as a separate species, although L. doubledayi occupies a much wider range, it still possesses more localised subspecies, even within continental Indochina, as shown in Figures 1 and 2. Each population represents a distinct evolutionary pool, while some of them only occupy a much smaller geographic range than others (e.g., ssp. cacharensis and ssp. putaoa). Since anthropogenic disturbance to most of its distribution range remains active to date [49][50][51][52][53], we think that the population status, vulnerability, and threat factors of each subspecies of L. doubledayi must be re-evaluated for future conservation.

Conclusions
The present study used DNA barcode data and morphological characters to unveil the relationship of the yellow and red marked Losaria coon across its distribution range. Our analyses support the separation of L. coon (yellow marked) and L. doubledayi (red marked) as two distinct species. Our findings also updated the geographic ranges of the two species: with L. coon restricted to southern Sumatra, Java, and Bawean Island, while L. doubledayi occurs widely in regions from North India to northern Sumatra, including Hainan and Nicobar Islands. The separation effectively made L. coon a narrow-ranged species which may require more conservation attention in the future, while that for L. doubledayi must also be revised due to its higher subspecies diversity reflecting multiple evolutionary pools associated with larval foodplant restriction and landscape heterogeneity.