Amblyscirtes arizonae H. Freeman, 1993

Zhang, Jing, Cong, Qian, Shen, Jinhui, Opler, Paul A. & Grishin, Nick V., 2020, Genomic evidence suggests further changes of butterfly names, The Taxonomic Report of the International Lepidoptera Survey 8 (7), pp. 1-41 : 31-36

Amblyscirtes arizonae H. Freeman, 1993 is a species distinct from Amblyscirtes elissa Godman, 1900

Megathymus violae D. Stallings & Turner, 1956 is a species distinct from Megathymus ursus Poling, 1902

Megathymus beulahae D. Stallings & J. Turner, 1958 (Fig. 31). The Fst/Gmin statistics for comparison of ursus and violae groups are 0.56/0.001 (note close to 0 gene exchange between these taxa). The COI barcodes of the M. ursus and M. violae holotypes differ by 1.8% (12 bp). For these reasons, we reinstate Megathymus violae D. Stallings & Turner, 1956 as a species-level taxon.

Discussion: genomic trees, branch lengths and genera

Near the end, coming back to the Introduction, we elaborate on and illustrate the reasons behind the classification decisions that we have chosen to make about genera. Traditionally, species were grouped into genera by phenotypic characters. For butterflies, these were mostly wing patterns and shapes, and genitalic morphology. When differences in these phenotypic aspects were deemed to be significant enough according to a subjective opinion of an individual researcher, they formed a basis for defining a genus. This system served its purpose until a consensus opinion was formed among taxonomists that each genus should be monophyletic. It is exceedingly difficult to predict monophyletic taxa from their phenotypes, and DNA-based phylogenetic trees provide the most reliable inference of monophyletic groups. Therefore, genera should be defined using phylogenetic trees constructed from DNA sequences.

Each individual feature of an organism can experience rapid evolution and fool researchers into making incorrect classification decisions. Genitalia that are commonly used in Lepidoptera classification are prone to such rapid changes as well. For instance, Steinhauser (1989) proposed a genus Thessia on the basis of unique shape of genitalic valvae. However, even a very short, 654 base pair region of DNA, such as the COI barcode, reveals the paraphyly of Achalarus Scudder, 1872 (as it was circumscribed at that time) with respect to Thessia ( Pfeiler et al. 2016) , suggesting that the unique valva is a result of accelerated evolution within Achalarus rather than a character originated after Thessia and Achalarus have (supposedly) diverged from each other. Therefore, a decision to erect the genus Thessia was a mistake, because Thessia is a subclade within (as it was then defined) Achalarus . Nevertheless, the barcode DNA region itself is a single feature, and as any other such feature, can experience evolutionary irregularities. To reduce such mistakes, it is better to use information from as many features as feasible. Complete genomes offer the ultimate DNA dataset for classification decisions. Genomic analysis suggests that Achalarus itself is a junior subjective synonym of the subgenus Thorybes Scudder, 1872 , and Thessia is actually a junior subjective synonym of the subgenus Murgaria E. Watson, 1893 ( Li et al. 2019).

Genomic trees summarize integral information about the entire organism, not just some of its features. For this reason, we use them to make decisions about classification of genera. Here, we explain how we arrive to these decisions using examples from this work and our previous publication ( Zhang et al. 2019c). A maximum likelihood tree constructed using IQ-TREE program (model GTR+I+G) ( Minh et al. 2020) from concatenated protein-coding regions of nuclear genomes is shown in Fig. 32. To best follow our logic, a reader may close the tree on the right ( Fig. 32b, the final result) and look only at the tree on the left ( Fig. 32a), which is the same as the tree on the right, but without the final results being marked in order not to bias the reader. This tree was constructed without assuming a molecular clock and reveals differences in evolutionary rates between species: i.e., species names are placed at difference distance from the left side of the page (=from the root of the tree). We see that Emesis evolved the fastest (the farthest from the left), and Ephyriades Hübner, [1819] evolved the slowest (closest to the left). In a tree, only horizontal (left-to-right) distances matter. Vertical (top to bottom) distances are arbitrary and are set to place species names evenly along vertical dimension, so that the names do not overlap and are not too far away from each other to save space.

Tree branches have different lengths. Again, only horizontal branches have evolutionary meaning, and vertical lines in the tree are set to avoid overlap of names and to connect branches to nodes. The length of a horizontal branch is proportional to the number of estimated changes in DNA (=fixed mutations) that happened along the branch. The tree has a scale bar near the bottom ( Fig. 32). The length of that bar, as indicated, corresponds to 6 changes per 100 base pairs (=0.06, or 6%). Using this bar, we can measure evolutionary distances between taxa in DNA changes. Long branches correspond to many changes in genomic DNA. Short branches correspond to few changes in genomic DNA. Because larger number of DNA changes are expected to result in larger number of phenotypic changes, longer branches correspond to more phenotypic changes on average. These are integral changes and some of them may be in genitalia, others may be in caterpillar morphology. Regardless of where these changes are, longer branches are more important than shorter branches. In addition to larger number of changes, longer branches are also more reliable and support clades that are more likely to be correct. The statistical reliability of every clade is indicated by a number next to each node. This number is a fraction of trees (out of 100 trees constructed from various subsets of genomic segments) that contain this node, e.g. a genome was divided into 100 segments and each segment was used to generate a tree. If a particular node is present in all 100 trees, the number by that node is 1. Therefore, this number measures consistency between trees constructed from different partitions of the data. If every DNA segment supports a clade, it has a number 1 next to it. If 94 out of 100 segments support the clade, the number is 0.94.

A genus should be a prominent, major clade in the tree that is above species level and below tribe and subtribe levels. Phenotypic features are difficult to quantify, and due to the possibly uneven speed of evolution, it is a challenge to determine which phenotypic changes correspond to major clades. Total genomic changes can be used as a yardstick to quantify each clade. The number of total genomic changes is proportional to branch lengths in genomic trees ( Fig. 32a). Therefore, the task of identifying genera may be viewed as a task of identifying prominent (i.e. supported by longer branches compared to surrounding branches) clades in genomic trees that on average correspond to how genera are defined currently (to avoid unnecessary taxonomic changes). Additionally, we believe that each genus should not be very different from another genus in terms of genetic differentiation of species placed in a genus, i.e. genera could be defined consistently, so that genera correspond to clades of approximately the same differentiation within. Defined consistently, the genus becomes a level (as meant by this word) of a classification instead of several varying levels, i.e., we can expect a genus to be a group of species bearing about the same relatedness among them as that in other genera. It would seem unnatural if one phylogenetic group is oversplit into genera, i.e. genera in that group correspond to very closely related species, but another group is undersplit, and genera in it correspond to species that are only distantly related. The measure of closeness as we use it, is overall genomic divergence.

Looking at the clade of Hesperiidae at the top of the tree ( Fig. 32a) we see three major clades, not two and not four. The first clade is Ephyriades and is sister to all other taxa. Then all others split into two clades of similar genetic differentiation within each clade. We see that each of these clades resembles a tight bush or a comb, rather than an evenly bifurcating tree, i.e. the internal branches in either clade are much shorter than a branch that supports the entire clade. The clade with Gesta bifurcates into two subclades, one consists of Gesta sensu stricto (s. s.). Species from the other subclade were called " Erynnis " previously (and are called Erynnis in the tree to facilitate communication): it is a subgenus Erynnides Burns, 1964 (type species Nisoniades propertius Scudder & Burgess, 1870 ). If we consider these two subclades to be major clades, then the Hesperiidae tree would consist of four major clades ( Ephyriades , Erynnis s. s., Erynnides and Gesta ). However, the branches supporting the two subclades ( Erynnides and Gesta ) are nearly three times shorter than the branches supporting the clades Erynnis s. s. and a clade combining Erynnides with Gesta . Therefore, the Hesperiidae subtree should not be partitioned into four major clades, because two of these clades ( Erynnides , Gesta ) would be minor compared to the other two, and more importantly, compared to the clade combining Erynnides with Gesta .

The remaining alternative to a three-clade partitioning would be a two major clade partition, where Erynnis s. s., Erynnides and Gesta are all joined together into Erynnis sensu lato (s. l.) The branch supporting this clade is only slightly shorter than the branch supporting Erynnis s. s., and therefore this clade is rather prominent in the tree. We reject this solution for the two reasons. First, Erynnis s. l. is not a homogenous group of species, which we think a genus should be, i.e. the Erynnis s. l. clade does not look like a bush or a comb. Instead, it splits into two major clades: Erynnis s. s. and Erynnides + Gesta , (we call this clade Gesta s. l. from now on) each of which individually looks more like a comb than when they are combined. In other words, Erynnis s. l. itself is composed of two major clades, and does not represent a single group of species, but two major groups of species.

The second reason stems from consistency between different genera, i.e. an idea that different genera should represent the same level in the classification ( Fig. 33). Being a level, genera should be groups of species with comparable divergence within each genus. In this tree ( Fig. 32a), where all branches are to scale, we can compare divergence between Erynnis s. s. and Gesta s. l. to the divergence in Nymphalidae previously placed in genera Aglais , Polygonia , Nymphalis , and Vanessa . These two subtrees ( Erynnis and Vanessa ) are illustrated in Fig. 33. Genetic differentiation of a clade is proportional to the average distance (average sum of branch lengths) from the last common ancestor of the clade (=node that supports the entire clade) to the leaves (=species) in the clade. In other words, it is a linear distance (in horizontal dimension) from the base of the clade to the tips of the tree. On the one hand, we see that Polygonia divergence is rather small, perhaps comparable to the divergence of the Erynnides subclade with horatius and juvenalis, and definitely smaller than the divergence within either Erynnis s. s., or Gesta s. l. On the other hand, the divergence of Erynnis s. l. is larger than the divergence of Aglais , Polygonia , Nymphalis and Vanessa combined. Therefore, having Erynnis s. l. as a genus is inconsistent with having Polygonia as a genus: these two groups represent different levels in the classification. Coming back to Nymphalidae , we see that branches supporting Aglais , Polygonia and Nymphalis individually are much shorter than the branches supporting Erynnis s. s. or Gesta s. l. Only the branch supporting Vanessa is somewhat comparable, although shorter. However, the branch supporting the first three clades together ( Nymphalis s. l.) is more prominent and is about the same as the branch supporting Vanessa .

In summary, Erynnis s. l. is comparable to Vanessa s. l. ( Nymphalis s. l. + Vanessa s. s.). A system of two genera ( Erynnis s. s. and Gesta s. l.) is comparable to two genera Nymphalis s. l. and Vanessa s. s. We attempt to choose an internally consistent solution that agrees the most with how these species are assigned to genera in the current classification. Therefore, we choose the 2-genus solution for both of these cases, as shown in Figs. 32b (colored clades E: Erynnis , G: Gesta , N: Nymphalis and V: Vanessa ) and 33 (shaded clades). These four genera represent a similar level in the classification and correlate with the current classification of these butterflies. The choice of Erynnis s. l. would correspond to a consistent choice of joining all four Nymphalidae genera in Vanessa , which may represent too much of a lump and more name changes ( Fig. 33).

Another point is that genetic differentiation can be used to estimate divergence times of these clades through the tree rescaling and calibration with fossils (primary calibration) ( Chazot et al. 2019) or other time-calibrated trees (secondary calibration) ( Zhang et al. 2019a). As we have seen in Hesperiidae ( Li et al. 2019) , the genus level typically corresponds to divergence between 10 and 15 million years ago (Mya). Divergence of Erynnis s. l. was estimated to be about 27 Mya, which is larger than the divergence between Vanessa s. s. and Nymphalis s. l., at about 22 Mya ( Zhang et al. 2019d). However, divergences within Gesta s. l. (~16 Mya), Vanessa s. s. (~16 Mya) and Nymphalis s. l. (~14 Mya) ( Zhang et al. 2019d) are very much comparable to each other, and these genera represent groups of about the same level. It should be noted that the divergence times are only approximate, should be considered with caution, and may have errors of possibly up to 50%, especially in groups with large differences in evolutionary rates. However, the relative comparison of divergence times estimated within the same tree using the same method is expected to be more accurate. Finally, a question arises about how these considerations of trees, branch lengths, divergence and geological times correlate with genera definition based on phenotypic characters. Because phenotypic characters are encoded by the genotype, longer branches in the tree that correspond to more changes in a genotype (these are integral genomic trees, not based on several gene markers) should translate to more changes in the phenotype. We advocate a method to delineate genera from genomic trees first, and then come back to phenotypic analysis to find the phenotypic characters that correspond to these genera. In the case of Erynnis and Gesta , the retrospective inspection of morphological characters yields substantial differences in male genitalia that have been noted and illustrated previously ( Evans 1953; Burns 1964). The uncus is asymmetric, terminally broad in Gesta , but is symmetric, extending into a "beak" in Erynnis .

The valvae are strongly asymmetric with at least one extended harpe in Gesta , but are more symmetric with shorter harpes in Erynnis . Other differences are stated in the diagnosis of Erynnides by Burns (1964).

Comparing the clades of other groups in Fig. 32a with Erynnis / Gesta and Nymphalis / Vanessa we see that divergence within Speyeria and Roeberella (a clade containing R. clavus and with Apodemia hypoglauca at its base), and divergence between Hypaurotis , Favonius and Habrodais is much smaller than that in the groups we define as genera. We also see that the colored clades (with letters denoting corresponding genera by each clade) in Fig. 32b are more or less equivalent to each other in terms of genetic differentiation (distance from the base of the clade to its tips) and prominence (length of the branch supporting the clade). For these reasons, we suggest that these clades can be treated as genera: they are prominent, consistent, and reasonably well correspond to how genera have been defined previously. The changes we suggest combine some more compact in terms of genetic (and phenotypic) differentiation genera into more internally diverse genera that become more consistent with the differentiation within many classic genera such as Emesis , Ministrymon , Vanessa , and Boloria .