Back to: Projects

Distribution of Afrotropical Kite Swallowtails

Our methods of study

On this page we discuss the features of the morphology of butterflies we used to provide the raw material for our studies, how we define our characters and how we analyse them.

We wanted to answer these questions:

  • How many species of Afrotropical kite swallowtails are there?
  • Are the Afrotropical Graphium a natural group, including all Afrotropical species and no others? In other words, is the subgenus Arisbe monophyletic?
  • Do the swordtailed species form a natural, monophyletic group?
  • Are the non-swordtailed species monophyletic?

Morphology

To provide the characters needed to produce a classification of the Afrotropical kite swallowtails we studied the morphology — the pattern and form — of the adult butterflies.
There are many features that could have provided further evidence, but were not readily available to us, either because of lack of material, lack of access to technical facilities, or both. The morphology of the early stages — egg, larva, pupa — can be useful, as can analysis of chemicals such as pheromones and DNA. Of course, this does not preclude further studies, by ourselves or others, as material and resources become available.

Wing pattern

The most obvious features of butterflies are their generally bright and showy wings. It is often said that 'butterflies wear their genes on their wings'.
The marked differences between the various types of pattern shown by the swordtailed and mimetic species meant that the number of characters we could define was limited.
Here is a reminder of some examples of the patterns.

Graphium policenes upperside Graphium taboranus upperside Graphium levassori upperside Graphium ucalegon upperside
G. policenes G. taboranus G. levassori G. ucalegon

Despite these gross differences we were able to define about 16 characters based on the shape and pattern of the wings.

Genitalia

After examining the wing pattern, entomologists traditionally turn their attention to the chitinous exoskeleton. In particular, we examine features towards the tip of the abdomen that are involved in mating. These features are often both highly complex and characteristic of species or groups. The features help prevent the insects from mating with the wrong species.

In the males, the organs are the aedeagus — the tube that enters the female body and through which the sperm is passed — and the surrounding and supporting organs. Most of the characters that we were able to define were on the valves. These articulated plates at the very tip of the abdomen are used to hold the female abdomen during mating and carry a number of structures that vary between species and between groups of species. These structures are believed to be important in mate recognition systems. We identified about 30 useful characters on the valves, with just another three on other parts: one each on the uncus, socius and saccus.

male genitalia

male valve

Dissected genitalia of male Graphium policenoides (left valve removed) to show the main organs. The tip of the abdomen is to the right.

Left valve male Graphium ridleyanus to show the main organs. The tip of the abdomen is to the left.

diagram of female genitalia imag of female genitalia

Diagram of the underside of the tip of the abdomen of female Graphium policenes. Redrawn from van Son (1949). The tip of the abdomen is to the top.

Electronic image of the vestibulum area of female Graphium policenes. The tip of the abdomen is to the top.

The female genitalia exhibit less obvious, or at least definable, variability. Nevertheless, we were able to describe eight characters.

The female genitalia of butterflies and related moths includes two distinct tubes: one for mating, the other for egg laying. Naturally, in order for the sperm to fertilise the egg, there is another, internal tube linking the two. The mating tube is the ductus bursae, leading to a chamber called the bursa copulatrix, where the sperm packets are stored. In Graphium the ductus opens into a cup-shaped depression called the vestibulum, a feature which helps define the genus (see higher classification). Surrounding the opening of the ductus (the ostium bursae) may be projections (usually a pair laterally and a single, central one), apparently standing guard. This may indeed be a function in some species, but, as in the male valves, they may be more important for tactile mate recognition. We call these projections the ostial lobes.

It was features of the ductus and its ostium and the ostial lobes that provided us with most of our female characters.

Unfortunately, females of many species are rare in collections; due to their behaviour they are caught less frequently than the males. We were unable to examine the females of eleven species, mainly in the policenes and adamastor groups. If it were possible to obtain these, we are sure our conclusions could be refined.

Defining characters

The problem with all characters, especially those derived from the genitalia and other features of the exoskeleton, is defining them.

The difficulty is to translate highly three-dimensional shapes we term 'organs' — formed by complex foldings and evaginations of the exoskeleton during metamorphosis — into simple linguistic statements that can be further translated into numerical form.
Due to the often smooth transition from plain body wall into these shapes, these organs require delimiting (often by gestalt perception) before characters can be defined or described. This represents a translation from what the philosopher Sir Karl Popper has described as 'World 1' of physical objects to his 'World 3' of 'objective contents of thought', specifically scientific thought.

Scientists may perceive and thus define and characterise the organs differently. It is necessary, therefore, to be as precise as possible in defining the characters and their states. These definitions are themselves, in effect, hypotheses about the characters we are using and should be as amenable to refutation as any other hypotheses, including those of the relationships we are trying to discover.

Analysis

We interpret our data and derive our hypotheses of relationship and classification by procedures known collectively as cladistic analysis.

Having diligently defined our characters and recorded their state or condition in our study organisms, we place these 'scores' in a matrix recording the state of each character for each taxon.

We attempted to define the characters as simple alternatives: the character either absent or present (scored as 0 and 1 respectively). In two cases, however, we were unable to make such a neat division as the organs concerned (the male valve ventral harpe and the female central ostial lobe — see above) were present in a number of different forms. For these two characters we therefore had a number of character states, ten in the case of the ventral harpe and six for the central ostial lobe.

Since there has been some controversy about the usefulness of these 'unordered multistate' characters, we carried out two sets of analyses: one with these characters included, the other with them excluded.

We can analyse the matrix using one (or more) of a number of available software packages. We used NONA (written by Pablo Goloboff) to analyse the data, and Kevin Nixon's Winclada to organise the matrix and interpret the results. Both of these packages are available as downloads via the Willi Hennig Society website.

The output of these programs is presented as a hierarchical branching diagram, technically known as a 'dendrogram', in which lowest level of taxa we are interested in — in our case the species or even subspecies — appear, once only, at the tips of the branches.

This is often referred to as a 'tree', but this is somewhat misleading as it does not represent a genealogical or historical family tree (the branching points do not represent ancestors), but is merely an efficient way of representing and mapping the pattern of changes of character states across the group of organisms under study. Although we use the term tree as a convenience, this consideration should be kept in mind.

Species are grouped by sharing character states that, ideally, are not found elsewhere (termed 'synapomorphy'). In practice, it does not work out as simply as this.

Different characters may link different sets of species in conflicting ways. In evolutionary terms, this implies that similar character states may have evolved independently on more than one occasion (termed 'homoplasy').

Another common cause for this conflict is that the animals may share a primitive character — one that has been modified or lost in more advanced groups. These 'symplesiomorphic' characters can artificially link taxa that are not really closely related. In the Afrotropical Graphium the presence of long tails appears to be just such a character, linking 13 species. Our results suggest, however, that the swordtails are just such a 'paraphyletic' grouping linked by the presence of a symplesiomorphic character.

By including as many independent characters as possible, we hope to overcome the bias that such obvious, but false, associations can create.

To help decide which characters are indeed symplesiomorphic we include a number of taxa believed to be somewhat more distantly related to our study group than they are to each other, based on previous studies. These are termed 'outgroup' taxa as opposed to the 'ingroup' subjects of our study.

In our case we chose nine species representing other, oriental subgenera of Graphium as well as other members of the tribe Leptocircini, based on the revisions by Munroe, Hancock and Miller (see references; — see also higher classification). As can be seen in our results, this choice can cause previously accepted classifications to be called into question.

So our method of analysis has to take into account the possibility of homoplasy and symplesiomorphy and the conflicting possibilities they may cause.

The number of possible trees rises steeply with increasing numbers of taxa. For 3 taxa there are just 3 possible bifurcating rooted trees. For 5 there are 105. By the time we reach 10 taxa, the number has risen to nearly 35 million, and for 20 taxa the number is:

8,200,794,532,637,891,559,375

For 39 taxa plus 9 outgroups, the number is astronomical (see Felsenstein, 1978).

Our analytical software has to find which of these trees best fit the pattern of distribution of the characters. The most accepted way of doing this — at least with morphological as opposed to DNA sequence data — is to apply the principle of 'parsimony'. That is the scientific principle of accepting the hypothesis that explains the data most simply and efficiently. Parsimony is sometimes referred to as 'Occam's razor'.

In cladistics this means preferring those trees which imply the fewest changes in character states — the shortest trees — and the software is designed to do just that.

Our initial analysis found 12,490 equally parsimonious trees. We then looked for those elements of the structure of the tree that were common to all those trees, the so-called 'strict consensus' tree.

Not surprisingly, very little structure survived. However, in all cases we recovered:

  • a clade composed of tynderaeus + latreillianus
  • a clade with G. (Pazala) mandarinus the sister of antheus + evombar
  • a clade comprising policenes, liponesco, biokoensis and policenoides as a 'polytomy' (all 4 arising from the same branching point or 'node')
  • a clade confirming the existence of an angolanus group composed of angolanus, endochus, morania, schaffgotschi, taboranus and ridleyanus, with morania sister to the remainder.

Clearly, we needed to refine our analysis to make sense of our data.

To do this, we reduced the number of taxa by excluding most of the policenes clade, most of the angolanus group, and those members of the otherwise well-characterised adamastor group where we lacked females, thus reducing the potential problems from missing data, that can affect the analysis.

This time, when we excluded the multistate characters we obtained 1,656 equally parsimonious trees. With the multistate characters included, just 756 equally parsimonious trees were generated. The strict consensus trees in each case retained a considerable amount of structure. Though there are some differences between them, they are sufficiently similar to allow us to draw some general conclusions and to discuss those differences.

We discuss those similarities and differences in our results section.

For more general information on the meaning and interpretation of cladistics, we would recommend Henry Gee's Deep Time. For details of theory and method, see Ian Kitching and colleagues' Cladistics.