Erratic Evolution of Glycerol-3-Phosphate Dehydrogenase in Drosophila, Chymomyza, and Ceratitis

We have studied the evolution of Gpdh in 18 fruitfly species by sequencing 1,077 nucleotides per species on average. The region sequenced includes four exons coding for 277 amino acids and three variable-length introns. Phylogenies derived by a variety of methods confirm that the nominal genus Zaprionus belongs within the genus Drosophila, whereas Scaptodrosophila and Chymomyza are outside. The rate of GPDH evolution is erratic. The rate of amino acid replacements in a lineage appears to be 1.0 × 10−10/site/year when Drosophila species are considered (diverged up to 55 million years ago), but becomes 2.3 × 10−10 when they are compared to Chymomyza species (divergence around 60 My ago), and 4.6 × 10−10 when species of those two genera are compared with the medfly Ceratitis capitata (divergence around 100 My ago). In order to account for these observations, the rate of amino acid replacement must have been 15 or more times greater in some lineages and at some times than in others. At the nucleotide level, however, Gpdh evolves in a fairly clockwise fashion.


Introduction
The nicotinamide-adenine dinucleotide (NAD)-dependent cytoplasmic glycerol-3-phosphate dehydrogenase (GPDH, EC 1.1.1.8) plays a crucial role in insect flight metabolism through its key position in the glycerophosphate cycle, which provides energy for flight in the thoracic muscles of Drosophila (O'Brien and MacIntyre 1978). In Drosophila melanogaster the Gpdh gene is located on chromosome 2 (O'Brien and MacIntyre 1972) and consists of eight coding exons (Bewley et al. 1989;von Kalm et al. 1989). It produces three isozymes by differential splicing of the last three exons (Cook et al. 1988). In the mouse, two different isozymes are encoded by two separate loci (Kozak et al. 1982).
The polypeptide chain of GPDH can be divided into two main domains: one is NAD binding, the other is catalytic. In the rabbit enzyme, the former is determined by the first 118 amino acids (Otto et al. 1980). The enzyme is known to be evolutionarily conserved (Bewley et al. 1989), displaying very low heterozygosity within or variation among Drosophila species (Lakovaara et al. 1977). The catalytic domain seems to be less conserved (Bewley et al. 1989). Here we present analysis of a Gpdh gene region comprising most of the coding sequence (768 bp out of 831 bp) of exons 3-6 in 18 species of Drosophila and related genera, corresponding to the whole catalytic domain plus 45 codons of the NAD-binding domain. We have also sequenced the intervening introns, which jointly have an average length of 309 nucleotides per species.
Sequence Analysis. Corrections for superimposed and back replacements (PAM; Dayhoff 1978) were made with the CLUSTAL V program (Higgins et al. 1992). Kimura two-parameter distances (K2) (Kimura 1980) and Jukes-Cantor distances (Jukes and Cantor 1969) were estimated with the MEGA 1.0 program (Kumar et al. 1993). Synonymous and nonsynonymous substitutions were calculated according to Li (1993) as implemented in programs written by A. Zharkikh. Neighbor-joining trees (Saitou and Nei 1987) were obtained with the MEGA program. The maximum-parsimony (Fitch 1971) and maximum-likelihood (Felsenstein 1981) trees were obtained and tested according to Templeton (1983) and Kishino and Hasegawa (1989), respectively, using the PHYLIP 3.57c package programs DNAPARS and DNAML (Felsenstein 1989). Ceratitis capitata a Raised to genus category in the revision by Grimaldi (1990)

Gene Structure and G+C Content
The Gpdh gene in Drosophila consists of eight exons with a total 1,100 nt of coding sequence. We have amplified a segment that extends from the seventh base of exon 3 to the third base from the end of exon 6, accounting for 823 of the 831 coding nucleotides (Fig. 1). We have also sequenced the intervening introns, amounting to 309 nt per species on the average. In Drosophila the lengths of exons 3 to 6 are, successively, 207, 373, 154, and 97 bp (Bewley et al. 1989;Tominaga et al. 1992;von Kalm et al. 1989). The PCR fragments have the same structure and identical exon length for the 17 drosophilid species in our sample. Ceratitis capitata has an additional intron that splits the coding sequence corresponding to exon 4 in the drosophilid gene. We have also found an intron at the homologous site in another dipteran species, Calliphora, as well as in the mouse, which suggests that this additional intron is the ancestral condition that has been lost in the family Drosophilidae. Figure 2 shows 768 bp of coding sequence (256 aa) for the Drosophilidae species and 729 bp for C. capitata (shorter owing to the different location of the 3Ј-primers L3 and FV, in Fig. 1). Table 2 gives the intron lengths, which are mostly short (between 54 and 83 bp), conforming to typical Drosophila intron sizes (Hawkins 1988;Mount et al. 1992), except for intron 4, which is substantially longer in the three species of the D. melanogaster group and in Ceratitis, and also, but not as long, in the three willistoni group species.
The G+C content of the Gpdh coding sequence is close to 50% in most species, but somewhat higher (55.8-56.7%) in the three melanogaster-group species (Table 3), which have particularly high G+C content in the third codon position (76.1-78.9%). All other species exhibit higher G+C content in the third than in the other two codon positions, but Ceratitis and the two Chymomyza species (as well as D. nebulosa) remain close to 50% in the third sites. Similar patterns have been observed in other Drosophila genes such as Adh (Starmer and Sullivan 1989) and Sod (Kwiatowski et al. 1992b), in which the species of the D. melanogaster group exhibit great G+C excess relative to other Drosophila species, and particularly relative to Chymomyza and Ceratitis (Kwiatowski et al. 1992b). The excess G+C content of Sod is, as in Gpdh, greater in the third than in the other two coding positions and particularly so in the melanogaster-group species (Kwiatowski et al. 1994). Table 3 shows the G+C content for each intron and exon. Drosophila introns have on average 17% more A+T content than adjacent exons (Csank et al. 1990;Mount et al. 1992); in the case of Sod the difference reaches 35% (Kwiatowski et al. 1992b). In the case of Gpdh, the A+T content of intron 3 is not very different from exon 3 (except in Ceratitis) and intron 5 is not very different from exon 6 (except in Chymomyza and Ceratitis), but exon 4 and exon 5 have on average much lower A+T content than their adjacent introns.
The different patterns of G+C content observed in Gpdh in Ceratitis and Chymomyza relative to Drosophila (and Zaprionus) and in the three melanogaster-group species relative to the other Drosophila species is consistent with earlier findings for other genes that codon use and G+C content have a strong species-specific component in fruit flies (Grantham et al. 1980;Kwiatowski et al. 1992b;Sharp et al. 1988;Starmer and Sullivan 1989).  Table 4 gives the number of nucleotide differences between species pairs based on the 729 bp sequenced in all 18 species (lower triangle) as well as the number of differences between the inferred amino acid sequences (upper triangle). We have also calculated genetic distances (not shown) between species following the method of Jukes and Cantor (1969) using the complete data set displayed in Fig. 2. A tree based on the Jukes-Cantor distances obtained by the neighbor-joining (NJ) method (Saitou and Nei 1987) is displayed in Fig. 3. A tree based on Kimura's twoparameter distance (see below) has identical topology and virtually identical bootstrap values. These trees show the well-established monophyly (100% bootstrap values) of each of the melanogaster, obscura, and willistoni species groups. These three groups are classified within the Sophophora subgenus, although the willistoni group diverged from the others shortly after the origination of the Sophophora clade (Kwiatowski et al. 1994). In Fig. 3, the D. willistoni group is less closely related to the melanogaster and obscura clades than to other non-Sophophora species, but this relationship is not statistically valid (28% bootstrap value).

Gpdh Evolution and Phylogeny
The two Chymomyza species and D. lebanonensis are outside the clade that includes the other Drosophila species and Zaprionus (70% bootstrap), consistent with results obtained with other genes and supporting the classification of lebanonensis (and other Scaptodrosophila) as a new genus (Grimaldi 1990;Kwiatowski et al. 1994). D. pictiventris (subgenus Hirtodrosophila) and Zaprionus tuberculatus are more closely related to species of the Drosophila subgenus than to the Sophophora subgenus, consistent with results obtained with the Sod gene (Kwiatowski et al. 1994), although in the present case, the relationships are not statistically robust.
The phylogeny represented in Fig. 3 leaves unresolved important relationships, such as whether Chymomyza and Scaptodrosophila (D. lebanonensis) are sister clades, and the phylogenetic relationships between Zaprionus and the subgenera Drosophila, Dorsilopha, and Hirtodrosophila. These relationships remain unresolved when genetic distances are estimated using a variety of additional methods, such as Kimura's two-parameter distance (Kimura 1980; see Table 5, lower triangle), Tamura and Nei's distance (1993), distances based only on transversions, only on transitions, or both, and when the phylogenies are reconstructed with methods other than NJ, such as maximum parsimony (Fitch 1971) or maximum likelihood (Felsenstein 1981).
The transition/transversion ratios for all interspecies comparisons are 1.7-2.0 between the three species of the melanogaster group, which diverged 5 My ago, and about the same for most other comparisons between Drosophila species, corresponding to divergence times between 40 and 60 My. It is somewhat lower, about 1.5, between Ceratitis and the rest of the species, corresponding to 100 My divergence (Fitch and Ayala 1994). We have used two statistical tests seeking to resolve phylogenetic relationships left unsettled by the bootstrap results: the maximum-parsimony test of Templeton (1983) and the maximum likelihood of Kishino and Hasegawa (1989), which compare phylogenies as wholes with one another. The alternative phylogenies examined are displayed in Fig. 4 and the results obtained with Templeton's method are given in Table 6; Ceratitis is not included since it is unambiguously an outgroup relative to the drosophilids. (The results obtained with the method of Kishino and Hasegawa are similar, except that fewer statistically different phylogenies are detected.) Tree 1 (Fig. 4) requires the fewest steps but, on the basis of the Gpdh data alone, is not statistically better than phylogenies 2-5. If we combine the data for Gpdh and Sod (Kwiatowski et al. 1994), tree 1 is again the best, but statistically no better than phylogenies 2-4. When the data for Adh are added, however, tree 1 is statistically better than any other phylogeny (Table 6; the Adh sequence is not available for D. busckii).
The topology of tree 1 clusters the subgenera Dorsilopha (D. busckii) and Hirtodrosophila (D. pictiventris) with each other and then successively with the Drosophila subgenus and Zaprionus. This phylogeny differs from phylogeny 2, the one proposed by Grimaldi (1990; see also DeSalle and Grimaldi 1991), based on cladistic analysis of morphological information, in that Grimaldi places D. pictiventris outside a cluster consisting of the genera Drosophila (except D. lebanonensis, which is appropriately classified in a different genus, Scaptodrosophila) and Zaprionus; and on this basis, he has raised Hirtodrosophila to the genus category. Moreover, Grimaldi considers the Sophophora subgenus more closely related to the subgenera Drosophila and Dorsilopha than to Zaprionus (see tree 2), whereas tree 1 shows that Sophophora is outside the cluster (((Dorsilopha + Hirtodrosophila) Drosophila) Zaprionus). Based on mitochondrial DNA sequence data, DeSalle (1992) has proposed phylogeny 7, which places Hirtodrosophila outside the cluster of the three genera Drosophila, Zaprionus, and Chymomyza, a hypothesis that is statistically rejected by the Gpdh data alone, or combined with Sod, or with Sod and Adh. Our results are also inconsistent with Throckmorton's (1975) proposal that Chymomyza is associated with the Sophophora radiation, but support his claim that the Sophophora radiation preceded the divergence of Zaprionus and the subgenera Drosophila, Dorsilopha, and Hirtodrosophila.

Rate of Evolution
The number of amino acid and nucleotide differences between species is given in Table 4. Values corrected for superimposed and back substitutions are given in Table  5. It is apparent that the rate of amino acid replacements is not uniform over time. Figure 5 gives the polymorphic amino acid sites and Fig. 6 (top left) displays the number of (corrected) amino acid replacements against time. Average values and times of divergence are given in Table  7. The number of amino acid replacements is zero between species of the same group, in all three species groups: melanogaster, obscura, and willistoni. The divergence time between species of the melanogaster group is 5-8 My, but it is 20-25 My between guanche and the other two obscura species or between nebulosa and the two other willistoni group species. The number of amino acid replacements for most other interspecific comparisons between species of the Drosophila genus is between zero and four, corresponding to divergence  times about 42-55 My. These differences suggest a slow rate of protein evolution. But, as is apparent in Tables 5  and 7 and Fig. 6 (top left), the number of replacements between Drosophila and Chymomyza is about three times as large as between the Drosophila subgenera, although the time elapsed is only slightly greater, 60-65 My. The discordance of evolutionary rates is also notable when comparisons are made between Ceratitis and Drosophila species, which diverged about 100 My ago and differ by seven to ten times more replacements than the Drosophila species. We have drawn in Fig. 6 (top left) three lines that correspond to the rate of amino acid replacement between Drosophila species (2.1 × 10 −10 replacements/site/year), between Chymomyza and Drosophila (5.2 × 10 −10 ), and between Ceratitis and all other species (9.1 × 10 −10 ). (The best-fit regression equation for the comparisons between Drosophila species is 0.004 + 2.75 × 10 −10 replacements/site/year, with a fit of r 2 ‫ס‬ 0.43. These rates are all replacements between species; the replacement rates per lineage will be one-half.) The rate of nucleotide substitutions is, nevertheless, approximately constant over time as can be seen by inspecting Table 5 (lower triangle), where Kimura's twoparameter distances are given, and Fig. 6 (lower left), where Kimura's distances are plotted against time. (The regression equation is 0.015 + 2.71 × 10 −9 substitutions/ site/year; r 2 ‫ס‬ 0.85.) We have used the method of Li (1993) to estimate K a and K s , the number of nonsynonymous (amino acid) and synonymous substitutions, respectively. The results plotted against time are displayed  Fig. 6 (right-hand panels). The rate of amino acid substitutions K a is clearly not constant over time; indeed, it seems to be increasing exponentially over time although this perception emanates primarily from the large K a values between Ceratitis and the other species. The average rate of K a substitutions is, however, more nearly constant over time, although with large variance. (The regression equation is −0.12 + 19.6 × 10 −9 synonymous substitutions/site/year; r 2 ‫ס‬ 0.67.) Throckmorton (1975), based primarily on biogeographical and morphological considerations, proposed that the subgenera Histodrosophila (here represented by D. pic-  tiventris) and Dorsilopha (here, D. busckii) are phylogenetically closely related to each other and to the genus Zaprionus. These taxa would have been part of a radiation that included the Drosophila subgenus. The statistical analysis presented in Fig. 4 and Table 6 supports clustering these taxa but shows Zaprionus as the sister taxon to the cluster of the subgenera Hirtodrosophila, Dorsilopha, and Drosophila (see 1 in Fig. 4), rather than being closer to the first two subgenera than to the Drosophila subgenus. According to Throckmorton (1975), the radiation of the four taxa just mentioned would have occurred after separation of their stem lineage from the subgenus Sophophora (represented in our paper by nine species; three from each of the melanogaster, obscura, and willistoni groups), a claim also supported by our analysis (Fig. 4 and Table 6). In a cladistic revision of the Drosophilidae, Grimaldi (1990) concludes that Sophophora is a sister clade to the doublet made of the Dorsilopha and Drosophila subgenera, whereas Zaprionus is sister to the previous three clades, and Hirtodrosophila is the sister clade of the previous four (see tree 2 in Fig. 4). On the basis of these proposed phylogenetic relationships, he accepts the status of Zaprionus as a genus distinct from Drosophila and raises Hirtodrosophila from subgenus to genus (see also DeSalle 1992;and DeSalle and Grimaldi 1991). Our analysis does not support these claims. If the  (Fig. 4)

Fig. 4. Eight trees showing alternative phylogenetic hypotheses.
Drosophila refers to the subgenus and includes D. hydei and D. virilis; the subgenus Sophophora includes three species of each of the melanogaster and willistoni groups; Chymomyza includes C. amoena and C. procnemis.
subgenus Sophophora is retained within the Drosophila genus, cladistic taxonomy requires that Hirtodrosophila be retained as a subgenus and that Zaprionus be downgraded from genus to subgenus. Throckmorton (1975) proposes that the traditional genus Chymomyza represents an early branch of the So-phophora radiation, a claim contradicted by our analysis, which statistically supports the position of Chymomyza outside the genus Drosophila, as proposed also by Grimaldi (1990). The phylogenetic position of Chymomyza outside the genus Drosophila (as in tree 1, Fig. 4) has been conclusively established by the presence of an extra intron in the Sod gene, which is also present in Scaptodrosophila and Ceratitis, but not in other Drosophila (or Zaprionus) (Kwiatowski et al. 1994). The discovery of this ancestral intron, deleted early in the evolution of the Drosophila genus, corroborates the phylogenetic position of Scaptodrosophila shown in tree 1 (Fig. 4) and thus that this taxon should be raised to the genus category, as done by Grimaldi (1990) and consistent also with Throckmorton's (1975) conclusions.
The pattern of evolution of GPDH is puzzling. There are no amino acid differences among the three members of any of the three Drosophila groups (Table 7), with times of divergence about 8-10 My between D. teissieri and the other two species of the melanogaster group, about 20 My between D. guanche and the other two species of the obscura group, and about 25 My between D. nebulosa and the two other species of the willistoni group. Barrio and Ayala (1997) have studied 253 GPDH amino acids (coded by exons 3-6) in 14 species of the melanogaster (Bewley et al. 1989) and D. virilis (Tominaga et al. 1992). Dots indicate amino acids identical to those on top.

for comparisons between
Drosophila species has been obtained by best-fit regression. The rates between Chymomyza and Drosophila (5.2) and between Ceratitis and the Drosophilids (9.1) are obtained by drawing straight lines from the origin to the relevant points.
affinis and obscura subgroups and observed very few polymorphisms: (1) One species differs from all others at each of two sites and (2) two species differ from all others at a third site (but are identical to each other, although one belongs to the affinis and the other to the obscura subgroup). This corresponds to an average of about one-half amino acid difference between any two species, or 0.002 per site, for species that diverged 20-25 My ago. These results are consistent with previous observations that GPDH is a slowly evolving enzyme (Bewley et al. 1989;Wells 1995Wells , 1996a. Comparisons among all Drosophila (including Zaprionus) species also manifest a slow rate of amino acid replacement. The number of amino acid differences between species from different groups of the Sophophora subgenus are two (corresponding to about 90 My of separate evolution, 45 My on each lineage) and between species from different subgenera are between zero and five (corresponding to about 110 My of separate evolution). This is a lineage rate of about 1 × 10 −10 replacements/site/year, comparable to the rates of the very slowly evolving histones; e.g., 1.7 × 10 −10 for H2A or H2B (Wilson et al. 1977).
The apparent rate of GPDH evolution increases, however, when Chymomyza is compared with Drosophila. The average number of amino acid differences is about eight (Tables 5 and 7), corresponding to 120 My of evolution, or a rate of 2.6 × 10 −10 replacements/site/year. When Ceratitis is compared with Drosophila or Chymomyza, the average number of replacements is about 23 (Table 4), corresponding to 200 My of evolution, or a rate of 4.6 × 10 −10 replacements/site/year, still slower than the rate of 6.7 × 10 −10 of cytochrome c (Wilson et al. 1977), a slowly evolving protein.
What is going on? One could argue that the rate of evolution in the Chymomyza lineage, as well as between nodes 3 and 4 in Fig. 7, is faster than within the genus Drosophila. But it would have to be much faster in order to account for the 2.6 times increase in the overall number of amino acid differences. If we assume 120 My of separate evolution between any Chymomyza and Drosophila species, 55 of the 120 My belong to the evolution of the Drosophila lineages; we are left with 65 My of evolution (55 My of the Chymomyza lineage, plus 10 My of the Drosophila lineage before the radiation of the genus, i.e., between nodes 3 and 4; see Fig. 7). But we know that the Chymomyza lineages have also been evolving slowly. The two Chymomyza species diverged about 42 My ago (Kwiatowski et al. 1994) but differ only by two amino acids (compared to about three between the somewhat older Drosophila subgenera) (Table 4). Thus, the time-length during which evolution would have happened faster is at a maximum 120 − 55 − 42 ‫ס‬ 23 My, or one-sixth of the time of separate evolution between any Chymomyza species and any Drosophila species. Thus, in order to account for the 2.6 times increase in amino acid differences between the two genera, we have to conclude that during those 23 My (represented by thicker lines in Fig. 7), the rate of GPDH was nine times faster than for the other 97 My of separate evolution between Chymomyza and Drosophila species.
This conclusion can be somewhat exacerbated by noticing that the average distance between Scaptodro-  Powell and DeSalle 1995, and references therein;Kwiatowski et al. 1994;Takezaki et al. 1995;Russo et al. 1995). Beverley and Wilson (1984) give somewhat larger estimates for the divergence between families (123 My vs our 120 ± 20 My) and between the subgenera Sophophora and Drosophila (62 My vs our 55 ± 10 My) sophila and Drosophila is about four amino acids, only slightly greater than the average between the Drosophila subgenera. Thus between nodes 2 and 4 (Fig. 7) the prevailing rate of evolution must have been the same as in the Drosophila genus. Therefore, the accelerated rate of Chymomyza evolution would have been restricted to the branch segment between nodes 3 and 5, when it would have been 12 times greater than at any other time in the evolution of Drosophila and Chymomyza. We noted earlier that the apparent rate of amino acid replacements between Ceratitis and Drosophila is about four times greater than within the Drosophila genus. Using the same logic as in the previous paragraph, the acceleration could have only occurred (Fig. 7) between nodes 1 and 2 and from node 1 to Ceratitis, corresponding to 140 My out of the total 200 My separating the two genera. The rate of evolution corresponding to those branches would need to be about six times faster than in the rest of the tree (excluding Chymomyza).
It may be noted that the disparate rates of GPDH evolution observed are not strongly dependent on the particular times of divergence assumed. We have noted in the Results (see also Fig. 4 and Table 6) that the topology displayed in Fig. 7 is statistically superior to other alternatives. This topology shows that Chymomyza is more closely related than Scaptodrosophila to the Drosophila species. Yet the mean number of amino acid replacements (calculated from Table 5) is much greater between Chymomyza and Drosophila (33.7 ± 0.8) than between Scaptodrosophila and Drosophila (17.9 ± 0.7). It is, in any case, now well established that the Scaptodrosophila and Chymomyza lineages diverged from the Drosophila lineages within a relatively short time interval. Changing their branching sequence would not by itself account for the large discrepancy in the GPDH differences. The same point can be made with respect to Ceratitis. The time of divergence between Ceratitis and Drosophila assumed in Table 7 is 100 ± 20 My, not quite double the time of divergence between different Drosophila subgenera (55 ± 10); yet the number of amino acid replacements is 92.0 ± 0.8, nearly six times as large. This discrepancy cannot be accounted for even if the divergence between the two families were as old as 150 My or the divergence between the subgenera as recent as 40 My, values which are outside those determined by various methods (see Powell and DeSalle 1995 and other references cited in Table 7).
The point made in the previous paragraph can also be made by reference to Fig. 6. The comparisons made and the times of divergence assumed are the same in all panels. Yet the nucleotide genetic distances (K2) as well as the number of synonymous substitutions (K s ) are approximately linear with respect to time (bottom panels), whereas the number of amino acid replacements (PAM) or of nonsynonymous substitutions (K a ) is not (top panels). Wells (1996b) has elucidated one idiosyncrasy of GPDH evolution: reversal and parallelism. He obtained the complete GPDH sequence of 13 Drosophila species and observed amino acid polymorphism at 17 of 363 sites, at nine of which two or more species were different from the rest. At four of these nine sites, parallel or reversed replacements have occurred. Wells considers these four sites hypervariable because they account for half of all interspecific amino acid replacements, even though only two different amino acids occur at these four sites in the 13 species. The degree of homoplasy, as measured by the global consistency index (Archie 1989), is significantly greater in Drosophila GPDH evolution than the average in ten additional proteins, one (ADH) from Drosophila and nine from vertebrates (Wells 1996b).
If we exclude Ceratitis and Chymomyza, we have observed eight polymorphic amino acid sites in 15 species of Drosophila (including Zaprionus) (Fig. 5). At three sites, only one species differs from all others; these sites are uninformative for the present purposes. Four of the remaining five sites exhibit parallel or reversed replacements, namely, sites 193, 195, 314, and 336. By reference to the phylogeny in Fig. 3, the minimum number of independent amino acid replacements would be: 1. Three at site 193, from ancestral E → D at the root of the genus and two parallel reversals D → E at the root of the willistoni group and of the Drosophila subgenus (virilis and hydei); the next most parsimonious alternative would be to assume that D is ancestral and that four D → E parallel replacements have occurred in Ceratitis, Chymomyza, the willistoni group, and the Drosophila subgenus. The first alternative, in addition to requiring only three rather than four replacements, is also favored when one notes that the muscoid fly Calliphora has E at this site (our unpublished data), similarly as Chymomyza and Ceratitis. 2. Three at site 195, all parallel from ancestral S → A in the Sophophora subgenus (which we assume to be monophyletic, comprising the willistoni, obscura, and melanogaster groups), in the Drosophila subgenus, and in pictiventris. (Calliphora, like Chymomyza and Ceratitis, has S at this site.) An equally parsimonious alternative is to assume ancestral S → A in the root of the Drosophila genus (including lebanonensis but not Chymomyza) and two parallel reversals A → S in D. busckii and Zaprionus; however, if we assume that the phylogenetic positions of Zaprionus and pictiventris are reversed in Fig. 3, only two S → A parallel replacements are required. 3. Two parallel replacements at site 314, ancestral G → K in the obscura and willistoni groups; alternatively, ancestral G → K in the root of Sophophora and one reversal K → G in the melanogaster group. 4. At 336 three parallel and one reverse replacement: ancestral K → N in the melanogaster and willistoni groups and in busckii + pictiventris (assuming these two are monophyletic) and N → K between the Fast and Slow alleles of D. melanogaster (the Slow allele is known to be more recent, Takano et al. 1993;Wells, 1996b); alternatively, ancestral K → N in the root of the Drosophila genus, and three parallel reversals N → K in the obscura group, the Drosophila subgenus, and D. melanogaster Slow.
Of the four homoplasious sites detected in our phylogeny, only site 336 is manifest in Wells's (1996) phylogeny, which only includes Drosophila species. The other homoplasious sites (28, 45, and 362) detected by Wells are not included within the Gpdh fragment sequenced by us. Wells (1996b) has pointed out that the high level of homoplasy makes GPDH unsuitable for reconstructing the phylogeny of Drosophila species. An instance of misleading similarity occurs between the melanogaster group species and D. pictiventris, which show no amino acid differences (Table 3), although they are phylogenetically remote. GPDH is unsuitable for reconstructing the phylogeny of Drosophila because of the slow rate of amino acid replacements within the genus. Moreover, as already noted, the detectable rate of amino acid replacements is very erratic within the set of fruitfly species surveyed in our study, which makes GPDH unsuitable for estimating times of divergence or making any other inferences dependent on the assumption of a molecular clock.
The evolution of Gpdh, however, appears to be much more nearly uniform through time at the nucleotide level than at the amino acid level. When the nucleotide distances between species are calculated using Kimura's (1980) two-parameter method (Table 5, below diagonal,  and Table 7), they approximate a uniform rate of evolution, as shown in Fig. 6 (lower left panel). This apparent regularity results, no doubt, from the confounding of two different rates corresponding to two sets of nucleotides. The rate of nonsynonymous nucleotide substitutions (K a ), calculated by the method of Li (1993), is clearly not a linear function of time, as shown in Fig. 6 (top right  panel), whereas the rate of synonymous substitutions (K s ) is more nearly linear (Fig. 6, bottom right), although much variation is apparent.