BRIEF COMMUNICATIONS
Evolution, 56(1), 2002, pp. 191–198
MOLECULAR CORRELATES OF REPRODUCTIVE ISOLATION
BENJAMIN M. FITZPATRICK
Population Biology Graduate Group, Section of Evolution and Ecology, University of California, Davis, California 95616
E-mail: benfitz@ucdavis.edu
Abstract. Evolution of reproductive isolation as a byproduct of genetic divergence in isolated populations is the
dominant (albeit not exclusive) mode of speciation in sexual animals. But little is known about the factors linking
speciation to general divergence. Several authors have argued that allopatric speciation should proceed more rapidly
if isolated populations also experience divergent selection. Reproductive isolation between allopatric populations is
not subject to direct selection; it can accumulate only by random drift or as a fortuitous byproduct of selection on
other traits. Here I present a novel analysis of published data, demonstrating that pre- and postmating isolation of
Drosophila species are more tightly correlated with allozyme divergence than with silent DNA divergence. Inasmuch
as proteins are more subject to the action of natural selection than are silent DNA polymorphisms, this result provides
broad support for a model of selection-mediated allopatric speciation.
Key words.
Allozymes, DNA hybridization, Drosophila, genetic distance, speciation, synonymous substitutions.
Received November 21, 2000.
Several authors have demonstrated that reproductive isolation is correlated with genetic divergence between species
(e.g., Ayala et al. 1974; Parker et al. 1985; Coyne and Orr
1989, 1997; Tilley et al. 1990; Knowlton et al. 1993; Gleason
and Ritchie 1998; Sasa et al. 1998). Such a correlation is to
be expected as both kinds of divergence should accumulate
over time (Muller 1942). Indeed, these studies have generally
conceived of molecular genetic divergence as a proxy for
divergence time and interpreted their results as demonstrating
gradual evolution of reproductive isolation. But we are well
aware that although some classes of molecular markers may
evolve according to a stochastic clock model, others are more
subject to natural selection and may evolve episodically at
different rates in different lineages (Gillespie 1991). Ideally,
we would like to control for time and see if there is any
residual correlation between reproductive isolation and certain kinds of molecular divergence. A residual correlation
could arise if both kinds of divergence are accelerated by
positive selection and/or retarded by stabilizing selection.
That is, episodes of adaptive divergence or population bottlenecks could have concordant influences on nonneutral molecular divergence and reproductive isolation.
Conventional wisdom has it that mutations causing changes in amino acid sequences of translated products are far more
‘‘visible’’ to selection than synonymous mutations that produce no change at the protein sequence level (Gillespie 1991;
Futuyma 1998). Thus, synonymous divergence is expected
to be more clock-like than amino acid divergence. Even the
architects of the neutral theory of molecular evolution agreed
that amino acid substitutions are more likely to be caused by
selection than silent substitutions (Ohta 1972; Kimura 1983).
This difference has been supported by several empirical analyses (Gillespie 1991; Golding 1994). DNA sequence analyses
provide some of the best evidence that amino acid substitutions are more sensitive to selection than are silent substitutions (McDonald and Kreitman 1991; Aguade et al. 1992;
Eanes et al. 1993). On the other hand, silent sequence variation and DNA hybridization divergence (which includes
Accepted September 4, 2001.
noncoding sequence divergence) appear to conform more
closely to predictions of the neutral theory (Gillespie 1991;
Ohta 1995; but see Zeng et al. 1998).
In this paper, I use DNA divergence from a variety of
sources as a proxy for time and divergence in electrophoretically detectable enzyme allele (allozyme) frequencies as
a proxy for divergence that could be affected by selection.
Once DNA divergence is taken into account, I evaluate the
residual correlation between reproductive isolation and allozyme divergence for Drosophila species-pairs. Some authors have cautioned against conflating species differences
with speciation (Templeton 1981; Wu and Hollocher 1998).
Reproductive isolating mechanisms are a particularly important and interesting class of species differences with a
significant and unique relationship to speciation. Differences
that cause reproductive isolation cause speciation (species
concepts differ only how they allow other kinds of differences
to determine species status). Following Wu and Hollocher
(1998) and Coyne and Orr (1998), I am satisfied that, by
learning about the evolution of reproductive isolation, I am
learning something of general importance to processes of
speciation.
The notion that natural selection can accelerate allopatric
speciation was adumbrated during the modern synthesis
(Dobzhansky 1937; Mayr 1942; Muller 1942), and has been
made more explicit by recent authors (Rice and Hostert 1993;
Rosenzweig 1995; Orr and Smith 1998; Schluter 1998). A
model of selection-mediated geographic speciation enjoys
some empirical support (Rice and Hostert 1993; Funk 1998),
but its theoretical basis has rarely been discussed. Mayr
(1970) asserted that most ecological adaptation involves several traits, most traits are polygenic, and most genes are pleiotropic. Therefore any response to selection may reverberate
through large portions of the genome and affect some loci
with pleiotropic effects on reproductive isolation. That is, the
footprint of divergent selection may be found in a large number of genes and in the degree of intrinsic reproductive isolation between taxa. By the same token, purifying selection
191
q 2002 The Society for the Study of Evolution. All rights reserved.
192
BRIEF COMMUNICATIONS
maintaining function in complex metabolic and developmental pathways ought to work against the accumulation of
intrinsic postzygotic isolation. These propositions, rooted in
the idea that organisms are composed of coadapted gene complexes (Dobzhansky 1937), lead to the prediction that the rate
of evolution of reproductive isolation should be more tightly
correlated with functional protein divergence than silent
DNA divergence. In addition to pleiotropy, hitchhiking due
to physical linkage and linkage disequilibria will contribute
to concordance in evolutionary rates (Rice and Hostert 1993).
Pleiotropic premating isolation is an obvious consequence
of adaptive changes in reproductive phenology and habitat
choice (Rice 1984; Rice and Salt 1988, 1990; Feder 1998;
Funk 1998. Pleiotropic postzygotic isolation will automatically result if ecologically intermediate hybrids find themselves unable to make a living (Rice and Hostert 1993; Hatfield and Schluter 1999). But Mayr’s argument implies that
the influence of pleiotropy/hitchhiking may be more subtle
and far reaching. It suggests that any functional genetic
change increases the chance of substituting an allele that
contributes to reproductive incompatibility. Further, if isolated populations are selected to remain in their mutual ancestral state (as in deleterious mutation theories), they should
retain reproductive compatibility longer than if they were
allowed to drift apart at random. This scenario is clearly
distinguished from parallel selection, which may contribute
to genetic and reproductive divergence if isolated populations
fix different changes in response to similar selection (Muller
1942; Cohan and Hoffman 1989).
Laboratory assays of reproductive isolation in Drosophila
have minimized ecological factors that could contribute to
environment-dependent reproductive isolation. Yet, I demonstrate that reproductive isolation of Drosophila species in
the laboratory is correlated with allozyme divergence even
after conditioning on divergence time, as estimated by silent
DNA sequence divergence. I argue that the residual variation
in allozyme divergence is best explained by natural selection,
and the residual correlation of reproductive isolation with
allozyme divergence is due to pleiotropy/hitchhiking.
METHODS
My central analysis is to test for correlation between allozyme divergence and reproductive isolation after factoring
out the dependence of both quantities on divergence time, as
estimated by silent and non-coding DNA divergence. This is
done by regressing both allozyme divergence and a measure
of reproductive isolation on silent DNA divergence, and then
testing for an association between the resulting residuals of
isolation and the residuals of allozyme divergence. All of the
data on reproductive isolation and allozyme divergence, DNei
(Nei 1987), used in this study were compiled from the literature by Coyne and Orr (1989, 1997). They calculated an
index of premating isolation as 1 2 (# heterospecific matings)/(# homospecific matings). Postmating isolation was
scored more coarsely. Potential offspring of interspecific matings were broken into four classes—male and female offspring from each of the two reciprocal crosses. The index
takes on values of 0, 1/4, 1/2, 3/4, or 1 when 0, 1, 2, 3, or
all 4 of the categories of F1 hybrids are completely inviable/
FIG. 1. Hypothetical phylogenetic tree used to illustrate phylogenetic correction method. Independent distances are AB, EF, (CA
1 CB)/2, and {[(EA 1 EB)/2 1 EC]/2 1 [(FA 1 FB)/2 1 FC]/
2}/2.
sterile. Both indices of reproductive isolation top out at one
(complete isolation), and therefore they are not linearly related to the actual number of genic incompatibilities. The
number of incompatibilities should ‘‘snowball’’ faster than
linearly with time (Orr 1995; Menotti-Raymond et al. 1997;
Orr and Turelli 2001), but no such expectation is warranted
for these simple indices of isolation.
Allozyme studies were typically based on 20–40 loci. DNA
hybridization divergences were reported as patristic distances
by Powell and DeSalle (1995; Powell 1997). All data were
checked against the original sources whenever possible. DNA
data from the mitochondrial cytochrome oxidase subunit I
(COI) and the nuclear alcohol dehydrogenase (Adh) and glycerol-3-phosphate dehydrogenase (Gpdh) were obtained from
Genbank. Exons from these loci were aligned in Clustal X
and adjusted by eye. Synonymous sequence divergence was
estimated using Nei and Gojobori’s (1986) method, as implemented in the computer program MEGA2.
Phylogenetically independent contrasts (PIC’s) for each
measure were calculated by Coyne and Orr’s (1989) modification of Felsenstein’s (1985) procedure (Fig. 1) using published phylogenies (MacIntyre and Collier 1986; Powell and
DeSalle 1995; Powell 1997). Some have advocated Mantel’s
matrix permutation test for these kinds of data. But Mantel’s
test only accounts for dependence due to redundancy and
ignores dependence due to common ancestry (Legendre et al.
1995). That is, Mantel’s test accounts for dependence between the distances AE and AF but not for the dependence
between AE and BF (Fig. 1). The PIC procedure accounts
for common ancestry by reducing all pairwise comparisons
across a node to a single comparison. Redundancy (e.g., taxon
A contributes to three of the four comparisons in Fig. 1)
BRIEF COMMUNICATIONS
193
FIG. 2. Relationships between allozyme divergence (DNei) and four DNA datasets. Vertical lines illustrate the residuals, which were
saved for further analysis.
should not be a problem if it is safe to assume that withinclade divergence is independent of between-clade divergence.
Further, properly nested averaging makes the method somewhat robust to violation of this assumption (Omland 1997).
The rate of evolution of premating isolation is considerably
accelerated by reproductive character displacement between
sympatric species, but no difference in the pattern of postmating isolation in sympatric versus allopatric taxa is expected or observed (Coyne and Orr 1989). Therefore, when
analyzing premating isolation, geographical relationships
must be taken into account. For the two largest datasets, Adh
and Gpdh, all sympatric comparisons were simply removed
prior to analysis (i.e., comparisons between sympatric species
pairs were not included when computing PICs—this may not
completely account for historical sympatry between clades).
This practice resulted in unacceptably small sample sizes for
the DNA hybridization and COI datasets, so I did not consider
prezygotic isolation in the analysis of these datasets. Preliminary analysis of prezygotic isolation without accounting for
geography gave results qualitatively similar to postzygotic
isolation (not shown).
For each of the phylogenetically corrected datasets (DNA
hybridization, Adh, Gpdh, and COI), DNei and reproductive
isolation were regressed on synonymous DNA divergence
and the residuals from these regressions were saved as linear
transformations of the dependent variables (Figs. 2 and 3).
In order to assure linear relationships and better fit the nor-
mality assumption of regression analysis, logarithmic and
arcsine transformations were employed as necessary (see axis
labels in Figs. 2–4). This purely phenomenological approach
to curve fitting is intended to avoid systematic biases that
may be introduced by assuming particular models of evolution. Relationships between residual DNei and residual reproductive isolation were assessed by Spearman rank correlation. Analysis of residuals is an excellent approach when
there is a priori knowledge of a confounding third variable
(Afifi and Clark 1996). Bivariate regressions of reproductive
isolation on both DNei and DNA divergence were used as
additional descriptive analyses.
I used rate heterogeneity comparisons to test whether silent
sequence divergence and DNei differed in performance as molecular clocks. Relative rates tests (Li 1997) on individual
datasets failed to reject a molecular clock null hypothesis
(data not shown). Nevertheless, one type of genetic distance
may be systematically more clocklike. I constructed unrooted
phylograms for independent species pairs with outgroups (no
species was included in more than one comparison). Let x
and y denote the genetic distances between the two in-group
species and their most recent common ancestor. A standardized measure of branch length heterogeneity was calculated
for both DNA and allozyme divergence: B 5 (x 2 y)/(x 1
y). Statistical properties of this index are discussed in the
Appendix. Heterogeneity indices for DNA and allozymes
were compared using Wilcoxon signed-rank tests for paired
194
BRIEF COMMUNICATIONS
FIG. 3. Relationships between postzygotic isolation and four DNA datasets. Vertical lines illustrate the residuals, which were saved for
further analysis.
data (Zar 1984), with each independent phylogram yielding
one pair of Bs. The DNA hybridization, Adh, and COI datasets
yielded 9, 11, and 9 independent phylograms, respectively.
If allozymes are generally neutral, population bottlenecks
can inflate DNei by increasing homozygosity (Chakraborty
and Nei 1977). Bottlenecks might also increase the rate of
evolution of reproductive isolation (Mayr 1970; Templeton
1981; but see Barton and Charlesworth 1984, Turelli et al.
2001). If an association between population size and homozygosity h as significantly influenced patterns of allozyme
differentiation, then relatively long branches should, on average, be associated with relatively low heterozygosities. I
used 26 unrooted phylograms, as above, to test for a correspondence between allozyme-based branch-length variation
and heterozygosity variation. That is the quantities (x 2 y)
for DNei and H should usually have opposite signs if population bottlenecks have been important in accelerating allozyme divergence. This was evaluated with a sign test for
deviation from the null expectation of having the same sign
half of the time. I also evaluated the correlation between DNei
and average log homozygosity. This follows from the definition of DNei (Nei 1987): DNei 5 2ln Jxy 1 (ln Jx 1 ln Jy)/
2, where Jxy is the probability of identity of alleles between
taxa x and y, and Jx and Jy are the within-taxon probabilities
of identity (homozygosities).
RESULTS
AND
DISCUSSION
Allozyme divergence was positively correlated with all
DNA-based divergences (Fig. 2). Reproductive isolation was
positively correlated with DNA hybridization and synonymous
Adh divergence but not significantly correlated with synonymous COI or Gpdh divergence (Figs. 3 and 4). Where analyzed, results for premating isolation were concordant with
those for postmating isolation (Tables 1 and 2, Figs. 3–5). In
all datasets, reproductive isolation was more strongly correlated with allozyme divergence than DNA divergence (Table
1). Knowlton et al. (1993) reported a similar result from species pairs of Alpheus shrimp. After regression on DNA divergence, residual reproductive isolation was significantly positively associated with residual allozyme divergence (Table 2,
Figs. 4 and 5). Conversely, after regression on allozyme distance, residual isolation was not significantly associated with
residual DNA divergence (Spearman and Pearson correlation,
all P-values . 0.05, not shown). Both kinds of genetic distance
are statistically adequate molecular clocks (relative rates tests,
not shown) and both are correlated, via a mutual dependence
on time, with degree of reproductive isolation. But there appears to be an additional source of association between allozyme divergence and reproductive isolation that is not shared
by silent DNA divergence.
195
BRIEF COMMUNICATIONS
FIG. 4. The top two panels present relationships between premating isolation and synonymous DNA divergence from Adh and Gpdh.
Vertical lines illustrate the residuals, which are plotted against residual allozyme divergence in the lower two panels.
DNA hybridization incorporates information from all single copy and some repetitive DNA in the genome (Powell
1997). According to Miklos and Rubin (1996), there is about
90Mb of single copy DNA in the Drosophila genome, comprising about 55% of the total genome. Exons comprise about
24.1Mb (Adams et al. 2000). According to data summarized
in Li (1997) an average of about 12% of substitutions in
codons actually change amino acids (based on comparisons
between the D. melanogaster and D. obscura groups). As-
suming (conservatively) that substitution rates in introns and
intergenic spacers are the same as for synonymous sites, we
can calculate an approximate proportion of single copy DNA
differences that are nonsynonymous as 0.12 3 (24.1 Mb 4
90 Mb) ø 0.032.
Thus, DNA hybridization represents the broadest sample
of the genome, and incorporates predominantly noncoding
differences. As such, the results from DNA hybridization are
probably the most compelling. Results from single sequences
TABLE 1. Bivariate regressions of reproductive isolation on DNA- and allozyme-based genetic distances. Coefficients are standardized so that
their relative magnitudes directly reflect their relative ‘‘strength’’ in the regression model. There is variation among datasets in the observed
association between allozyme divergence (DNei) and reproductive isolation because no two datasets included exactly the same pairs of taxa.
Allozyme, DNA hybridization, and Adh distances were ln transformed.
Dataset
DNA hybrid
Adh
COI
Gpdh
Independent
variable
Post
Post
Preb
Post
Post
Preb
DNAa coefficient
0.0541
0.1175
0.4840*
20.5795
20.4193
20.0192
Allozyme coefficient
0.8502**
0.7604*
0.5085*
1.1545**
0.8107*
0.3740
r2
F
P
0.7862
0.7260
0.7953
0.6019
0.3753
0.1286
18.3847
13.2506
21.3757
6.8040
3.9053
1.1072
0.000447
0.001543
0.000162
0.01585
0.04697
0.3560
All DNA divergence, except for DNA hybridization, is based solely on synonymous sequence differences (Nei and Gojobori 1986).
Prezygotic isolation was analyzed using only allopatric species pairs.
Significance of individual coefficients (t-test): * P , 0.05; ** P , 0.01.
a
b
196
BRIEF COMMUNICATIONS
TABLE 2. Results of Spearman rank correlation analyses corresponding to Figures 2–5. Variation among datasets is largely attributable to the
fact that each includes a slightly different subset of the pairs of taxa in Coyne and Orr (1989, 1997). ‘‘Post’’ and ‘‘Pre’’ refer to post- and
premating reproductive isolation, respectively.
Dataset
DNA hybridization
Adh
Gpdh
COI
Variables
Residual
Residual
Residual
Residual
Residual
Residual
Post vs. Residual DNei
Post vs. Residual DNei
Pre vs. Residual DNei
Post vs. Residual DNei
Pre vs. Residual DNei
Post vs. Residual DNei
might be attributed to the fact that DNei is based on a large
sample of loci and, therefore, might provide a better molecular clock than single loci. To verify that allozymes are not
more clocklike than DNA, I used rate heterogeneity comparisons (see Methods) to test branch length variation in phylograms constructed from the different molecular datasets.
Allozyme-based branch lengths were more heterogeneous
than DNA-based branch lengths in the majority of comparisons, although the difference was statistically significant
only in the DNA hybridization dataset (Wilcoxon signedrank tests, Table 3).
My interpretation of the central analysis of the paper (Table
rs
P
0.7253
0.7253
0.6879
0.5912
0.2143
0.8322
0.005023
0.005023
0.006540
0.1588
0.6103
0.000785
2, Figs. 4 and 5) relies on the important differences between
synonymous DNA and allozyme divergence discussed previously. Silent DNA sequence divergence should behave much
more like a stochastic clock than allozyme divergence. Allozyme divergence (as measured by DNei) is likely to be significantly affected by selection and, possibly, bottleneck effects
(which perturb the interaction between stabilizing selection and
drift). Regression of DNei on synonymous DNA divergence is,
coarsely, regression of allozyme divergence on time. The residual variation in allozyme divergence not accounted for in
such a regression should be due to selection and/or changes in
effective population size (e.g., founder effects). Similarly, re-
FIG. 5. Residual relationships between postmating isolation and allozyme divergence for four DNA datasets; lines were fitted by least
squares.
197
BRIEF COMMUNICATIONS
TABLE 3. Results of Wilcoxon signed-rank tests for paired branchlength heterogeneity indices estimated from independent three-taxon
phylograms (see text). BD and BDNA are branch length heterogeneities
based on allozymes (DNei) and DNA, respectively.
Dataset
DNA hybridization
Adh
Gpdh
COI
Number
Number
with
with
Rank sum BD . BDNA BD , BDNA
41
33
9
30
7
8
2
6
2
3
5
4
Exact
probability
0.027
0.175
0.469
0.846
sidual variation in reproductive isolation after regression on
DNA divergence should reflect these same factors.
My analyses provide no evidence that variance in heterozygosity has had a systematic effect on variance in DNei. Taxa
with longer branches had lower H in only 10 of 26 comparisons (sign-test P 5 0.3269) and DNei was not significantly
correlated with average log homozygosity (Spearman rank
correlation 5 0.0182, P 5 0.960). These results suggest that
bottleneck effects on neutral allozyme variation do not provide a satisfactory explanation for variance in DNei.
However, there are other ways that variation in effective
population size could influence DNei. If most electrophoretically detectable variation evolves according to deleterious
mutation theories, small population size could increase the
rate of divergence without substantially affecting heterozygosity (Ohta 1972, 1992; Gillespie 1999). By contrast, if
isolated populations are continually substituting different advantageous alleles, divergence will be faster in large populations, but heterozygosity will still be insensitive to population size (Gillespie 1999). In neutral, overdominant, and
fluctuating selection models, substitution rates are expected
to be fairly constant while heterozygosity increases with population size (Gillespie 1999).
Significant residual correlation between allozyme divergence and reproductive isolation after factoring out divergence time (as estimated by synonymous DNA) is consistent
with the venerable proposition that ‘‘positive selection’’ can
accelerate speciation via pleiotropy/hitchhiking. It is equally
consistent with deleterious mutation theories, which suggest
that ‘‘negative selection’’ would tend to retard divergence.
Both models predict ‘‘correlated progression’’ among different parts of the genome over long spans of time (Bell
2000). My results should not be taken as evidence against
alternative models that invoke fluctuating selection or idiosyncratic variation in modes of evolution among different
loci. They simply suggest that correlated progression—beyond what is expected due to time—is common enough to
leave a statistical footprint in comparative allozyme and reproductive isolation data from Drosophila.
This is not merely another confirmation of the fact that
genetic divergence is associated with speciation. Rather, my
analysis shows that one kind of genetic divergence (electrophoretically detectable enzyme differences) is more tightly
correlated with reproductive isolation than is another (silent
DNA differences). This is all the more noteworthy given that
DNA sequences are now the preferred source of data for
addressing questions about taxonomic status, species boundaries, and modes of speciation. The result is indicative of a
pervasive role of natural selection in influencing the rate of
evolution of reproductive isolation.
ACKNOWLEDGMENTS
I would like to thank the participants in the evolution discussion group at the University of California, Davis for constructive criticism. M. Turelli, H. B. Shaffer, J. H. Gillespie,
D. J. Begun, and two anonymous reviewers provided excellent advice and editorial comments. My research is supported
by an NSF Grant (DEB 0089716) to M. Turelli and an EPA
STAR Graduate Fellowship (U-91572401).
LITERATURE CITED
Adams, M. D., S. E. Celniker, R. A. Holt, C. A. Evans, J. D.
Grocayne et al. 2000. The genome sequence of Drosophila melanogaster. Science 287:2185–2195.
Afifi, A. A., and V. Clark. 1996. Computer-aided multivariate analysis. Chapman and Hall, London.
Aguade, M., N. Miyashita, and C. H. Langley. 1992. Polymorphism
and divergence in the Mst26A male accessory gland gene region
in Drosophila. Genetics 132:755–770.
Ayala, F. J., M. L. Tracey, D. Hedgecock, and R. C. Richmond.
1974. Genetic differentiation during the speciation process in
Drosophila. Evolution 28:576–592.
Barton, N. H. and B. Charlesworth. 1984. Genetic revolutions,
founder effects, and speciation. Annu. Rev. Ecol. Syst. 15:
133–164.
Bell, M. A. 2000. Bridging the gap between population biology and
paleobiology. Evolution 54:1457–1461.
Chakraborty, R., and M. Nei. 1977. Bottleneck effects on average
heterozygosity and genetic distance with the stepwise mutation
model. Evolution 31:347–356.
Cohan, F. M., and A. A. Hoffman. 1989. Uniform selection as a
diversifying force in evolution: evidence from Drosophila. Am.
Nat. 134:613–637.
Coyne, J. A., and H. A. Orr. 1989. Patterns of speciation in Drosophila. Evolution 43:362–381.
———. 1997. Patterns of speciation in Drosophila revisited. Evolution 51:295–303.
———. 1998. The evolutionary genetics of speciation. Philos.
Trans. R. Soc. Lond. B 353:287–305.
Dobzhansky, T. 1937. Genetics and the origin of species. Columbia
Univ. Press, New York.
Eanes, W. F., M. Kirchner, and J. Yoon. 1993. Evidence for adaptive
evolution of the G6pd gene in the Drosophila melanogaster and
Drosophila simulans lineages. Proc. Natl. Acad. Sci. USA 90:
7475–7479.
Feder, J. L. 1998. The apple maggot fly, Rhagoletis pomonella flies
in the face of conventional wisdom about speciation? Pp. 130–
144 in D. J. Howard and S. H. Berlocher, eds. Endless forms:
species and speciation. Oxford Univ. Press, New York.
Felsenstein, J. 1985. Phylogenies and the comparative method. Am.
Nat. 125:1–15.
Funk, D. J. 1998. Isolating a role for natural selection in speciation:
host adaptation and sexual isolation in Neochlamsis bebbianae
leaf beetles. Evolution 52:1744–1759.
Futuyma, D. J. 1998. Evolutionary biology. Sinauer Associates,
Sunderland, MA.
Gillespie, J. H. 1991. The causes of molecular evolution. Oxford
Univ. Press, Oxford, U.K.
———. 1999. The role of population size in molecular evolution.
Theor. Popul. Biol. 55:145–156.
Gleason, J. M., and M. G. Ritchie. 1998. Evolution of courtship
song and reproductive isolation in the Drosophila willistoni species complex: do sexual signals diverge the most quickly? Evolution 52:1493–1500.
Golding, B. 1994. Non-neutral evolution. Chapman and Hall, New
York.
Hatfield, T., and D. Schluter. 1999. Ecological speciation in sticklebacks: environment-dependent hybrid fitness. Evolution 53:
866–873.
198
BRIEF COMMUNICATIONS
Kimura, M. 1983. The neutral allele theory of molecular evolution.
Cambridge Univ. Press, Cambridge, U.K.
Knowlton, N., L. A. Weight, L. A. Solorzano, D. K. Mills, and E.
B. Bermingham. 1993. Divergence in proteins, mitochondrial
DNA, and reproductive compatibility across the Isthmus of Panama. Science 260:1629–1632.
Legendre, P., F.-J. Lapointe, and P. Casgrain. 1995. Modeling brain
evolution from behavior: a permutational regression approach.
Evolution 48:1487–1499.
Li, W-H. 1997. Molecular evolution. Sinauer Associates, Sunderland, MA.
Lynch, M., and B. Walsh. 1998. Genetics and the analysis of quantitative traits. Sinauer Associates, Sunderland, MA.
MacIntyre, R. J., and G. E. Collier. 1986. Protein evolution in the
genus Drosophila. Pp. 39–146 in M. Ashburner, H. L. Carson
and J. N. J. Thompson, eds. The genetics and biology of Drosophila. Vol. 3e. Harcourt Brace Jovanovich, London.
Mayr, E. 1942. Systematics and the origin of species from the
viewpoint of a zoologist. Columbia Univ. Press, New York.
———. 1970. Populations, species, and evolution. Harvard Univ.
Press, Cambridge, MA.
McDonald, J. H., and M. Kreitman. 1991. Adaptive protein evolution at the Adh locus in Drosophila. Nature 351:652–654.
Menotti-Raymond, M., V. A. David, and S. J. O’Brien. 1997. Pet
cat hair implicates murder suspect. Nature 386:774.
Miklos, G. L. G., and G. M. Rubin. 1996. The role of the genome
project in determining gene function: insights from model organisms. Cell 86:521–529.
Muller, H. J. 1942. Isolating mechanisms, evolution and temperature. Biol. Symp. 6:71–125.
Nei, M. 1987. Molecular evolutionary genetics. Columbia Univ.
Press, New York.
Nei, M., and T. Gojobori. 1986. Simple methods for estimating the
number of synonymous and nonsynonymous nucleotide substitutions. Mol. Biol. Evol. 3:418–426.
Ohta, T. 1972. Evolutionary rate of cistrons and DNA divergence.
J. Mol. Evol. 1:150–157.
———. 1992. The nearly neutral theory of molecular evolution.
Annu. Rev. Ecol. Syst. 23:263–286.
———. 1995. Synonymous and nonsynonymous substitutions in
mammalian genes and the nearly neutral theory. J. Mol. Evol.
40:56–63.
Omland, K. E. 1997. Correlated rates of molecular and morphological evolution. Evolution 51:1381–1393.
Orr, H. A. 1995. The population genetics of speciation: the evolution of hybrid incompatibilities. Genetics 139:1805–1813.
Orr, H. A., and M. Turelli. 2001. The evolution of postzygotic
isolation: accumulating Dobzhansky-Muller incompatibilities.
Evolution 55:1085–1094.
Orr, M. R., and T. B. Smith. 1998. Ecology and speciation. Trends
Ecol. Evol. 13:502–506.
Parker, H. R., D. P. Philipp, and G. S. Whitt. 1985. Gene regulatory
divergence among species estimated by altered developmental
patterns in interspecific hybrids. Mol. Biol. Evol. 2:217–250.
Powell, J. R. 1997. Progress and prospects in evolutionary biology:
the Drosophila model. Oxford Univ. Press, Oxford, U.K.
Powell, J. R., and R. DeSalle. 1995. Drosophila molecular phylogenies and their uses. Evol. Biol. 28:87–138.
Rice, W. R. 1984. Disruptive selection on habitat preference and
the evolution of reproductive isolation: a simulation study. Evolution 38:1251–1260.
Rice, W. R., and E. E. Hostert. 1993. Laboratory experiments on
speciation: what have we learned in 40 years? Evolution 47:
1637–1653.
Rice, W. R., and G. W. Salt. 1988. Speciation via disruptive selection on habitat preference: experimental evidence. Am. Nat.
131:911–917.
———. 1990. The evolution of reproductive isolation as a correlated character under sympatric conditions: experimental evidence. Evolution 44:1140–1152.
Rosenzweig, M. L. 1995. Species diversity in space and time. Cambridge Univ. Press, Cambridge, U.K.
Sasa, M. M., P. T. Chippindale, and N. A. Johnson. 1998. Patterns
of postzygotic isolation in frogs. Evolution 52:1811–1820.
Schluter, D. 1998. Ecological causes of speciation. Pp. 114–129 in
D. J. Howard and S. H. Berlocher, eds. Endless forms: species
and speciation. Oxford Univ. Press, New York.
Templeton, A. R. 1981. Mechanisms of speciation: a population
genetic approach. Annu. Rev. Ecol. Syst. 12:23–48.
Tilley, S. G., P. A. Verrell, and S. J. Arnold. 1990. Correspondence
between sexual isolation and allozyme differentiation: a test in
the salamander Desmognathus ochrophaeus. Proc. Natl. Acad.
Sci. USA 87:2715–2719.
Turelli, M., N. H. Barton, and J. A. Coyne. 2001. Theory and speciation. Trends Ecol. Evol. 16:330–343.
Wu, C.-I., and H. Hollocher. 1998. Subtle is nature: the genetics
of species differentiation and speciation. Pp. 339–351 in D. J.
Howard and S. H. Berlocher, eds. Endless forms: species and
speciation. Oxford Univ. Press, New York.
Zar, J. H. 1984. Biostatistical analysis. 2d ed. Prentice-Hall, Englewood Cliffs, NJ.
Zeng, L.-W., J. M. Comeron, B. Chen, and M. Kreitman. 1998. The
molecular clock revisited: the rate of synonymous vs. replacement change in Drosophila. Genetica 102/103:369–382.
Corresponding Editor: S. Edwards
APPENDIX
Here I use the delta method (Appendix 1 in Lynch and Walsh
1998) to approximate the expectation and variance of B under a
Poisson molecular clock. The result indicates that, under the Poisson
clock, the rate heterogeneity comparisons employed in this paper
would be biased against finding greater rate heterogeneity for allozymes versus DNA.
Let x and y denote the branch lengths, in units of actual substitutions per site (or codon substitutions per locus for allozymes),
from the most recent common ancestor to species 1 and 2, respectively. Define
B5
x2y
.
x1y
(A1)
The delta method gives approximate expressions for the expectation
and variance, assuming Cov(x, y) 5 0:
1
E(B) ù V(x)
2)
]2B
]2B
1 V(y)
2
2]x
2]y 2
5 0,
(A2)
E(x),E(y)
assuming the expectations and variances of x and y are equal. And
[
2
1 ]x 2
V(B) ù V(x)
]B
]
1 ]y 2 )
1 V(y)
]B
2
5
E(x),E(y)
V(x 1 y)
.
[E(x 1 y)] 2
(A3)
If the clock is Poisson, then E(x 1 y) 5 V(x 1 y) 5 2lt, so
V(B) ù
1
2l t
(A4)
where l is the substitution rate and t is the number of generations
since the most recent common ancestor. Under a Poisson molecular
clock, all measures of genetic distance used in this paper are estimates of 2lt. DNei is in units of codon substitutions per enzyme
locus (Nei 1987), while DNA distances are in numbers of nucleotide
substitutions per site (Nei and Gojobori 1986; Nei 1987). DNA
hybridization distance, measured by DTM (Powell 1997), is linearly
related to the proportion of nucleotide differences p 5 c(DTM),
where c is between 0.010 and 0.015 (Nei 1987). Thus, DNA hybridization distances estimate 2lt/c. The numerical value of DNei
for a given pairwise comparison is always greater than the numerical
values of the DNA-based estimates of 2lt in my datasets. Thus,
under a Poisson clock null model, B calculated from DNei is expected
to have a lower variance than B based on DNA distances. This
would bias my rate heterogeneity comparisons toward finding that
DNei is a better clock than DNA. Despite this bias, I found modest
evidence that DNei has more variable rates than DNA distances
(Table 3).