does fish ecology predict dispersal across …...2e-mail: [email protected]...

ORIGINAL ARTICLE

doi:10.1111/j.1558-5646.2008.00377.x

DOES FISH ECOLOGY PREDICT DISPERSALACROSS A RIVER DRAINAGE DIVIDE?Christopher P. Burridge,1,2 Dave Craw,3 Daniel C. Jack,4 Tania M. King1 and Jonathan M. Waters1

1Department of Zoology, University of Otago, PO Box 56, Dunedin, New Zealand.2E-mail: [email protected]

3Department of Geology, University of Otago, PO Box 56, Dunedin, New Zealand4Department of Conservation, PO Box 5244, Dunedin, New Zealand

Received November 29, 2007

Accepted February 19, 2008

Obligate freshwater taxa are frequently distributed among catchments isolated by marine and terrestrial barriers. Such distributions

can arise through vicariant changes in drainage geometry, or dispersal via intermittent freshwater connections. We employed two

adjacent rivers in southern New Zealand to test for interdrainage dispersal while controlling for historical drainage geometry,

and analyzed four ecologically distinct freshwater-limited fish taxa to assess any relationship with habitat preference. Individuals

from the Mararoa and Oreti catchments (n >100 per species) were sequenced for a minimum of 1297 bp of mitochondrial DNA

(cytochrome b and control region). Phylogeographic relationships were consistent with ecological expectations of interdrainage

dispersal capability, with the two obligate riverine taxa each exhibiting reciprocal monophyly between catchments, whereas

the two facultative swamp dwellers revealed paraphyletic relationships, one of which shared a haplotype between catchments.

Statistical phylogeography, accommodating taxon-specific mutation rates and the known age of the last major riverine connection

between these catchments, rejected complete isolation of populations for one of the swamp dwellers. Therefore, dispersal across a

young (145–240 kyr) drainage divide is inferred for one species, and can be predicted to some extent by species ecology. Moreover,

our study highlights the importance of historical drainage geometry when assessing the causes of contemporary genetic structuring

in freshwater taxa.

KEY WORDS: Comparative phylogeography, Galaxias, geomorphology, river capture, statistical phylogeography, time-

dependency.

Freshwater-limited taxa are commonly distributed among

drainages isolated from one another by marine and terrestrial habi-

tats (Banarescu 1990). However, given their reliance on the fresh-

water environment, how do these taxa develop such distributions?

Possible mechanisms can be divided into two classes: loss of in-

tervening freshwater environment (vicariance), or chance move-

ment between otherwise isolated habitats (dispersal). Craw et al.

(2007b) summarized several geomorphological mechanisms that

could mediate vicariance or dispersal in freshwater-limited taxa.

Vicariance could involve capture of river tributaries by neighbor-

ing drainages (i.e., the displacement of stream sections between

adjacent catchments; Fig. 1A; Banarescu 1990; Bishop 1995), or

glacial-eustatic changes in sea level drowning stream junctions

(Fig. 1B; Banarescu 1990). Alternatively, dispersal could be fa-

cilitated by divide inundation during flooding (Fig. 1C; Bishop

1995; Thacker et al. 2007), such as the Rupanuni floodplain re-

gion of Guyana, intermittently connecting the Amazon catchment

with a Caribbean drainage (Hamilton et al. 2002). Tributaries

near drainage divides may also periodically discharge to different

catchments at different times, depending on the vagaries of sedi-

ment accumulation or erosion near the divide (Fig. 1D). Likewise,

a lake or swamp on a drainage divide may discharge alternately be-

tween different catchments—or flow into both simultaneously—

depending on the integrity of the impounding walls, providing

1484C© 2008 The Author(s) . Journal compilation C© 2008 The Society for the Study of Evolution.Evolution 62-6: 1484–1499

ECOLOGY AND INTERDRAINAGE DISPERSAL

Figure 1. Six geomorphological processes (A-F) by which

freshwater-limited taxa could have become distributed in catch-

ments presently isolated from one another by marine and terres-

trial barriers (modified from Craw et al. 2007b).

opportunities for either intermittent or continuous dispersal be-

tween catchments (Fig. 1E–F; Banarescu 1990). Dispersal could

also be mediated by biological vectors, such as the hypothetical

“eggs on ducks feet” mechanism (Unmack 2001). Regardless, it

seems likely that the ability of freshwater-limited species to ex-

ploit these dispersal mechanisms will be dictated to some extent

by their ecological attributes and habitat requirements (Hurwood

and Hughes 2001; Thacker et al. 2007).

Although several studies have examined intrinsic (Tibbets

and Dowling 1996; Turner et al. 1996; Turner and Trexler 1998;

Whiteley et al. 2004; Turner and Robison 2006; Cook et al.

2007) and extrinsic (Currens et al. 1990; Shaw et al. 1994; King

and Wallis 1998; Esa et al. 2000; McGlashan and Hughes 2000;

Costello et al. 2003; Taylor et al. 2003; Roy et al. 2006) factors in-

fluencing dispersal of freshwater-limited fish within catchments,

comparatively few researchers have examined movement of such

taxa between catchments. Those to address this issue have typ-

ically invoked ad hoc historical changes to drainage geometry,

often in the absence of any geological evidence (e.g., Waters et al.

1994; Durand et al. 1999; Kreiser et al. 2001; Froufe et al. 2003).

As stated by Bishop (1995), while “drainage rearrangements may

be an evidently obvious and readily invoked explanation for ge-

netic differences . . . it may not necessarily be the correct expla-

nation.” The perceived biogeographic significance of river cap-

ture may, therefore, have been overestimated by previous studies

(Bishop 1995; Unmack 2001; Smith and Bermingham 2005). In-

terdrainage dispersal via intermittent “wet connections,” such as

swamps on low drainage divides, may represent an important alter-

native mechanism of range expansion in freshwater-limited taxa

(Hurwood and Hughes 1998; Unmack 2001; Craw et al. 2007b;

Thacker et al. 2007).

To assess the biogeographic significance of intermittent “wet

connections” between otherwise isolated catchments, it is neces-

sary to choose a geographic system with a well-characterized his-

tory. In particular, it is essential that any past changes in drainage

geometry have been identified and dated using geological tech-

niques, or can be safely refuted. Otherwise, genetic relationships

suggestive of recent dispersal between isolated catchments via

intermittent “wet connections” (e.g., catchment paraphyly) could

actually represent signatures from a period during which a main

channel connection existed, particularly if such a connection was

recent and effective population sizes were large (Neigel and Avise

1987; Hudson and Turelli 2003; Rosenberg 2003). In addition,

to determine whether ecological attributes specifically moderate

interdrainage dispersability, it is preferable to perform compara-

tive analyses on closely related taxa that are likely to vary only

in the ecological attributes of interest, rather than on unrelated

taxa that differ in any number of ecological or physiological traits

(Williams 1994; Turner and Trexler 1998; Dawson et al. 2002;

Whiteley et al. 2004; Hickerson and Cunningham 2005). The taxa

compared should also be broadly overlapping in distribution, such

that they are exposed to the same environments (Dawson et al.

2002; Froufe et al. 2003). Comparative studies of unrelated taxa,

by contrast, are better suited to questions of biogeographic con-

gruence (e.g., Carstens et al. 2005; DeChaine and Martin 2006).

The dynamic geological history of rivers in New Zealand’s

South Island has been well characterized (e.g., Turnbull 1980;

McIntosh et al. 1990; McAlpin 1992; Craw et al. 1999; Turnbull

2000; Craw and Norris 2003; Craw and Waters 2007; Craw et al.

2007a,b), and thus represents a valuable system for studies of in-

tercatchment dispersal in freshwater-limited taxa. River histories

are particularly well documented due to the informative geolog-

ical record preserved in local river terraces. River terraces are

remnants of old river gravel deposits preserved on valley walls,

and their gradients reflect paleodrainage direction. River terraces

often occur as a series of “steps” up a valley wall, reflecting the

past series of glaciations, against which the ages of paleodrainage

geometries can be determined.

This study focuses on two neighboring rivers in southern

South Island, the Mararoa and Oreti rivers (Fig. 2). These rivers

EVOLUTION JUNE 2008 1485

CHRISTOPHER P. BURRIDGE ET AL.

Figure 2. Sampling sites of Galaxias (G. paucispondylus, G. gollumoides, G. “southern”) and Gobiomorphus breviceps collected from

southern New Zealand (collection details in Appendix). White- and black-filled circles correspond to the contemporary Mararoa and Oreti

catchments, respectively. Arrows indicate the paleochannels linking the Mararoa and Oreti, and the chronology of these connections is

described in the insert (see Craw et al. 2007b).

originated with the rise of the mountains in their headwaters dur-

ing the Pliocene (2–5 Myr BP), and drained toward the southwest

(Fig. 2, inset A), depositing large volumes of gravel—the Prospect

Formation (Turnbull 2000). This drainage pattern persisted until

the middle Quaternary (< 1 Myr BP), when major glaciations in

the mountains to the west diverted drainage toward the east, and

glacial outwash gravels inundated intervening ridges (Fig. 2, in-

set B). High gravel terraces as old as 340 kyr are still preserved

from this stage (Turnbull 2000; Craw et al. 2007b). The Mararoa

and Oreti rivers became progressively incised into these old grav-

els, but drainage toward the east persisted, and the two rivers were

joined where underlying bedrock ridges were low (Fig. 2, inset C).

Eventually, incision into gravels became so deep that the Mararoa

River returned to southwestward flow, as at present, and became

separated from the Oreti River (Fig. 2, inset D). The lowest parts of

the divide between the two rivers (arrowed, Fig. 2) are defined by

1486 EVOLUTION JUNE 2008


gravel terrace remnants (Turnbull 2000; Craw et al. 2007b). The

terrace level that forms these divides has not been dated because

suitable material is lacking, but it lies approximately halfway be-

tween a higher terrace that is ca. 240 kyr, and a lower terrace that

is ca. 145 kyr (Craw et al. 2007b). Hence, the age of separation is

most likely to be ca. 190 kyr, with separation ages becoming less

likely toward the upper and lower terrace age limits.

The Oreti and Mararoa are presently separated by low re-

lief topography (50–70 m above valley floor) in the two regions

that correspond to paleochannel connections (Fig. 2; Craw et al.

2007b), and this relief would have been even lower in the past,

owing to main channel incision that has occurred on either side.

Swamps or headwater streams on these low divides may have facil-

itated intercatchment dispersal via intermittent “wet connections”

in species able to exploit such habitats (Fig. 1D–F). Although the

Oreti River has experienced recent connections with neighbor-

ing catchments to the north (Von River, 70–11 kyr BP; Turnbull

1980, 2000; Craw and Norris 2003; Craw et al. 2007b), south

(Aparima River, 12–24 kyr BP; Turnbull and Allibone 2003) and

east (Mataura River, 350–120 kyr BP; McIntosh et al. 1990), none

of these additional systems has been connected to the Mararoa

more recently than the Oreti (Turnbull and Allibone 2003; Craw

et al. 2007b). For the sake of clarity, therefore, the current study

focuses exclusively on the adjacent Mararoa and Oreti systems.

The aim of this study is to compare phylogeographic rela-

tionships among the four freshwater-limited fish taxa that occupy

the Mararoa and Oreti systems, and test for interdrainage dispersal

subsequent to drainage divide formation 145–240 kyr BP. We also

test the hypothesis that interdrainage dispersal is predictable on the

basis of fish ecology, as we study three closely related but ecolog-

ically divergent galaxiid species (Galaxiidae; Waters et al. 2001b;

Waters and Wallis 2001a,b), and a completely unrelated taxon

(Eleotridae). Galaxias paucispondylus and G. “southern” (the lat-

ter previously synonymous with G. depressiceps, Waters and Wal-

lis 2001b) occupy fast flowing high-order streams over gravel-

boulder substrates, typically representing significant tributaries

or main channels of large rivers (McDowall 2000; Crow 2007). In

contrast, Galaxias gollumoides occupies steep headwater streams

over gravel-boulder substrates, swamps, and small, slow, low gra-

dient streams over soft sediment substrates (McDowall and Chad-

derton 1999; McDowall 2000; Waters et al. 2001b; Crow 2007).

The completely unrelated species, Gobiomorphus breviceps, of-

ten occurs in swampy conditions, but also inhabits more riverine

habitats excluding steep headwater streams (Jowett and Richard-

son 1995; McDowall 2000). Despite the contrasting ecologies of

these species, fine-scale habitat heterogeneity permits all four to

occur in close proximity in the Mararoa and Oreti systems.

We predict that swamp-dwelling taxa G. gollumoides and

Go. breviceps will exhibit the genetic signature of dispersal

across the Mararoa–Oreti drainage divide, as they should be able

to traverse swampy or low order stream connections between

catchments (Fig. 1D–F). In contrast, we predict that riverine taxa

Galaxias “southern” and G. paucispondylus will show evidence

of complete isolation across the drainage divide, as these species

do not occur in swampy habitats or low-order streams. These

predictions are consistent with preliminary genetic data collected

for two of these species as part of an unrelated study investigating

the affinities of Von River Galaxias; G. “southern” exhibited

reciprocal monophyly of Mararoa and Oreti mitochondrial

haplotypes, where as G. gollumoides exhibited paraphyletic

relationships between these drainages (Burridge et al. 2007).

However, here we undertake more intensive sampling of these

species in the vicinity of the Mararoa–Oreti divide, with matching

sampling of G. paucispondylus and Go. breviceps. We apply

statistical phylogeographic analyses (Knowles and Maddison

2002; Knowles 2004) to rigorously assess whether population

isolation under neutral coalescence can be discounted for any

discordance between empirical gene trees and contemporary

catchment boundaries, given the explicit temporal framework of

main channel connection and isolation.

Materials and MethodsSAMPLING

In geographic terms, our sampling program targeted two low

divides: potential sites for interdrainage dispersal between the

Mararoa and Oreti river systems. Incidentally, these low relief

regions also correspond to paleochannel connections between

the Mararoa and Oreti rivers (Fig. 2; Craw et al. 2007b). Eight

“zones” in each of the Mararoa (zones A–H) and Oreti (zones I–P)

systems were sampled, with the aim of obtaining approximately

five individuals of each species per zone (Fig. 2; Appendix). Out-

groups for Mararoa and Oreti G. gollumoides, G. paucispondylus,

and Go. breviceps were provided by conspecific individuals from

the Nevis and Clutha rivers, whereas the outgroup for Galaxias

“southern” was provided by Clutha G. sp. “D”. Fish samples were

collected using a backpack electric fishing apparatus. Continuous

current was applied in a manner to draw out and stun fish present.

These were captured using dip nets in pools and backwaters, or

using a pole net in flowing water. Fish were anesthetized with 2-

phenoxy-ethanol and then preserved on dry ice or in 95% ethanol.

MITOCHONDRIAL DNA SEQUENCE ANALYSIS

Protocols employed for total DNA extraction, PCR amplification,

and sequencing of galaxiid mitochondrial cytochrome b gene (cyt

b) and control region (CR) followed Burridge et al. (2006). The

complete cyt b of galaxiids was amplified with primers cytb-Glu

and cytb-Thr (Waters and Wallis 2001a), and sequenced with cytb-

Glu, yielding a minimum of 764 bp, whereas CR was amplified

with primers P3 and SPHE (Waters and Wallis 2001a), and



sequenced with P3, yielding a minimum of 763 bp. Cyt b of

Go. breviceps was amplified using primers L14724 (Paabo et al.

1990) and H16498 (Meyer et al. 1994) under the same conditions

as galaxiid cyt b, and sequenced with L14724 (and occasionally

H15149; Kocher et al. 1989), yielding 898 bp. PCR amplification

and sequencing of Go. breviceps CR employed the protocol of

Smith et al. (2003), yielding 399 bp. Length variation was absent

from both regions with the exception of a single nucleotide indel

in Go. breviceps CR.

Measures of haplotype (h) and nucleotide (�) diversity (Nei

1987) were calculated for each catchment using Arlequin 2.001

(Schneider et al. 2000). Evidence for non-neutrality of mitochon-

drial DNA variation or recent population expansion was assessed

based on Tajima’s D (Tajima 1989) and Fu’s Fs (Fu 1997) statis-

tics and the number of segregating sites, calculated using DNASP

4.10.9 (Rozas et al. 2003). Confidence limits for Tajima’s D were

obtained following equation (47) of Tajima (1989), and for Fu’s

Fs via 10,000 neutral coalescent simulations (constant population

size) based on the sample size and number of segregating sites

(Ramos-Onsins and Rozas 2002). We combined cyt b and CR

datasets during phylogenetic analyses, consistent with the facts

that the mitochondrial genome is predominantly nonrecombining,

and that separate analyses of the two regions were largely congru-

ent. Phylogenetic relationships among mtDNA sequences were re-

constructed via maximum parsimony, maximum-likelihood, and

Bayesian analysis following the methods detailed in Burridge et al.

(2007), with minor differences. During Bayesian analysis, three

of the Markov chains were heated according to “Temp = 0.05”

for G. paucispondylus to improve mixing among chains rather

than “Temp = 0.1.” For all taxa, the attainment of asymptotes

(stationarity) for LnL and nucleotide substitution model param-

eters during Bayesian analysis was achieved within the first 105

generations. Convergence of duplicate Bayesian runs (107 gen-

erations) was indicated by an average standard deviation of split

frequencies between runs of approximately 0.01 or less; starting

values were between 0.10 and 0.14, values less than 0.02 were

attained within the first 0.6 × 106 generations. Consequently, the

first 104 trees sampled (106 generations) were discarded as “burn-

in” before the calculation of a consensus topology and bipartition

posterior probabilities.

STATISTICAL PHYLOGEOGRAPHY

Coalescent simulation was employed to determine whether em-

pirical gene trees are consistent with an explicit model of pop-

ulation isolation (online Supplementary Fig. S1). A critical fac-

tor in the reliability of statistical phylogeography is the applica-

bility of population model topology and demographic parame-

ters to the actual history (Knowles 2004; Knowles and Carstens

2007). In this instance, the geological framework of Mararoa and

Oreti main channel isolation provides a simple and accurate model

population topology (Craw et al. 2007b). In addition, the age of

main channel isolation is well constrained by dated river terraces

(Craw et al. 2007b), and female generation times appear to be 1–2

years for freshwater-limited Galaxias, and 2 year for Go. brevi-

ceps (Hopkins 1971; Staples 1975; McDowall 2000; Hamilton and

Poulin 2001). Fluctuate1.4 (Kuhner et al. 1995) was employed for

the calculation of � under the assumption of constant population

size.

We modeled stochasticity in both gene lineage coalescence

and the nucleotide substitution processes using the Coalescence

and Batch Architect modules of Mesquite 1.11 (Maddison and

Maddison 2005, 2006; Maddison 2006). First, we simulated 1000

genealogies under neutral coalescence for the observed number

of haplotypes in each population, along the demographic model

of population connectivity and effective sizes (online Supplemen-

tary Fig. S1). These genealogies simulate patterns of mother–

daughter descent (going forward in time) and haplotype inher-

itance. Second, we simulated a nucleotide dataset along each of

the simulated genealogies, to accommodate the fact that an empir-

ical genealogy is only an estimate of the actual genealogy based

on DNA sequence variation, and this reconstruction is subject

to error. Nucleotide datasets were simulated under the model of

nucleotide substitution inferred for the combined cyt b and CR

regions by ModelTest (Posada and Crandall 1998). The “scaling

factor” was optimized such that the mode of the minimum tree

lengths obtained from parsimony analysis of the simulated nu-

cleotide datasets (PAUP∗, hsearch addseq = random, nrep = 1,

maxtrees = 100, timelimit = 10) approximated the minimum tree

length derived from the empirical data. The parsimony trees de-

rived from the simulated datasets were then compared to the con-

temporary catchment boundaries to construct null distributions

of the s statistic (Slatkin and Maddison 1989), against which the

score from the empirical topology was compared.

We varied effective population sizes (Ne)—assuming con-

stant ratios between populations that matched those of �—to de-

termine the Ne at which 95% of the simulated s values were less

than the empirical s (N e critical). Below this Ne the discordance

between the empirical genealogy and catchment boundaries is

greater than could be expected under neutral coalescence given the

hypothesized history of population isolation. We then assessed un-

der what combinations of mutation rate and � the N e critical could

be realized, and hence whether the null hypothesis of neutral coa-

lescence for the observed discrepancy between the empirical gene

tree and contemporary catchment boundaries is rejectable. Al-

though statistical phylogeographic studies have predominantly re-

lied on perceived “universal” nucleotide substitution rates (u), of-

ten derived from distantly related taxa, here we employ rates from

a series of galaxiid and Go. breviceps calibrations (Burridge et al.

2008), which are obviously more appropriate to the current study

(Arbogast et al. 2002). In addition, as the majority of these rates



have been calibrated for freshwater isolation events within the

last million years, they enable us to apply values applicable to the

timescale of Mararoa–Oreti isolation given that appropriate rates

may be time- as well as taxon-dependent (Ho et al. 2005; Burridge

et al. 2006, 2008; Ho and Larson 2006; Waters et al. 2007).

ISOLATION WITH MIGRATION ANALYSIS

Coalescent analyses were also performed under the isolation

with migration model of Nielsen and Wakeley (2001). This

model involves analysis of a single nonrecombining locus in

two closely related populations that were geographically isolated

some time in the past, but may have experienced subsequent gene

flow. Using the IM software (http://lifesci.rutgers.edu/∼heylab/

HeylabSoftware.htm), genealogical topologies were simulated

and updated along a Markov chain Monte Carlo during which

six model parameters were recorded (divergence time, ancestral

and two contemporary �, forward and reverse migration rates).

The mitochondrial cyt b and CR were combined during analysis

as justified above, and to maximize the information that could

be obtained from each simulated genealogical history. Nucleotide

substitution was assumed to follow the HKY model—IM’s closest

approximation of the models selected by ModelTest. Where possi-

ble, upper bounds of parameter priors were set such that posterior

distributions were fully contained within them, with peaks then

representing maximum-likelihood estimates. A minority of pos-

teriors for � and divergence time contained infinite nonzero tails,

but the peaks in these distributions appear insensitive to analyses

under alternate priors. Markov chain Monte Carlo searches em-

Table 1. Sampling intensity (n) and mitochondrial genetic variation for freshwater-limited species of Galaxias and Gobiomorphus within

the Mararoa and Oreti catchments from southern New Zealand. Divergences represent uncorrected p-distance, and models represent

those selected by ModelTest under the hierarchical likelihood ratio test.

G. paucispondylus G. “southern” G. gollumoides Go. breviceps

Mararoa Oreti Mararoa Oreti Mararoa Oreti Mararoa Oreti

n 61 56 59 55 43 60 67 36Number of haps 24 29 25 27 14 20 6 14Haplotype diversity 0.9355 0.9221 0.8153 0.9455 0.8682 0.8701 0.3524 0.7797Nucleotide diversity 0.0023 0.0019 0.0012 0.0025 0.0051 0.0035 0.0004 0.0015Tajima’s (1989) D −0.682 −2.096∗ −2.149∗ −1.708 1.040 −1.016 −1.575 −0.956Fu’s (1997) Fs −10.778∗ −23.335∗ −24.434∗ −16.511∗ 0.910 −3.500 −3.841∗ −4.968Variable sites: cyt b 38 (4.3%) 59 (7.7%) 29 (3.8%) 13 (1.4%)Variable sites: CR 19 (2.5%) 25 (3.2%) 23 (3.0%) 7 (1.8%)Max. intracatchment divergence 0.0068 0.0092 0.0164 0.0023Min. intercatchment divergence 0.0048 0.0256 0.0039 0.0000Model: cyt b TrN+� TrN TrN+I TrN

CR HKY+I+� HKY+I F81 F81Combined HKY+I+� TrN+I+� TrN+I+� TrN

∗P significant at � = 0.05.

ployed four chains, three of which were incrementally heated to

promote broader searching of parameter space. After a burn-in

period of 105 generations, parameter trend line plots were ex-

amined for the attainment of stationarity, and parameter values

were subsequently recorded every 10 generations. Independence

of samples collected along a chain was assessed via their auto-

correlation statistics, and runs were continued until the Effective

Sample Size exceeded 100 for all parameters. Independent runs

using different random number seeds were conducted to assess

convergence upon the true stationary distribution.

ResultsVARIATION WITHIN AND AMONG CATCHMENTS

All DNA sequences collected during this study are available from

GenBank (accession numbers EU512434-893). The number of

haplotypes observed for each species, the number of variable nu-

cleotide sites, haplotype, and nucleotide diversity, and levels of

uncorrected sequence divergence within and between catchments,

are listed in Table 1. Cyt b exhibited a greater proportion of vari-

able sites than CR in the galaxiids, but not in Go. breviceps (Table

1). Maximum intracatchment divergence exceeded minimum in-

tercatchment divergence for all taxa except G. “southern” (Table

1). Only one haplotype was shared between the Mararoa and Oreti

catchments, observed for Go. breviceps. Genetic diversity was

similar between catchments for each galaxiid species, whereas

Go. breviceps diversity in the Mararoa was less than half that of

the Oreti (Table 1). Haplotype diversity was higher in obligate



riverine taxa (G. paucispondylus, G. “southern”) relative to

“swamp-dwellers” (G. gollumoides, Go. breviceps) (Table 1). In

contrast, the “swamp-dwelling” taxa exhibited higher nucleotide

diversity than the obligate riverine taxa (with the exception of

Go. breviceps in the Mararoa) (Table 1). Neutrality of mtDNA se-

quence variation was not rejected in the Mararoa and Oreti for G.

gollumoides, or for Go. breviceps in the Oreti (Tajima’s D and Fu’s

Fs; Table 1). Neutrality was rejected, however, for Go. breviceps

in the Mararoa (Fu’s Fs only, Table 1), and in both catchments for

G. paucispondylus and G. “southern” (Fu’s Fs for each catchment,

Tajima’s D in the Oreti and Mararoa for G. paucispondylus and

G. “southern,” respectively; Table 1).

PHYLOGEOGRAPHIC RELATIONSHIPS

Maximum-likelihood, maximum parsimony, and Bayesian anal-

yses were concordant with respect to the composition and sup-

port for major clades (Fig. 3). Mararoa and Oreti haplotypes rep-

resented well-supported monophyletic assemblages for G. pau-

cispondylus and G. “southern” (1.00 Bayesian bipartition proba-

bility [pP], > 96% parsimony bootstrap proportion [BP]; Fig. 3).

In contrast, G. gollumoides exhibited paraphyly among catch-

ments, with a small number of Mararoa and Oreti haplotypes

forming a well-supported clade (0.99 pP, 96% BP) intermedi-

ate to “pure” and well-supported Mararoa and Oreti assemblages

(Fig. 3). In comparison to the galaxiids, Go. breviceps exhibited

much shallower divergence among haplotypes in general (Fig. 3).

One Go. breviceps haplotype was shared between the Oreti (fre-

quency = 0.13) and Mararoa (frequency = 0.81).

STATISTICAL PHYLOGEOGRAPHY

As G. gollumoides and Go. breviceps did not exhibit recipro-

cal monophyly across the Mararoa–Oreti drainage divide, we as-

sessed the effective population sizes (Ne) under which the ob-

served magnitudes of gene- and population-tree discordance were

no longer compatible with population isolation under neutral co-

alescence. The Ne at which 0.95 of the gene trees from simulated

datasets exhibited less discordance with the population tree than

was observed for the empirical gene tree (Ne critical) approxi-

mated 133,000 and 199,000 for Mararoa and Oreti populations of

G. gollumoides, and 37,000 and 107,000 for corresponding pop-

ulations of Go. breviceps. Combinations of � and u that yield Ne

greater than or less than Ne critical are shown in Fig. 4.

Under the slowest of 13 galaxiid mtDNA rate calibrations (5

Myr isolation event, Burridge et al. 2008), Ne exceeds Ne critical

for all plausible values of � (Fig. 4), and therefore the magnitude

of empirical gene- and population-tree discordance observed in

this taxon can be explained by population isolation under neu-

tral coalescence. Along similar lines, molecular rates that appear

applicable to time scales in excess of 100–200 kyr in galaxiids

(“asymptotic” rate; Burridge et al. 2008) do not reject population

isolation under neutral coalescence unless � are assumed to be

close to their lowest plausible estimates (Fig. 4). Therefore, the

null hypothesis of no interdrainage gene flow in G. gollumoides

cannot be rejected unless we are willing to accept much faster

molecular rates or smaller values of � (Fig. 4). These analyses

employed a 1-year female generation time. Employing a 2-year

female generation time suggested by some life-history studies of

freshwater-limited galaxiids would make rejection of null hypoth-

esis even more unlikely.

Three substitution rate calibrations are available for Go. bre-

viceps at present (Burridge et al. 2008). Under the slowest substitu-

tion rate (5 Myr event, Burridge et al. 2008), Ne exceeds Ne critical

for all plausible values of � in the Oreti population, and for the vast

majority of � in the Mararoa, and hence population isolation un-

der neutral coalescence is unlikely to be rejected. However, using

an intermediate substitution rate based on a calibration similar in

age to Mararoa–Oreti main channel isolation (70–130 kyr event,

Burridge et al. 2008), only � appreciably larger than the maximum

likelihood estimates yield Ne greater than Ne critical (Fig. 4), and

hence the null hypothesis of population isolation under neutral

coalescence is rejected in the majority. Once substitution rates

exceed ∼1.2 × 10−7 subs./site/generation the null hypothesis is

rejected exclusively, assuming that plausible values of � only exist

within the estimated 95% confidence limits (Fig. 4).

In general, it is apparent that different combinations of plau-

sible nucleotide substitution rates and values of � will give differ-

ent assessments of the null hypothesis in both species, but under

the most realistic conditions (calibrations derived from approxi-

mately the same time period, maximum-likelihood estimates of

�) it appears that the null hypothesis of population isolation under

neutral coalescence is rejected for Go. breviceps, but not for G.

gollumoides.

ISOLATION WITH MIGRATION ANALYSIS

Galaxiid migration rate posteriors were consistent with complete

isolation of populations since their divergence; highest marginal

posterior probabilities were associated with negligible migration

(the bin comprising lowest migration rates), and then rapidly

declined to zero probability with increasing migration rate (on-

line Supplementary Fig. S2). In contrast, Go. breviceps exhibited

a peak in probability for non-negligible migration, representing

movement from the Mararoa to the Oreti (online Supplementary

Fig. S2). Posterior probabilities for migration in the opposite

direction were highest for the bin of smallest migration rates,

but did not decline to zero probability with increasing migra-

tion rate as was observed for the galaxiids (online Supplementary

Fig. S2). The peak in the posterior distribution of the divergence

time parameter was also more recent for Go. breviceps than any

of the galaxiids (online Supplementary Fig. S2). Although the

time parameters obtained from reciprocally monophyletic datasets



Figure 3. Bayesian inference majority-rule consensus topologies derived from mitochondrial DNA sequences (cyt b and CR) of Galaxias

(G. gollumoides, G. paucispondylus, G. “southern”) and Gobiomorphus breviceps collected from southern New Zealand. White-filled

circles indicate haplotypes present in the Mararoa, and black-filled circles indicate Oreti haplotypes. Values above branches represent

Bayesian bipartition posterior probabilities, whereas those below branches represent parsimony bootstrap percentages, when either of

the corresponding values were at least 0.95 (Bayes) or 70% (parsimony). Root placement based on Nevis and Clutha River outgroup

individuals is depicted by the arrow. Branch lengths are proportional to the scale bar for each tree, with G. “southern” and G. gollumoides

shown under identical scales, but G. paucispondylus and Gobiomorphus under scales that are an order of magnitude finer and coarser,

respectively. Shaded boxes indicate scores of the empirical trees for the s statistic of Slatkin and Maddison (1989), quantifying discordance

with the drainage divide.



Figure 4. Statistical phylogeographic analyses of population isolation subsequent to drainage-divide formation in species exhibiting

paraphyletic relationships of individuals between catchments. Solid lines represent the critical values of effective population size (Ne)

for Mararoa and Oreti catchments that will yield 0.95 of simulated genealogies with less gene tree–population tree discordance than

observed for the empirical datasets. Open and filled circles represent the 95% confidence limits and ML estimates of � for each population,

respectively. Dashed lines demarcate the plausible combinations of � and substitution rate (u) for each population that will yield the

critical Ne. (A) Galaxias gollumoides. (B) Gobiomorphus breviceps. If a given combination of � and u yields an effective size lower than

the critical value (below the solid line), population isolation under neutral coalescence can be rejected for the observed level of empirical

gene- and population-tree discordance.

(G. paucispondylus, G. “southern”) are expected to suffer from

strong negative correlation with ancestral � (Wakeley and Hey

1997), such correlations were not observed, and the peak in time

parameter for G. gollumoides, which was not reciprocally mono-

phyletic between catchments, was certainly older than that for Go.

breviceps.

DiscussionThe aim of this study was to test for intercatchment dispersal in

a group of codistributed freshwater-limited fish, and to test for a

link with habitat preference. Although distributions of freshwater-

limited species among catchments presently isolated from one

another by marine or terrestrial barriers have predominantly been

considered the results of historical changes in drainage geometry

such as river capture (Banarescu 1990), the biogeographic signifi-

cance of such drainage rearrangements may have been overstated

(Bishop 1995; Unmack 2001; Smith and Bermingham 2005), and

intercatchment dispersal could also play a role (Hurwood and

Hughes 1998; Unmack 2001; Craw et al. 2007b; Thacker et al.

2007). Despite the evolutionary importance of allopatric distribu-

tions as potential initiators of cladogenesis and speciation (Coyne

and Orr 2004), and the somewhat paradoxical commonality of

such distributions in freshwater-limited taxa (Bilton et al. 2001),

until now there have been no formal tests of intercatchment dis-

persal in freshwater-limited fish, or whether the potential for such

dispersal is influenced by species ecology.

EVIDENCE FOR INTERCATCHMENT DISPERSAL

AND A RELATIONSHIP WITH ECOLOGY

We investigated four freshwater-limited fish species with differ-

ent ecological attributes: G. paucispondylus and G. “southern”

are obligate riverine taxa, occupying swift main channel habi-

tats, whereas G. gollumoides and Go. breviceps occupy swamps,



slow-flowing streams, and in the case of G. gollumoides, steep

headwaters. Phylogeographic analyses were performed in the con-

text of a well-characterized geographic system, where the main

channels of the Mararoa and Oreti rivers have been isolated since

145–240 kyr BP (Craw et al. 2007b). We predicted complete ge-

netic isolation across the drainage divide for the two obligate

riverine taxa, but incomplete genetic isolation for the swamp-

dwelling taxa that could have traversed the drainage divide via

intermittent “wet connections,” such as tributaries or swamps on

low saddles that periodically discharge to different catchments at

different times (Fig. 1D–F). The riverine taxa G. paucispondy-

lus and G. “southern” exhibited strongly supported reciprocally

monophyletic relationships between the Mararoa and Oreti, con-

sistent with genetic isolation (Fig. 3). In contrast, swamp-dwelling

taxa G. gollumoides and Go. breviceps exhibited paraphyletic re-

lationships across the drainage divide, with one haplotype shared

between these catchments in the case of Go. breviceps (Fig. 3).

Although these initial observations are congruent with eco-

logically based predictions of dispersal across the drainage divide,

statistical phylogeographic analysis indicated that the degree of

discordance between the empirical gene tree and the population

tree for G. gollumoides does not exceed that which could be ex-

pected by chance under a history of complete population isolation

and neutral coalescence during the last 190 kyr, given estimates of

� and plausible nucleotide substitution rates. Migration rate pos-

teriors from IM analysis were also consistent with complete isola-

tion of G. gollumoides populations during the last 190 kyr (online

Supplementary Fig. S2). However, as predicted, statistical phylo-

geographic analysis of the swamp-dwelling taxon Go. breviceps

rejected complete isolation across the Mararoa–Oreti drainage di-

vide during the last 190 kyr. IM analysis also revealed a peak

in marginal posterior probability for non-negligible Mararoa-to-

Oreti migration, although its probability was only slightly higher

than another peak for negligible migration, and hence this anal-

ysis is compatible with both possibilities (online Supplemen-

tary Fig. S2). This ambiguity may reflect inadequate information

from a single locus to inform the IM analysis (e.g., Carstens and

Knowles 2007), which is parameter-rich relative to the statisti-

cal phylogeographic analysis based on parametric bootstrapping

under an explicit population history. IM analysis repeated with

fewer parameters (all � equal, migration rates symmetrical) re-

vealed only a posterior probability peak for non-negligible migra-

tion, hence favoring incomplete isolation of populations (online

Supplementary Fig. S2).

Although some Gobiomorphus species are diadromous, lar-

vae of Go. breviceps are found instream (McDowall 2000), and

there is no evidence of larvae returning from the sea. Likewise,

there is no evidence for recent low sea stand connection be-

tween the Waiau catchment (containing the Mararoa) and systems

further east (Craw et al. 2007b). It is unlikely that vector-mediated

dispersal (e.g., eggs on ducks feet) could be responsible for in-

terdrainage dispersal, as Go. breviceps lays eggs under large rocks

(Hamilton et al. 1997). We also argue that the distribution of hap-

lotypes between catchments, and the relationships among them

(online Supplementary Fig. S3), are unlikely to reflect anthro-

pogenic translocation. Consequently, this study provides the first

strong genetic evidence of intercatchment dispersal via intermit-

tent “wet connections” in a system in which drainage history has

been explicitly considered. We note that the findings of this study

rely on data from a single locus, and it is possible that our as-

sumptions about the relationship between genetic diversity and

N e may not hold (Bazin et al. 2006). A more robust interpretation

awaits a multilocus assessment, which will also better inform IM

analysis.

COLONIZATION VERSUS GENE FLOW

Unlike Go. breviceps, the swamp-dwelling taxon G. gollumoides

does not provide any evidence for gene flow across the Mararoa–

Oreti divide. This finding is surprising, given that in addition to

the habitats occupied by Go. breviceps, G. gollumoides also oc-

cupies steep headwater streams over gravel–boulder substrates

(McDowall 2000), systems that may facilitate intermittent dis-

persal across axial valleys (Fig. 1D). Indeed, our sampling pro-

gramme encountered G. gollumoides on the low saddles between

catchments, occupying “perched” tributaries above the main river

channel (e.g., sites Ob, Oc).

One possible explanation for the inferred lack of gene flow

for G. gollumoides is that our statistical phylogeographic analysis

may have been biased by inappropriate estimates of �, nucleotide

substitution rate, or time since river isolation. However, IM anal-

ysis, which accommodates uncertainty in � and time since popu-

lation isolation, returned highest marginal posterior probabilities

for negligible migration rates in G. gollumoides (online Supple-

mentary Fig. S2). It should be noted that repeated statistical phy-

logeographic analysis with a 240 kyr separation time and 1-year

generation time—effectively maximizing the chance of recipro-

cal monophyly among catchments—did reject the null hypothesis

of population isolation under neutral coalescence across all plau-

sible substitution rates and � (not shown). However, population

isolation of this age is considered unlikely, and is more likely to

be halfway between the two glacial dates bounding the change

in drainage geometry based on the relative elevations of the low

saddles and river terraces. The inference of a 2-year generation

time for G. gollumoides would also remove the significance of the

240 kyr test result.

An alternative explanation for the lack of evidence for inter-

drainage dispersal in G. gollumoides is that such dispersal does

not always leave a genetic signature, or may not be detected un-

der a given sampling regime. We have previously hypothesized

that genetic signatures of immigrants to a river system are most



likely to be preserved when they are not exposed to conspecific

individuals, either because the species was not previously repre-

sented in that system (i.e., colonization; Burridge et al. 2006), or

because immigrants are isolated from resident populations by the

presence of gorges or waterfalls (see also McGlashan and Hughes

2000; Taylor et al. 2003; Poissant et al. 2005; Burridge et al. 2007),

which are known to inhibit gene flow in many fish (e.g., Currens

et al. 1990; Costello et al. 2003; Taylor et al. 2003; Crispo et al.

2006), including freshwater-limited Galaxias (King and Wallis

1998; Esa et al. 2000). Otherwise, the genetic signature of immi-

grants could be lost during hybridization or genetic drift (Dowling

et al. 1989; Epifanio and Philipp 2000; Esa et al. 2000; Kreiser

et al. 2001; Lu et al. 2001; Ruber et al. 2001; Good et al. 2003;

Carson and Dowling 2006; Sousa-Santos et al. 2006), not with-

standing competitive exclusion prior to reproduction (Marchetti

et al. 2004). In this instance, however, features such as large

gorges that could isolate immigrating individuals from residents

are lacking.

We suggest that the signature of interdrainage dispersal may

be particularly strong in Go. breviceps as this species may have

arrived in the Mararoa only recently. Although this is an ad hoc

argument, the pattern of genetic variation in this species bears the

hallmarks of “colonization”—in this case of the Mararoa—rather

than just “gene flow.” Specifically, haplotype diversity was appre-

ciably higher in the Oreti (0.78) than the Mararoa (0.35), in con-

trast to galaxiid patterns (Table 1), and cannot be explained by dif-

ferences in elevation (Shaw et al. 1994; Castric et al. 2001; Froufe

et al. 2003). One haplotype was also shared between catchments

but at a much higher frequency in the Mararoa (0.81 vs. 0.13), and

all other Mararoa haplotypes were linked back to this haplotype

prior to any other haplotypes observed in the Oreti (online Sup-

plementary Fig. S3). The significance of Go. breviceps Fu’s Fs in

the Mararoa, but not the Oreti, could also be explained by recent

population expansion into the former (Fu 1997; Ramos-Onsins

and Rozas 2002), although significant results are also obtained

for some of the obligate riverine galaxiids for which we do not

infer recent “colonization.” Although IM analysis appears to fa-

vor Go. breviceps migration in the direction opposite to this theory

(Mararoa to Oreti; online Supplementary Fig. S2), it may be biased

if the only gene flow between catchments was a colonization event,

as opposed to that occurring after initial population isolation, for

which this analysis was developed (Nielsen and Wakeley 2001).

Consistent with recent colonization of the Mararoa, Go. brevi-

ceps has a relatively shallow phylogenetic history in the southern

South Island, exhibiting low sequence divergence not only within

the Mararoa and Oreti systems (Fig. 4; Table 1), but also broadly

throughout the region in comparison to that observed for galaxiids

(e.g., Southland vs. Clutha genetic distances; Waters and Wallis

2001b; Smith et al. 2005; Burridge et al. 2007). Also in contrast

to galaxiids, Go. breviceps appears to lack multiple, distinct lin-

eages within catchments in southern New Zealand (Waters et al.

2001a; Waters and Wallis 2001b; Smith et al. 2005; Burridge et al.

2007).

EVOLUTIONARY AND ECOLOGICAL SIGNIFICANCE OF

INTERCATCHMENT DISPERSAL

There is an ongoing debate regarding the significance of drainage

rearrangement as a mechanism for the formation of allopatric

distributions in freshwater-limited taxa (Bishop 1995; Unmack

2001; Smith and Bermingham 2005). Our study is the first to

present strong genetic evidence for intercatchment dispersal in a

freshwater-limited taxon while controlling for drainage history,

but we only observed it in one of the two taxa we expected based

on habitat preferences. It should be noted that the current study

encompasses a fairly short geological time scale, constrained

by the dynamic nature of drainage rearrangement in the New

Zealand landscape (e.g., McAlpin 1992; Turnbull 2000; Waters

et al. 2001a; Craw et al. 2007a). Contrasting results might be

expected in different settings, or over different periods of geo-

logical time. In more stable geological systems, for instance, one

might predict a more significant role for intercatchment dispersal

relative to changes in drainage geometry. Hence, correlations be-

tween fish ecology and interdrainage dispersal might be stronger

in Australia, for example, than in the current New Zealand setting.

Indeed, levels of molecular divergence across the Great Dividing

Range in Australia appear to appreciably postdate the uplift of

this landscape and putative ages of drainage reorientation events

(Musyl and Keenan 1992; Rowland 1993; Musyl and Keenan

1996; Hurwood and Hughes 1998; McGlashan and Hughes 2001;

Thacker et al. 2007), lending support to a greater role of dispersal.

But in such geologically stable regions the major challenge is to

document (and preferably date) or discount past drainage reori-

entations, such that the role of intercatchment dispersal can be

determined. The analysis of greater numbers of taxa—preferably

closely related—will also be desirable to increase the strength of

ecological conclusions.

THE IMPORTANCE OF A GEOLOGICAL FRAMEWORK

IN FRESHWATER BIOLOGY

The incorporation of relevant geological information is an in-

creasingly important component of ecological and evolutionary

analysis. However, relatively few phylogeographic studies have

considered the possibility that genetic patterns suggestive of dis-

persal across drainage divides may actually reflect paleodrainage

connections (e.g., Fetzner and Crandall 2003; Kauwe et al. 2004),

or vice versa (e.g., Hurwood and Hughes 1998; Thacker et al.

2007). Although genetic patterns in certain taxa (e.g., odonatans,

anurans) might be most likely to reflect inter-catchment disper-

sal via aerial or terrestrial environments, without consideration of

paleohydrology the outcomes of future studies will merely reflect



these prior beliefs, and not represent unbiased assessments of in-

tercatchment dispersal for a species. It is also clear that failure to

control for river history may confound the ability to understand

the partitioning of genetic diversity within and among rivers (e.g.,

Stream Hierarchical model vs. the Death Valley or Headwater

models; Meffe and Vrijenhoek 1988; Finn et al. 2007), and the

effects of contemporary hydrology on instream genetic structur-

ing (e.g., Shaw et al. 1994; Turner and Trexler 1998; Castric et al.

2001; Costello et al. 2003; Poissant et al. 2005).

ACKNOWLEDGMENTSR. Allibone, S. Crow, N. Dunn, D. Rowe and the N.Z. Department ofConservation assisted with the collection of fish specimens. H. Edmonds,S. Charteris, and E. Edwards (N.Z. Department of Conservation) facili-tated the provision of collection permits. Marsden Contract UOO-0404(Royal Society of New Zealand) provided financial support for the re-search. Three anonymous reviewers provided comments that improvedthe manuscript.

LITERATURE CITEDArbogast, B. S., S. V. Edwards, J. Wakeley, P. Beerli, and J. B. Slowinski.

2002. Estimating divergence times from molecular data on phylogeneticand population genetic timescales. Annu. Rev. Ecol. Syst. 33:707–740.

Banarescu, P. 1990. Zoogeography of fresh waters. AULA-Verlag, Wiesbaden.Bazin, E., S. Glemin, and N. Galtier. 2006. Population size does not influence

mitochondrial genetic diversity in animals. Science 312:570–572.Bilton, D. T., J. R. Freeland, and B. Okamura. 2001. Dispersal in freshwater

invertebrates. Annu. Rev. Ecol. Syst. 32:159–181.Bishop, P. 1995. Drainage rearrangement by river capture, beheading and

diversion. Prog. Phys. Geog. 19:449–473.Burridge, C. P., D. Craw, and J. M. Waters. 2006. River capture, range expan-

sion, and cladogenesis: the genetic signature of freshwater vicariance.Evolution 60:1038–1049.

———. 2007. An empirical test of freshwater vicariance via river capture.Mol. Ecol. 16:1883–1895.

Burridge, C. P., D. Craw, D. Fletcher, and J. M. Waters. 2008. Geologicaldates and molecular rates: fish DNA sheds light on time-dependency.Mol. Biol. Evol. 25:624–633.

Carson, E. W., and T. E. Dowling. 2006. Influence of hydrogeographic historyand hybridization on the distribution of genetic variation in the pupfishesCyprinodon atrorus and C. bifasciatus. Mol. Ecol. 15:667–679.

Carstens, B. C., S. J. Brunsfeld, J. R. Demboski, J. M. Good, and J. Sullivan.2005. Investigating the evolutionary history of the Pacific Northwestmesic forest ecosystem: hypothesis testing within a comparative phylo-geographic framework. Evolution 59:1639–1652.

Carstens, B. C., and L. L. Knowles. 2007. Shifting distributions and speciation:species divergence during rapid climate change. Mol. Ecol. 16:619–627.

Castric, V., F. Bonney, and L. Bernatchez. 2001. Landscape structure andhierarchical genetic diversity in the brook charr, Salvelinus fontinalis.Evolution 55:1016–1028.

Cook, B. D., S. E. Bunn, and J. M. Hughes. 2007. Molecular genetic and stableisotope signatures reveal complementary patterns of population connec-tivity in the regionally vulnerable southern pygmy perch (Nannopercaaustralis). Biol. Conserv. 138:60–72.

Costello, A. B., T. E. Down, S. M. Pollard, C. J. Pacas, and E. B. Taylor.2003. The influence of history and contemporary stream hydrology onthe evolution of genetic diversity within species: an examination of mi-

crosatellite DNA variation in bull trout, Salvelinus confluentus (Pisces:Salmonidae). Evolution 57:328–344.

Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer Associates, SunderlandMA.

Craw, D., and R. J. Norris. 2003. Landforms. Pp. 17–34 in J. Darby, R.E. Fordyce, A. Mark, K. Probert and C. Townsend, eds. The natu-ral history of Southern New Zealand. Univ. of Otago Press, Dunedin,New Zealand.

Craw, D., and J. Waters. 2007. Geological and biological evidence for regionaldrainage reversal during lateral tectonic transport, Marlborough, NewZealand. J. Geol. Soc. Lond. 164:785–793.

Craw, D., J. H. Youngson, and P. O. Koons. 1999. Gold dispersal and placerformation in an active oblique collisional mountain belt, Southern Alps,New Zealand. Econ. Geol. Bull. Soc. 94:605–614.

Craw, D., L. Anderson, U. Rieser, and J. M. Waters. 2007a. Drainage reorienta-tion in Marlborough Sounds, New Zealand, during the Last Interglacial.NZ. J. Geol. Geophys. 50:13–20.

Craw, D., C. Burridge, L. Anderson, and J. M. Waters. 2007b. Late Quaternaryriver drainage and fish evolution, Southland, New Zealand. Geomorphol-ogy 84:98–110.

Crispo, E., P. Bentzen, D. N. Reznick, M. T. Kinnison, and A. P. Hendry. 2006.The relative influence of natural selection and geography on gene flowin guppies. Mol. Ecol. 15:49–62.

Crow, S. 2007. Evolutionary ecology of non-diadromous galaxiid fishes(Galaxias gollumoides and G. ‘southern’) in Southern New Zealand.Department of Zoology. Univ. of Otago, Dunedin, New Zealand.

Currens, K. P., C. B. Schreck, and H. W. Li. 1990. Allozyme and morphologicaldivergence of rainbow trout (Oncorhynchus mykiss) above and belowwaterfalls in the Deschutes River, Oregon. Copeia 1990: 730–746.

Dawson, M. N., K. D. Louie, M. Barlow, D. K. Jacobs, and C. C. Swift. 2002.Comparative phylogeography of sympatric sister species, Clevelandiaios and Eucyclogobius newberryi (Teleostei, Gobiidae), across the Cal-ifornia Transition Zone. Mol. Ecol. 11:1065–1075.

DeChaine, E. G., and A. P. Martin. 2006. Using coalescent simulations totest the impact of quaternary climate cycles on divergence in an alpineplant-insect association. Evolution 60:1004–1013.

Dowling, T. E., G. R. Smith, and W. M. Brown. 1989. Reproductiveisolation and introgression between Notropis cornutus and Notropischrysocephalus (family Cyprinidae)—comparison of morphology, al-lozymes, and mitochondrial DNA. Evolution 43:620–634.

Durand, J. D., A. R. Templeton, B. Guinand, A. Imsiridou, and Y. Bouvet.1999. Nested clade and phylogeographic analyses of the chub, Leuciscuscephalus (Teleostei, Cyprinidae), in Greece: implications for BalkanPeninsula biogeography. Mol. Phylogenet. Evol. 13:566–580.

Epifanio, J., and D. Philipp. 2000. Simulating the extinction of parental lin-eages from introgressive hybridization: the effects of fitness, initial pro-portions of parental taxa, and mate choice. Rev. Fish Biol. Fisher. 10:339–354.

Esa, Y. B., J. M. Waters, and G. P. Wallis. 2000. Introgressive hybridizationbetween Galaxias depressiceps and Galaxias sp D (Teleostei: Galaxi-idae) in Otago, New Zealand: secondary contact mediated by water races.Conserv. Genet. 1:329–339.

Fetzner, J. W., and K. A. Crandall. 2003. Linear habitats and the nested cladeanalysis: An empirical evaluation of geographic versus river distancesusing an Ozark crayfish (Decapoda: Cambaridae). Evolution 57:2101–2118.

Finn, D. S., M. S. Blouin, and D. A. Lytle. 2007. Population genetic structurereveals terrestrial affinities for a headwater stream insect. FreshwaterBiol. 52:1881–1897.

Froufe, E., S. Alekseyev, I. Knizhin, P. Alexandrino, and S. Weiss. 2003. Com-parative phylogeography of salmonid fishes (Salmonidae) reveals late to



post-Pleistocene exchange between three now-disjunct river basins inSiberia. Divers. Distrib. 9:269–282.

Fu, Y. X. 1997. Statistical tests of neutrality of mutations against populationgrowth, hitchhiking and background selection. Genetics 147:915–925.

Good, J. M., J. R. Demboski, D. W. Nagorsen, and J. Sullivan. 2003. Phylo-geography and introgressive hybridization: Chipmunks (genus Tamias)in the northern Rocky Mountains. Evolution 57:1900–1916.

Hamilton, W. J., and R. Poulin. 2001. Parasitism, water temperature and lifehistory characteristics of the freshwater fish Gobiomorphus brevicepsStokell (Eleotridae). Ecol. Freshw. Fish 10:105–110.

Hamilton, W. J., M. K. Stott, and R. Poulin. 1997. Nest site characteristics andmale reproductive success in the upland bully, Gobiomorphus breviceps(Eleotridae). Ecol. Freshw. Fish 6:150–154.

Hamilton, S. K., S. J. Sippel, and J. M. Melack. 2002. Comparison of inun-dation patterns among major South American floodplains. J. Geophys.Res.-Atmos. 107:D208038.

Hickerson, M. J., and C. W. Cunningham. 2005. Contrasting Quaternary his-tories in an ecologically divergent sister pair of low-dispersing intertidalfish (Xiphister) revealed by multilocus DNA analysis. Evolution 59:344–360.

Ho, S. Y. W., and G. Larson. 2006. Molecular clocks: when times are a-changin’. Trends Genet. 22:79–83.

Ho, S. Y. W., M. J. Phillips, A. Cooper, and A. J. Drummond. 2005. Timedependency of molecular rate estimates and systematic overestimationof recent divergence times. Mol. Biol. Evol. 22:1561–1568.

Hopkins, C. L. 1971. Life history of Galaxias divergens (Salmonoidea: Galaxi-idae). NZ J. Mar. Fresh. 5:41–57.

Hudson, R. R., and M. Turelli. 2003. Stochasticity overrules the “three-timesrule”: genetic drift, genetic draft, and coalescence times for nuclear lociversus mitochondrial DNA. Evolution 57:182–190.

Hurwood, D. A., and J. M. Hughes. 1998. Phylogeography of the freshwaterfish, Mogurnda adspersa, in streams of northeastern Queensland, Aus-tralia: evidence for altered drainage patterns. Mol. Ecol. 7:1507–1517.

———. 2001. Nested clade analysis of the freshwater shrimp, Caridina zebra(Decapoda: Atyidae), from north-eastern Australia. Mol. Ecol. 10:113–125.

Jowett, I. G., and J. Richardson. 1995. Habitat preferences of common, riverineNew Zealand native fishes and implications for flow management. NewZeal. J. Mar. Fresh. 29:13–23.

Kauwe, J. S. K., D. K. Shiozawa, and R. P. Evans. 2004. Phylogeographic andnested clade analysis of the stonefly Pteronarcys californica (Plecoptera:Pteronarcyidae) in the western USA. J. N. Am. Benthol. Soc. 23:824–838.

King, T. M., and G. P. Wallis. 1998. Fine-scale genetic structuring in endemicgalaxiid fish populations of the Taieri River. NZ J. Zool. 25:17–22.

Knowles, L. L. 2004. The burgeoning field of statistical phylogeography. J.Evol. Biol. 17:1–10.

Knowles, L. L., and B. C. Carstens. 2007. Estimating a geographically explicitmodel of population divergence. Evolution 61:477–493.

Knowles, L. L., and W. P. Maddison. 2002. Statistical phylogeography. Mol.Ecol. 11:2623–2635.

Kocher, T. D., W. K. Thomas, A. Meyer, S. V. Edwards, S. Paabo, F. X. Vil-lablanca, and A. C. Wilson. 1989. Dynamics of mitochondrial DNAevolution in animals—amplification and sequencing with conservedprimers. Proc. Natl. Acad. Sci. USA 86:6196–6200.

Kreiser, B. R., J. B. Mitton, and J. D. Woodling. 2001. Phylogeography of theplains killifish, Fundulus zebrinus. Evolution 55:339–350.

Kuhner, M. K., J. Yamato, and J. Felsenstein. 1995. Estimating effective pop-ulation size and mutation rate from sequence data using Metropolis-Hastings sampling. Genetics 140:1421–1430.

Lu, G., D. J. Basley, and L. Bernatchez. 2001. Contrasting patterns of mito-chondrial DNA and microsatellite introgressive hybridization betweenlineages of lake whitefish (Coregonus clupeaformis); relevance for spe-ciation. Mol. Ecol. 10:965–985.

Maddison, W. P. 2006. Coalescence package for Mesquite. Version 1.1.http://mesquiteproject.org.

Maddison, D. R., and W. P. Maddison. 2005. Batch Architect: automation ofsimulations and replicated analyses. A package of modules for Mesquite.Version 1.06.

Maddison, W. P., and D. R. Maddison. 2006. Mesquite: a modular system forevolutionary analysis. Version 1.11 http://mesquiteproject.org.

Marchetti, M. P., P. B. Moyle, and R. Levine. 2004. Alien fishes in Californiawatersheds: characteristics of successful and failed invaders. Ecol. Appl.14:587–596.

McAlpin, J. P. 1992. Glacial geology of the upper Wairau Valley, Marlborough,New Zealand. NZ J. Geol. Geop. 35:211–222.

McDowall, R. M. 2000. The reed field guide to New Zealand freshwater fishes.Reed, Auckland New Zealand.

McDowall, R. M., and W. L. Chadderton. 1999. Galaxias gollumoides(Teleostei: Galaxiidae), a new fish species from Stewart Island, withnotes on other non-migratory freshwater fishes present on the island. J.R. Soc. New Zeal. 29:77–88.

McGlashan, D. J., and J. M. Hughes. 2000. Reconciling patterns of geneticvariation with stream structure, earth history and biology in the Aus-tralian freshwater fish Craterocephalus stercusmuscarum (Atherinidae).Mol. Ecol. 9:1737–1751.

———. 2001. Genetic evidence for historical continuity between popula-tions of the Australian freshwater fish Craterocephalus stercusmuscarum(Atherinidae) east and west of the Great Dividing Range. J. Fish Biol.59:55–67.

McIntosh, P. D., D. N. Eden, and S. J. Burgham. 1990. Quaternary deposits andlandscape evolution in northeast Southland, New Zealand. Palaeogeogr.Palaeocol. 81:95–113.

Meffe, G. K., and R. C. Vrijenhoek. 1988. Conservation genetics in the man-agement of desert fishes. Conserv. Biol. 2:157–169.

Meyer, A., J. M. Morrissey, and M. Schartl. 1994. Recurrent origin of a sexuallyselected trait in Xiphophorus fishes inferred from a molecular phylogeny.Nature 368:539–542.

Musyl, M. K., and C. P. Keenan. 1992. Population genetics and zoogeographyof Australian fresh-water Golden Perch, Macquaria ambigua (Richard-son 1845) (Teleostei, Percichthyidae), and electrophoretic identificationof a new species from the Lake Eyre basin. Aust. J. Mar. Fresh. Res.43:1585–1601.

———. 1996. Evidence for cryptic speciation in Australian freshwater eel-tailed catfish, Tandanus tandanus (Teleostei: Plotosidae). Copeia. 526–534.

Nei, M. 1987. Molecular evolutionary genetics. Columbia Univ. Press, NewYork.

Neigel, J. E., and J. C. Avise. 1987. Phylogenetic relationships of mitochondrialDNA under various demographic models of speciation. Pp. 515–534 inE. Nevo and S. Karlin, eds. Evolutionary processes and theory. AcademicPress, New York.

Nielsen, R., and J. Wakeley. 2001. Distinguishing migration from isolation: aMarkov chain Monte Carlo approach. Genetics 158:885–896.

Poissant, J., T. W. Knight, and M. M. Ferguson. 2005. Nonequilibrium con-ditions following landscape rearrangement: the relative contribution ofpast and current hydrological landscapes on the genetic structure of astream-dwelling fish. Mol. Ecol. 14:1321–1331.

Posada, D., and K. A. Crandall. 1998. MODELTEST: testing the model ofDNA substitution. Bioinformatics 14:817–818.



Paabo, S., K. Dew, B. S. Frazier, and R. H. Ward. 1990. Mitochondrial evolu-tion and the peopling of the America. Am. J. Phys. Anthropol. 81:277–277.

Ramos-Onsins, S. E., and J. Rozas. 2002. Statistical properties of new neu-trality tests against population growth. Mol. Biol. Evol. 19:2092–2100.

Rosenberg, N. A. 2003. The shapes of neutral gene genealogies in two species:probabilities of monophyly, paraphyly, and polyphyly in a coalescentmodel. Evolution 57:1465–1477.

Rowland, S. J. 1993. Maccullochela ikei, an endangered species of freshwatercod (Pisces: Percichthyidae) from the Clarence River system, NSW andM. peelii mariensis, a new subspecies from the Mary River system, Qld.Rec. Aust. Mus. 45:121–145.

Roy, D., D. W. Kelly, C. Fransen, D. D. Heath, and G. D. Haffner. 2006.Evidence of small-scale vicariance in Caridina lanceolata (Decapoda:Atyidae) from the Malili Lakes, Sulawesi. Evol. Ecol. Res. 8:1087–1099.

Rozas, J., J. C. Sanchez-DelBarrio, X. Messeguer, and R. Rozas. 2003. DnaSP,DNA polymorphism analyses by the coalescent and other methods.Bioinformatics 19:2496–2497.

Ruber, L., A. Meyer, C. Sturmbauer, and E. Verheyen. 2001. Populationstructure in two sympatric species of the Lake Tanganyika cichlid tribeEretmodini: evidence for introgression. Mol. Ecol. 10:1207–1225.

Schneider, S., D. Roessli, and L. Excoffier. 2000. Arlequin ver. 2.001: a soft-ware for population genetics data analysis. Genetics and Biometry Lab-oratory. Univ. of Geneva, Switzerland. (http://lgb.unige.ch/arlequin/)

Shaw, P. W., G. R. Carvalho, A. E. Magurran, and B. H. Seghers. 1994. Factorsaffecting the distribution of genetic variability in the guppy, Poeciliareticulata. J. Fish Biol. 45:875–888.

Slatkin, M., and W. P. Maddison. 1989. A cladistic measure of gene flowinferred from the phylogenies of alleles. Genetics 123:603–613.

Smith, S. A., and E. Bermingham. 2005. The biogeography of lowerMesoamerican freshwater fishes. J. Biogeogr. 32:1835–1854.

Smith, P. J., S. M. McVeagh, and R. Allibone. 2003. The Tarndale bully re-visited with molecular markers: an ecophenotype of the common bullyGobiomorphus cotidianus (Pisces: Gobiidae). J. R. Soc. NZ. 33:663–673.

———. 2005. Extensive genetic differentiation in Gobiomorphus brevicepsfrom New Zealand. J. Fish Biol. 67:627–639.

Sousa-Santos, C., M. J. Collares-Pereira, and V. C. Almada. 2006. Evidenceof extensive mitochondrial introgression with nearly complete substi-tution of the typical Squalius pyrenaicus-like mtDNA of the Squaliusalburnoides complex (Cyprinidae) in an independent Iberian drainage.J. Fish Biol. 68:292–301.

Staples, D. J. 1975. Production biology of upland bully Philypnodon brevicepsStokell in a small New Zealand lake.1. Life-history, food, feeding andactivity rhythms. J. Fish Biol. 7:1–24.

Tajima, F. 1989. Statistical method for testing the neutral mutation hypothesisby DNA polymorphism. Genetics 123:585–595.

Taylor, E. B., M. D. Stamford, and J. S. Baxter. 2003. Population subdivisionin westslope cutthroat trout (Oncorhynchus clarki lewisi) at the northernperiphery of its range: evolutionary inferences and conservation impli-cations. Mol. Ecol. 12:2609–2622.

Thacker, C. E., P. J. Unmack, L. Matsui, and N. Rifenbark. 2007. Compara-tive phylogeography of five sympatric Hypseleotris species (Teleostei:Eleotridae) in south-eastern Australia reveals a complex pattern of

drainage basin exchanges with little congruence across species. J. Bio-geogr. 34:1518–1533.

Tibbets, C. A., and T. E. Dowling. 1996. Effects of intrinsic and extrinsicfactors on population fragmentation in three species of North Americanminnows (Teleostei: Cyprinidae). Evolution 50:1280–1292.

Turnbull, I. M. 1980. Sheet E42AC—Walter Peak (West). Geological Map ofNew Zealand 1:50 000. Department of Scientific and Industrial Research.Wellington, New Zealand.

———. 2000. Geology of the Wakatipu Area. New Zealand Institute of Geo-logical and Nuclear Sciences Limited. Lower Hutt, New Zealand.

Turnbull, I. M., and A. H. Allibone. 2003. Geology of the Murihiku area.Institute of Geological and Nuclear Sciences. Lower Hutt, New Zealand

Turner, T. F., and H. W. Robison. 2006. Genetic diversity of the caddo mad-tom, Noturus taylori, with comments on factors that promote geneticdivergence in fishes endemic to the Ouachita highlands. Southwest. Nat.51:338–345.

Turner, T. F., and J. C. Trexler. 1998. Ecological and historical associations ofgene flow in darters (Teleostei: Percidae). Evolution 52:1781–1801.

Turner, T. F., J. C. Trexler, D. N. Kuhn, and H. W. Robison. 1996. Life-historyvariation and comparative phylogeography of darters (Pisces: Percidae)from the North American central highlands. Evolution 50:2023–2036.

Unmack, P. J. 2001. Biogeography of Australian freshwater fishes. J. Biogeogr.28:1053–1089.

Wakeley, J., and J. Hey. 1997. Estimating ancestral population parameters.Genetics 145:847–855.

Waters, J. M., and G. P. Wallis. 2001a. Cladogenesis and loss of the marine life-history phase in freshwater galaxiid fishes (Osmeriformes: Galaxiidae).Evolution 55:587–597.

———. 2001b. Mitochondrial DNA phylogenetics of the Galaxias vulgariscomplex from South Island, New Zealand: rapid radiation of a speciesflock. J. Fish Biol. 58:1166–1180.

Waters, J. M., M. Lintermans, and R. W. G. White. 1994. Mitochondrial DNAvariation suggests river capture as a source of vicariance in Gadopsisbispinosus (Pisces, Gadopsidae). J. Fish Biol. 44:549–551.

Waters, J. M., D. Craw, J. H. Youngson, and G. P. Wallis. 2001a. Genesmeet geology: fish phylogeographic pattern reflects ancient, rather thanmodern, drainage connections. Evolution 55:1844–1851.

Waters, J. M., Y. B. Esa, and G. P. Wallis. 2001b. Genetic and morphologicalevidence for reproductive isolation between sympatric populations ofGalaxias (Teleostei: Galaxiidae) in South Island, New Zealand. Biol. J.Linn. Soc. 73:287–298.

Waters, J. M., D. L. Rowe, S. Apte, T. M. King, G. P. Wallis, L. Anderson,R. J. Norris, D. Craw, and C. P. Burridge. 2007. Geological dates andmolecular rates: rapid divergence of rivers and their biotas. Syst. Biol.56:271–282.

Whiteley, A. R., P. Spruell, and F. W. Allendorf. 2004. Ecological and lifehistory characteristics predict population genetic divergence of twosalmonids in the same landscape. Mol. Ecol. 13:3675–3688.

Williams, C. F. 1994. Genetic consequences of seed dispersal in three sym-patric forest herbs. II. Microspatial genetic structure within populations.Evolution 48:1959–1972.

Associate Editor: K. Crandall



Appendix. Details of freshwater-limited fish analyzed from the Mararoa and Oreti rivers in South Island, New Zealand. Map references

correspond to the NZMS 260 Topographic Series 1:50000 (easting, northing).

Collection Zone Coordinates Site Galaxias Gobiomorphuscode “southern” gollumoides paucispondylus breviceps

MARAROAElm Tree (A) 2111300 5499300 Aa 6

2109500 5505500 Ab 4 1 6Princhester (B) 2114400 5506800 Ba 5 8 6

2114400 5507200 Bb 6Station Br (C) 2117400 5510700 Ca 7

2117800 5510800 Cb 72117800 5510800 Cc 5 52116900 5509000 Cd 7

Haycock (D) 2120600 5512100 Da 222121900 5512800 Db 1 6 12121700 5512800 Dc 5 5

Wash (E) 2127000 5518000 Ea 13 52125500 5520500 Eb 82122900 5514000 Ec 3

Hikuraki (F) 2127400 5527000 Fa 5 5 5Boundary Hut (G) 2131100 5552300 Ga 9 4 6

2131500 5552200 Gb 2Upper (H) 2133300 5553800 Ha 12 4 5

ORETIMossburn (I) 2139900 5494600 Ia 2

2140000 5494600 Ib 5 6 62136900 5496200 Ic 4

Morris (J) 2135100 5502800 Ja 13 112131900 5502100 Jb 7 13

Windley (K) 2130300 5508500 Ka 82132433 5507407 Kb 32131695 5507108 Kc 2 22132600 5507500 Kd 5 8

Centre Hill (L) 2133500 5507800 La 22129232 5519241 Lb 22129798 5517643 Lc 52128200 5515900 Ld 3 3 11

Upper Centre Hill (M) 2129300 5518700 Ma 92131500 5522800 Mb 4 5 9

Ashton Burn (N) 2131900 5522200 Na 1 12136106 5527184 Nb 22134900 5527900 Nc 2 5 4 6

Mt Nicholas Br (O) 2135400 5527600 Oa 22134679 5530581 Ob 92134100 5530700 Oc 72134400 5532200 Od 3 5 7

Gorge Burn (P) 2134500 5531400 Pa 112138000 5534200 Pb 5 72138000 5534300 Pc 13

TOTAL 114 103 117 103



Supplementary MaterialThe following supplementary material is available for this article:

Figure S1. Model employed during statistical phylogeography.

Figure S2. Marginal posterior distributions of population divergence time (green line) and migration rate (red line = Mararoa to

Oreti; blue line = Oreti to Mararoa) parameters from isolation with migration analysis. Analyses repeated with equal ancestral

and contemporary θ, and forward and reverse migration rates, are indicated by black (migration rate) and grey (population

divergence time) lines.

Figure S3. Statistical parsimony network of Gobiomorphus breviceps haplotypes (mtDNA cyt b and CR) from the Mararoa and

Oreti catchments.

This material is available as part of the online article from:

http://www.blackwell-synergy.com/doi/abs/10.1111/j.1558-5646.2008.00377.x

(This link will take you to the article abstract).

Please note: Blackwell Publishing is not responsible for the content or functionality of any supplementary materials supplied by

the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.


does fish ecology predict dispersal across …...2e-mail: [email protected]...

Documents