Rapid and convergent evolution of parental care in hydrobiid gastropods from New Zealand


  • M. HAASE

    1. National Institute of Water and Atmospheric Research, Centre for Biodiversity and Ecology Research, Department of Biological Sciences, The University of Waikato, Hamilton, New Zealand
    Search for more papers by this author

Dr. Martin Haase, Muséum National d'Histoire Naturelle, Département Systématique et Évolution, UMS 2700 Taxonomie-Collections [Mollusques], case postale n ° 51, 55 rue Buffon, F-75231 Paris Cedex 05, France.
Tel.: 00 33 (0)1 40 79 31 04; fax: 00 33 (0)1 40 79 57 71;
e-mail: martin_haase@excite.com


Although parental care occurs in most phyla encompassing a wide array of forms, little is known about its evolution in invertebrates. Two types of egg capsules have been known among ovoviviparous New Zealand hydrobiid gastropods, elastic capsules and simple membranes. Based on a phylogenetic analysis using two mtDNA sequence fragments, I asked whether the second state was derived from the first or whether brooding had multiple origins. The evolution of ovoviviparity was also investigated in the context of habitat transition between brackish and freshwater. Maximum parsimony and Markov chain models of character state transformations in a maximum likelihood framework suggested that hydrobiids have invaded freshwater three times independently. Two of these invasions were followed by the evolution of ovoviviparity, probably in adaptation to changing water levels during periods of irregular precipitation. The syntopy of two congeneric species, one oviparous and the other one brooding, indicated that the transition between reproductive modes must have occurred rapidly.


Parental care encompasses a wide array of behavioral, morphological, and physiological traits, which have evolved in most phyla in order to increase the offspring's chances of survival. However, most investigations have focused on vertebrates and many invertebrate taxa remain un- or understudied with respect to ecology and evolution of parental care (Clutton-Brock, 1991). In gastropods, parental care includes provisioning embryos with nurse eggs, brooding eggs under the foot, carrying egg capsules attached to the shell, preparation of oviposition sites, retention of eggs in the mantle cavity, ovoviviparity, and viviparity (Purchon, 1977; Tompa et al., 1984; Baur, 1994). Ovoviviparity occurs in various aquatic and terrestrial gastropods (Tompa et al., 1984) and has apparently evolved numerous times independently. In general, ovoviviparous species often live under ecologically harsher conditions than their egg-laying relatives (Clutton-Brock, 1991). The transition from egg laying to ovoviviparity involves changes in maternal morphology, physiology, and behavior. It is thus surprising that such complex changes have occurred several times convergently. Even closely related taxa such as different, notably freshwater species of the Cochliopidae (Hershler & Thompson, 1992) or Pachychilidae (Köhler & Glaubrecht, 2003; Köhler et al., 2004) appear to have switched to brooding independently. However, these hypotheses have either not been rigorously tested using phylogenetic and comparative methods (Cochliopidae), or the analysis is somewhat ambiguous (Pachychilidae). Based on the criterion of maximum parsimony, Köhler et al. (2004) concluded that ovoviviparity has evolved three times independently within Asian Pachychilidae, but they did not discuss the equally parsimonious solution including a single origin of brooding and two reversals to egg laying.

Among gastropods of the family Hydrobiidae from New Zealand, a monophyletic radiation (Haase, unpublished data), ovoviviparity is the predominant mode of reproduction. Two types of egg capsules can be distinguished. In most species the capsule is reduced to a thin membrane without defined shape, whereas the brooding species of the genus Potamopyrgus have elastic, ovoid egg capsules (Haase, unpublished data). It is tempting to assume that the latter is the plesiomorphic state from which the membranous condition has been derived. Alternatively, ovoviviparity may have evolved along two different evolutionary pathways. According to Climo's (1974, 1977) classification, which was not based on a phylogenetic analysis, ovoviviparity would even have evolved three times independently among New Zealand hydrobiids.

Practically all ovoviviparous hydrobiids from New Zealand occur in freshwater, mainly in springs, small streams or groundwater habitats. Only one brooding species, the ubiquitous Potamopyrgus antipodarum, has a broad ecological valence occurring in all types of freshwater as well as brackish lagoons and estuaries. The four brackish water species, on the other hand, are oviparous. This suggests that reproductive mode and habitat are evolutionarily linked. Brooding may be conceived as preadaptation for marine or brackish water species to invade osmotically challenging limnetic habitats as originally suggested for thiarid and pachychilid gastropods (Glaubrecht, 1996; but see Köhler & Glaubrecht, 2003). Alternatively, ovoviviparity might be advantageous in habitats with unstable conditions such as fluctuating discharge and water level in springs and streamlets.

In order to investigate whether brooding has evolved more than once among New Zealand hydrobiid gastropods, a phylogenetic analysis was conducted based on sequence data of two mitochondrial fragments, viz. cytochrome oxidase subunit I (COI) and 16S rDNA. Ancestral states were reconstructed applying the criteria of maximum parsimony (MP) and maximum likelihood (ML) and the resulting scenarios compared in a ML framework developed by Pagel (1994, 1999). Similarly, I analysed habitat transitions between brackish and freshwater and examined, whether transitions between modes of reproduction and habitats were evolutionarily correlated. Furthermore, the temporal sequence of the transitions was inferred (Pagel, 1994).

Material and methods

Taxonomic remarks

Most freshwater gastropods belonging to the Rissooidea and their related brackish water species from around the world were once regarded as monophyletic group, viz. Hydrobiidae or Hydrobioidea (see Wilke et al., 2001). Based on detailed morphological investigations, Davis (1979) recognized the polyphyletic origin of the group and coined the term hydrobioid to denote taxa similar but not necessarily related to Hydrobia. Later, sequence data have demonstrated deep splits between selected hydrobioid lineages and confirmed the notion of polyphyly (Wilke et al., 2001). The hydrobioid radiation of New Zealand does almost certainly not belong to the Hydrobiidae sensu strictu. However, in the absence of a sound suprageneric classification I continue to use the name Hydrobiidae for the New Zealand taxa for the sake of communication.

In a recent revision of New Zealand hydrobiid gastropods (Haase, unpublished data), a total of 64 species belonging to 15 genera were recognized. Forty-six species and nine genera were newly introduced. In the present account, the new taxa are formally treated as nomina nuda.

Sampling and DNA extraction

The comprehensive phylogenetic analysis was based on sequence fragments of COI and 16S rDNA from one specimen each of 27 species representing all but three genera [these specimens were selected from the foregoing systematic account (Haase, unpublished data) including two or more animals per species where available]. In two additional species, P. doci and P. troglodytes, only the 16S rDNA fragment could be sequenced. These two species and P. oppidanus, morphologically distinct and each known from a single locality on the North Island (Haase, unpublished data), were genetically very similar to the ubiquitous P. antipodarum, which was also associated with P. doci and P. oppidanus. For a separate analysis of these species (see below) based on 16S rDNA, in which P. antipodarum was represented by seven populations from the North Island, South Island and Stewart Island, up to four specimens per population were used. Three and four specimens, respectively, were also sequenced – again the 16S rDNA fragment – of the sympatric sister species Opacuincola johannstraussi and O. josefstraussi occurring only in two adjacent caves in the north of the South Island. Most New Zealand species have very restricted ranges and several are known from only very few sites or even a single spring (Haase, unpublished data), a pattern well known from hydrobioid radiations in other parts of the world (e.g. Radoman, 1983; Hershler & Thompson, 1992; Ponder et al., 1993; Haase & Bouchet, 1998). The exception among the freshwater species is P. antipodarum, which occurs throughout New Zealand as well as on its offshore islands. In addition, P. antipodarum is extremely polymorphic (see Haase, 2003) and occurs in diploid, outcrossing and triploid, parthenogenetic strains (e.g. Dybdahl & Lively, 1995). Locality data for the samples analysed are given in Table 1 and more details in Haase (unpublished data).

Table 1.  Specimens analysed genetically. Numbers following acronyms indicate specimens: e.g. PantipodarumWa124 is short for specimens 1, 2 and 4 of P. antipodarum from Wadestown. More detailed locality data and catalogue numbers of voucher specimens deposited at the Museum of New Zealand Te Papa Tongarewa are given in Haase (unpublished data).
AcronymSpeciesLocalityGenBank acces. no.
CmatapangoCr1Catapyrgus matapangoCrazy Paving CaveAY631072AY634050
HngataanaMa1Hadopyrgus ngataanaMaitai CaveAY631073AY634051
HpagodulusSp1Halopyrgus pagodulusSpirits BayAY631113AY634091
HpupoidesSp1Halopyrgus pupoidesSpirits BayAY631117AY634095
LmanneringiWa3Leptopyrgus manneringiWaikaretuAY631074AY634052
LmelbourniAw1Leptopyrgus melbourniAwakino GorgeAY631075AY634053
LtainuiKa1Leptopyrgus tainuiKawhiaAY631077AY634055
MmuaukauDu1Meridiopyrgus muaupokoDunedinAY631083AY634061
MmurihikuBr1Meridiopyrgus murihikuBrownsAY631084AY634062
OalpinusBr1Obtusopyrgus alpinusBroken River Ski Club RoadAY631088AY634066
OdeliraCr1Opacuincola deliraCrazy Paving CaveAY631090AY634068
OdulcinellaPo1Opacuincola dulcinellaPohara BeachAY631091AY634069
OjohannstraussiBa1O. johannstraussiBallroom CaveAY631093AY634071
OjohannstraussiBa2  AY634072
OjohannstraussiBa4  AY641543
OjosefstraussiBa1Opacuincola josefstraussiBallroom CaveAY631095AY634073
OjosefstraussiBa23  AY634074
OjosefstraussiBa4  AY634110
OmeteRe1Opacuincola meteReeftonAY631097AY634075
OngatapunaAn1Opacuincola ngatapunaAnatori RiverAY631098AY634076
OpermutataKe2Opacuincola permutataKennedy's SinkAY631100AY634078
PantipodarumBr1Potamopyrgus antipodarumBrownsAY634104
PantipodarumHo12 Horseshoe BayAY634106
PantipodarumMor1 Morere SpringsAY631101AY634079
PantipodarumMot1 Motu RiverAY631102AY634080
PantipodarumRu23 Ruakuri CaveAY634109
PantipodarumTu1 N of Awakino GorgeAY634090
PantipodarumWa124 WadestownAY634090
PantipodarumWa3  AY634107
PdociRu13Potamopyrgus dociRuakuri CaveAY634105
PdociRu2  AY634108
PestuarinusRa1Potamopyrgus estuarinusRaglanAY631103AY634081
PkaitunuparaoaMo1Potamopyrgus kaitunuparaoaMokauAY631105AY634083
PnanumMa2Paxillostium nanumN of Mata RoadAY631111AY634089
PoppidanusWa234Potamopyrgus oppidanusWadestownAY631112AY634090
PtroglodytesAr123Potamopyrgus troglodytesAriaAY634103
RcresswelliGo2Rakiurapyrgus cresswelliGolden BayAY631081AY634059
SkutukutuTu1Sororipyrgus kutukutuN of Awakino GorgeAY631109AY634087
SmarshalliPi1Sororipyrgus marshalliPipinui ReserveAY631110AY634088
SrakiSe1Sororipyrgus rakiSE of TwinbridgesAY631120AY634098
TkohitateaSp2Tongapyrgus kohitateaSpring HillAY631124AY634102

DNA was extracted from homogenates of whole snails using the DNeasy Tissue Kit of QIAGEN. A 658 bp fragment of the mitochondrial subunit I of the cytochrome oxidase (COI) was amplified using Folmer et al.'s (1994) primers LCO1490 and HCO2198, the latter modified after Wilke & Davis (2000). A second, ca. 500 bp long mitochondrial fragment of the 16S rDNA was amplified with the primer pair 16Sar-L and 16Sbr-H (Palumbi et al., 1991). PCR volumes comprised 50 μL containing 20–100 ng of genomic DNA, 4.5 μL of each 10X buffer (Roche) and MgCl2 (25 mM), 10 μL dNTP (1 mM), 2 μL BSA (10 mg mL−1), 1.5 μL of each primer (10 μM), 1 μL = 1 unit Taq polymerase (Roche), and H2O. The PCR conditions were an initial denaturation for 2 min at 94 °C followed by 40 cycles comprising denaturation for 1 min at 94 °C, annealing for 1.5 min at 48 °C (COI) and 45 °C (16S rDNA), respectively, and extension for 1 min at 72 °C. The final step was an additional extension for 5 min at 72 °C. PCR products were cleaned with QIAGEN's QIAquick PCR Purification Kit. If secondary products were present, the fragments of desired length were excised from 1% agarose gels and cleaned using the MinElute Gel Extraction kit of QIAGEN. For sequencing on a MegaBACE 500 DNA Analysis System (Amersham Biosciences, Little Chalfont, UK) PCR products were submitted to the sequencing facility of the University of Waikato. Sequence editing and alignment of COI fragments were performed using BioEdit (Hall, 1999). 16S rDNA sequences were aligned in ClustalX 1.81 (Thompson et al., 1997) using default settings. Subsequently, this alignment was manually refined in BioEdit. The final alignment comprised 644 bp of COI and 486 bp of 16S rDNA. All sequences were submitted to GenBank (Table 1). Tatea sp. from southeastern Australia and Hemistomia winstonefi from New Caledonia served as outgroup.

Sequence analysis

Base frequencies were homogeneous across the whole set as well as within each partition. These tests included the outgroup but constant sites were removed (whole set: inline image = 100.91, n.s.; COI: inline image = 97.43, n.s.; 16S rDNA: inline image = 48.32, n.s.). Thus, the condition for substitution model based phylogenetic reconstruction methods, viz. stationarity, was not violated. I conducted maximum likelihood (ML) and Bayesian analyses (BA) using the programs PAUP*4.0b10 (Swofford, 1998) and MrBayes 3.0b4 (Huelsenbeck & Ronquist, 2001), respectively. The substitution models for ML and BA were fitted to the partitions and the whole set using Modeltest 3.5, which compares 56 models applying hierarchical likelihood ratio tests (Posada & Crandall, 1998). The models selected are given in Table 2. The optimal ML tree was searched for heuristically applying TBR branch-swapping in ten replicates. The starting tree was obtained via stepwise addition and the starting seed for the random addition sequence was 1580119401. This setup was run with and without molecular clock enforced. Clock-like sequence evolution was tested by comparing the likelihood scores with a likelihood ratio test. Computation time for bootstrapping was prohibitive. The models selected for the separate partitions by Modeltest are special cases of the general time reversible (GTR) model. Therefore, separate GTR models were estimated for each partition in the BA by MrBayes. Four simultaneous chains were run over 106 generations of which every 100th was sampled. The likelihood scores converged after about 60 000 generations. Conservatively, the first 1000 samples were discarded for the computation of the 50% majority rule consensus (MJR) tree. Combinability of the partitions was indicated by the lack of conflicting groupings supported by a posterior probability of >95% each when analyzing the partitions separately. Saturation of substitutions at third codon positions of the COI partition was controlled for by comparing Xia et al.'s (2003) indices using the program DAMBE (Xia & Xie, 2001). A symmetrical tree shape was assumed and the proportion of constant sites (0.0744) used as a proxy for the proportion of invariant sites.

Table 2.  Optimal models of substitution fitted to partitions and whole data set by Modeltest 3.0. Pinvar, proportion of invariant sites.
PartitionModel –lnLBase frequenciesSubstitution ratesPinvar (I) Gamma (∝)
COITVM + I + GA0.3045A-C0.8450I = 0.5866
5100.99C0.1111A-G33.7411∝ = 0.7915
16S rDNATIM + I + GA0.3613A-C1.0000I = 0.4635
2378.65C0.1317A-G10.3151∝ = 0.4104
COI + 16S rDNAGTR + I + GA0.3336A-C0.6547I = 0.5733
7566.45C0.1140A-G19.1489∝ = 0.5945

For the group of species very closely related to P. antipodarum a network was reconstructed based on the 16S rDNA fragment applying statistical parsimony (Templeton et al., 1992) using the program TCS (Clement et al., 2000).

Parental care and habitat transition

The evolution of ovoviviparity was analysed using two approaches, both ignoring the outgroup. First, the character states brooding/nonbrooding (Fig. 1) were optimized onto the MJR tree resulting from BA according to the criterion of maximum parsimony (MP). The second approach to reconstruct ancestral states used Markov models of discrete character state transitions and the maximum likelihood framework implemented in the program Discrete 4.0 developed by Pagel (1994,1999). For these analyses O. permutata, which formed a polytomy with O. dulcinella and (O. ngatapuna, (O. johannstraussi, O. josefstraussi)), had to be pruned from the tree manually, because Discrete cannot handle polytomies. Since O. permutata is ovoviviparous like all its congeners this omission did not influence the analyses. The most likely set of ancestral character states was determined globally (independent menu) as well as locally at each separate node (graphics menu). For the question of how often ovoviviparity has evolved among New Zealand hydrobiids, the likelihoods of different scenarios were compared by fixing the ancestral state at respective nodes accordingly. These analyses did not involve the nonbrooding P. doci, for which only 16S rDNA sequences were available.

Figure 1.

Phylogeny of hydrobiid gastropods from New Zealand. Bayesian analysis based on an alignment of sequence fragments of COI and 16S rDNA comprising 1130 bp. The outgroup was pruned from the tree. Posterior probabilities (pp) are given above branches, root pp = 100. Nodes are numbered according to output of Discrete. Acronyms of species are given in Table 1. Symbols denote alternative hypotheses of character evolution comprising 2–4 steps: reproductive mode is indicated in grey above branches, habitat in black below branches. Scenarios with more than four transitions between habitats are only discussed in the text. Hypotheses denoted by squares were the most likely reconstructions of character state evolution.

After an analogous analysis of habitat transitions [states indicated in Fig. 1; P. antipodarum was scored as freshwater species, because this is certainly its preferred habitat (Winterbourn, 1970; see also Discussion)], I tested whether reproductive mode and habitat transition were evolutionarily correlated and inferred the evolutionary pathway by comparing the transition rates (Pagel, 1994). In case alternative hypotheses were nested, significance was determined with likelihood ratio tests. In the analysis of correlated evolution, differences of transition rates were tested for significance by comparing the likelihoods of the full model and a model where the transition rates in question were set equal (Pagel, 1994). For the reconstruction of ancestral states, a difference between the likelihood scores (∂L) exceeding 2 log units was considered significant support at α = 0.05 as preferring one state over the other at a certain node (Pagel, 1999).


Phylogenetic reconstruction

The observed index of substitutions was significantly smaller than the critical index (Iss = 0.55, Iss.c = 0.68, P < 0.01). This indicates only little saturation not detrimental to phylogeny reconstruction. BA and ML resulted in almost identical tree topologies (Fig. 1). Halopyrgus was the basal-most clade. (Potamopyrgus, Sororipyrgus) was sister clade to the remaining genera, which formed two lineages consisting of Rakiurapyrgus, Catapyrgus, Meridiopyrgus, and Opacuincola, and Hadopyrgus, Paxillostium, Leptopyrgus, Obtusopyrgus, and Tongapyrgus, respectively. The only difference between BA and ML was the relationship of Obtusopyrgus alpinus and Tongapyrgus kohitatea. In BA they formed the sister clade to Leptopyrgus, whereas the ML topology was (T. kohitatea, (O. alpinus, (Leptopyrgus))). All nodes received a posterior probability ≥56 and only 6 of the total of 25 nodes had support of <80 (Fig. 1). Since base substitutions were modeled more accurately in BA, this analysis was used in the reconstruction of ancestral states. However, the results would be identical using the ML tree as input for Discrete. The concatenated dataset evolved according to a molecular clock indicated by the insignificantly lower likelihood for a ML reconstruction with molecular clock enforced (lnL = -7581.60 vs. −7564.85, inline image = 33.51, n.s.).

The network based on 16S rDNA sequences alone inferring the relationships among several populations of P. antipodarum from the North, South and Stewart Island as well as P. doci, P. oppidanus and P. troglodytes revealed very low variability, which was partly homoplastic as indicated by the reticulation (Fig. 2). Maximum distance between specimens from different populations of P. antipodarum was four mutational steps. P. doci differed by one to three mutational steps from its sympatric population of P. antipodarum and P. oppidanus was identical to or one step away from P. antipodarum occurring at the same locality. Similarly, the 16S rDNA sequences of the sympatric species O. johannstraussi and O. josefstraussi– three and four specimens, respectively – had only four variable sites and the maximum pair-wise divergence was three steps.

Figure 2.

Haplotype network inferring relationships of four species of Potamopyrgus based on the 16S rDNA fragment. Each branch equals one mutational step. Empty circles represent nonsampled haplotypes. Acronyms of species are slightly truncated from Table 1.

Reconstruction of ancestral character states

Brooding species appeared in two clades and the most parsimonious reconstruction suggested its independent evolution at node 16 and within Potamopyrgus at node 19, respectively (Fig. 1). Two times convergent evolution of ovoviviparity would also have been the case in a three step scenario, where brooding would have evolved at node 23, got lost at 22 and would have been regained at 19. Another three-step scenario involves an ovoviviparous ancestor, transition to oviparity twice at nodes 24 and 22, and a reversal to brooding at node 19. Alternatively, parental care could have appeared three times independently among New Zealand hydrobiids at nodes 6, 15 and 19, respectively. The assumption of a single origin of ovoviviparity, viz. at node 23, required at least four steps including independent reversals to egg-laying of Sororipyrgus, P. estuarinus (Winterbourn, 1970) and P. kaitunuparaoa (Fig. 1).

For the likelihood analyses of character evolution, oviparity and occurrence in brackish water were coded as zero and ovoviviparity and freshwater as one. An initial exploration indicated that the backward transitions were not different from zero. The likelihoods of the unrestricted model and the model with backward transitions restricted to zero were practically identical indicating unidirectional evolution in both traits, or, in other words, that transitions from brooding to egg-laying and freshwater into brackish water were unlikely. Consequently, the backward transitions were set to 10−8 (setting to zero may cause a floating point error) in all following analyses. Varying kappa, the branch length scaling parameter, between 10−8 and two affected the returned likelihood values only behind the 11th decimal place indicating that traits have evolved independently of branch lengths in a punctuational mode.

According to the global ML reconstruction of ancestral states, brooding has evolved twice at nodes 16 and 19, respectively, as suggested by the criterion of MP. The local reconstruction was only ambiguous with respect to the state at node 16, where brooding had a higher likelihood, albeit not significantly (∂L = 1.93), indicating that ovoviviparity could have evolved twice or three times convergently among New Zealand hydrobiids with equal likelihood. All other nodes had significantly higher likelihoods for the respective state in accordance with MP. Since backward transitions were highly unlikely, hypotheses with reversals including the single origin hypothesis of ovoviviparity were excluded from further considerations. The most likely scenario supported by MP, the global ML reconstruction of ancestral states and the high, although not significant, likelihood for brooding at node 16 remained the hypothesis suggesting two independent origins of ovoviviparity among New Zealand hydrobiids at nodes 16 and 19.

The reconstruction of habitat transitions was more ambiguous. There were five equally parsimonious scenarios comprising three steps. The first one suggested three independent invasions of freshwater at nodes 16, 18 and 19. Alternatively, the first transition into freshwater could already have happened at node 23. Either the common ancestor of P. estuarinus and P. kaitunuparaoa or both species independently would then have returned to brackish conditions. In the former case, the third step back into freshwater would have happened at node 19. The remaining two three-step scenarios were similar to the preceding ones, but had the root, i.e. node 25, already in fresh water and also Halopyrgus, i.e. node 24, moving into brackish water (Fig. 1).

All hypotheses except for the first scenario involved reversals and had therefore to be rejected according to the criterion of ML. The most likely set of ancestral states determined in the global ML analysis was identical to the first hypothesis and suggested three independent invasions of freshwater at nodes 16, 18 and 19, respectively. Also in the local ML reconstruction these three nodes had higher likelihoods for freshwater, however, significant only at node 19 (∂L = 4.37). Even nodes 6 and 4 had only insignificantly higher likelihood scores for freshwater. All remaining nodes from 1–17 scored significantly for freshwater. Nodes 20–25, the root, had significantly higher likelihoods for brackish water. The ancestor at node 18 was certainly already in freshwater. Otherwise, species of Sororipyrgus would have invaded freshwater independently. This possibility was not further considered. However, the ambiguity at nodes 4, 6 and 16 suggested that four or more transitions into freshwater were not unlikely. Only fixing node 4 to oviparity resulted in a likelihood significantly lower than that for the model in which nodes were free to vary (∂L = 2.08). Thus, although the scenario suggesting three independent invasions of freshwater was most likely, four invasions could not be ruled out. However, since the solution including only three transitions from brackish into freshwater did receive the highest likelihood, which was supported by MP, and, on the other hand, no auxiliary criterion supported four transitions, it is conservatively assumed that New Zealand hydrobiids entered freshwater three times independently.

The dependent analysis with the root fixed to the plesiomorphic states (0/0 = oviparous/brackish water; the derived states 1/1 denote ovoviviparity/freshwater) indicated that the evolution of both traits has been correlated (inline image = 10.35, P < 0.05). Since reversals were considered unlikely, also the reverse transition rates q21, q31, q42 and q43 (see Fig. 3) were restricted to 10−8 (see above). This model could not be distinguished from the former one (inline image = 0.43, n.s.) and was hence used as basis for the following tests. To address the question of which trait evolved first, the transition rates q12 and q13 leading alternatively from 0/0 to 0/1 and 1/0, respectively (Fig. 2), were compared, but could not be distinguished (inline image = 0.37, n.s.). Both rates were different from 0, particularly q12 (inline image = 35.16, P < 0.01; inline image = 4.86, P < 0.05). This may be taken as a weak argument for the transition to freshwater preceding the evolution of ovoviviparity. Because of this vagueness, further tests were not considered meaningful. The assumption that brooding evolved after populations had adapted to freshwater is supported by the fact that there are oviparous freshwater species, viz. the three species of Sororipyrgus and P. doci, but no brooding species restricted to brackish water.

Figure 3.

Flow diagram of correlated character transitions. 0/0 = oviparity/brackish water, 1/1 = ovoviviparity/freshwater; qij = transition rate from node i to node j.


Ovoviviparity is a rare phenomenon among hydrobioid (sensu Davis, 1979) gastropods. Of the more than 10 families from temperate and tropical regions around the world (Wilke et al., 2001, unpublished data) and more than 1000 species (Hershler & Ponder, 1998), it is only known from Indopyrgus nevilli (Thiele, 1928) from the Andamans, American cochliopids, and the New Zealand radiation (Thiele, 1928; Hershler & Ponder, 1998). Despite this rarity, brooding dominates and has evolved independently twice among New Zealand hydrobiids, as the present analyses have clearly shown. The sister clade of (Potamopyrgus, Sororipyrgus) switched to ovoviviparity much earlier than Potamopyrgus. The egg membrane of the former certainly represents an advanced stage of eggshell reduction compared to the elastic shell in Potamopyrgus. However, this advanced reduction is explained by the earlier origin and not as derivative of the latter. A phylogenetic analysis of cochliopids, which are morphologically much more diverse, is outstanding, but multiple convergent origins of ovoviviparity are assumed as well (Hershler & Thompson, 1992).

Among New Zealand hydrobiids, the switch to brooding seems to be linked to the invasion of freshwater. Combining the ML and MP approaches suggested that freshwater was invaded three times independently. Transitions of both reproductive mode and habitat were correlated. The temporal sequence of these transitions could only be inferred using auxiliary criteria in addition to ML inference, probably because the total number of transitions was relatively low along the tree and the character states tightly coupled (see Pagel, 1994). Nevertheless, the most likely scenario appears to be that two of the invasions of freshwater were followed – probably shortly afterwards (see below) – by the adoption of ovoviviparity as if there had been a predisposition.

It seems unlikely that freshwater per se was the selective agent for the evolution of brooding when considering that the vast majority of hydrobioids do well in its absence. Rather some peculiarities of the freshwater situation in New Zealand, especially that of small streams, such as e.g. unequal distribution of precipitation over the year, must have been responsible for the success of brooding. A female carrying the eggs with her can actively react to changing water levels and thus prevent their desiccation. Such a situation is at least plausible for the cold periods of the Late Pliocene and Pleistocene, during which ovoviviparity may have evolved in the lineage of Potamopyrgus as molecular clock calibrations suggest (Haase, unpublished data). Whether the common ancestor of the remaining brooding taxa, which dates back at least to the Middle Miocene, has experienced similarly unstable conditions is probably impossible to infer.

The association of the ubiquitous P. antipodarum and the closely related single site endemic P. doci in a spring in the Waitomo area in the west of NI allowed to shed light on spatial and temporal scale of the evolution of ovoviviparity. A principal component analysis based on eight shell parameters showed clear morphological separation indicating that mitochondrial similarity was due to common ancestry and not caused by introgression through hybridization, which would have been indicated if intermediate shells were present (for more details see Appendix I). The differences in reproductive biology between P. doci and P. antipodarum, the former is oviparous and the latter brooding, indicate that these species are probably not true sister species or, rather, that one is not an immediate stem species of the other as the network (Fig. 2) may suggest. This would only hold assuming that P. doci has returned to oviparity after splitting off from P. antipodarum. Such a reversal seems unlikely per se and because all freshwater species, excepting the three primarily oviparous species belonging to Sororipyrgus, are successful brooders. It is more likely that P. doci is the sister species of the immediate, nonbrooding ancestor of P. antipodarum. The intraspecific genetic variability of P. doci either represents an inherited polymorphism or parallels that of P. antipodarum. This scenario indicates that morphological change as well as the transition to ovoviviparity have occurred very rapidly. It furthermore suggests that P. antipodarum, which occurs throughout New Zealand, has originated in the Waitomo area in the western North Island and that it was originally a freshwater species that has extended its range into brackish water secondarily. These and the following conclusions are based on the more conservative 16S rDNA partition. However, this is hardly derogating given that the genetic distances were very small.

Introgression as cause for genetic similarity can also be excluded for the remaining two pairs of congeneric species occurring in sympatry, viz. P. antipodarum and P. oppidanus as well as O. johannstraussi and O. josefstraussi (Appendix I). The spatial situation of the former pair –P. oppidanus is a single site endemic whereas P. antipodarum practically occurs everywhere – even suggests sympatric speciation. Only if one assumes that the ubiquitous P. antipodarum has colonized the stream in Wadestown/Wellington after P. oppidanus had sufficiently differentiated and reproductive isolation been established could allopatry be invoked. This appears to be unlikely considering the apparent high potential of passive dispersal of the stem species. However, the theoretical possibility of (Gavrilets, 2003; Arnegard & Kondraschov, 2004; Bolnick, 2004; Kirkpatrick & Nuismer, 2004) as well as finding empirical evidence (Barraclough & Vogler, 2000; Losos & Glor, 2003) for sympatric speciation are contentious issues. In the absence of further supporting data it is refrained from a deeper elaboration.

The difficulties in overseeing the biogeographic situation of the two stygobiontic species of Opacuincola in detail – both are known from only two adjacent caves in the Aorere Valley – prevent an unambiguous assessment of the mode of speciation. Too little is known about the extent and connectivity of the local groundwater system.

Some of the deeper nodes of particular importance to the reconstruction of ancestral character states were only weakly supported, viz. nodes 6, 16 and 23. In the foregoing, more comprehensive phylogenetic analysis including more specimens per species nodes 16 and 23 had higher posterior probabilities of 71 and 73, respectively. And node 6 was supported by an anatomical synapomorphy (Haase, unpublished data). However, the reconstruction of ancestral character states would hardly be affected if the relationships within the clade supported by node 16 were somewhat different, because the taxa in question have identical character states. The relatively low support in the present analysis was probably due to slight saturation of substitutions in third codon positions of the COI partition, which, however, was not prohibitive to phylogenetic reconstruction as indicated by Xia et al.'s (2003) indices.

The present analysis was based on all species available for sequencing. However, a total of 64 are known from New Zealand and more, notably creno- and stygobiontic species, probably remain to be discovered. Three genera, all occurring in springs or the groundwater, were not represented at all. Kuschelita is very deviant and its allocation to the family is doubtful. The anatomical information available for the monotypic genera Rakipyrgus and Platypyrgus suggests that they belong to the large ovoviviparous clade whose common ancestor is represented by node 16 in Fig. 1. Similarly, species, which could not be included in the analysis but share the states of their represented congeners, will not change the conclusions. Only those species with insufficient or no anatomical information, all of them creno- or stygobionts (Haase, unpublished data), as well as new discoveries might change the conclusion of two-fold convergent evolution of brooding derived from the present analysis. Nevertheless, this picture will not revert to the single origin hypothesis. It can only get more complex, i.e. homoplasy might even increase.


The combined approach of MP and ML allowed the unambiguous reconstruction of transitions between reproductive modes and habitats. An analysis based on either of these criteria would have resulted in considerable ambiguity, especially in the case of MP. Accordingly, New Zealand hydrobiids have invaded freshwater three times independently and two of these invasions were followed by the evolution of ovoviviparity, possibly as a consequence of irregular precipitation regimes and changing water levels in streams and springs during drier geological periods. In the lineage of Potamopyrgus it could be demonstrated that this switch from egg laying to brooding must have occurred very rapidly. Similarly rapid were changes in shell morphology. The resulting picture of punctuated equilibrium against a background of gradual mitochondrial evolution was also indicated by the independence of ML estimates of character state evolution from branch length and the apparent paraphyly of P. antipodarum (Eldredge & Gould, 1972; Gould & Eldredge, 1993). The punctuational mode of evolution is also in accordance with theories of speciation predicting short durations of speciation relative to the waiting times for speciation (Gavrilets, 2003). In one case, speciation may have occurred in sympatry.


I thank Kevin Collier and Mike Scarsbrook for their multifaceted support throughout the project, which was initiated by Winston Ponder. Hsiu-Ping Liu and Robert Hershler generously provided the outgroup sequences. Kevin Collier and Ian Hogg as well as anonymous referees gave valuable comments. Financial support was received from the Foundation for Research, Science and Technology, program number CO1X0219, and the Department of Conservation, contract DOC032.


Appendix I

Control for hybridization

Mitochondrial similarity of a pair of species may be due to recent common ancestry or introgression by hybridization. Information from nuclear genes may serve as a control. If hybridization has occurred recently, one would expect to find morphologically intermediate specimens, provided the alleles of one parental species controlling shell shape and size are not dominant over those of the other species (but see Ballard & Whitlock, 2004). The likelihood of such one-sided dominance decreases with the number of genes influencing morphology. Morphologies of sympatric species were compared by Principal Component Analyses (PCA) conducted with Primer 5 (Primer-E, 2000). PCAs were based on correlation matrixes of eight normalized shell parameters (Table A1). There are two morphs of P. antipodarum living in the stream in Ruakuri, a large and a small one. Only the small one was sequenced, because it is more similar in size to P. doci. But shells of both morphs were measured for the PCA. Both species of Opacuincola occur in both Ballroom and Aorere Cave. For sequencing, only material from Ballroom Cave was available. Since I had only few specimens of O. johannstraussi from Ballroom Cave I also measured specimens from Aorere Cave in order to represent the morphospace more comprehensively in the PCA.

Table A1.  Shell morphometry.
Species/Locality (N) shswahawbwwsh/swsh/ahw
  1. ah, aperture height; aw, aperture width; bww, width of body whorl (=penultimate whorl); N, number of specimens; s, standard deviation; sh, shell height; w, number of whorls; measurements in mm.

P. antipodarum, largeMean5.043.172.341.962.611.592.165.94
Ruakuri (6)s0.430.
P. antipodarum, smallMean2.901.761.251.081.491.652.324.79
Ruakuri (6)s0.
P. dociMean2.071.551.
Ruakuri (20)s0.
P. antipodarumMean4.332.271.651.341.951.912.646.55
Wadestown (12)s0.410.
P. oppidanusMean2.851.210.840.711.042.353.365.98
Wadestown (40)s0.530.
O. johannstraussiMean1.671.611.040.931.181.041.613.42
Ballroom Cave (14)s0.
O. johannstraussiMean1.841.711.100.981.281.081.673.70
Aorere Cave (10)s0.
O. josefstraussiMean2.231.461.050.931.221.532.133.99
Aorere Cave (20)s0.

The PCAs showed clear patterns, i.e. well separated morphologies (Fig. A1). Only the plot of the Opacuincola species appeared to reveal a single intermediate specimen. However, this was most likely a teratological shell rather than that of a hybrid indicated by the partly detached final whorl. Details of the PCAs are summarized in Table A2. The first principal component (PC) of the PCA comparing the two morphs of P. antipodarum and P. doci was mainly determined by size and number of whorls, whereas the shape parameters had the highest weights in PC2. Similarly distributed were the loads in the PCA of P. antipodarum and P. oppidanus. Only the number of whorls weighed higher in PC2 than in PC1. By contrast, for O. johannstraussi and O. josefstraussi PC1 was largely composed of shell height, the shape parameters and the whorl count and PC2 was mainly influenced by the remaining size variables. These different compositions of the PCs suggest that shell size and shape are determined by several loci suggesting that complete dominance of morphology controlling alleles of one species over those of another one is very unlikely should hybridization have occurred. Ruling out introgression indicates that mitochondrial similarity in all three pairs of sympatric, congeneric species was due to recent common ancestry.

Figure A1.

Plots of first and second principal components (PC) from principal component analyses based on eight shell parameters. (a) Potamopyrgus antipodarum (s, small; l, large) and P. doci (d) from Ruakuri. (b) P. antipodarum (w) and P. oppidanus (o) from Wadestown/Wellington. (c) Opacuincola johannstraussi (a, Aorere Cave; b, Ballroom Cave) and O. josefstraussi (f, Ballroom Cave).

Table A2.  Principal component analyses based on matrixes of correlations of eight normalized shell parameters.
  1. First three principal components. (A) Potamopyrgus antipodarumP. doci; (B) P. antipodarumP. oppidanus; (C) Opacuincola johannstraussiO. josefstraussi. Load values dominating a principal component are highlighted in bold. ah, aperture height; aw, aperture width; bww, width of body whorl; cumv, cumulative variance in %; E, Eigenvalue; sh, shell height; sw, shell width; v, variance in %; w, number of whorls.

VariableLoads of parameters in Eigenvectors