Most specimens examined in this study were collected in 2012 at 22 localities in New Caledonia (Table 1, Fig. 3; Zielske and Haase 2015). Presorting of the collected animals to morphospecies was made by eye and in case of population 46 deviated from our previous paper. 37 paratypes (Haase and Bouchet 1998) of H. cockerelli included in morphometric analyses were borrowed from the Museum National d’Histoire Naturelle Paris (MNHN-IM-2012-2694). Altogether 324 individuals from 24 sites were used for morphometric analyses. 108 thereof from 20 sites were used for phylogenetic analyses. The COI-sequence of a specimen from population 18 was taken from Zielske and Haase (2015). Sequences were submitted to GenBank and received accession numbers KT203603 - KT203710. Specimens included in the recent study are encoded as follows: “Population.Specimen Number”, for instance “1A.01” represents specimen number one from population 1A.

Figure 3. 

Map of New Caledonia showing sampling localities. Samples “misplaced” in phylogenetic analyses identified by population numbers.

All 324 individuals were investigated for the presence/absence of a denticle 1/3 whorl behind the outer lip, which has been described as diagnostic for H. cockerelli (Haase and Bouchet 1998), under a dissecting microscope. For documentation, four specimens were selected for micro-computed X-ray tomography (μCT) - one paratype of H. cockerelli, one topotype of H. fabrorum, and one specimen each from morphologically intermediate populations 9A and 46 (see below). Scans were performed using an XRadia Micro XCT-200 (Carl Zeiss X-ray Microscopy Inc., Pleasanton, USA). The samples were placed in pipette tips glued on an insect pin. Each shell was scanned for 1 h at 40 kV and 8 W at four times magnification. Image stacks were processed and three dimensional surface models constructed using the 3D analysis software AMIRA v. 5.6.0 (FEI, Visualization Science Group).

For geometric morphometric investigations (Zelditch et al. 2012), shells were photographed at 30-times magnification using a Nikon SMZ 800 stereoscopic microscope (Nikon Corporation, Tokyo, Japan) equipped with a Nikon DS-2M camera and NIS-Elements AR v. 3.2 software (Nikon, Tokyo, Japan). Snails were placed onto a silicone surface for easier positioning, all orientated with the longitudinal axis (columella) parallel to the y-axis. Images were converted into tps format using TPSUTIL v. 1.58 (Rohlf 2013a). 17 landmarks (Fig. 4) were digitized for each individual with TPSDIG v. 2.17 (Rohlf 2013b). In order to check the repeatability of the procedure, it was repeated for ten specimens on two consecutive days. The comparison with Goodall’s F-test in TWOGROUP v. 8 (a program of the IMP suite written by David Sheets: http://www.canisius.edu/~sheets/morphsoft.html) was not significant (p = 0.59) indicating that positioning the shells and placing the landmarks was highly repeatable (cf. Haase and Misof 2009).

Figure 4. 

Seventeen landmarks placed on a shell of Hemistomia fabrorum from population 39.

In order to visualize the morphological variation, a principal component analysis (PCA) was performed using MORPHOJ v. 1.06a (Klingenberg 2011). COORDGEN v. 8 belonging to the IMP suite was used in order to generate input-data for further analyses with other IMP programs. Based on the PCA, we identified individuals with uncertain morphospecies allocation for further analyzes by canonical variates analysis (CVA) and assignment tests conducted with the IMP program CVAGEN v. 8. For this identification, equal frequency confidence ellipses with a probability of 95% were plotted on the PCA graph. Individuals localized in the intersection of both ellipses and its neighborhood were classified as uncertain and treated as “unknown specimens” in assignment tests including jackknife-assignment (Webster and Sheets 2010). The CVA axes determined by the two known groups were used to assign the unknown specimens to one of the known groups. A pair-wise comparison between the two morphospecies including all specimens previously assigned to either morphospecies with a p-value higher than 0.05 was performed with Goodall’s F-test in TWOGROUP v. 8.

DNA was isolated using Qiagen’s DNeasy Blood & Tissue Kit (QIAGEN GmbH; Hilden, Germany) in compliance with the manufacturer’s protocol except that we eluted only in 20 μl of AE-buffer. A fragment of the mitochondrial cytochrome c oxidase subunit I gene (COI) was amplified using the primers LCO1490 and HCO2198 (Folmer et al. 1994), the latter modified at position 12 (G→A; Zielske et al. 2011). Polymerase chain reactions were performed in 12.5 μl containing 1.1 μl 10x BH4 buffer (BIOLINE GmbH; Luckenwalde, Germany), 4.4 mM MgCl, 0.3 μM of each primer, 0.2 mM dNTP, 0.5 μl BSA (1%), 0.25 U DNA-Polymerase (BIOLINE), 5–50 ng DNA and water. The PCR conditions were 95 °C for 5 min for denaturation and 40 cycles starting at 94 °C for 60 s, followed by annealing for 90 s with an initial touchdown from 55 to 46 °C with a drop of one degree per cycle, an extension step at 72 °C for 60 s, and 10 min final extension at 72 °C. Amplification products were purified using 4 μl PCR-product and 1 μl Exo-SAP [0.04 μl Exonuclease I (20.000 U/ml; New England BioLabs GmbH; Frankfurt/Main, Germany), 0.15 μl 10x Shrimp Alkaline Phosphatase Buffer (Promega; Madison, WI, USA), 0.81 μl ddH₂O] per sample. This mix was incubated at 37 °C for 15 min followed by 85 °C for another 15 min. Cycle sequencing was conducted using the Big Dye Terminator Ready Reaction Mix v3.1 (Applied Biosystems, (ABI); Carlsbad, CA, USA) and the PCR primers. The cycle sequencing products were purified using Agencourt’s® CleanSEQ® Dye-Terminator Removal (Beckman Coulter; Beverly, MA, USA) before sequencing on an ABI 3130xl Genetic Analyzer.

COI sequences were edited using the software DNA BASER v. 4.16 (DNA Baser Sequence Assembler v. 4.16 2014). After addition of two outgroup sequences of H. nyo and H. andreae (see Haase and Zielske 2015; Zielske and Haase 2015), the alignment was generated with Clustal W implemented in MEGA v. 6.06 (Tamura et al. 2013) and trimmed to a length of 658 bp. The alignment was screened for potential stop-codons with the software DAMBE v. 5.5.1 (Xia 2013) to check for potential editing errors and nuclear pseudogenes. jMODELTEST v. 2.1.4 (Darriba et al. 2012) identified HKY + I + Γ as best-fitting DNA substitution model according to the Bayesian information criterion. Three phylogenetic analyses were conducted, a maximum likelihood analysis (ML), a bio-neighbor-joining analysis (BNJ), and a neighbor-net analysis. ML was performed using Garli 2.01 (Zwickel 2006). Both optimal tree and bootstrap support were inferred from 500 replicates. The BNJ tree including 5000 bootstrap replicates was constructed in PAUP* v. 4.0b10 (Swofford 2002). Bootstrap support was considered significant if > 75. The neighbor-net was computed with the software SPLITSTREE v. 4.13 (Huson and Bryant 2006) based on the K3ST-model. The optimal model (see above) could not be applied because the program issued undefined distance values.

The first two axes of the PCA comparing the morphospecies explained 76.2% of the total morphological variation. The nominal species were fairly well separated along axis 1, however, the 95% confidence ellipses were overlapping. Nineteen specimens within the area of overlap and its neighborhood as defined in Figure 5 were treated as specimens with unknown identity (“unknown”) in further analyses. The CVA defined the set of axes allowing for the greatest possible discrimination of the two groups and axis 1 was highly significant (p < 0.001). These axes were used to assign the unknown specimens. Eight of these specimens were classified as H. cockerelli and 11 as H. fabrorum. All assignments were correct with respect to the initial classification by eye, however, only nine significant. Twelve of the 305 shells with - according to the PCA - unambiguous identity were also assigned correctly but not significantly.

Figure 5. 

Landmark-based principal component analysis with 95% confidence ellipses. A Variation of visually determined H. cockerelli and H. fabrorum B Definition of specimens treated as “unknown” in assignment test.

A jackknife test of assignment, a cross-validation procedure, was performed a posteriori to assess the robustness of the CV axes and comprised these 305 specimens. In 10,000 replicates, 10% of the specimens (31) were randomly selected as unknown data and assigned to one of the resulting groups of the following CVA analysis based on the remaining 274 shells. 99% of the assignments were correct and significant, 1% correct and not significant, and no shell was incorrectly assigned.

In a subsequent Goodall’s F test both morphospecies were highly significantly (p < 0.001) distinguished. This test included all specimens unambiguously (p > 0.05) assigned to one of the nominal species in the previous CVA-based tests.

The μCT-based three dimensional shell surface models revealed, as expected, presence and absence of the palatal denticle in the paratype of H. cockerelli and the topotype of H. fabrorum, respectively (Figs 1, 2). In the specimen from population 9A with shells unusually broad for H. cockerelli, the denticle was well developed, while it was entirely absent in the shell from population 46, a slender H. fabrorum, also confirming the initial classification.

The phylogenetic tree reconstructions inferred very similar relationships. Figure 6 emphasizes the fully resolved BNJ topology, because it had higher bootstrap support than the ML tree except for the clade of population 46. In order to show the general congruence of both reconstructions, the ML topology is given in Figure 6B. The main differences between both optimal topologies were the different positions of the clade of population 46 and individual 6B.10. In the BNJ tree population 46 was nested among specimens allocated to H. cockerelli and 6B.10 sister species of a clade containing sequences from populations 25, 28, 38, and 39. In the ML tree, population 46 was sister group to nominal H. cockerelli and 6B.10 clusterd with other individuals from population 6B. However, when unsupported nodes including those concerning both aforementioned lineages were collapsed, the topologies were practically identical. The ingroup was well supported. It consisted of two main clades of which only the one containing the majority of specimens identified as H. fabrorum received significant bootstrap support. This H. fabrorum-clade included three specimens that were morphologically identified as H. cockerelli, one from population 18 nested with a short branch among individuals from populations 1A and 10, and the two snails from population 9A forming a well supported clade on a long branch. In turn, population 46, morphologically identified as H. fabrorum, clustered with specimens allocated to H. cockerelli. Within both morphospecies, not all populations were monophyletic. The neighbor-net revealed basically the same topology including the shallow position of the specimen from population 18 and the rather basal connections of the snails from populations 9A and 46, respectively (Fig. 7). The network also illustrated the conflicting signal within the two major clades most likely due to lack of data and homoplasy, which explain e.g. the lack of bootstrap support for the H. cockerelli-clade in the tree reconstructions (Fig. 6).

Figure 6. 

Phylogenetic reconstructions. A Bio-neighbor-joining tree based on COI fragment with bootstrap support values from both maximum likelihood and neighbor-joining analyses (ML/BNJ) B Maximum likelihood topology. Asterisks indicate “misplaced” individuals. In B only the “misplaced” individuals and one individual (6B.10), whose position differs considerably in both trees, are indicated. Otherwise, only the clade composition is given. Crossed branches are shortened by 50%. The outgroup was pruned from the tree. Scale bars: substitutions per site.

Figure 7. 

Neighbor-net illustrating conflict in phylogenetic signal of sequence data. Arrows indicate “misplaced” individuals. Scale bar: substitutions per site.