Molecular phylogeny and trait evolution of Madeiran land snails: radiation of the Geomitrini (Stylommatophora: Helicoidea: Geomitridae)

The Geomitrini is the most species‐rich group of land snails in the Madeiran Archipelago. The phylogeny of the group is reconstructed based on mitochondrial and nuclear genetic markers. The timing of diversification, the colonisation history of the islands of the Madeiran Archipelago and the evolution of characters of the dart apparatus are studied. The results of the phylogenetic analyses confirm the sister group relationship of Geomitrini and Cochlicellini, but also show that several previously accepted genus‐group taxa are not monophyletic. A new classification for the Geomitrini is proposed, including the description of two new genera, Domunculifex Brozzo, De Mattia, Harl & Neiber, n. gen. and Testudodiscula Brozzo, De Mattia, Harl & Neiber, n. gen. The onset of diversification of Geomitrini was dated in our analysis at 13 Ma, which largely coincides with the emergence of the present‐day islands. The ancestral state estimation recovered the presence of two appendiculae in the reproductive system as the ancestral state in Geomitrini. One appendicula was lost three times independently within the tribe and is even missing completely in one group. The ancestral area estimation suggests recurrent colonisations of Madeira (and the Ilhas Desertas) from the older island Porto Santo.


Introduction
The Madeiran Archipelago (Fig. 1) is located about 800 km west of Morocco and 900 km southwest of the Iberian Peninsula. It includes the main island Madeira (741 km 2 ) with an estimated age of 4.6 Ma and Porto Santo (69 km 2 ), which has an estimated age of 14.3 Ma and is surrounded by various satellite islets, as well as the three lesser islands 10 km to the southeast of Madeira, the uninhabited Ilhas Desertas (Ilh eu Chão, Deserta Grande and Bugio; together approximately 15 km 2 ) with an estimated age of around 3.6 Ma (Geldmacher et al., 2000).
In contrast to the currently emerged islands, the Madeiran Archipelago has its origins about 65-67 Ma ago, close to the Cretaceous/Paleogene boundary. The Madeiran Archipelago forms the end of a volcanic hotspot chain that connects the archipelago with the Iberian Peninsula to the north-east by a chain of isolated volcanic seamounts, which probably also formed islands and submerged in the course of time due to erosion (Geldmacher et al., 2000Czajkowski, 2002;Kl€ ugel et al., 2005). Within the region forming the Canary Islands, van den Bogaard (2013) found even older volcanic structures from the beginning of the Cretaceous. Since that time, a continuous production of new as well as a constant loss of older islands within the archipelagos is assumed (Fern andez-Palacios et al., 2011), which presumably made the process of colonisation for both, the land flora and fauna (from, e.g., the Iberian Peninsula or other islands) much easier than today. Fern andez-Palacios et al. (2011) emphasise the potential role of the archipelagos of Palaeo-Macaronesia as refuges and radiation centres of palaeo-endemic taxa of plant and animal groups and as stepping stones for the colonisation of the geologically younger Madeiran Islands and Canary Islands. In the Madeiran Archipelago, a closer biogeographical connection to Europe rather than Africa is supported by the presence of many lineages that can be traced back to the European mainland, but which often almost vanished there with the onset of the Pleistocene climate cooling (Baez, 1993;Fern andez-Palacios et al., 2011;Caro et al., 2019). Endemism in the Madeiran Archipelago is particularly high in animal groups that have a low capacity for dispersal, me such as Diplopoda (Cook, 2008) and land snails (Wald en, 1983).
The Madeiran Archipelago harbours an astonishingly rich land snail fauna. The high number of endemic taxa is remarkable, and many of these endemic taxa are only found on one single island within the archipelago, in a small part of an island or even a single locality (Bank et al., 2002;De Mattia et al., 2018a). From the 19th to the beginning of the 21st century, this rich land snail fauna has caught the interest of many biologists, leading to various descriptions, monographs, revisions and checklists (e.g. Lowe, 1831;de Paiva, 1867;Wollaston, 1878;Mandahl-Barth, 1950;Wald en, 1983;Bank et al., 2002;Seddon, 2008;Bank, 2009;Groh et al., 2009;De Mattia et al., 2018a, b).
Family-group and genus-group systematics of Helicoidea Rafinesque, 1815, to which Geomitridae belong, traditionally rested on the morphology of the reproductive organs, especially the presence or absence and morphology of accessory genital appendages such as the dart apparatus (Hesse, 1921(Hesse, , 1931(Hesse, , 1934Schileyko, 1970Schileyko, , 1972aSchileyko, ,b,1978aSchileyko, ,b,1991Schileyko, , 2004Schileyko, , 1991Giusti and Manganelli, 1987;Nordsieck, 1987Nordsieck, , 1993. Molecular analyses in the past 15 years have led to numerous systematic rearrangements in helicoid land snails (Wade et al., 2007;Hugall and Stanisic, 2011;G omez-Moliner et al., 2013;Razkin et al., 2015;Sei et al., 2017). Moreover, it could be shown that reconfigurations, transformations and reductions of the dart apparatus have evolved many times in parallel in this superfamily (Hirano et al., 2014;K€ ohler and Criscione, 2015;Walther et al., 2016Walther et al., , 2018Neiber et al., , 2018aChueca et al., 2018), and it may even differ considerably among closely related taxa or within a single species (Kor abek et al., 2015;Neiber and Hausdorf, 2015;Kruckenhauser et al., 2017;Zopp et al., 2017;Neiber et al., 2018b). In Geomitrini, the calcareous dart has been lost and the dart sac or the accessory sac has been transformed to a hollow tube, the appendicula. In some groups of Geomitrini, two appendiculae are present, while in others only a single appendicula is developed or appendiculae are entirely missing ( Fig. 3; see also Mandahl-Barth, 1950;Schileyko, 2006b). De Mattia et al. (2018a) noted that molecular genetic analyses did not always support the separation of genera based on shell characters and to some extent on genital anatomy. These authors also started to investigate the genital anatomy of Geomitrini in detail, resulting in a revised taxonomy of several genera and the description of a new genus and several (sub-)species. Internal structures of the genital organs proved useful for distinguishing closely related species, but in some cases also allowed the distinction of genus-group taxa. A comprehensive molecular analysis of the Geomitrini is, however, currently lacking. Therefore, the present contribution aims to (i) reconstruct a backbone phylogeny of the tribe based on of a representative sample of species covering nearly the entire generic diversity of the group as currently accepted, (ii) shed light on the evolution of the dart apparatus in this group, (iii) date the radiation of Geomitrini in the Madeiran Archipelago based on a molecular clock approach and (iv) use the molecular genetic analyses to discuss the role of in situ radiations on single islands or groups of islands versus recurrent inter-island dispersal events.

DNA sampling
DNA extractions of specimens previously used in the study of De Mattia et al. (2018a), as well as additional tissue samples of other Geomitrini taxa, were used for the analyses. Specimens were chosen to cover almost all described genera in the Geomitrini. Craspedaria Lowe, 1852 is possibly extinct (Seddon, 2008(Seddon, , 2011Cuttelod et al., 2011;Neubert et al., 2019) and Moreletina de Frias Martins, 2002 from the Azores (de Frias Martins, 2002) could not be included in the analyses. Altogether, 41 specimens of the Geomitrini were used, plus eight individuals belonging to other genera representing the related subfamilies/tribes within the Geomitridae as well as one specimen, Hygromia (Hygromia) cinctella (Draparnaud, 1801), as a representative of the Hygromiidae Tryon, 1866, and one specimen, Canariella (Canariella) giustii Ib añez & Alonso, 2006, as a representative of the Canariellidae Schileyko, 1991 as an outgroup. The classification, sampling sites and extraction numbers of specimens are compiled in Table 1.

DNA extraction and amplification
For the DNA extraction, foot muscle tissue stored in 100% isopropanol was used. Whenever fresh material was unavailable, foot muscle tissue of museum specimens stored in 70% ethanol was used for DNA extraction.
Total genomic DNA was extracted from the foot tissue using a slightly modified version of the protocol of Sokolov (2000) as described in Scheel and Hausdorf (2012) with additional modifications: Tissue samples were incubated in 500 lL lysis buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 10 mM EDTA, 1% sodium dodecyl sulphate) and 20 lL proteinase K at 56°C until complete digestion. Then, 50 lL saturated KCl solution was added to the lysate. The samples were centrifuged at 16 000 g for 15 min. The supernatant was transferred to a clean tube and 500 lL of icecold isopropanol (100%) and 50 lL 3 M sodium acetate solution was added. DNA was precipitated overnight at À20°C. The samples were then centrifuged at 16 000 g for 15 min, and the pellet was washed in 70% ethanol and air-dried. The pellet was subsequently re-suspended in 80 lL of ddH 2 O. For older specimens, stored in 70% ethanol, 300 lL lysis buffer, 30 lL KCl and 30 lL sodium acetate was used instead.
Partial sequences of the mitochondrial cytochrome c oxidase subunit 1 (cox1) and of the 16S rRNA (16S) gene, as well as a part of the nuclear ribosomal RNA gene cluster including the 3ʹ end of the 5.8S rRNA (5.8S) gene, the complete internal transcribed spacer 2 (ITS2) and the 5ʹ end of the 28S rRNA (28S) gene, were amplified by polymerase chain reaction (PCR). The primer pairs used were LCO1490 plus HCO2198 (Folmer et al., 1994) for cox1, 16Scs1 (Chiba, 1999) plus 16S_MN3R  for 16S and LSU1 plus LSU3 (Wade and Mordan, 2000) as well as LSU2 plus LSU4 (Wade and Mordan, 2000) for the nuclear ribosomal rRNA gene cluster, respectively. For a subset of samples, especially older museum material stored in 70% ethanol, other primer combinations were used, since the amplification had been difficult and/or incomplete. In these cases, previously published (Palumbi, 1991;Uit de Weerd, 2008) as well as newly designed internal forward and reverse primers were used in combination with one of the above-listed primers (see Supporting Information, Table S1).
Amplifications were performed in 25 lL volumes containing 18.3 lL ddH 2 O, 2.5 lL Dream Taq TM Green Buffer (Thermo Fisher Scientific, Waltham, MA, USA), 1.0 lL of a dNTP mix (5 mM each), 1.0 lL of each primer (10 lM), 0.2 lL DreamTaq TM DNA polymerase (Thermo Fisher Scientific) and 1.0 lL of the template DNA under the following reaction conditions: an initial denaturation step at 94°C for 2 min, 35-45 PCR cycles (94°C for 30 s, primer specific annealing temperature for 30 s, 72°C for 30 s) and a final extension step at 72°C for 5 min. Both strands of the amplified products were sequenced at Macrogen Europe Laboratory (Amsterdam, The Netherlands).

Phylogenetic analyses
Forward and reverse sequences were assembled using ChromasPro 1.7.4 (Technelysium, Tewantin, Australia). The protein-coding mitochondrial sequences were aligned with MUSCLE (Edgar, 2004) as implemented in MEGA X (Kumar et al., 2018) with the default settings. The nuclear sequences and the sequences coding for the 16S rRNA gene were aligned with MAFFT 7 (Katoh et al., 2017), using the Q-INS-i strategy and otherwise default settings (see Supplementary Files, Data S1). Bayesian inference (BI), maximum likelihood (ML) and maximum parsimony (MP) analyses were used to reconstruct phylogenetic relationships.
Partitions and evolutionary models were evaluated separately for mitochondrial and nuclear sequences. The cox1 sequences were initially divided into three partitions based on codon positions (1st, 2nd and 3rd position), while the 16S sequences were not further subdivided. The nuclear sequences were initially divided into three partitions (5.8S, ITS2 and 28S). PartitionFinder 2.1.1 (Lanfear et al., 2017) was used to search for the best evolutionary models and partitioning schemes, conducting exhaustive searches on the basis of the mitochondrial and nuclear data sets, respectively, with a separate estimation of branch lengths for each partition and with the Bayesian information criterion to select among models and partitions. The models were limited to those available in MrBayes 3.2.6 ( Ronquist et al., 2012). For the cox1 and 16S sequences, the PartitionFinder analysis suggested three partitions, the first containing the 1st and 2nd codon positions of cox1 (GTR + G model), the second containing the 3rd codon positions of cox1 (GTR + I + G model) and the third containing the 16S sequences (HKY + I + G model). For the nuclear sequences, the PartitionFinder analyses suggested a single partition and the HKY + I + G model.
The BI analysis was performed using MrBayes. Metropolis-coupled Monte Carlo Markov chain (MC 3 ) searches were run with four chains in two separate runs with 50 000 000 generations with default priors, trees sampled every 1000 generations under default heating using the partitions and evolutionary models for the mitochondrial and nuclear data sets as suggested by the PartitionFinder analyses. The first 500 000 generations of each run were discarded as a burnin.
The ML analysis was performed using GARLI 2.1 (Zwickl, 2006) with the partitions and models suggested by PartitionFinder and otherwise default settings from the standard configuration file (http://www.bio.utexas.edu/faculty/antisense/garli/Garli.html), except for setting the number of replicates to 100. Support values were calculated by bootstrapping with 1000 replications. For comparison of support values, an additional bootstrap analysis (1000 non-parametric bootstrap replications) was conducted using IQ-TREE (Chernomor et al., 2016;Minh et al., 2020) using the same partitions and evolutionary models as in the analysis with GARLI.
Heuristic MP searches were conducted with PAUP* 4.0b10 (Swofford, 2002) with unordered characters, 100 random sequence addition replicates, tree bisection reconnection (TBR) branch swapping, and gaps treated as missing data. Support for internal branches was assessed in PAUP* by bootstrapping with 1000 replications, using full heuristic searches with 10 random addition sequence replicates, TBR branch swapping, and one tree held at each step during stepwise addition. For comparison of support values, an additional bootstrap analysis (1000 non-parametric bootstrap replications) was conducted using TNT (Goloboff et al., 2008) with 10 random addition sequence replicates, TBR branch swapping, and one tree held at each step during stepwise addition.
Bootstrap support (BS) values from the ML and MP analyses as well as posterior probabilities (PP) from the Bayesian analysis were mapped on the BI 50% majority-rule consensus tree with SumTrees 3.3.1, which is part of the DendroPy 3.8.0 package (Sukumaran and Holder, 2010). PP ≥ 0.95 and BS ≥ 70 were interpreted as positive support for nodes.

Molecular dating
To infer a time frame for diversification patterns of Geomitridae, we used the Bayesian algorithm implemented in Beast 2.5.2 (Bouckaert et al., 2019) based on mitochondrial and nuclear sequence data assuming the same partitions and nucleotide substitution models as in the ML and BI analyses. A strict molecular clock was rejected at a = 0.05 by the test implemented in MEGA X for each of the different partitions. Therefore, a linked uncorrelated relaxed lognormal molecular clock was used for the Beast analysis assuming the birthdeath model as tree prior. The analysis was run for 50 000 000 generations with a sampling frequency set to 10 000. Tracer v1.7.1 (Rambaut et al., 2018) was used to assess convergence and to check that the estimated effective population sizes (ESS) for all estimated parameters were above 200. 10% of generations were discarded as burn-in. A maximum clade credibility tree with median node heights was calculated with Treeannotator 2.1.3 (included in the Beast 2.5.2 software suite) from the post-burn-in samples.
We used Loganiopharynx rarus (Boissy, 1840) from Ypresian deposits (Early Eocene, 56-47.8 Ma) in France for the calibration of the tree. Nordsieck (2017) classified this taxon in Hygromiidae. However, it cannot be assigned unambiguously to one of the subfamilies of the crown group of Hygromiidae. We assume that it is a representative of the stem group taxon of Hygromiidae. Actually, it was assigned to the stem group of Hygromiidae in the analysis of Razkin et al. (2015: fig. 3). Thus, this taxon can be used to put a minimum age on the divergence of Hygromiidae and their sister group, Geomitridae plus Canariellidae, according to Ho and Phillips (2009: fig. 1). We used a lognormal-distributed prior for this calibration.

Ancestral character state and ancestral area estimation
For the ancestral state estimation, Mesquite v.3.4 (Maddison and Maddison, 2018) was used, tracing the state of the appendicula (missing, single or double) over the MP tree calculated in the previous step by using the parsimony approach and otherwise default settings.
The biogeographic history of the group, especially with regard to inter-island dispersal events within the Madeiran Archipelago (island hopping) was estimated on the basis of our mitochondrial and nuclear sequence data set using the Bayesian binary MCMC method implemented in RASP 3.02 (Yu et al., 2015). For the analysis, the calibrated maximum clade credibility tree from the Beast analysis was used together with a matrix, in which the distribution of the sequenced species in the following seven geographical regions was listed: (A) Madeira, (B) Porto Santo and its satellite islets, (C) Ilhas Desertas, (D) Canary Islands, (E) Iberian Peninsula, (F) Europe (excl. the Iberian Peninsula), (G) North Africa. The analysis was run on 100 randomly selected post-burn-in trees from the Beast analysis to account for statistical uncertainty and otherwise default settings.

Phylogenetic analyses
The concatenated alignment of mitochondrial and nuclear sequences obtained from 51 individuals comprised a total number 2698 base pairs (bp) (655 bp cox1, 515 bp 16S and 1528 bp 5.8S + ITS2 + 28S). The three different methods (BI, ML and MP) for the reconstruction of phylogenetic relationships resulted in trees with identical topologies. The results of the bootstrap analyses with IQ-Tree (ML) and TNT (MP) were very similar to the results of the analyses with GARLI (ML) and PAUP* (MP) (Fig. 4). For brevity, only the results obtained with GARLI and PAUP* are presented in the following, but see Fig. 4 for all support values.
The Geomitridae in the sense of Razkin et al. (2015) and  were recovered in our  Razkin et al. (2015) and  were also recovered with strong support. Within the Helicellinae, the relationships of the representatives of the Cernuellini, Helicellini, Helicopsini and Trochoideini were only supported in the BI analysis but not in the ML and MP analyses (Fig. 4).
Within the Geomitrinae, the Ponentinini represented by Ponentina cf. revelata (Michaud, 1831) was recovered as the sister group of a maximally supported clade including the Cochlicellini and the Geomitrini. were also recovered as monophyletic groups with high support (Fig. 4), with Wollastonaria as the sister group of a maximally supported clade including Hystricella and Callina. However, some genera in the sense of, e.g., Seddon (2008) and/or Bank (2009) Table 1 Collection numbers, voucher numbers, GenBank accession numbers and geographical coordinates for the specimens used in the molecular phylogenetic analyses Taxon Wollastonaria was only significantly supported in the BI analysis.

Molecular dating
Assuming an Early Eocene (47.8 Ma) split of Hygromiidae and the clade including Geomitridae plus Canariellidae, the split of the latter clade was dated at 38.9 Ma (95% highest posterior density interval (HPD): 27.5-48.7 Ma) in the Late Eocene (Fig. 5). The split of Geomitrinae and Helicellinae in the sense of Razkin et al. (2015)  The basal relationships within Geomitrini were not well-resolved in the phylogenetic analyses. However, a sister group relationship of Geomitra and Actinella, Helicomela and Pseudocampylaea, as well as Serratorotula and Domunculifex, were supported in the Bayesian analysis (Fig. 5) and the respective splits of these sister groups were all dated at 9.3-9.4 Ma in the Late Miocene (Fig. 5). Similarly, the separation of Caseolus and a clade including Disculella, Discula, Spirorbula, Steenbergia, Testudodiscula, Lemniscia, Hystricella, Callina and Wollastonaria was dated at 9.2 Ma (HPD: 5.9-12.7 Ma). The split of Disculella from Madeira and the remaining genera in the latter clade was dated at 7.9 Ma (HPD: 5.1-11.0 Ma), and the split of Discula and the clade including Spirorbula, Steenbergia, Testudodiscula, Lemniscia, Hystricella, Callina and Wollastonaria was dated at 7.1 Ma (HPD: 4.6-9.9 Ma), with the onset of diversification of Discula dated at 3.3 Ma (HPD: 1.1-5.1 Ma). Steenbergia and Spirorbula were dated to have diverged 4.6 Ma (HPD: 2.7-6.7 Ma), with the two together separating from the lineage leading to Testudodiscula, Lemniscia, Hystricella, Callina and Wollastonaria 6.7 Ma (Ma: 4.2-9.3 Ma). Testudodiscula and Lemniscia were recovered to have split 2.1 Ma (HPD: 1.0-3.5 Ma) and the origin of the clade including Hystricella, Callina and Wollastonaria was dated at 2.9 Ma (HPD: 1.9-4.2 Ma), while the split of these two clades was dated at 3.8 Ma (HPD: 2.4-5.6 Ma). Hystricella, Callina and Wollastonaria all originated according to our analysis within the last two million years (Fig. 5).

Ancestral character state and ancestral area estimation
The ancestral state estimation recovered the presence of two appendiculae as the ancestral state of the Geomitrini (Fig. 6). The analysis suggested that one appendicula was lost three times independently within the tribe. One occurred in Pseudocampylaea, a second in Serratorotula and a third loss in the lineage leading to the clade including Disculella, Discula, Steenbergia, Spirorbula, Lemniscia, Testudodiscula, Callina, Hystricella and Wollastonaria. In the lineage leading to Steenbergia and Spirorbula, the second appendicula was subsequently also lost, resulting in a complete absence of these vaginal appendages in these two taxa. However, glandulae mucosae are present in Spirorbula, while these are lost in Steenbergia.
The ancestral area estimation (Fig. 6) implies that the Madeiran Geomitrini originated most likely on Madeira. The analysis further suggests an inter-island dispersal event from Porto Santo to Madeira for the lineage leading to the clade including Geomitra and Actinella and another inter-island dispersal event for the lineage leading to Disculella, also from Porto Santo to Madeira (Fig. 6). Moreover, the ancestral area estimation suggests a dispersal event from Porto Santo to Madeira within Discula, i.e. the lineage leading to D. polymorpha probably colonised Madeira from Porto Santo (Fig. 6).
Remarks: Testudodiscula testudinalis (Lowe, 1852) is very similar to Callina bulverii (Fig. 2c) with regard to the shell, the keel along the body whorl is however not bent downwards and usually situated higher on the body whorl. Anatomically it differs from C. bulverii by having a relatively shorter epiphallus with less numerous longitudinal folds on the inner wall. From the phylogenetically closely related L. michaudi (Fig. 2h), T. testudinalis is easily distinguished by size and shell form. Anatomically, Testudodiscula differs from Lemniscia (Fig. 3m)  Etymology: Masculine, from domuncula (= small house) and -fex (maker/builder); Latinisation of the German vernacular word "H€ auslebauer" meaning constructor/builder of a small house.
Description: Shell depressed-globular, creamcoloured, usually with two narrow brown spiral bands, ornamented with irregular growth lines, without hairs, 9-10 mm wide, 6-7 mm high, with approximately 5 whorls; aperture elliptical, peristome simple, without thickened lip that is only slightly reflected on the columellar side, partly covering the narrow umbilicus. Vagina longer than penis, slightly shorter than epiphallus, with two appendiculae that are less than one quarter the length of the unramified glandulae mucosae; at the distal end of the vagina a distinct bulge is visible from the outside that corresponds to the location of a very conspicuous, fleshy fold that runs along the inner wall of the vagina from just distal of the level of the insertion points of the appendiculae to the distal end of the vagina, slightly reaching into the lumen of the genital atrium. Duct of the bursa copulatrix stout, approximately as long as vagina. Flagellum very short and slender.
Remarks: Domunculifex differs from Serratorotula (Fig. 2g), its sister taxon in our phylogenetic reconstructions, by the depressed-globular shell that is ornamented with irregular growth lines, whereas the shell of Serratorotula is depressed-lenticular with prominent, often wing-like expanded radial ribs. Anatomically Domunculifex differs from Serratorotula (Fig. 3g) by the presence of two appendiculae and the presence of a very conspicuous, fleshy fold on the inner wall of the vagina, which is completely lacking in Serratorotula.
With regard to shell shape, Domunculifex is similar to Plebecula and Helicomela. It differs from Helicomela (Figs 2r, 3f) in the smaller, less globular shell, the vestigial flagellum, a shorter duct of the bursa copulatrix in relation to the length of the vagina (much longer in Helicomela) and the presence of a very distinct, fleshy, longitudinal fold on the inner wall of the vagina. Plebecula (Figs 2m, 3s) differs from Domunculifex in the presence of hairs and a more globular, usually larger shell, the presence of a transverse fold in the atrium and a longer flagellum.

Phylogeny, classification and character evolution
The phylogeny based on mitochondrial and nuclear sequence data presented here (Fig. 3) is the first to cover almost the entire genus-level diversity of the Geomitrini. It thus allows insights into the phylogenetic relationships of this morphological highly diverse group of land snails. Our results confirm incongruences between the traditional classification based on shell morphology and anatomy on the one hand and the molecular-based analysis on the other hand, as has also observed in other groups of helicoid land snails (Hirano et al., 2014;K€ ohler and Criscione, 2015;Neiber andHausdorf, 2015, Neiber andWalther et al., 2016Walther et al., , 2018Neiber et al., , 2018aChueca et al., 2018).
The phylogenetic relationships obtained here were largely congruent with the results obtained by Razkin et al. (2015), De Mattia et al. (2018a) concerning family-group taxa, though within the Helicellinae the phylogenetic relationships were slightly different compared to the work of Razkin et al. (2015). Since these discrepancies only occurred in relation to lineages that were only supported in the BI analysis (Fig. 3), the classification of Razkin et al. (2015) remains, however, unopposed. Within a previous work, De Mattia et al. (2018a) were able to show the polyphyly of the genera Discula and Hystricella in the sense of Bank et al. (2002), Seddon (2008), Groh et al. (2009) or Bank (2009. The subgenus Discula (Callina) was raised to genus level, while Hystricella was split into Hystricella and Wollastonaria (De Mattia et al., 2018a,b). These results were confirmed here since the former subgenus Callina formed a clade with the two remaining Hystricella taxa and this clade was recovered as sister to Wollastonaria (Fig. 3). The sister-group relationship between Testudodiscula (Figs 2l, 7a-e) and Lemniscia (Figs 2h, 3m) was one of the most surprising results of this work because the shell morphology of the representatives of these two taxa is very different. Testudodiscula testudinalis exhibits a flatter, discoidal shell, while the shell of L. michaudi has a conical shape. Both species also differ markedly in size (Fig. 2h,l). Anatomically L. michaudi and T. testudinalis share some anatomical features like the short flagellum and the relatively long epiphallus, but L. michaudi differs, for example, by the less ramified glandulae mucosae (Figs 3m, 7b,c). Testudodiscula testudinalis and Callina bulverii were both previously classified in Discula s. str. (Bank, 2009), the type species of which, D. discina (Lowe, 1852), is conchologically similar to both taxa, suggesting that either discoidal shell shapes evolved independently in several lineages of the Geomitrini or that the ancestor of Hystricella, Callina, Lemniscia, Wollastonaria, Spirorbula,  Other genera, e.g. Actinella and Caseolus in the sense of Bank (2009), were also shown to be polyphyletic (Fig. 3). The two species that were studied here and classified by Bank et al. (2002), Seddon (2008), Groh et al. (2009) or Bank (2009 in the subgenus Plebecula of Actinella were not grouped within Actinella. Furthermore, they did not even form a clade together (Fig. 3). The sister-group relationship of Domunculifex littorinella (Mabille, 1883) (Fig. 2k) and Serratorotula (Fig. 2g) remains somewhat questionable because it only received significant support from the BI analysis (Fig. 3) and both groups differ in their number of appendiculae (Fig. 4). The nominal species Helix littorinella Mabille, 1883 differs from P. nitidiuscula in the structure of the genital organs, most markedly in the internal structure of the vagina, which is equipped with a very conspicuous and large, fleshy fold in H. littorinella (Fig. 8b-d). Taken together with the results of our phylogenetic analyses, the introduction of a new genus, Domunculifex Brozzo, De Mattia, Harl & Neiber, for H. littorinella appears therefore justified. A strongly developed vaginal fold is not known from any other Geomitrini and is here regarded as an apomorphy of the new genus.
Helicomela (Fig. 2r), which contains only one extant species, was hitherto regarded as a subgenus of Caseolus, from which it is distinguishable, e.g., by the much larger, globular shell. It was resolved, however, as the sister group of Pseudocampylaea, although only supported in the BI analysis. Helicomela differs from Pseudocampylaea not only by the more globular shell but also in the number of appendiculae inserting into the vagina. While Helicomela possesses two appendiculae, one appendicula has been lost in Pseudocampylaea, which suggests the previous classification of Helicomela based on anatomical characters (e.g. Seddon, 2008) was probably misled by the high variability of this trait (Fig. 6). It is noteworthy that species within the clade including Helicomela and Pseudocampylaea have evolved exceptionally large adult body sizes, with P. lowii and H. bowdichiana (F erussac, 1832) being the largest representatives of these lineages, respectively.
The absence of larger predators on Porto Santo and surrounding islets in the past may have favoured the evolution of larger body sizes in these snails as has also been put forward as an explanation for other cases of island gigantism (e.g. Barahona et al., 2000). Both taxa are meanwhile thought to be extinct. While H. bowdichianus probably went extinct shortly after human settlement in the 15th century, P. lowii survived at least until the end of the 19th century on the small islet Ilh eu de Cima off Porto Santo (Seddon, 2008(Seddon, , 2019. Causes for the decline of the species are unknown, but may possibly be the consequence of habitat destruction or alteration and the introduction of predators. Aside from high variability of shell morphology and shell size, the estimation of the ancestral state of the number of appendiculae (Fig. 6) showed also a high variability of this trait within the Geomitrini, as it is also common within other Helicoidea (Razkin et al., 2015and Chueca et al., 2018. Of the three losses of one appendicula, one probably occurred along a branch which was positively supported in the MP analysis as well as in the BI analysis which is leading to a clade including Disculella, Discula, Steenbergia, Spirorbula, Lemniscia, Testudodiscula, Callina, Hystricella and Wollastonaria (Figs 3  and 6). Within this clade, the appendiculae were entirely lost along the branch leading to the positively supported clade including Steenbergia and Spirorbula, with the glandulae mucosae being also lost in Steenbergia (Figs 3 and 6). The other two losses of one appendicula occurred along branches that were only supported in the BI analysis, which therefore requires further research to confirm these results.

Timing of diversification and ancestral area estimation
The onset of diversification of Geomitrini was dated in our analysis at 13 Ma (Fig. 5). It has to be noted that the estimates of node ages have to be interpreted with caution because HPD intervals are often relatively large and in part overlapping. Accepting the fossil calibration, diversification patterns of Geomitrini largely coincide with the emergence of the present-day islands (Geldmacher et al., 2000). The oldest known fossil, referred to Caseolus (Leptostictea) sp. by Groh (1984) because of conchological similarity, originates from > 13 Ma old Miocene deposits of Ilh eu de Cima off the south-eastern coast of Porto Santo. Whether this fossil can be assigned to Caseolus or represents a different, possibly extinct lineage close to the most recent common ancestor of Geomitrini can currently not be answered with certainty. This is due to the overlapping HPD intervals (Fig. 5) and the depressedglobular shell with a blunt keel which, according to Groh (1984), could represent the ancestral shell form of Geomitrini ("Grundplanvertreter"). The split of the representatives of Caseolus s. str. and C. (Leptostictea) was only dated at 1.9 Ma (HPD: 0.8-3.5 Ma) rendering an assignment of the fossil taxon to Leptostictea at least questionable.
The split of some lineages (Plebecula clade, Actinella + Geomitra clade, Disculella clade) occurring on Madeira (and on the Ilhas Desertas) from their respective sister groups on Porto Santo and surrounding satellite islets predates, however, the formation of the subaerial parts of these islands. Whether this dating is the result of insufficient phylogenetic resolution, the uncertainty of age estimates or may be attributed to several independent dispersal events to the younger islands and subsequent extinctions of the respective lineages on the older islands cannot be answered with certainty at the moment. However, there are many species and subspecies only known as fossils and many examples of formerly somewhat wider geographic ranges exist suggesting that range contractions/expansions and extinctions must have been rather frequent during the evolution of Geomitrini (Wald en, 1983;Groh, 1984;Cook et al., 1993;Goodfriend et al., 1994Goodfriend et al., , 1995Cameron et al., 1996;Bank et al., 2002;Seddon, 2008;Groh et al., 2009;De Mattia et al., 2018a).
Although our ancestral area estimation places the origin of the present-day diversity of Geomitrini in the Madeiran Archipelago (Fig. 6), the estimation of Madeira as the ancestral area is contradictory to the sequence of island emergence (Porto Santo-Madeira-Ilhas Desertas; Geldmacher et al., 2000). This is interpreted as an artefact of the analysis here as the algorithms implemented in RASP 3.02 do not allow temporal layering to be taken into account (Yu et al., 2015). Furthermore, the decisive position of Plebecula from Madeira was only supported in the BI analysis (Fig. 4). However, it appears likely that extinction rates on Porto Santo were higher because of the smaller size of that island so those old lineages might have been lost on Porto Santo. Because of the lack of Moreletina from the Azores in the phylogenetic analyses, with vaginal fold, epiphallus and penis with exposed penial papilla. (d) Inner structure of vagina with exposed vaginal fold and penis with exposed penial papilla. we can also not clarify whether the Madeiran Archipelago or the Azores was the ancestral area of Geomitrini. Our ancestral area estimation points towards an origin of Geomitrinae in Europe, more precisely the Iberian Peninsula (not shown) probably as a consequence of recovering Ponentinini as the sister group of Cochlicellini plus Geomitrini. Outgroup sampling is, however, not dense enough to allow a definite conclusion on whether the Geomitrini reached the Madeiran Archipelago from Europe or North Africa. The genus Cochlicella F erussac, 1821 occurs autochthonously in the Western Mediterranean region of Europe, Africa and the Canary Islands (Richter et al., 2019), while Monilearia Mousson, 1872 andRipkeniella Hutterer &Gittenberger, 1998 are restricted to the Canary Islands and Obelus Hartmann, 1842 is known from the Canary Islands and the adjacent coastal regions of northwestern Africa (Mousson, 1872;Wollaston, 1878;Schileyko and Menkhorst, 1997;Hutterer and Gittenberger, 1998;Ib añez et al., 2003Ib añez et al., , 2006Groh, 2014;Neiber, 2015;Neiber et al., 2016).
Further investigations will have to show, whether Cochlicellini originated in Europe or north-western Africa and colonised the Canary Islands from either of these regions. This will have implications for the origins of Geomitrini because an origin of Cochlicellini in north-western Africa would suggest an African affinity for Geomitrini as well. An origin of Cochlicellini in Europe, on the other hand, would also indicate an origin of Geomitrini in that region.
Affinities with north-western Africa have been suggested for the helicid genus Theba Risso, 1826. However, it is not entirely clear whether Africa and subsequently south-western Europe have been colonised by this genus from the Canary Islands or the other way around (Greve et al., 2010;Haase et al., 2014;B€ ockers et al., 2016). The colonisation of the Atlantic archipelagos from south-western Europe has been suggested for the helicid genus Hemicycla Swainson, 1840 (Canary Islands; Chueca et al., 2015) and the hygromiid genus Leptaxis Lowe, 1852 (Madeiran Archipelago, Azores, Cape Verde Islands and fossil also from the Canary Islands; Caro et al., 2019). The analyses of Caro et al. (2019) suggest that Leptaxis have colonised the Azores from the Madeiran Archipelago. Moreletina from the Azores is the only taxon in the Geomitrini that is not known to occur in the Madeiran Archipelago. It would be interesting to test whether this taxon also reached the Azores from the Madeiran Archipelago as appears the case for Leptaxis. Unfortunately, no material was available for study.
Although taxon sampling is not complete, it is already clear from our analyses that inter-island dispersal must have played an essential role in shaping the present-day distribution of taxa in the Geomitrini ( Fig. 6) and especially that Madeira (and the Ilhas Desertas) has been colonised recurrently from Porto Santo (Fig. 6).

Radiation of the Geomitrini
Gittenberger (1991) defined radiation relatively vaguely as the evolution of a relatively large, monophyletic group of species or higher taxa. That author further distinguishes adaptive and non-adaptive radiations, with the latter being the evolutionary diversification from a single clade, not accompanied by relevant niche differentiation. This kind of radiation may, however, involve sexual selection (Sauer and Hausdorf, 2009;Sauer and Hausdorf, 2010). Feh er et al. (2018) showed that the rate of niche differentiation appears to have lagged behind compared to phylogenetic divergence in rock-dwelling door snails (Clausiliidae) and viewed this as strong support for the hypothesis that speciation in this group of land snails was driven primarily by non-adaptive processes. Contrarily, adaptive radiation in the sense of Schluter (2000) is the evolution of ecological and phenotypic diversity within a rapidly multiplying lineage involving the differentiation of a single ancestor into an array of species that inhabit a variety of environments and that differ in the morphological and physiological traits used to exploit those environments.
Our phylogenetic analyses show that Geomitrini form a monophyletic group (Fig. 4), the diversification of which probably started shortly after the emergence of the present-day islands of the Madeiran Archipelago in the Miocene (Fig. 5). In that sense, the Madeiran Geomitrini represents a radiation according to the above-cited definition of Gittenberger (1991). The timing of the diversification of Geomitrini in the Madeiran Archipelago suggests a gradual pattern of diversification for the Geomitrini as a whole (Fig. 5) rather than only an initial burst of lineage diversification. It has to be noted, however, that extinctions have been frequent in the land snail fauna of the archipelago as evidenced by the numerous fossil taxa that have been described (Bank et al., 2002;Teixeira et al., 2019;Walther and Groh, 2020). Therefore, it cannot be ruled out that some of the comparatively long internal branches in the dated phylogeny (Fig. 5) may actually be the result of extinction rather than the absence of diversification. Cook and Pettitt (1979) suggested for Discula from Madeira a direct adaptive component, i.e. they viewed the development of a keel and increased granulation as likely to increase shell strength and give protection from wear and conceivably predation. Somewhat in contrast, Cameron et al. (1996) concluded, based on data in part obtained from Geomitrini, that adaptive responses have occurred in the land snails from Porto Santo, but that much of the species richness can be interpreted as non-adaptive in the sense of Gittenberger (1991). Although the broad range of shell sizes and different shell forms and ornamentations in Geomitrini (Fig. 2) points towards an adaptive component in the Geomitrini radiation, a detailed interpretation of the contribution of adaptive and/or non-adaptive processes in the sense of Gittenberger (1991) and Schluter (2000) is currently not possible as detailed data on phenotype-environment correlations or the utility of traits are not available for these land snails. The Geomitrini may, however, serve as an excellent model system to study the putatively complex interplay of adaptive and non-adaptive processes, for which the results presented here can provide the basis.