Species distribution, hybridization and connectivity in the genus Chionodraco: Unveiling unknown icefish diversity in antarctica

The species of the genus Chionodraco (Notothenioidei) are the most abundant icefish on the continental shelf of the Weddell Sea. While previous studies indicated that only Chionodraco hamatus and Chionodraco myersi inhabit the Weddell Sea, the third Chionodraco species, Chionodraco rastrospinosus, was recently sampled in the area. As C. rastrospinosus is supposed to be found only at the Antarctic Peninsula and Scotia Arc, this study aimed at confirming the species classification of C. rastrospinosus by molecular methods and identifying its putative source population. Given the documented evidence of introgression among the three species, we tested whether the newly found C. rastrospinosus shared any genetic variability with the other Chionodraco species. To explain the pattern of distribution of the Chionodraco species, we aimed at estimating the hydrodynamic connectivity between the Antarctic Peninsula and the Weddell Sea.


| INTRODUC TI ON
The continental shelf of the Weddell Sea (Antarctica) is still largely unprotected. This region is one of the most pristine and relatively unexplored hotspots of marine biodiversity in Antarctica, and it is less affected by sea surface warming compared with other, more exposed, areas as the Antarctic Peninsula (Teschke et al., 2016).
Under a climate change scenario, Griffiths et al. (2017) hypothesized that the Weddell Sea continental shelf will be likely a sink area for species' range shifts and a feasible refugium for highly cold-adapted benthic organisms. The Weddell Sea hosts different marine habitats that support a rich fish community dominated by the suborder Notothenioidei. Notothenioids comprise 142 species, 80% of them endemic to Antarctica (calculated from the list of notothenioid species curated by J. T. Eastman and R. R. Eakin, version 15 December 2016, personal communication). They represent ~ 50% of all vertebrate species of the Antarctic continental shelf and constitute the 90%-95% of fish biomass, thus dominating the ichthyofauna of the Southern Ocean (Matschiner et al., 2015;Peck, 2018).

| The diversity of Chionodraco species in the Weddell Sea
Two notothenioid species of the icefish genus Chionodraco, C. hamatus and C. myersi (family Channichthyidae) are among the most abundant icefish species on the Weddell Sea continental shelf (Eastman & Eakin, 2000;Fischer & Hureau, 1985). Chionodraco hamatus and C. myersi are found on the continental shelf all around Antarctica with the exception of the Antarctic Peninsula and Scotia Sea islands. These two species prefer slightly different but overlapping depth ranges (4-972 m for C. hamatus and 99-926 m for C. myersi ;Eastman, 2017).
The genus Chionodraco includes a third species, C. rastrospinosus, only found off the Antarctic Peninsula and around the South Shetland and South Orkney islands at up to 1,000 m depth (Eastman, 2017;Gon & Heemstra, 1990;Kock, 1992). While previous studies indicated that C. hamatus and C. myersi are the only Chionodraco species inhabiting the Weddell Sea, a recent expedition on the continental shelf of the south-eastern Weddell Sea has recorded, for the first time, the presence of several individuals identified as C. rastrospinosus in sympatry with the other two congeneric species (Knust & Schröder, 2014).
The three species are morphologically very similar, and their identification is based on subtle differences, such as the presence/ absence of a rostral spine and number and morphology of gill rakers (Fischer & Hureau, 1985;Gon & Heemstra, 1990). The three Chionodraco species are phylogenetically very close, and their divergence was estimated to have occurred ~ 2 mya (Near et al., 2012).
Chionodraco myersi probably diverged earlier, around 1.8 mya, from the clade that includes the two sister species C. hamatus and C. rastrospinosus (Near et al., 2012). Given these premises, the first aim of this study was to formally confirm by molecular methods the taxonomic identification of C. rastrospinosus specimens collected on the continental shelf of the south-eastern Weddell Sea and to identify the putative source population of this pool of individuals.

| Incomplete reproductive isolation among Chionodraco species in the Weddell Sea
The palaeoclimatic events at which notothenioids have been exposed after the acquisition of the ability to synthesize antifreeze glycoproteins (AFGPs) are generally invoked to explain the mechanisms that could have driven species diversification in these fish (Near et al., 2012). On one hand, repeated cycles of glaciation have periodically isolated populations in remaining ice-free refugia providing a mechanism for allopatric speciation (Barnes et al., 2006;Hewitt, 1997;Rogers, 2007). On the other hand, interglacial times may have provided context for population expansions subsequent to glacial retreat (Janko et al., 2007;Matschiner et al., 2009;Zane et al., 2006) and for secondary contacts between recently diverged species . Marino et al. (2013) suggested that secondary contacts between Chionodraco species have led to natural events of hybridization that left signatures of introgression in their genomes. The authors found mixed ancestry and similar allelic frequencies at 18 microsatellites, providing molecular evidence of past and present hybridization among the three species.
Moreover, approximate Bayesian computation (ABC) simulations suggested that the shared alleles could bear the signature of introgression occurred most likely during the last two major interglacial times . Despite evidence of interspecific gene flow, no specific hypothesis has been brought up so far for where a geographic contact zone for the three species could be located. If genetically confirmed, the presence of C. rastrospinosus in the Weddell Sea would suggest the hypothesis that this area may be a possible contact zone for the three Chionodraco species. Therefore, the second aim of this study was to investigate whether the newly found C. rastrospinosus specimens may carry any trace of introgression from other Chionodraco species and to describe hypotheses need to be tested in future studies about the mechanisms maintaining the interspecific connectivity in Chionodraco spp.

K E Y W O R D S
Antarctic continental shelf, clustering analysis, connectivity, gene flow, hybridization, Lagrangian modelling, microsatellite, species distribution, Weddell Sea patterns of contemporary genetic differentiation and hybridization at interspecific scale.

| Hydrodynamic connectivity
The current system in the area of the Antarctic Peninsula and Weddell Sea is dominated by the Antarctic Circumpolar Current (ACC, Figure 1). The ACC flows eastward over the slope off the western Antarctic Peninsula, moves seaward offshore of the South Shetland Islands (Figure 1) (Ryan et al., 2016) and continues east to form the northern border of the Weddell Gyre (WG, Figure 1), a mainly wind-driven, cyclonic ocean gyre (Ryan et al., 2016). In contrast to the ACC, the Antarctic Slope Current (ASC, Figure 1) is a near-circumpolar, westward feature that forms a narrow, rapid flow across broad sections of the continental shelf. In the Weddell Sea, the ASC is thought to form the southern border of the WG before reaching waters north of the Antarctic Peninsula (Caccavo F I G U R E 1 Map of sampling locations for the three Chionodraco species and schematic representation of the main marine currents occurring in the region considered in this study. The figure represents the regions of the Antarctic Peninsula and the Weddell Sea. The dots show the sampling locations of the specimens analysed in this study (red for C. hamatus, yellow for C. myersi and blue for C. rastrospinosus).
The arrows indicate the main marine currents in the area: yellow for the Antarctic Circumpolar Current (ACC), orange for the Weddell Gyre (WG), blue for the Antarctic Slope Current (ASC) and green for the Antarctic Coastal Current (AACC). The map was initially generated with the software environment R ver. 4.0.0, using packages ggplot2 (Wickham, 2016) and sf (Pebesma, 2018) for plotting, and the package rnaturalearthdata (South, 2017) to access and download Natural Earth (http://www.natur alear thdata.com) map dataset, and refined by hand Meredith & Brandon, 2017;Orsi et al., 1995;Ryan et al., 2016;Thompson et al., 2018). Inshore, the Antarctic Coastal Current (AACC, Figure 1) occupies the upper part of the water column, transports shelf waters between glacial trough systems of the Weddell Sea and merges with the ASC off the eastern Antarctic Peninsula (Graham et al., 2013).
Such transport pathways, coupled with a relatively long pelagic passive larval stage (of up to five months) (La Mesa et al., 2013), can be expected to contribute to the exchange of Chionodraco individuals by connecting habitats of the Antarctic Peninsula and Weddell Sea.
The contribution of adult movement to dispersal is expected to be smaller than via passive larval dispersal as adults of the Chionodraco species are bentho-pelagic (Eastman & Lannoo, 2004;Eastman & Sidell, 2002). Therefore, the third and final aim of this study was to assess the hydrodynamic connectivity of the three species in terms of preferential routes of Chionodraco larval dispersal between the Antarctic Peninsula and the Weddell Sea. We wanted to test whether the dispersal of C. rastrospinosus occurs preferentially along The two samples differ in size as it was not possible to sequence the mtDNA D-loop for 4 individuals.
the ACC, with larvae that get entrained in the Weddell Gyre and are transported from the Antarctic Peninsula to the Weddell Sea.
We wanted also to verify whether the pelagic passive larval dispersal of C. hamatus and C. myersi could occur along the shelf from the south-eastern Weddell Sea towards the Antarctic Peninsula driven by coastal currents (AACC and ASC) and the Weddell Gyre.
To meet the aims of this study, we genotyped a panel of 19 microsatellites and sequenced 326 bp of the mitochondrial D-loop of 560 individuals. For the three species, we simulated more than 3 million drifters and analysed the dispersal routes estimated by the model starting from different release areas in the Antarctic Peninsula and in the Weddell Sea.
Specimens were collected by bottom trawl, weighted and measured and assigned to species according to morphological characteristics reported in Gon andHeemstra (1990) andFisher &Hureau (1985). The total length range for sampled specimens was 14.5-49 cm for C. hamatus, 24.5-38 cm for C. myersi and 7-54 cm for C. rastrospinosus. A piece of muscle or fin clip was collected from each specimen and preserved in ethanol 95% for the subsequent molecular analysis. Sample collection conducted on specimens obtained during R/V Polarstern cruises and analysed for the first time in this study was approved by the competent national authority for Antarctic research (Umweltbundesamt, UBA, Germany).
Genomic DNA was extracted from 456 specimens following a standard salting out protocol (Patwary et al., 1994

| Microsatellite amplification and genotyping
A total of 456 individuals were newly genotyped in this study for a panel of 19 microsatellites. All loci were previously characterized for the three Chionodraco species by Agostini et al. (2013).
Detailed amplification conditions follow Marino et al. (2013) and references therein (see also

| Identification of outlier loci
To detect possible outlier markers showing extremely low or high divergence among species, we used the approach implemented in the software bayeScan ver. 2.1 (Foll & Gaggiotti, 2008). The following parameters were used: burn-in = 100,000, thinning interval = 100, sample size = 10,000, number of pilot runs = 50, length of each pilot run = 10,000 and prior odds = 10. To estimate the number of clusters in our dataset, and the extent of admixture among clusters, we used STrucTure ver. 2.3.4 (Pritchard et al., 2000). For each run, we set a burn-in of 100,000 steps followed by 1,000,000 MCMC iterations, using the admixture model and independent allele frequencies. We tested from 2 to 6 K putative clusters, with 10 runs for each K following indications from Evanno et al. (2005) (see also Porras-Hurtado et al., 2013 for a review). The most probable value of K was identified with the method described in Evanno et al. (2005) and implemented in STrucTure harveSTer ver. 0.6.94 (Earl & vonHoldt, 2012). The different runs for each K were averaged with clumPak (Kopelman et al., 2015) and the result plotted with the R package "Pophelper" ver. 2.3.0 (Francis, 2017).

| Detection of hybrids
We ran STrucTure with the USEPOPINFO model (burn-in of 100,000 steps, 1,000,000 MCMC iterations, independent allele frequencies) to identify hybrids or migrants and the extent of introgression when pre-defined groups (in our case the three Chionodraco species) correspond almost exactly to STrucTure clusters. To estimate the optimal threshold values to distinguish an individual as admixed and to test the power of our dataset to discriminate between pure and hybrid individuals, we followed a simulation approach as performed by Vähä and Primmer (2006). For each species, we selected the individuals with a Q-value > 0.99 (pure individuals) from the former STrucTure analysis and used these individuals as input for the software hybriDlab ver. 1.0 (Nielsen et al., 2006). These pure individuals were virtually crossed randomly in a simulation with hybriDlab to generate 810 simulated new pure individuals for each species and 30 simulated new admixed individuals for each of the following crosses (F1 hybrid groups): C. hamatus × C. rastrospinosus, C. myersi × C. rastrospinosus, and C. hamatus × C. myersi. We also generated six simulated new groups by backcrossing individuals of the three F1 hybrids with one of the two parental species. Sample sizes for every group followed indications by Vähä and Primmer (2006) and aimed to obtaining a 10% of admixed individuals. This simulated dataset was analysed with STrucTure applying the same settings as for the former run. The resulting Q-values for the real and simulated datasets were plotted in a box plot with boxPloTr (http://shiny.chemg rid.org/boxpl otr/), and the distributions were compared.
Admixture was analysed also with the software neWhybriDS ver. 1.1 (Anderson & Thompson, 2002). Differently from STrucTure, which treats the Q-value as a continuous variable to assign individuals to a number of clusters specified by the user, neWhybriDS calculates the probability that each individual belongs to these categories: pure, F1, F2 (F1 x F1) and the two types of backcross (F1 × parental pure species). In nature, many degrees of admixture are expected, but this approach aims to investigate the early stages of hybridization (F1, F2 and backcrosses). As neWhybriDS works with two species at a time, we ran three pairwise comparisons. For each run, a burn-in of 100,000 steps, 1,000,000 MCMC iterations and uniform priors were set. We ran neWhybriDS also with the simulated dataset gen-

| Mitochondrial DNA amplification and sequencing
A 326 bp D-loop sequence was obtained for 560 individuals, including 457 specimens newly analysed in this study and 103 originally used by Marino et al. (2013) but never sequenced for this mitochondrial marker ( Table 1). The number of samples genotyped for microsatellites (N = 564) and sequenced for the mitochondrial marker (N = 560) differs as it was not possible to sequence the D-loop for 4 individuals (see Table 1 for more details). The mitochondrial D-loop was used in this study as previous studies (Damerau et al., 2012;Deli Antoni et al., 2019;Hüne et al., 2015;Matschiner et al., 2009;Patarnello et al., 2003;Zane et al., 2006) already demonstrated that this marker carries sufficient variability to resolve the population genetic structure and allows the species differentiation in notothenioids. Patarnello et al. (2003) clearly distinguished the three Chionodraco species and identified population clusters within the three species by means of a 249 bp D-loop marker.

| Mitochondrial DNA variability
D-loop sequences were aligned with cluSTal omega (Sievers et al., 2011), and descriptive statistics (segregating sites, number of haplotypes, haplotype diversity and nucleotide diversity) for each species and population sample were obtained with DnaSP ver. 6.12.03 (Rozas et al., 2017). To identify possible patterns of genetic structure within each species, a haplotype network was built with PoParT ver. 1.7 (Leigh & Bryant, 2015) using the TCS algorithm (Clement et al., 2002).
Individuals were assigned to their maternal parental species according to their position in the network. Patterns of population structure and species differentiation were investigated by estimating F ST and average number of pairwise differences (Pi) with arlequin.

| Connectivity analyses
To estimate the possible routes of dispersal of each Chionodraco spp.
from the Weddell Sea and from the Antarctic Peninsula under the assumption that connectivity may be mainly driven by pelagic passive larval dispersal, we used the mathematical model coherenS ver. 2 (Luyten, 2011). coherenS includes a Lagrangian particle tracking module able to compute the trajectory of particles that are virtually released into the ocean. In this study, it was assumed that each particle (as a proxy for a virtual larvae) passively drifts following the surrounding water masses. The Lagrangian approach included The model domain covers the Southern Ocean with northern and southern boundaries at 45° south and 80° south, respectively. The grid resolution is 1/12 degree in the horizontal and 50 vertical sigma levels (from 0 to 5,500 m). Once the particle leaves the model domain, it is lost and does not come back. The model time step is 5 min.
The dispersal pattern of the pelagic larval stage was modelled in the context of several life history traits of the three Chionodraco species. For every species, we simulated the sites of particle release (virtual larvae) from locations on the continental shelf where the specimens analysed in this study were caught (see Table 1 for corresponding sampling campaigns and Table S2 in Supporting Information Appendix S1 for the coordinates of the release locations) as a proxy for spawning areas because precise data on spawning areas for the three species are not known. As specimens analysed in the genetic section of this study were all big enough for not being passive dispersers, we did not distinguish between juvenile and adult stage in the definition of potential spawning areas for the hydrodynamic connectivity section. As a proxy for spawning areas, we also considered where some larvae of C. rastrospinosus were recently sampled (cruise PS112, M. La Mesa, personal communication, see Table S2 in Supporting Information Appendix S1 for details, samples not available for the genetic analyses performed in this study). Three locations were very close to the coast (see Table S2 in Supporting Information Appendix S1 for details) and were adapted according to the model land/sea mask.  (Ekau, 1991;Kock, 2005;La Mesa et al., 2013;Vacchi et al., 1996) Out of 361 probability tests for HWE, 23 showed a significant departure from HWE after correction for multiple tests (Table S4 in Supporting Information Appendix S1). Five out 513 comparisons showed the presence of linkage disequilibrium after correction for multiple tests. None of these tests involved the same pair of loci in the three species, and none of these loci were in linkage in previous studies so these results were attributed to chance.  Appendix S1).
The genetic assignment with STrucTure attributed 12 specimens (out of 179 Chionodraco individuals collected during the same cruise in the Weddell Sea) to the species C. rastrospinosus. These individuals were also morphologically identified as C. rastrospinosus during the sampling cruise.
The most probable number of clusters inferred from STrucTure is three (Figure 2). These groups corresponded to the three species under study and confirmed the original morphological identification.
Some individuals showed admixed ancestry.

| Detection of hybrids
Three putative hybrids, one with C. hamatus and C. rastrospinosus ancestry and two with C. myersi and C. rastrospinosus ancestry, were iden- and are more likely F2 or a backcross. Appendix S1).  Figure 4).

Interspecific differentiation tests based on mitochondrial data
were all statistically significant (Tables S9 and S10 in Supporting Information Appendix S1 population samples for the three species are reported in Tables S11-S16 in Supporting Information Appendix S1.

| Connectivity analyses
Particles simulating larvae of C. rastrospinosus were virtually released in the model from the Antarctic Peninsula (blue dots in Figure 5a) every day during the period from August to November. Individuals are assumed to passively follow the water masses and to drift during a 5-month period after which they die. Model simulations were run for a time range of 10 years to account for interannual variability.
All simulations were combined, and resulting dispersal patterns of C.
rastrospinosus are presented in Figure 5b. Most individuals of C. rastrospinosus consistently moved north-eastward, following the ACC and did not cross the sub-Antarctic front nor reach the Weddell Sea.
Only a minority of individuals first drifted southward before being transported north-eastward.
Particles simulating larvae of C. hamatus and C. myersi were released from the south-eastern Weddell Sea (where specimens used in this study for genetic analysis were caught, see Table 1, and see red and yellow dots in Figure 5a). According to the sim-

| D ISCUSS I ON
This study provides the first evidence of C. rastrospinosus on the con- Hatch marks on the links between circles represent the number of mutations between haplotypes of the allelic and haplotypic frequencies at the temporal and geographic scale. Genetic drift could act with different strength on the two markers used in this study justifying the stronger differences indicated by the mitochondrial marker (F ST and Pi, Tables S11-S14 in Supporting Information Appendix S1) compared with the microsatellites (F ST , Tables 2 and S6 in Supporting Information Appendix S1).
It is also possible that C. rastrospinosus has always been present in the Weddell Sea and has gone undetected due to its low abundance We can, however, speculate that other mechanisms maintain the connectivity and species distribution pattern observed in this study. Some distance could be covered by actively swimming individuals (juveniles and adults), or larvae may be transported further F I G U R E 5 Passive larval dispersal patterns estimated for each Chionodraco species and based on coherenS particle module. Panel a: location of spawning grounds from where the passive dispersal phase of the larvae starts (red for C. hamatus, yellow for C. myersi, blue for C. rastrospinosus) (see Table S2 in Supporting Information Appendix S1 for the exact coordinates); panels b, c and d: geographic dispersal patterns of the fish larvae during their pelagic phase as estimated from model results generated

| Incomplete reproductive isolation among Chionodraco species in the Weddell Sea
Besides documenting a potentially wider distribution than previously known for C. rastrospinosus, our study shows that the three Chionodraco species can be clearly separated and that pairwise interspecific distances based on microsatellites and D-loop reflected the classical phylogenetic relationships with C. myersi being generally more distant from the two sister species C. hamatus and C. rastrospinosus (Near et al., 2012). Nonetheless, the admixture analyses with STrucTure and neWhybriDS showed instances of interspecific introgression and hybridization. These results are in agreement with previous observations by Marino et al. (2013) although the specimens described as hybrids in this study are different. This could be due to our five times larger and more recent sample set, which allowed a more precise estimation of allele frequencies. This result could also depend on the different criterion used by Marino et al. (2013) for the classification of the pure and introgressed individuals. In fact, Marino et al. (2013) identified 29 hybrid individuals based on the Q-value < 0.95 (Table S17 in Supporting Information Appendix S1) while we decided to apply a more conservative approach and suggested as putative hybrids only those individuals identified as hybrids by the USEPOPINFO option in STrucTure and with a Q-value non-overlapping with the pool of pure individuals based on our simulations ( Figure 3). Therefore, considering our approach, none of the individuals identified by Marino et al. (2013) would be considered as putative hybrids and only seven would result as admixed (Table S17 in Supporting Information Appendix S1).
In addition, previous observations by Marino et al. (2013) of hybridization between C. rastrospinosus and the other two Chionodraco species were mainly explained by past events of secondary contacts during interglacial periods and were less supported by the geographic distribution of the species as it was known before this study. In fact, the possibility of hybridization events and introgression also at present temporal scale finds additional support given our evidence of sympatric occurrence of the three Chionodraco species in the Weddell Sea. Some theoretical and methodological caveats limit our power to identify hybrids and introgressed individuals with absolute certainty.
In fact, in recently diverged species with a history of isolation during glacial periods and secondary contacts in post-glacial times (Near et al., 2012), it is often difficult to separate shared polymorphisms due to secondary admixture from those owing to common ancestry (incomplete lineage sorting-ILS), and because both processes frequently co-occur (Allendorf et al., 2001;Qu et al., 2012). In this study, we have attempted to solve this complexity by analysing two marker types with different evolutionary pace and inheritance Considering the above caveats, we decided to adopt a conservative approach and considered as hybrids only three individuals that showed a Q-value completely outside the range of pure individuals and that, at the same time, were identified as hybrids in STrucTure using the USEPOPINFO function. Although possibly underestimated, the percentage of individuals with admixed ancestry (three hybrids and three specimens with nuclear-mitochondrial discordance; see below) that we found (~1%) is in line with the level of abundance reported in literature for sympatric species (1/100 or 1/10,000; Mallet, 2005). Moreover, independently from the adopted marker, after few generations the signal of introgression fades away, becoming less than 1% after seven generations. Thus, in many situations only the discordance between nuclear DNA and mitochondrial DNA can detect an individual with admixed ancestry (Weisrock et al., 2005). In our dataset, we also identified three specimens from the Antarctic Peninsula, which, according to STrucTure, were assigned to C. rastrospinosus with ~ 100% probability but exhibited a mitochondrial D-loop haplotype from C. hamatus. This case of nuclear-mitochondrial discordance indicates that these specimens are descendants of hybrid individuals and that they or their ancestors must have migrated or been transported from the Weddell Sea to the Antarctic Peninsula. This conclusion suggests that, despite the indication of the Lagrangian simulation, opportunities of connection might occur between the two areas.
According to STrucTure, two hybrid individuals resulted from the cross between C. myersi and C. rastrospinosus and one derived from C. hamatus × C. rastrospinosus. Thus, a reproductively active C. rastrospinosus population in the Weddell Sea may be postulated.
The specimens of C. rastrospinosus collected in the south-eastern Weddell Sea and analysed in this study have a heterogeneous composition with both males and females at different stages of maturity (but none of them were larval stages). Among these, a special instance of a gravid C. rastrospinosus female (gonadosomatic index of 24% of total weight and 39% of gutted weight, maturity stage 4 following the scale by Kock & Kellermann, 1991) (Kock & Kellermann, 1991;Papetti et al., 2007;Ruzicka, 1996).
Therefore, it could be hypothesized that some individuals of C. rastrospinosus have adopted a similar change in the reproductive time in the south-eastern Weddell Sea.
Another pre-zygotic barrier can consist of differences in reproductive behaviour that allow for intraspecific recognition. Males of C.
hamatus display nesting behaviour: they prepare nests on the sea floor where females lay eggs subsequently guarded by males until hatching (Ferrando et al., 2014). Although nest construction has never been demonstrated conclusively in C. rastrospinosus and C. myersi, evidence of hybridization among Chionodraco species may represent an indirect proof of a similar nesting behaviour. Desvignes et al. (2019) also mentioned that the characteristics of icefish nests and the reproductive behaviour are usually species-specific. In our study, none of the three hybrid individuals carry a C. rastrospinosus genetic contribution as a maternal species and three individuals from the Antarctic Peninsula, indicated as pure C. rastrospinosus with microsatellite data, have a C.
hamatus D-loop haplotype (black arrows in Figure 4). This preliminary observation leads to speculate that if interspecific mating occurred among Chionodraco spp., it might preferentially involve males of C. rastrospinosus with females of the other two species.
This remains a speculation owing to the very limited number of observations in our study, and an alternative explanation would imply post-zygotic barriers to hybridization as, for instance, differential em-

| Conclusions
Connectivity among populations is fundamental to shape the genetic diversity and demography of species and the stability of ecological communities (Clobert et al., 2012). In the case of Chionodraco, future sampling and a constant monitoring are needed to fully understand the pattern of connectivity and the evolution of the population of C. rastrospinosus and the hybrids recently discovered in the Weddell Sea. Similar studies should target also other icefishes to understand the extent of interspecific gene flow at the whole family scale and whether this is promoted by environmental changes. Experiments in vitro of interspecific crossing among Chionodraco spp. could also help testing survival of hybrid larvae and assessing fecundity of hybrid adults. Regular surveys may also help discover changes in species richness and early arrivals of allochthonous species. This becomes even more relevant when it comes to propose and support the implementation of marine protected areas.

ACK N OWLED G EM ENTS
The

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13249.

DATA AVA I L A B I L I T Y S TAT E M E N T
The microsatellite datasets supporting the conclusions of this article are available in Genepop format as additional files in DRYAD