Getting off to a good start? Genetic evaluation of the ex situ conservation project of the Critically Endangered Montseny brook newt (Calotriton arnoldi)

Sample collection and genotyping

Tissue samples were taken from 42 individuals (19 founders and 23 descendants) from the captive breeding facility (Table 1), and compared to results previously published from the wild populations (Valbuena-Ureña et al., 2017). Of the 19 founders, nine were from the eastern sector and belonged to population A1, and 10 were from the western sector and belonged to populations B1 and B2 (Fig. 1; see Valbuena-Ureña, Amat & Carranza (2013) for population codification). All F1 descendants were genotyped and are comprised of 17 F1 individuals from the eastern captive line, and six F1 individuals from the western captive line, all born in 2007. These descendants were interesting, as they originated from founder females that were already gravid when sampled from the wild. Therefore, they could exhibit genotypes from unsampled wild males. Due to spatial limitations in the facilities and readjustments of breeding methodology, only F1 experimental matings were designed; F2 is not yet available for genotyping. In total, 26 captive individuals from the eastern breeding line were analysed and compared to 77 genotyped wild animals from this sector (Valbuena-Ureña et al., 2017). Similarly, 16 captive individuals from the western breeding line were genotyped and compared to 83 wild animals from the western sector. The collection of all samples was conducted under the licenses required by the Catalan Government, with the permit numbers SF/298 and SF/469.

Tissue samples consisted of tail tips or fingers preserved in absolute ethanol. Genomic DNA was extracted using QiagenTM (Valencia, CA, USA) DNeasy Blood and Tissue Kit following the manufacturer’s protocol. For consistency, all 42 individuals were genotyped for the same 24 polymorphic microsatellite loci used to examine the wild populations (Drechsler et al., 2013; Valbuena-Ureña, Steinfartz & Carranza, 2014; Valbuena-Ureña et al., 2017). Details of the PCR-multiplex combinations of loci and specific amplification procedures are given in Valbuena-Ureña et al. (2017).

Data analysis

MICRO-CHECKER (Van Oosterhout et al., 2004) was used to check for potential scoring errors, large allele dropout and the presence of null alleles. Pairwise linkage disequilibrium between loci and deviations from Hardy-Weinberg equilibrium in each grouping and for each locus were checked using the software GENEPOP version 4.2.1 (Rousset, 2008). Genetic diversity was measured for each grouping as the mean number of alleles (A), observed (H_O), and expected heterozygosity (H_E) using FSTAT 2.9.3.2 (Goudet, 1995). Formal analyses of raw allelic richness values were not performed, because the number of alleles detected at a locus can be influenced by sample sizes. Therefore a rarefied estimate of allelic richness (Ar) was obtained with HP-RARE (Kalinowski, 2005). The nonparametric Wilcoxon signed-rank test (Wilcoxon, 1945) was used to test for differences in diversity levels between captive and wild groups using locus-specific values of A, Ar and H_E as paired replicates. In analyses based on allelic richness, separate rarefactions were performed for each sector grouping, to account for the smallest sample size from a particular group. The observed number of private alleles (PA), defined here as alleles found in a single population throughout the study region, for each locus and each grouping was calculated with GDA (Lewis & Zaykin, 2000), and the ratio of presence compared to wild PA was computed. Further, overall and intrapopulation subdivision coefficients (F_IS) for all markers and groupings were calculated using FSTAT 2.9.3.2, which calculates the estimator f of Weir & Cockerham (1984) for each marker, as well as a multilocus estimate. Genetic differentiation between the captive and the wild populations was derived from two parameters, F_ST and R_ST (Slatkin, 1995), values were obtained using ARLEQUIN 3.5.1.2 (Excoffier & Lischer, 2010).

Relatedness among individuals, effective population size, and bottleneck detection

Pairwise values of genetic relatedness (r_xy) were calculated using two estimators: the r_qg89 estimator (Queller & Goodnight, 1989) and the r_lr99 estimator (Lynch & Ritland , 1999). These estimated indices were compared following the Blouin et al. (1996) criteria implemented in iREL (Gonçalves da Silva & Russello, 2009), using the population allele frequencies estimated for each sample. The cut-off values were applied following the method described in Blouin et al. (1996), in order to classify individuals into relationship categories: unrelated (UR), half-siblings (HS), full siblings (FS) and parent–offspring (PO). Using this method, there is a greater than zero probability of misclassifying individuals if observed values of relatedness fall outside theoretically expected values (Russello & Amato, 2004). To minimize this error, the cut-off values specific for our samples were calculated using the Monte Carlo simulation procedure implemented in Blouin et al. (1996). In total, 10,000 dyads in each of the four relatedness categories were randomly generated using genotypes and allele frequencies specific for each eastern and western captive population as input data; r_qg89 and r_lr99 estimates for each simulated dyads were then computed. Expected misclassification rates were computed using the cut-off values described in Blouin et al. (1996) (the midpoints between the means of the distributions of pairwise relatedness estimates of each simulation category). Then, the index at which the overlap in the empirical distribution of unrelated and half-sibling dyads was minimized was selected. The relationship category compatible with the observed r_xy value was then determined for each individual pair. The observed distribution of pairwise relatedness among individuals for each population was plotted against those of the four simulated relatedness categories.

The effective population size (N_e) for each breeding line based on one-sample N_e estimator method was calculated with ONeSAMP (Tallmon et al., 2008). ONeSAMP employs approximate Bayesian computation. The analyses were run specifying a minimum N_e value of 2 and a maximum of 100. To detect evidence of a recent bottleneck, we tested for significant heterozygosity excess or deficit with the program BOTTLENECK, version 1.2.02 (Piry, Luikart & Cornuet, 1999). This approach is effective for detecting bottlenecks that have occurred recently, within the past 0.2–4.0N_e generations (Luikart & Cornuet, 1998). The stepwise mutation (SMM) and two-phase mixed (TPM) models were tested with default parameters. We also sought signatures of genetic bottlenecks by calculating Garza and Williamson’s (Garza & Williamson, 2001) M-ratio for each breeding line using ARLEQUIN. The M-ratio is the mean ratio of the number of alleles to the range of allele size. When alleles are lost from a population, the number of alleles decreases at a faster rate than the allelic size range, so small (<0.68) M-ratio values are indicative of populations that have gone through a genetic bottleneck at some time in the past.

Random subsampling of empirical datasets

In order to estimate how many captive individuals would be needed for each population to accurately match the level of the genetic diversity present in wild populations, comparisons of genetic diversity estimates between real and distinct sample-sized simulated populations were conducted following the methodology from Hale, Burg & Steeves (2012). A simulated dataset consisting of 100 replicates for each cluster was constructed, with the following sample sizes: 5, 10, 15, 20, 25, 30, 35, 40, 45 and 50 individuals. Each replicate contained a random subset of individuals from the empirical dataset, created using a macro in Excel. This was designed to assign a random number (between 1 and 10,000) to each individual from the empirical dataset, and then sort the dataset by this number. The macro selects the first five (or 10, 15, etc. depending on the sample size category) to a new worksheet, 100 times, resulting in 100 simulated ‘populations’ that are independent, i.e., random subsamples of the empirical dataset, at each sample size. Empirical datasets comprise all individuals genotyped, including wild and captive specimens. Sampling was done without replacement, so no individual was present more than once in the same replicate (as in a real population genetic dataset). As replicates were independent from each other, the same individual could be present in more than one replicate of the simulated dataset at each sample size. The mean expected heterozygosity (H_E) and the mean percentage of alleles—including all alleles, common alleles (frequency >0.05) and rare alleles (frequency <0.05)—detected at each sample size (mean of the 100 random replicates per size) were then compared to each cluster (eastern sector, western sector, cluster A1A2 and cluster B1B2B4). The mean pairwise F_ST between the 100 random replicates at each sample size and the empirical dataset was also calculated for each dataset. In order to further examine variations between the simulated and empirical estimations, ANOVA with Tukey HSD post hoc tests were performed separately on data from each cluster using Statistica v.7.

Relatedness among individuals, effective population size, and bottleneck detection

Comparison of the relatedness indices of r_qg89 and r_lr99 suggests that the latter performed better at discriminating between unrelated and half-sibling dyads, with fewer unrelated individuals being misclassified as half-sibling dyads following Blouin et al. (1996) (Table S3). All further analyses were therefore based upon the index of Lynch & Ritland (1999). The observed distribution of pairwise relatedness among individuals for each captive population was log normal, with a peak around zero (Fig. 2). Using the empirical distribution derived from 10,000 simulated dyad for each relatedness category, a dyad with a relatedness value of ≥0.1881 and 0.2295 for the eastern and western breeding line respectively, was determined to have a ≤ probability of 0.05 of being unrelated (Table S4). The proportion of dyads classified as unrelated for the eastern and western founder population is 94.44% and 88.89%, respectively (Fig. 3).

The estimated N_e for the eastern captive breeding line was 38.96 (median, 95% CI [34.27–52.25]), and N_e = 17.62 (median, 95% CI [15.30–24.05]) for the western captive breeding line (Table 2). Founders N_e estimates from both breeding lines were roughly 10, while effective population sizes estimated from the F1 were 17.03 and 7.97 for eastern and western captive founder stocks, respectively. No captive population showed any signs of a recent bottleneck under the TPM (one tailed test for heterozygote excess) or SMM models (Wilcoxon one-tailed test for excess of heterozygosity; P > 0.05 for all tests). In contrast, M-ratio tests suggest that both breeding lines experienced bottlenecks, with a Garza-Williamson M-ratio of 0.27 and 0.30 for the eastern and western breeding lines, substantially lower than the threshold of 0.68 considered indicative of a bottleneck.

Size of the captive lines needed to accurately represent the genetic diversity in wild populations

The mean percentages of allele detection were significantly different across the range of sample sizes (P < 0.001, in all four datasets) when considering all alleles, rare alleles, and common alleles (Table S5). In the two former cases, the Tukey HSD posthoc comparisons indicated significant differences among all sample size groups tested. The differences found when considering 20 individuals were not significant when only the common alleles were analysed. This is also the minimum number of individuals needed to detect at least the 99% of the common alleles (Fig. 4). The mean expected heterozygosities were far more precise (lower standard deviation) and accurate with increasing sampling size (ANOVA P < 0.001, in all four datasets), much of the increase occurring when the sample size grew from five to 20 individuals (Tukey HSD posthoc comparisons, Table S5; Fig. 5). Increasing sample size above 20 individuals seemed to have little impact on the mean H_E and its standard deviation for all four datasets. As above, the genetic distance between simulated and empirical datasets decreased as the sample size was increased, but again the decrease was not linear. ANOVAs indicated that Pairwise F_ST between each replicate and the empirical dataset were significantly different between sample sizes (P < 0.001, in all four datasets, Table S5). The difference was mainly from samples 5 to 20–25. No significant differences in pairwise F_ST was detected from 25 individuals upwards (Tukey HSD posthoc comparisons, Table S5), suggesting that there is little to be gained from sampling more than 25 individuals per captive population. In all four datasets, the 0.5% of genetic differentiation from wild populations is achieved with up to 20 individuals (Fig. 6), showing that the captive stock matches 99.5% of the wild genetic composition.

Q1: estimation of the genetic diversity present in the captive stock

The levels of the genetic diversity measured in the captive dataset indicate that the individuals that form the current captive program are good, but not optimal, representatives of the genetic diversity present in the wild. Genetic theory indicates that the founder population should capture 97.5% of the genetic diversity present in the wild populations (Leus, Traylor-Holzer & Lacy, 2011). The current (initial) captive stock of Calotriton arnoldi captures roughly 93–95% of the total genetic diversity previously observed in the wild. Moreover, the percentage of unrelatedness among dyads from both breeding lines does not exceed 95%. Therefore, in order to increase genetic diversity, new genetic material should be incorporated by introducing new unrelated individuals or their sperm. Since this species present sperm storage (Alonso, 2013) and multiple paternity (Valbuena-Ureña et al., 2017), an ideal strategy to maximize the gene pool of the captive population while minimizing the impact on wild populations would be to incorporate wild-caught gravid females. Thus, descendants may retain the genetic diversity from individuals not kept in captivity.

Another issue to consider is that the genetic diversity of the captive stock should be maintained throughout the entire duration of the breeding program. A drop of the effective population size is detected on the eastern captive line (N_e = 39) when compared to the eastern wild populations (N_e = 86; Valbuena-Ureña et al., 2017), as well as in the western captive line (N_e = 18) compared to the western wild populations (N_e = 80). Although it has been shown that this species is able to survive in the wild with very low values of N_e, the possibility of inbreeding depression in captivity should be considered, since the strategies used by this species to maintain high levels of genetic diversity in the wild despite the relatively low number of individuals are not known, or may not be possible in captivity. Genetic drift erodes genetic variation as N_e decreases (Allentoft & O’Brien, 2010; Montgomery et al., 2000), elevating the probability of fixation of deleterious alleles, and reducing the effectiveness of selection. This may lead to a reduction of overall fitness by limiting adaptive responses (Hare et al., 2011). Animal models suggest that, in captivity, a N_e > 50 is needed to avoid the risk of inbreeding depression (Frankham, Ballou & Briscoe, 2010; Traill et al., 2010). Thus, efforts to enhance the N_e of the captive population of C. arnoldi captive should be undertaken.

In keeping with the wild situation, a higher allelic richness observed in the eastern dataset when compared to the western one (Valbuena-Ureña, Amat & Carranza, 2013; Valbuena-Ureña et al., 2017). Further, the high levels of genetic differentiation observed between the eastern and western wild and captive populations suggests that the maintenance of two separated breeding lines is needed. Although outcrossing the two management units may help to increase the genetic diversity (Frankham, Ballou & Briscoe, 2010), we do not recommend it at this point for two main reasons: (a) the importance of maintaining the two genetically and morphologically differentiated groups; and (b) the potential threat to the existing populations due to outbreeding depression (OD) (Allendorf & Luikart, 2007). Existing conservation guidelines might focus on maintaining high levels of genetic diversity within the two distinct breeding lines. In the case that captive-bred individuals are introduced into brooks currently not occupied by this species, it is important to consider the eastern or western geographical position of the brook, in order to avoid possible future interbreeding as a result habitat reconnection following floods in the rivers of each sector.

Q2: increasing the captive breeding population

Increasing the sample size clearly results in a better representation of alleles and expected heterozygosities between the simulated and the empirical datasets. The genetic distance between these two datasets also decreases when sample size is increased. The probability that a particular allele will be included among the founders is determined by its frequency in the population. Although the presence of rare alleles does not contribute much to the global genetic diversity of a certain species, they a enhance populations distinctiveness and adaptive potential. However, as seen in Fig. 4, a very large number of founders (>40 founders by sector/cluster) would be required to represent 80% of the rare alleles. In order to avoid extra costs associated with maintaining oversized captive stocks, it is crucial to detect the point at which increasing the sample size will not result in a significant benefit in terms of allelic representation (e.g., retaining all common alleles and the maximum number of rare alleles, or maximizing the mean expected heterozygosity of the founders; Lacy, 1994). This sample size is defined as the minimum captive stock size to meet the genetic goals.

Our results suggest that a minimum of 20–25 individuals for each captive breeding line should be kept in order to maintain good representation of the genetic diversity. Our results are in accordance with the Amphibian Ark guidelines (Schad, 2008) and Lacy (1994), which recommend breeding at least 20 unrelated founders by sector, with an equal sex ratio.

The expected heterozygosity (H_E) of the eastern captive stock is significantly lower than that of the wild populations (p = 0.0082), indicating that the eastern captive stock only partially represents the gene pool of the eastern wild populations. Captive individuals of this breeding line originated from a single population (A1). As suggested by previous analyses of both mitochondrial and nuclear data, the eastern populations are grouped into two clusters, population A3 being the most differentiated and genetically diverse, and populations A1 and A2 clustering together in the microsatellite analysis (Valbuena-Ureña et al., 2017). Therefore, we recommended to introduce further individuals captured mainly from population A3.

In the western breeding line, in order to reduce the relatedness detected (Fig. 3), and to avoid the possibility of inbreeding, an increase of breeding individuals from this sector is also recommended. According to microsatellite loci data, the western sector clusters into two groups, the main one formed by populations B1, B2 and B4, and the other by population B3, which is separated into a distinct cluster (Valbuena-Ureña et al., 2017). Whereas the overall H_E of the western wild populations is 0.352, the heterozygosity increases up to 0.423 when population B3 is excluded. The low genetic diversity (H_E = 0.197) detected in this latter population biases the mean H_E value of the western wild populations. Currently, the western captive line comprises no founders from B3, but as a result of its low genetic diversity, it is unadvisable to include breeders from this site. Therefore, if additional founders are incorporated into this breeding line, individuals from B1, B2 and B4 should be prioritized.

A need for periodic re-evaluation

Similar values of rarefied allelic richness are found among founders and descendants, which indicates that the descendants inherited almost all the microsatellite alleles from the founders. Eleven and two new PAs were found in the descendant group of the eastern and western sectors respectively; these were not detected in the founders. This could be explained by the fact that some founder females were already gravid when they were caught in the wild, so the genotype of the males was unknown. Also, some females may have fertilized their eggs with sperm from wild males stored in their reproductive apparatus, using a strategy suspected within C. arnoldi (Alonso, 2013) and present in many urodel species (Kühnel, Reinhard & Kupfer, 2010). Therefore, it is highly recommended that the F2, F3, etc. descendants are genotyped in order to ensure that genetic diversity is maintained over time.

Although no signs of a recent bottleneck are detected by BOTTLENECK (more effective at detecting very recent bottlenecks of low magnitude), M-ratio results (best suited for detecting more severe and older bottlenecks) suggested that a past bottleneck occurred in both breeding lines. Thus, it is crucial to keep monitoring for population bottlenecks in managed species since a reduction in N_e may enhance the rate of inbreeding, loss of genetic variation, and fixation of deleterious alleles considerably; thereby increasing the risk of population extinction (Luikart et al., 1998). As the captive breeding program of the Montseny brook newt was established very recently (less than 10 years ago) it is essential that signs of reduced genetic variation or a founder effect are monitored. Therefore, the captive stock and subsequent cohorts should be scrutinised over the long-term in order to effectively preserve genetic variation.

The current breeding protocol widely used by many captive breeding programs to minimize loss of genetic diversity involves pairing individuals according to the mean kinship value (MK) (Willoughby et al., 2015), which is based on the number and degree of relatedness of relatives that an individual has within the captive stock. The use of molecular data has proved useful to investigate the relatedness among individuals, to allow for the design of optimal mating groups, as well as allowing the identification of the natural population from which the founders originated. This underlines the importance of using molecular markers to evaluate genetic management of captive breeding programs. Therefore, further analyses on the resolution of the pedigree are needed to avoid possible mating of closely related individuals, allow deterministic parental assignment of offspring, and design optimal mating groups for maximizing diversity and avoiding genetic erosion in captivity.

Finally, a further problem associated with small founder populations is their significant rapid adaptation to captivity (Araki, Cooper & Blouin, 2007; Frankham, 2008; Gilligan & Frankham, 2003; Griffiths & Pavajeau, 2008; Witzenberger & Hochkirch, 2011). A possible solution would be to occasionally renew the captive populations with previously genotyped individuals from the wild, thereby reducing adaptation to captivity to a minimum (Williams & Hoffman, 2009; Montgomery et al., 2010).