Hatchery-reared enhancement program for silver carp (Hypophthalmichthys molitrix) in the middle Yangtze River: monitoring the effectiveness based on parentage analysis

Sampling

Juveniles of silver carp were captured from Yanwo and Sanzhou sections in the middle Yangtze River and then reared for 3–5 years in the National Original Breeding Farm (NOBF) located in Jianli (29°35′N, 113°1′E) and Shishou (29°50′N, 113°12′E) Counties, Hubei Province. The matured individuals were selected as hatchery-reared fish and released in Sanzhou section (29°33′N, 112°57′E) of the middle Yangtze River (Fig. 1). The release activity was chosen in late April (closed fishing season) which is one month before the spawning season and aims to reduce the risk of being captured and allows enough time to adapt to natural conditions so that the hatchery-reared fish could spawn in the river. A total of 702 and 827 hatchery-reared fish were released in 2016 and 2017, respectively. The parameters of sex, body length (BL) and body weight (BW) (Table 1) were recorded before release. Meanwhile, the little fin of each fish was clipped and stored in a solution of absolute alcohol for further DNA tests.

The annual spawning season for silver carp in the Yangtze River is from May to June (Duan et al., 2009). In the spawning season, most of the matured silver carp migrated to spawning ground in Yidu section (Fig. 1) for reproduction and the fertilized eggs hatch with the flowing water. A station was set up at the releasing site to capture the larvae with a ring net and a trap net as described earlier (Duan et al., 2009). A total of 376 and 493 larvae were collected in 2016 and 2017, respectively, and the collected larvae were preserved in a solution of 95% ethanol for the extraction of DNA and subsequent analysis.

DNA extraction and microsatellite genotyping

Total genomic DNA was extracted from the alcohol-preserved fins and larvae using a salt-extraction method (Martínez et al., 1998). All specimens were investigated with thirteen microsatellite loci: BL145, BL55, BL109, BL18, BL66, Hym133, Hym230, Hym267, Hym159, BL65, BL106, Hym364, and Hym284 (Table 2), following multiplex PCR protocols as described previously (Li, Cheng & Wang, 2012). Amplification products were genotyped using an ABI 3500XL Genetic Analyzer (Applied Biosystems, Foster City, CA, USA), and the alleles were obtained with GENEMAPPER v. 4.1 (Applied Biosystems).

Parentage assignment

Cervus version 3.0 (Marshall et al., 1998) was used to calculate the number of observed alleles (k), observed heterozygosity (Ho), expected heterozygosity (He), polymorphic information content (PIC) and the null allele frequency. Genotyping error rate was set at 1% (as default), and the typing errors were set as 5% in the parentage assignment procedures, which help to reduce the impacts of mutations and null alleles on the analysis of parent-larvae relationships (Marshall et al., 1998). It was more efficient using the single-parent mode than the pair-parent mode in Cervus v 3.0 (Li et al., 2013). According to the recorded gender information (Supplemental Information 1), the parentage analysis was conducted at first by the mode of maternity, followed by the mode of paternity.

The number of potential parents could influence the precision of the assignment to one correct parent. Also, in this study, the number of hatchery-reared fish in 2016 and 2017 were all less than 900. The parentage assignment simulation was carried out using the software in both the modes of paternity and maternity, and the number of candidate parents was set to 900. Ten thousand cycles of simulated assignments were carried out using 95% confidence intervals. After analysis, the assignment rate was still kept at 100% with 900 candidate parents.

In this study, the river-reared potential parents were not sampled and thus if a larva was not assigned to at least one parent, then it could be assumed that non-hatchery-reared fish produced it. Paternity exclusion is a commonly used method for parentage assignment, which is based on Mendelian segregation of alleles, and it allows moderate mismatching and genotyping errors between a progeny and its parental alleles (Vandeputte & Haffray, 2014). For our purpose, larvae with one pair of loci mismatching (LOD > 1) were also considered to be the offspring of hatchery-reared fish.

Population genetic analysis

To explore the fine genetic diversity and genetic structure, each year the larvae were divided into three populations based on the results of parentage analysis: the R larvae population having larvae with both parents being hatchery-reared fish; the H larvae population with only a male or a female parent being hatchery-reared fish; and the W larvae population for the remaining larvae. The hatchery-reared fish population was also divided into two: confirmed breeders population having hatchery-reared fish with the detection of larvae, and unconfirmed breeders population with the remaining larvae. There were three larvae and one confirmed breeder population in 2016 and 2017 respectively, and overall eight populations were studied.

Box plots were drawn by Origin 6.0 (And, 2010) software to display the distribution of the length and weight of the bodies of two hatchery-reared fish populations. T-test (George & Mallery, 2003) was used to evaluate the significant differences between two hatchery-reared fish populations for body length and weight, with the same sex.

Genetic diversity was characterized by the number of alleles (A), expected heterozygosity (He) and observed heterozygosity (Ho), which were calculated by using PopGene32 (version 1.32) (Yeh et al., 1997) software. The allelic richness (Ar) was calculated to correct the variations in the sample sizes by FSTAT ver 2.9 (Goudet, 2002). Regression analysis (George & Mallery, 2003) was used to determine significant differences. Departures from Hardy-Weinberg equilibrium (HWE) at each locus and pairwise fixation index (F_ST) were performed using the Arlequin v 3.0 package (Excoffier, Laval & Schneider, 2005). The analysis of molecular variance (AMOVA) was also performed using Arlequin v 3.0 package. Bonferroni adjustments to the P-values of F_ST were applied whenever multiple tests were performed. Among populations, the possibility of their structure was examined by clustering techniques based on Bayesian method using STRUCTURE v2.2 (Pritchard, Stephens & Donnelly, 2000). For each value of K (K = 2–7), ten runs were implemented with a burn of 20,000 in length, followed by 100,000 Markov chain Monte Carlo iterations. ΔK was calculated with the obtained standard deviation, which was used to identify the appropriate number of clusters (Evanno, Regnaut & Goudet, 2005). The optimal K values were selected by STRUCTURE HARVESTER (http://taylor0.biology.ucla.edu/structureHarvester/). Besides, Principal Component Analysis (PCA) was performed by PCAGEN v 1.2.1 (http://www2.unil.ch/popgen/softwares/pcagen.htm), to examine the patterns of genetic differentiation in the populations.

Parentage assignment

A total of 1,529 hatchery-reared fish and 869 larvae were genotyped at thirteen microsatellite loci. The number of allele (k) per locus ranged from 12 in BL145 to 31 in Hym284 (Table 3). The average observed heterozygosity (Ho) and expected heterozygosity (He) were 0.849 and 0.852, respectively. All markers used in the study were highly informative with a high average PIC value (0.835). The absolute values of null allele frequencies were between 0.3% and 4.8%. The combined exclusion power (CPE) achieved 99% correctness and 100% assignment with as few as six loci for no available information from any parents (Fig. 2).

The results of parentage analysis showed that 39 larvae in 2016 could be unambiguously assigned to hatchery-reared fish, which included six R larvae and 33 H larvae. Fifty seven larvae in 2017 could be unambiguously assigned to hatchery-reared fish, which included one R larva with two breeders in 2016, one R larva with two breeders in 2017, four R larvae with one breeder each in 2016 and 2017, 19 H larvae with one breeder in 2016, and 32 H larvae with one breeder in 2017 (Table 4). The sum of R and H larvae accounted for 10.37% and 11.56% of the larvae in 2016 and 2017, respectively. In this study, the remaining larvae were produced by non-hatchery-reared fish.

Differential parental contributions of hatchery-reared fish to larvae and their growth performance

In 2016, 63 larvae were produced from 59 confirmed breeders, which consisted of 36 females and 23 males, whereas in 2017, 37 larvae were produced from 37 confirmed breeders, which consisted of 23 females and 14 males, as shown in Table 4. In this study, among all the studied hatchery-reared fish, 1,433 unconfirmed breeders did not contribute to larvae (93.72% which included 763 females and 670 males), and 87 confirmed breeders contributed to one larva (5.68% which included 54 females and 33 males), and 7 confirmed breeders contributed to two larvae (0.45% which included four females and three males), and two confirmed breeders contributed to over two larvae (0.13% which included 1 female and 1 male) (Fig. 3).

In case of all confirmed breeders, the observed variations in the sex, body length and body weight are shown in Table 1. Box plots were used to display the distribution of body length and body weight of all the confirmed and unconfirmed breeders’ populations (Fig. 4). T-test results showed statistically no significant difference (p > 0.05) between two populations with the same sex.

Genetic variation and differentiation among confirmed breeders and larvae populations

The obtained results based on eight populations were analyzed at 13 microsatellite loci (Table 5). W larvae populations had the highest number of alleles (A) per locus, while R larvae populations had the lowest number of alleles per locus. The discrepancy between them was statistically significant (p < 0.05, Supplemental Information 2). The values of average allelic richness (Ar) supported the difference in the sample size among populations which varied between 5.8 in two R populations to 6.9 in W2017. The mean expected heterozygosity (He) values were similar in all eight populations, ranging from 0.782 in R2017 to 0.863 in W2017. No significant difference was found among eight populations in terms of Ar and Ho (p > 0.05). All the average values of observed heterozygosity (Ho) revealed high levels of genetic variation, and the R larvae population had the highest number, followed by H larvae population and then by the W larvae population per year.

Low F_ST values were observed among the samples of silver carp, suggesting that there was high homogeneity among all populations (Table 6). W2016 had the highest F_ST value which was significantly different from R2017 (F_ST = 0.03, P = 0.00). The differences were also significant between W larvae populations and most of the other populations. In particular, W2017 had low but significantly different F_ST values from other populations except for R2016. The results of AMOVA analysis showed that 97.98% of the genetic variations occurred within populations while only 2.02% of the genetic variation occurred across eight populations, and a significant but low genetic differentiation was found (F_ST = 0.020 and P < 0.001, Table 7).

The STRUCTURE analysis showed the number of genetic clusters, and the obtained results revealed that the highest ΔK value was obtained for K = 3 (Fig. 5), with four populations in 2016 forming one cluster and fish from populations in 2017 assigned among other two clusters. A visual inspection of the STRUCTURE results showed that W2017 and other populations in 2017 did not cluster together and this is confirmed from the proportion of eight populations in each of the three inferred clusters (Table 8).

The Principal Component Analysis (PCA) results are shown in Fig. 6. The first component (68.76% of the overall variation) and the second component (17.59% of the overall variation) analysis separated four populations in 2016, W2017 and the rest of the sample populations (Fig. 6A). Meanwhile, the first and third components (4.8% of the overall variation) analysis demonstrated a similar result (Fig. 6B). The results obtained from the second and third components’ analyses (Fig. 6C) were not considered as the proportion of their overall variation was low (<25%). Hence, the analysis based on three main components disclosed three different groups. It could be seen that the results from PCA were in good agreement with that of STRUCTURE. Moreover, these groupings were supported by the results of AMOVA analysis as shown in Table 7 (F_CT = 0.021 and P = 0.002, F_SC = 0.001 and P = 0.015).