Carps

Software was developed to deal with SLAF-seq data. Procedures are shown in Figure S1. All SLAF pair-end reads with clear index information were clustered based on sequence similarity. To reduce computing requirements, identical reads were merged together, and sequence similarity was detected using one-to-one alignment by BLAT [23] (link) (-tileSize = 10 -stepSize = 5). Sequences with over 90% identity were grouped in one SLAF locus.
Alleles were defined in each SLAF using the MAF evaluation. To prevent false positive results, the sequence error rate was estimated using the rice data as a control. These were obtained using the same sequencing scheme as that used with common carp (Figure S1B). True genotypes had markedly higher MAF values than genotypes containing sequence errors. Tags with sequence errors were corrected to the most similar genotype to improve data efficiency. In mapping populations of diploid species, one locus can contain at most 4 genotypes, so the groups containing more than 4 tags were filtered out as repetitive SLAFs. SLAFs with sequence depth less than 213 were defined as low-depth SLAFs and were filtered out of the following analysis. Only groups with suitable depth and fewer than 4 seed tags were identified as high-quality SLAFs, and SLAFs with 2–4 tags were identified as polymorphic SLAFs.
To evaluate the accuracy of our genotyping objectively, a Bayesian approach was proposed. Using the coverage of each allele and the number of single-nucleotide polymorphism, we calculated a posteriori conditional probability that a given individual would have a specific genotype at a corresponding locus. We proceeded as follows. Supposing there were alleles at any given locus, denoted as . For a diploid species, the number of all possible genotypes was equal to and is less than five regardless of the type of segregation of the loci. We assign a priori probability to each genotype according to the theoretical frequencies with which these genotypes would occur in such a finite probability space. For a homozygous genotype, this priori probability would equal , but it would be double that for a heterozygous genotype. Consider a pair of distinguished alleles and , the probability of sequencing one allele to another can be calculated using the following formula: Here is the average ratio of sequencing error. In our model it took on a value of 0.015 for the Illumina sequencing platform, and we used to represent the length of reads and for number of single-nucleotide polymorphisms. Based on this, we obtained the probability of allele conditioned on the genotype. , denoted as . The depth observation of allele was assumed to be , and the conditional probability of observation of each genotype can be illustrated as follows: In this way, we determined the probability of assigned genotype conditioned on the following coverage observation: The probability was translated to a genotyping quality score finally using: The final genotyping quality score value indicated the confidence with which the genotype had been called. In particular, when the difference in depth between both alleles exceeded 1∶5, the score value could be modified directly using formula (1) due to systematic bias. The upper bound of the score is 30.
This genotyping quality score was used to select qualified markers and individuals for subsequent analysis. This was a dynamic optimization process. Briefly, we counted low-quality markers for each SLAF marker and for each individual and deleted the worst markers or individuals. We repeated this process, deleting one individual or marker each time. We ceased when the average genotyping quality score of all SLAF markers reached the cutoff value, which was 13.

We recorded grass carp locations in rkm upstream from the dam to characterize large-scale upstream movements. We associated locations in wider parts of the reservoir or adjacent coves to the closest point to the river channel using the rkm at that point. This method assumes the shortest distance a fish could have traveled, although the distance could be longer if a fish traveled in a nonlinear path. Distance between locations was calculated as the shortest path a fish could take through the reservoir and tributaries upstream.
We cannot confirm grass carp that made large-scale upstream movements spawned or were intent on spawning, but seasonal upstream migrations to areas suitable for spawning has been observed in grass carp [39 (link)] and other invasive carp species [40 (link)]. As such, we assumed that upstream movements during the spring/summer seasons were associated with spawning activity in our analyses. Time between detections was irregular, making it difficult to statistically compare movement behavior among individual fish. Therefore, we scaled individual fish movement to regular 12-h time steps and estimated the distance moved (rkm) using the ‘redisltraj’ function in the package ‘adehabitatLT’ [41 (link)] in R [42 ]. We used 12-h steps because consecutive detections at different locations were rarely < 12-h and intervals greater than this may miss quick movements associated with the start of a potential migratory run. Following Acre et al. [43 (link)], we identified individual 12-h steps as part of a migration when distance moved was ≥ 85^th percentile of all observed distances moved over a 12-h period. Each movement step was then binomially categorized as part of a migration (1) or not (0).
We quantified river conditions based on data from the Osage River gage at rkm 116.7 (USGS gage 06918250). Temperature (°C), discharge (m³/s), and stage (m) are measured every 15 minutes at this location. We are aware that conditions were unlikely to be exactly homogeneous throughout the study area; however, as it was not feasible to measure the entire river, we make the necessary simplifying assumption that measurements at this location were representative of general conditions of discharge and temperature. River conditions along movement paths were based on gauge readings at the time of each 12-h interpolated location. The importance of these variables for successful spawning of grass carp is well documented, including ideal temperatures for optimal ripening and development of eggs post fertilization as well as higher velocities and turbulence associated with increased discharge and river stage to keep eggs suspended prior to hatching [36 (link), 44 (link)]. In addition, these variables have been associated with grass carp spawning movements in previous studies [45 , 46 ]. Although other environmental variables may play a role in grass carp movements (i.e., turbidity, food availability), it was not feasible to quantify them on a large spatial scale.

Deduced amino acid sequences of eomes genes were downloaded from the Ensembl genome browser (release 95, January 2019) for representative vertebrates, including human (Homo sapiens), mouse, chicken, and zebrafish. Sequence data of three other fishes were downloaded from species-specific genome databases, including grass carp (Ctenopharyngodon idella) (http://www.ncgr.ac.cn/grasscarp/), goldfish (https://research.nhgri.nih.gov/goldfish/), and common carp (http://www.fishbrowser.org/database/Commoncarp_genome/) (S2 Table).
Protein sequences of eomes gene from aforementioned species were aligned by ClustalW2 with default parameters [35 (link)]. A phylogenetic tree was built, using a JTT substitution model with maximum likelihood (ML) method with 1,000 bootstrap replications in MEGA6 software package [36 (link)].

Two copies of eomesa genes were identified in the genome of common carp. The eomesa1 and eomesa2 cDNA sequences of goldfish (Carassius auratus), Japanese silver crucian carp (Carassius auratus langsdorfii) and common carp downloaded from GenBank were used to identify and confirm the eomesa gene sequences and structures in the common carp genome (http://www.fishbrowser.org/database/Commoncarp_genome) using BLAST program (S1 Table).
The genomic sequences of common carp eomesa1 and eomesa2 were compared. Regions with low similarity and uniqueness to each gene were selected to design specific primer pairs to amplify eomesa1 and eomesa2 separately (Table 1, S1 Fig). The amplified fragments were then sequenced with the Sanger method. The sequencing results showed that these primers could be used to specifically amplify eomesa1 and eomesa2 of color common carp.