Parental (P) fish for genetic crosses were collected from New Bedford Harbor (NBH), Massachusetts, USA, and Block Island (BI), Rhode Island, USA, and were maintained as previously described [4] . Following a classic F2-intercross design, BI males were crossed with NBH females to generate the first filial generation (F1). Once sexually mature, F1 siblings were crossed (F1 × F1) to generate the recombinant F2 generation. F2 larvae from each of three independent pedigrees were collected and used to construct an integrated consensus map of the killifish genome. All research including fish collection, culture, and breeding have been approved and adhere to the guidelines set forth by USEPAs Institutional Animal Care and Use Committee (IACUC).

Genomic libraries enriched for microsatellites were developed following the methods of Hamilton et al. [12] and Glenn et al. [13] . Briefly, genomic DNA extracted from eight larval fish (DNeasy, Qiagen) was pooled and digested with Hae III and Rsa I (NEB) and ligated to double-stranded linkers (SNX-f-5’-CTAAGGCCTT- GCTAGCAGAAGC-3’and SNX-r-5’-pGCTTCTGCTAGCAAGGCCTTAGAAAA-3’). Microsatellite probes (5’-biotinylated) were used to capture fragments containing microsatellite repeats on streptavidin-coated beads. Microsatellite-enriched single stranded DNA was amplified using the SNX-f primer to recover double-stranded DNA. PCR amplification products of clones (TOPO 2.1, Invitrogen) were sequenced (ABI 3730 Genetic Analyser) and screened for microsatellites. Primers for candidate microsatellites were designed using Primer 3 [14] .

Parental fish (n = 6) and 95 F2 embryos from each of three genetic crosses (total n = 291) were genotyped with the microsatellites developed herein, as well as with previously published markers including microsatellites [7] - [9] , and SNPs [10] [11] . Polymerase chain reactions and genotyping were performed according to Waits and Nebert, 2011 [15] , for microsatellite markers, and according to Williams and Oleksiak, 2011 [10] , for SNPs.

Linkage groups were identified based on estimated two-point logarithm of odds (LOD) scores and Haldane distances. For each genetic cross, a minimum LOD score of 3.0 and a maximum recombination rate (rf) of 0.5 were used as threshold values to determine linkage groups [16] . Within each linkage group, marker order was determined by examining all pairs of markers, starting with the pair having the highest maximum likelihood and incrementally adding markers (most strongly linked first) always choosing the highest log likelihood and the best insertion point . Linkage maps were then improved and validated by perturbation of the initial maps in a systematic manner [16] . First, the “flips” algorithm was used to evaluate all possible permutations of marker order in a sliding window of fixed size (window = 5) on the initial map. Initial maps were replaced with improved maps as identified [16] . These maps were then interrogated by “polishing,” which displaces each marker in all possible intervals in search of a map having a better log-likelihood [16] . For the consensus map, markers in common among the three families were used to anchor an integrated map. Linkage groups were identified and an initial framework map was built using stringent criteria (LOD > 10, rf < 0.2), which were then incrementally relaxed (to LOD ≥ 3.0, rf ≤ 0.5). Marker order within each consensus linkage group was determined as described above.

To identify microsatellite and SNP marker identities, an annotated draft of the F. heteroclitus genome (supported by National Science Foundation grant, DEB-1120512) was queried. Specifically, marker sequences were assigned to annotated killifish gene models by performing nucleotide BLAST searches against the killifish genome sequence. Markers were assigned a gene name when the marker sequence fell within an annotated gene model and the match between the marker sequence and the gene model sequence spanned at least 150 bp and exceeded 96% identity.

Syntenic relationships were explored for the consensus linkage map against two other teleost species for which an annotated genome assembly is available: medaka (Oryzias latipes) [17] and zebrafish (Danio rerio) [18] . Synteny with Tilapia (Oreochromis niloticus) was also explored, however, the structure of tilapia genome is less well defined precluding the comparison. Orthologs were assigned by reciprocal BLAST and predicted proteins from the annotated killifish gene models were compared to the zebrafish and medaka non-redundant protein databases (NCBI). Orthology was confirmed if the BLAST search of zebrafish/medaka genes against the killifish genome returned the original killifish gene model as the primary hit and exhibited an e-value < 1e−5. To validate and augment the list of orthologs, an additional query was performed by blasting the individual marker sequences against sequence data for D. rerio and O. latipes (Ensembl).

Approximately 1200 clones were sequenced from the microsatellite-enriched libraries and surveyed for tandem repeats with sufficient flanking sequence for oligonucleotide design. One-third of the sequences contained di-, tri-, or tetra-nucleotide repeats that met these criteria. Of these, 330 candidate loci were selected for primer design and evaluated for amplification success and polymorphism in the parental generation used to establish genetic crosses. In the parental generation of the three genetic crosses, 59, 65, and 94 loci, respectively, proved polymorphic and Informative. A X² (chi-square) goodness-of-fit test for deviation from expected 1:2:1 Mendelian ratios identified a small proportion of markers showing segregation distortion (Supplementary Data, Table 1(a)) in the 285 F2 larvae genotyped (following Bonferonni correction, α = 0.00015). Moderate departures from these frequencies are not unusual and may indicate the presence of partially lethal alleles. Gross departures from these frequencies may indicate problematic markers or null alleles [19] .

Linkage maps were constructed for each of the three mapping families. A total of 75, 88, and 105 loci clustered into 16, 21, and 23 linkage groups (Lg), respectively (Supplementary Data, Table 1(b)). Genetic loci incommon among the three families were used to anchor and join the individual maps into one integrated consensus map (Figure 1). At a threshold of LOD ≥ 3.0 and a recombination frequency ≤ 0.5, twenty-four linkage groups were identified. Together, the 24 linkage groups contain 240 loci covering 2300 cM total. Individual linkage groups range from 35 to 195 cM with an average marker resolution of 9.5 cM.

Overall, the integrated consensus map is in good agreement with the maps based on individual genetic crosses. Marker order is conserved between the consensus map and individual maps with few exceptions. Exceptions include the relative locations of Fhe_2179 and Fhe_495 in cross A, which are reversed relative to cross B, C, and the consensus map. This discrepancy is likely due to heterogeneity in recombination rates between source popula- tions and the extreme segregation distortion of Fhe_2179 (p = 2.27E−10). Other loci with inconsistent positioning due to segregation distortion in at least one of the three genetic crosses include: Fhe_2254 (p = 5.88E−19), Fhe_2277 (p = 7.31E−06), and Fhe_2027 (p = 2.50E−06). Inter-marker distances in the consensus map are

also consistent (±5 cM) with individual based maps barring three exceptions: xTC15606_316 and xTC19271_848 (Lg-9 consensus, Lg-7 genetic cross B) are separated by 35.6 cM on the consensus map and 15.5 cM on the individual map; Fhe_2151 and Fhe_2133 (Lg-19 consensus, Lg-13 genetic cross C) are separated by 84 cM on the individual map and 49.6 cM on the consensus map; and xTC21107_340 and Fhe_2069 (Lg-5 consensus, Lg-9 genetic cross C) are separated by 20.5 cM on the consensus map and 13.8 cM on the individual map.

Of the 240 loci comprising the consensus linkage map, 181 loci fell within a gene model in the draft Fundulus genome (Supplementary Data, Table 1(a)) and were assigned identities. Four additional loci were named based on significant hits to the zebrafish and medaka genome assemblies. In order to perform a comparative genome analysis, zebrafish and medaka orthologs and their genomic coordinates were identified (NCBI; non- redundant protein database) based on the predicted protein sequences of the Fundulus gene models. The identified homologous genes were used to determine homologous chromosomes in medaka and zebrafish (Figure 2). Comparing killifish and medaka, 10 killifish linkage groups (2, 20, 10, 9, 6, 11, 21, 3, 16, 1) exhibit 1:1 relationships with medaka chromosomes and for 7 other linkage groups (5, 19, 7, 4, 17, 15, 8, 22), >75% of killifish loci fall on a single medaka chromosome with only a single locus on a second medaka chromosome. For LG4, 7 loci fall on medaka CHR10, 2 on CHR18, and one on CHR21. Only one medaka chromosome (CHR2) does not share homologous loci with the consensus Fundulus linkage map. A comparative coordinate analysis of loci on the Fundulus linkage map and medaka genome assembly reveal conservation of gene order for approximately 70% of shared loci as well as multiple rearrangements (Figure 3).

Homologous chromosomes in zebrafish are less apparent. Three killifish linkage groups (12, 8, 22) exhibit 1:1 relationships with zebrafish chromosomes and for 7 other linkage groups (2, 9, 15, 4, 7, 16, 10), >65% of killifish loci fall on a single medaka chromosome with multiple loci located on up to three additional zebrafish chromosomes (Figure 2). Overall, at least 22 of 25 zebrafish chromosomes contain loci from between two and five Fundulus linkage groups. A comparative coordinate analysis of loci of the Fundulus linkage map and zebrafish genome assembly reveals a higher proportion of rearrangements in gene order (Figure 3) than observed when comparing to Medaka.