Microgeographic Variation and Inter-Riffle Migration of Rhinichthys atratulus (Pisces: Cyprinidae) in a Small Connecticut Stream, United States

We examined the genetic diversity on a microgeographic scale of Rhinichthys atratulus (Eastern Blacknose Dace) in Allyn Brook, a small tributary in the upper Coginchaug River drainage in Connecticut. By looking at gene flow on a microgeographic scale among populations that had no physical barriers to migration, we tested the null hypothesis that the populations should be homogeneous. We resolved seven polymorphic microsatellite loci and one mitochondrial gene, nd2, in three adjacent populations (<0.5 km) in Allyn Brook and compared these populations to the two closest populations (>5 km) in the Coginchaug River. A dam from the 1920’s in lower Allyn Brook has isolated Allyn-Brook populations from Coginchaug-River populations. Allyn Brook was selected because there are only three riffle habitats in the brook and, therefore, there can be no immigration from upstream populations. Each population has private (i.e., unique) alleles and haplotypes, and there are significant genetic differences between all sites. The Allyn Brook populations are almost as different from one


Introduction
One of the prominent features of the Anthropocene has been the disruption of continuous habitat for the world's biota. Understanding how fragmented landscape features affect the maintenance and production of genetic diversity across space and time is critical for conservation biology Anderson, Epperson, Fortin, Holderegger, James, Rosenberg, Scribner and Spear [1] [2]. Riverine systems are ideal to measure how landscape features, such as dams, affect gene flow because of the linear nature of streams. The degree of connectivity among habitats within a watershed is a critical driver of the genetic population structure of freshwater fishes, because barriers to movement limit gene flow and may lead to loss of genetic diversity or even the production of novel genetic signatures [3] [4] [5]. The degree of divergence among populations should increase relative to the number and relative strengths of barriers, and/or the geographic distance that isolates populations [3] [6] [7] [8]. Isolation by distance is expected in riverine fishes but is not always found [9]. Furthermore, freshwater fishes generally have low effective population sizes, which can contribute to high genetic differences among populations in different river drainages [10].
Population genetic structure and migration behavior of riverine fishes have also been studied for ecosystems in which no obvious physical barriers were present. Genetic variation among Poecilia reticulata (Guppy) in Trinidad correlated positively with geographical distance; populations separated by waterfalls were more genetically divergent than those that were not [7]. In a study of the inter-riffle (i.e., rocky habitat where the rocks break the water's surface) movement of five common North American fishes from three families, long corridors between pool habitats, water depth and current were found to limit the amount and rate of fish movement [11].
In our studies of the population structure and ecology of the small riffle-dwelling fish, Rhinichthys atratulus (Eastern Blacknose Dace), we have found large degrees of eco-genetic variation among populations in rivers in Connecticut and in the eastern United States [12] [13]. At a more local scale, populations of R. atratulus differed significantly in a western Connecticut river and its tributaries [14].
The present study examines genetic differences among and between populations of R. atratulus on a microgeographic scale ( Figure 1). The question is, at a microgeographic scale, will the ecology of R. atratulus-that is, their easy ability to swim to adjacent habitats-prevent the establishment of novel genetic diversity?
At this geographic scale, we expect populations to be genetically homogeneous because there are no obvious physical barriers to gene flow and movement of fishes among the habitats should prevent the establishment of novel genetic diversity. This is critical because we must be able to distinguish the effects of a naturally patchy environment on a local level from anthropogenic fragmentation of ecosystems.
In this paper we: 1) compare the genetic diversity among R. atratulus from adjacent habitats that are isolated in a single small stream (see below); 2) we estimate the migration and mutation rate for each of the populations; 3) we estimate whether the number of individuals in the populations are stable; and 4) we compare the populations from the study sites ( Figure 1) to the closest two populations in the river for which the study-site stream is a tributary. The latter is important in order to provide a baseline for an older isolation due to an anthropogenic factor (see below).

Field Sites
Allyn Brook, a tributary of the Coginchaug River in central Connecticut, USA ( Figure 1), is a small stream with a wetted width less than 4 m wide (Appendix 1) in the area of sampling. We sampled at three riffles, each less than 3 m long composed primarily of gravel and sand (Appendix 1) in June 2017. The coordinates of the three riffles -AB, AB1 and AB2 -are given in Appendix 1. The three riffles span a distance of 0.47 km ( Figure 1) and there are no physical barriers in between the riffles to prevent or hinder the movements of fishes. A dam in the lower part of Allyn Brook from the 1920's [15] has prevented emigration of R. atratulus from the Coginchaug River; the backwater of the dam has removed all riffles below our study sites. Above our study sites Allyn Brook changes character and lacks riffle habitats. Thus, the three riffle habitats are the only place in Allyn Brook that the species is found [12].
Physical and chemical characterizations of the three habitats are given in Appendix 1. Conductivity, pH, dissolved oxygen, and water temperature were measured using a Yellow Springs Instrument (YSI) probe Model 556 MPS. At three points along each riffle canopy cover was estimated using a GRS Densitometer. Stream flow velocity was estimated with a Price-Gurley meter at 60% of the depth. Flow and depth were recorded approximately every meter along the riffle. The riffle slope was measured using a tripod and surveyors rod with a vertical tolerance of ±5 cm. There were no significant differences among the three sites in physical or chemical characteristics.

Study Species
Rhinichthys atratulus ( Figure 2) is a small (<12.5 cm) freshwater riffle-dwelling minnow that lives east of the Appalachian Mountains from Virginia to Nova Scotia [12] [16]. R. atratulus tend to live in large populations [16]. They prefer rocky riffles and adjacent pools and of small to medium streams, and can be found in most small streams across Connecticut [17]. They reach maximum length in two to three years, and are an important link in the trophic chain [16] [18]. Rhinichthys atratulus eat primarily invertebrates and plankton, and are important prey to larger fish, especially large brook trout, and fish-eating birds [16] [18]. Rhinichthys atratulus are not broadcast spawners, they bury their eggs in the river substrate adjacent to riffles during spawning [19]. Because of their ubiquitous distribution and trophic importance, they are a useful model organism.

Molecular Methods
DNA was extracted from fin clips using the DNeasy blood and tissue kit (QIAGEN Sciences, MD, USA) using the protocol from [12]. The methods and primers used for amplification of the mitochondrial gene nd2 are from Tipton, Gignoux-Wolfsohn, Stonebraker and Chernoff [20]. The fully sequenced nd2 is 1042 base pairs long. After PCR, 7 μl of PCR product was mixed with 1 μl of Gel Loading Dye and run for 30 minutes on a 2% agarose gel with 5 μl of SYBRsafe (Invitrogen, Thermo Fischer Scientific). Samples were sequenced at Yale University and purified using their Exo-AP purification protocol. Sequences were aligned using ClustalW multiple alignment in BioEdit v.7.1.7 [21] and manually curated based on chromatograms viewed in FinchTV 1.4 [22]. Curated sequences are available on GenBank under accession numbers MH341914-MH341923, with haplotype 14 added later under MH378685. While Tipton, Gignoux-Wolfsohn, Stonebraker and Chernoff [20] labelled haplotypes by letter, we designate haplotypes by number. Haplotypes 1 -38 correspond to haplotypes all of Tipton, Gignoux-Wolfsohn, Stonebraker and Chernoff [20]; haplotypes not discovered in the previous study [20] begin sequentially with 39. Private haplotypes and microsatellite alleles are defined as those found at only one study site and nowhere else in the distribution of the species [12] [20].
The primers and PCR protocol for amplification of 14 microsatellite loci from non-coding regions are from Kraczkowski and Chernoff [12]. Screening of the microsatellite loci identified seven that were suitably polymorphic for inclusion in this study. Optimization temperatures for loci are as follows: Rhca15b -60˚C; Rhca16 -59˚C; Rhca20 -60˚C; BD165 -47˚C; Ca3 -45˚C touchdown procedure; Ca12 -50˚C; and BD174 -60˚C. PCR products were run on a 2% agarose gel to test for amplification. Samples with definitive bands were sent to the DNA analysis facility of Yale University for fragment analysis. Samples were viewed in Peak Scanner v2.0 [23] to determine the length of microsatellites and possible heterozygotes. Microsatellites were added to GenBank under sequences MH341907-MH34913.

Statistical Analyses
DnaSP v5 10.01 was used to calculate nucleotide and haplotype diversity as well as neutrality tests Tajima's D, Fu and Li's D and Fu's F [24]. Haplotype diversity (H) is the probability of that two randomly-selected haplotypes within a sample are different [24]. Equation (1) is where 2 i x is the frequency of each haplotype within a sample and N is the sam-ple size. Nucleotide diversity (π) is the average number of nucleotide differences per nucleotide site between two DNA sequences in all possible pairs in the sample population [24]. Equation (2) is where x i and x j are the respective frequencies in the i th and j th sequences, and τ ij is the number of nucleotide differences per nucleotide site in the i th and j th sequences and summed over all sequences in the sample. Arlequin 3.5 (with input files generated in DnaSP) was used to generate a haplotype network and to test null hypotheses of genetic homogeneity with analyses of molecular variance (AMOVAs).
The programs PGDSpider [25] and CONVERT [26] were used to generate input files. Arlequin 3.5 [27] and POPGENE [28] generated tests for Hardy-Weinberg equilibrium, Garza-Williamson statistics and AMOVAs. Inbreeding coefficients were calculated as one minus the ratio of the observed over the expected heterozygote frequencies [29].
We used STRUCTUREv2.3.4 to generate a-posteriori Bayesian-likelihood classifications for each individual based upon the genetic signatures of all seven microsatellite loci [30]. STRUCTURE parses individuals into k groups, where k is the number of populations into which individuals are classified. Structure was run three times (one run per k = 1 to 3, the number of Allyn Brook populations) with 10,000 burn-ins and 10,000 generations under the admixture model. STRUCTURE was also run with Allyn Brook sites and two samples from the Coginchaug River (k = 1 -5), one upstream and one downstream of Allyn Brook (CR and CRA in Figure 1). STRUCTURE does not allow the user to assign colors to the genetic groups ( Figure 4); it is automatic.

Estimation of Migration Rates
The program Migrate v3.2.1 was used to estimate: the number of migrants per generation [31]; the mutation rate scaled by population size, θ; and [31] the mutation-scaled immigration rates, M [32]. The number of individuals migrating per generation is estimated by M × θ [32] [33]. M is a measure of the relative importance of immigration over mutation to bring new variants into the population [32] [33]. Mitochondrial results were run with the DNA sequence model, while microsatellite results were run with the Brownian Motion Model, which is a quicker approximation to the Ladder Model [32]. For both mitochondrial and microsatellite data, the program was run 10 times, and the model with the highest maximum likelihood parameter was selected for further analysis.

Genetic Diversity -nd2
Eleven haplotypes were identified from the 72 individuals that were sequenced for the nd2 gene from the three Allyn Brook populations. Haplotype 1, hypothesized as the ancestral haplotype of the populations that recolonized the Con-  [20], was present at each of the sites but was common in only the upstream locality (Table 1, Figure 3). Three additional haplotypes (numbers: 9, 10 and 44) were found at all sites (     For seven or more loci, values less than 0.68 indicate that the population has gone through a recent reduction in size [38]. The means and standard deviations of the modified for each locus and each population (

Population Differences -Global AMOVA's
Global AMOVA's were calculated among Allyn Brook sites as well as between the two Coginchaug River sites and Allyn Brook sites (

Bayesian Classification of Individuals from Microsatellite Data
The maximum likelihood classification model for Allyn Brook populations with the best fit was for two genetic groups (k = 2; Figure 4    The following are the percentage of individuals with a > 50% probability of belonging to the one genetic group (shown as red in Figure 4 upper): AB = 11.76%; AB1 = 50%, and AB2 = 81.48%. We calculated the classification probabilities by adding microsatellite data from populations in the nearby Coginchaug River (CR and CRA), the two closest sampled populations to Allyn Brook populations; the Coginchaug and the Allyn Brook populations have been separated from each other for at least 80 years because of the dam (Figure 1). There were two genetic groups (k = 2) ( Figure 4). The results identified two distinct genetic groups -one with the Co-

Estimation of Migration Rates
The number of migrants per generation was estimated from both the nd2 and microsatellite data (Table 4). Among the sites, migrants per generation for microsatellites ranged from 1.5 to 6.6, while migrants per generation for nd2 ranged from 1.1 to 10.7. The average of the downstream direction for migrants per generation was 5.0 and 2.3 for microsatellites for nd2, respectively. The upstream average was 3.3 and 4.9 for microsatellites and nd2, respectively. Potential upstream or downstream biases in migration were tested using a two-tailed T-Test, and no significant difference in direction of stream flow was found.
Using AB1 as an example, this means that we expect there will be 9.1 mutations per 100 individuals in the next generation over all sampled loci. AB and AB2 had the highest θ's for microsatellites (θ = 0.098); AB1 was highest for nd2 (θ = 0.0.0042). The θ's for microsatellites were an order of magnitude larger than those for nd2, indicating more rapid evolution of the former.
The values of M are relative within data type (Table 4) and estimated the importance of immigration of genes over mutation to introduce new variants into a population. Gene flow due to immigration was more important than mutation for introducing: 1) microsatellite alleles into AB1 from AB2; and 2) nd2 haplotypes into AB1 from AB.

Discussion
We examined the structure among populations of R. atratulus, in order to assess gene flow on a microgeographic scale. Importantly due to a dam placed near the mouth of Allyn Brook in the 1920's [15] the populations have been isolated from We discovered a number of private alleles and haplotypes suggesting that migration and gene flow among populations was insufficient to prevent in situ evolution and the establishment of novel genes in the populations. We expected the middle site, AB1, to display the greatest effects of migration and geneflow. In fact, AB1 has no private haplotypes, which could have been due to small sample size. But AB1 had private alleles though the fewest (Table 1). Site AB has the largest sample size and the most private alleles and haplotypes, which could indicate that the number of private alleles and haplotypes are a function of sample size. Rarefaction analyses did not show that to be the case. Also, AB's estimated population-scaled mutation rate, θ, was an order of magnitude higher than AB1 or AB2, indicating that either in situ evolution or differential extinction was primarily responsible for novel alleles in AB.
Maximum-likelihood analyses of migration provide evidence for limited gene flow through the exchange of less than 11 individuals per generation among all sites, with slightly fewer individuals on average migrating upstream than downstream (Table 4). Although the estimated migration of nd2 among populations is, on average, higher than estimations for microsatellites, we suggest that the estimates (Table 4) are overestimates because the analyses of migration ignore historical/phylogenetic information about genes that are currently in high frequency in populations. For example, because Haplotype 1 is the ancestral haplotype for R. atratulus in the Connecticut River drainage [20]. The presence of Haplotype 1 in Allyn Brook populations is not result of current gene flow but rather that of post-glacial colonization. A similar case inheres for haplotypes 44 and 10 that had the highest frequencies in all populations ( Figure 3, Table 1).

The results of AMOVAs and Bayesian a posteriori classifications of Allyn
Brook individuals show that all three populations differed significantly in their genetic signatures consistent with low levels of estimated migration and gene flow among populations (Table 3, Figure 4). Furthermore, Allyn Brook populations differed significantly from nearby populations in the Coginchaug River (Table 3, Figure 4) to the extent that 100% of all individuals from the two rivers had an almost 100% probability (mode = 99%) of belonging to a genetic group unique to each river. The results indicate that: 1) gene flow among Allyn Brook populations is limited and not sufficient to prevent in situ differentiation; and 2) that the isolation of Allyn Brook populations from Coginchaug River populations over the last 80+ years has resulted in larger genetic distinctions among populations. These results are consistent with findings from several species of Rhinichthys [12] [20] [39] [40] [41] and other related North American cyprinids, e.g., Phoxinus and Gila [42] [43] that revealed cryptic genetic diversity and inter-or intra-basin genetic diversity, suggesting little or no current gene flow. However, these studies were performed on larger spatial scales, on the order of km's. In microgeographic studies (≤1 km), high levels of genetic diversity and There is only weak evidence to suggest a relationship between geographic distance and genetic differences. At the largest geographic scale, the genetic distances between Allyn Brook populations and Coginchaug River populations would be consistent with an isolation by distance model. Within Allyn Brook, there are only three populations and mathematically no relationship could be calculated. However, the transition between the two microsatellite genetic signatures identified in the Bayesian classification analysis (Figure 4 upper) varied clinally with distance. Isolation by distance has been difficult to identify in river fishes [9], depending upon such factors as stream size and body size [44] as well as habitat preference. Rhinichthys atratulus is a relatively small minnow < 12.5 cm total length, living in a small stream, <2.5 m wide with limited proclivity for migrating.
We present six hypotheses for why gene flow is low even on such a small geographic scale: 1) The fish are not moving between populations; 2) fish are moving but not mating and, therefore, not exchanging genetic material; 3) migrating fish are preyed upon in the long sandy channels between riffle habitats; 4) fish migrate but upon some density threshold of encounters with neighboring dace return to the original site (F. Cohan, pers. comm.); 5) only individuals that possess common haplotypes or microsatellite alleles migrate, or individuals with private haplotypes or alleles have minimal rates of migration, such that we cannot detect migration with genetic data; or 6) all fish are migrating, but private haplotypes and alleles are selected against in the new habitat. We dismiss hypothesis 6 as unlikely because statistical tests failed to identify non-random processes (Appendix 2) and the microsatellite loci are for the most part in Hardy-Weinberg equilibrium (Appendix 4). The four significant heterozygote deficits were likely due to inbreeding (inbreeding coefficients ranged from 0.13 to 0.80). Because the riffle habitats of the Allyn Brook sites are virtually identical in size, substrate, chemistry, forest overhang and flow, we dismiss hypothesis 6-there should not be significant differences in any selection regimes among riffle sites. At this time there is no evidence to support or refute hypotheses 2, 4 and 5.
There is some support in the literature for hypothesis 1. The restricted movement paradigm states that resident stream fishes tend to be relatively sedentary [45]. However, there are data to show that some stream species move more often and farther than thought previously when movements are measured directly [46] [47]. In estimating movements, mark and recapture studies can be ambiguous due to low capture rates. While low recapture rates may imply significant movement, high turnover rates or large population sizes in a restricted area would yield similar results [48]. Gerking [45]) proposed a movement range of 20 Open Journal of Ecology in the majority of trout populations studied by Rodríguez [48] fell within this range: a median of ca. 5 m in Oncorhynchus clarki (Cutthroat Trout) and up to 90 m in Salvelinus fontinalis (Brook Trout). In our study, sample sites were 80 -470 m apart. Perhaps that these sites are too far apart for R. atratulus to move through open channels. There were no individuals of R. atratulus found between the three riffles and these are the only riffles in Allyn Brook.
Lonzarich, Lonzarich and Warren Jr [11] hypothesized that long riffles and water velocity regimes can limit the daily inter-habitat movement of stream fishes. Two sympatric congeners from New Zealand, Galaxias gollumoides with strong habitat preference for discontinuous habitats of low-velocity water had larger mean DNA sequence divergence and higher among-population differentiation than an undescribed congener that lives in faster-moving water [49]. The migratory Luciobarbus comizo (Iberian Barbel) adjusted their behavior in response to small stream barriers, and remained within 1000 m of initial capture sites even though the barbels could easily pass the barriers [50]. Rhinichthys atratulus is a riffle-dwelling species, preferring slightly faster moving water over rocky substrates, and seems to regard the runs (long, slow-moving river segments) as barriers to movement. We have never captured dace in between riffles despite multiple samplings.
It is also possible that R. atratulus are preyed upon in the runs between riffle habitats (hypothesis 3). In observations of the fishes in Allyn Brook, there were high numbers of large (>18 cm) Semotilus corporalis (Fallfish), known as a predator of R. atratulus [51], swimming actively in the runs. Furthermore, sites in Allyn Brook were fairly open, and not shaded by trees, exposing fishes to piscivorous birds [52]. In Allyn Brook, three piscivorous bird predators, Megaceryle alcyon (Belted Kingfishers), Ardea herodias (Great Blue Heron) and Butorides virescens (Green Heron) are common (B. Chernoff, pers. observ.). Predation by birds may often be overlooked. Predators such as Great Blue Herons and Belted Kingfishers have the potential to significantly impact community dynamics and trophic structure of aquatic prey species [53].
Larson, Hoffman et al. 2002 inferred in a study over several that R. obtusus (Western Blacknose Dace; the sister group of R. atratulus) (Kraczkowski and Chernoff 2014) had established a robust population in a neighboring creek in Arkansas. They labelled the suspected emigres as "dispersers". Clearly, R. atratulus is capable of migrating, even long distances, for example, the post-glacial colonization of the Connecticut River basin from south of New York [20]. But it is also clear that on a microgeographic scale, something important is limiting gene flow among populations in Allyn Brook allowing for separate population-genetic structures to establish.
The genetic differentiation of populations on a microgeographic scale has important implications for conservation of biodiversity. The Evolutionary Significant Unit (ESU), first proposed by Ryder (1986) has been expanded upon importantly by Moritz [54] and Casacci, Barbero and Balletto [55]. The idea is to  [56]. If the haplotype network ( Figure 3) is rooted at Haplotype 1 as per Tipton, Gignoux-Wolfsohn, Stonebraker and Chernoff [20], then there are at least three distinct lineages; the divergence of nuclear microsatellite alleles is well described above and the populations are distinct from their nearest neighbors in the Coginchaug River (Table 3, Figure 4 lower). As more species are examined at microgeographic scales, such as that by Nash, Kraczkowski and Chernoff [8], we predict that divergence and structuring patterns of the type demonstrated herein will become the norm rather than the exception. If we recognize evolutionary potential as suggested by Tim, Anne, Boyd, Micah, Douglas, Aaron and Bonnie [56], then populations meeting the other criteria of Moritz [54] will be recognized as ESU's-a situation we would agree with but may be impractical. We do suggest that focusing upon habitat preservation at landscape levels can also help protect unique and valuable genetic diversity.