Briefings in Functional Genomics, 15(5), 2016, 342–351

doi: 10.1093/bfgp/elw008
Advance Access Publication Date: 4 April 2016
Review paper

Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019

Ecological population genomics in the marine
environment
Douglas L. Crawford and Marjorie F. Oleksiak
Corresponding author: Douglas L. Crawford, University of Miami, RSMAS, Marine Biology and Ecology, 4600 Rickenbacker Causeway, Miami, FL 33149.
E-mail: dcrawford@rsmas.miami.edu

Abstract
Marine species live in a wide diversity of environments and yet, because of their pelagic life stages, are thought to be well-
connected: they have high migration rates that inhibit significant population structure. Recent innovations in sequencing
technologies now provide information on nucleotide polymorphisms at thousands to tens of thousands of loci based on
whole genomes, reduced representative portions of genomes (0.1–1%) or a majority of expressed mRNAs. Data from these
genomic approaches are used to define and quantify single-nucleotide polymorphisms (SNPs). These SNP data tend to agree
with data from older technologies (allozymes or microsatellites), which support well-connected populations with few gen-
etic differences among populations. However, these studies also find few percentages of SNPs (1–5%) that readily distinguish
genetic differences among populations on relatively small geographic scales. The magnitudes of the genetic differences (FST
values) suggest that hundreds of loci with significant differences are due to positive selective pressures. Thus, these data
suggest that natural selection is effectively altering allele frequencies at 100s of loci in marine populations. In this manu-
script, we provide examples of these studies, the strengths and weaknesses of different genomic approaches as well as im-
portant technical aspects associated with genomic approaches.

Key words: GBS, RAD, RNA-Seq, evolution, population genetics

Importance of genomics
The world’s oceans provide diverse organisms and environmental
habitats to better explore ecological genomics: how genomic vari-
ation and its expression enhance an organism’s probability of suc-
cess. Marine habitats range from freezing polar regions to warm
tropical waters, from shallow coastal estuaries to deep-sea hydro-
thermal vents that rely on chemosynthesis rather than photosyn-
thesis, and adaptation to these environments likely requires
many different genes in many different pathways. Using genomic
approaches, which query the whole genome rather than a tar-
geted gene set, we are more likely to discover the evolved changes
responsible for this variation in life than if we explore only a few
genes. Additionally, because genomic approaches provide so
many loci, many of which are evolving by neutral processes, we
can use neutral-demographic variation to define how populations
are connected. This is important because many marine organisms
have a pelagic larval stage, and the oceans’ currents can transport
larvae thousands of kilometers from where they hatched. We also
can contrast neutral loci to genetic changes that deviate from the
neutral patterns and are most likely evolving by natural selection
to better understand local adaptation. Marine environments pro-
vide clines along a coast, within an estuary and within a tidal
shore, and altered environments at vastly different time and spa-
tial scales to investigate the potential for adaptive change.
Further, changing habitats due to factors such as global climate
change, anthropogenic pollution and habitat destruction are af-
fecting all the world’s oceans. Genomic approaches allow us to ex-
plore the rate of adaptive change on a scale that was not
previously possible. Combining genomics with research on marine
organisms that inhabit diverse habitats provides opportunities to
address important questions concerning how marine organisms

Douglas L. Crawford is a Professor in the Department of Marine Biology and Ecology at the Rosenstiel School of Marine and Atmospheric Science,
University of Miami. His research focuses on evolutionary genomics integrating physiology, mRNA expression and nucleotide polymorphisms.
Marjorie F. Oleksiak is an Associate Professor in the Department of Marine Biology and Ecology at the Rosenstiel School of Marine and Atmospheric
Science, University of Miami. Her research uses genomic tools with physiological and toxicological phenotypes to enhance our understanding of evolu-
tionary processes.

C The Author 2016. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com
V

342

and populations interact with their environments and adapt to
environmental change, and genomic approaches provide the tools
to address the genetic basis of these ecological interactions with a
breadth of coverage that has not previously been available.
Genomic studies interrogate across the genome and include
whole-genome analyses, quantifying gene expression and ana-
lyses of genetic variation (e.g. SNP, single-nucleotide polymorph-
ism). This essay focuses on nucleotide variation; quantitative
differences in gene expression are reviewed elsewhere [1–11].
Common to all genomic studies is the breadth of the data that
are interrogated. Whole-genome sequences, all expressed genes
and thousands of genetic markers are interrogated versus a few
sequences, transcripts or microsatellites that have been used pre-
viously. Both the genomic breadth and random sampling of a
large number of genetic markers across the genome allow one to
establish neutral expectations and provide greater confidence in
estimating neutral population parameters, such as effective
population size and migration rate [12]. Demography and evolu-
tionary history should affect these neutral loci differently from
loci under selection or outlier loci, and consequently, population
genomics also provides the ability to identify adaptive or func-
tionally important loci and genes [13]. This is the fundamental
difference between recent genomic approaches and many earlier
studies: with many 1000s of markers, one can readily define neu-
tral and demographic patterns and contrast these with DNA poly-
morphism patterns that significantly differ from the neutral
expectation. It is the discovery of the rapidity and breadth of non-
neutral changes that will greatly enhance our understanding of
life and how it accommodates to changes in our biosphere.
Progress in ecological genetics
Genetic markers have been used to study population structure
in natural populations since the early 1960s. The first genetic
marker used was hemoglobin, which was used to detect inter-
species polymorphisms in whiting and cod populations [14].
Hemoglobin studies were soon followed by allozyme studies.
Allozymes are amino acid protein variants in enzymes that can
be differentiated by non-denaturing gel electrophoresis (starch
gels) using enzyme-specific stains, and initial allozyme studies
uncovered substantial protein polymorphisms in natural fly
populations [15] and humans [16]. These two studies were fol-
lowed by many others that substantiated these initial findings
[17, 18]: natural populations showed too many protein poly-
morphisms (30% of proteins were polymorphic) based on the
concept of genetic load [19, 20]. These discoveries led to the
neutral theory of molecular evolution [21]. The neutral theory of
molecular evolution suggests that most molecular polymorph-
isms have no effect on fitness (are not adaptively important)
and has since been used extensively as a null hypothesis to test
whether the variations within and among species are biologic-
ally important because they affect fitness [22, 23].
With the advent of DNA methods, genetic markers shifted
from protein-based markers to DNA-based markers. These DNA-
based markers include restriction fragment length polymorph-
isms (RFLPs) and the closely related amplified fragment length
polymorphisms (AFLPs), microsatellites and SNPs and are well-
summarized in other reviews [24]. Table 1 lists key characteristics
of both protein- and DNA-based genetic markers.
SNP discovery with next-generation sequencing
Until recently, most studies investigating environmental impacts
on genetic variation examined less than 100 loci [26]. Yet, in the
Ecological population genomics | 343
past 5 years, several approaches have been developed to survey
1000s to 10 000s of loci by sequencing whole genomes or reduced
portions of the genome (transcriptome sequencing or reduced rep-
resentative libraries [27]). These approaches are discussed below.
While whole-genome sequencing has been widely used to
discover SNPs in model organisms such as humans and flies,
there are fewer examples in non-model, marine organisms. A
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
few examples of SNP discovery using whole-genome sequenc-
ing in marine organisms include 5.9 million SNPs identified in
three-spined stickleback (Gasterosteus aculeatus) [28], 3.8 mil-
lion SNPs identified in Pacific oyster (Crassostrea gigas) [29], 1
million SNPs identified in Atlantic cod (Gadus morhua) [30] and
335 000 SNPs identified in acorn barnacle (Semibalanus bal-
anoides) [31]. These whole-genome approaches require a refer-
ence genome. An alternative approach is to assemble a partial
genome by aligning genomic sequences to an assembled tran-
scriptome (full-length RNA sequences that represent a majority
of expressed genes). This creates a partial genome assembly of
both exons and flanking regions. This approach identified
440 817 SNPs in Atlantic herring (Clupea harengus) [32]. A similar
approach was taken with European anchovy (Engraulis encrasico-
lus) [33]. Although feasible for many laboratories, whole-gen-
ome sequencing for non-model species is not trivial because
genome assembly requires high-performance computing and
extensive bioinformatics. Thus, until whole-genome bioinfor-
matics methods (e.g. starting with whole-genome assembly and
annotation followed by variant detection) for hundreds of indi-
viduals become standard for non-model organisms, SNPs for
non-model organisms are most likely to be identified by
sequencing a reduced subset of a species’ genome. Even if
whole-genome bioinformatics methods were standard, the ex-
perimental question may be more appropriately answered
using a reduced representative library because it is a less labori-
ous approach for sampling 10 000s of loci in many individuals.
There are two main ways to reduce the complexity of a spe-
cies’ genome for population genomic analyses across many in-
dividuals: (1) focus on expressed genes, and (2) construct a
reduced representative genomic DNA library. A big advantage of
both these methods is that using the same data, one can simul-
taneously discover SNPs and genotype individuals at those
same SNPs [34] as long as individuals are not pooled. If individ-
uals are pooled, then population allele frequencies can be deter-
mined but one forgoes knowing individual genotypes. A final
advantage of these two methods is that they can be used on
species with few or no genomic resources, which is often the
case for marine organisms, especially those of little economic
importance. A third method, exon capture [35], provides a gene-
targeted approach to identify SNPs. However, this approach re-
quires prior knowledge of the genes to be captured based on the
species of interest or a related species and thus is more difficult
to implement for many marine species.
Focusing on transcriptomics or expressed genes (mRNAs se-
quences) for SNP discovery significantly reduces the sequence
target size (only 1–5% of vertebrate genomes are transcribed
[36]) and has been used to identify SNPs in a variety of marine or-
ganisms. SNPs have been identified from expressed sequence tag
(EST) collections in Atlantic cod (Gadus morhua) [37, 38], shrimp
(Litopenaeus vannamei) [39, 40], sole (Solea solea) [41], stickleback
(Gasterosteus aculeatus) [28], killifish (Fundulus heteroclitus) [42] and
eastern oyster (Crassostrea virginica) [43]. SNPs also have been
identified using 454 cDNA sequence data in copepods (Tigriopus
californicus) [44], marine gastropods (Littorina saxatilis [45] and
Concholepas concholepas [46]), soft clam (Mesodesma donacium) [47],
Atlantic herring (Clupea harengus) [48], silver-lipped pearl oyster
 344 | Crawford and Oleksiak

Table 1. Common genetic marker characteristics for diploid organisms

Characteristics Allozymes RFLPs/AFLPs Microsatellites SNPs

Codominance (genotypic data) Yes No Yes Yes

Neutrality Some Yes (mostly) Yes Transcriptome-based: uncertain; genomic-based: mostly
Number of variable loci typically assayed 10–50 10–1000 5–20 100s–10 000s
Number of alleles/locus 1–5 2 1–50 2

Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019

Based on [25].

Table 2. Useful terms and their description for this article

Term Description

Whole-genome sequencing
RRGS: Reduced representative
genome sequencing (RAD, RAD-tag, GBS)
Transcriptomics and RNA-seq
Capture sequencing, exon-capture,
exomic sequencing, targeted sequencing
454 sequencing technology
Illumina sequencing technology
ABI SOLiD sequencing technology
Sequences that capture nearly all of the genome and are assembled into large (10s–100 kb)
scaffolds. Because whole genomes can be 0.1 to 10 Gbps, whole-genome approaches require
much more sequencing than GBS, RNA-seq or exon-capture. Thus, few individuals are
typically analyzed.
Genome sequencing that only sequences a selected ‘reduced’ portion of the genome. Typically
this approach sequences 0.1% to 1% of the genome derived from specific restriction sites.
The power of this approach is that the same loci in the genome of approximately 50–100 bp
are sequenced in many (100s of) individuals.
Transcriptomics is the quantitative and qualitative analysis of RNA expression. RNA-seq is a
transcriptomic approach that uses sequences of expressed RNA and provides both quantita-
tive measures of expression and identifies nucleotide variation in expressed DNA (i.e. in
RNA).
Capture sequencing, as the name implies, is sequencing selected (captured) genomic DNA tar-
gets. The most general approach is exon-capture or exomic sequencing that uses PCR,
hybridization probes or cDNAs to capture exons or DNA that is expressed as RNA. Yet, the
selected targets can be any parts of the genome where there are probes, or primers to specify
a portion of the genome.
454 was one of the initial next-generation sequencing approaches that captures individual
DNA fragments on beads and sequences each of these fragments in separate wells in high-
density well plates; 454 sequencing produces 300–500 bp sequences (reads) yielding 0.5
Gbp/run.
Illumina sequencing captures DNA fragments on a solid surface and sequences these
fragments on the surface. There are many different platforms, and the High-Seq platform
that typically produces 50–200 bp reads yields 50 to 1000 Gbp per run
Sequencing by Oligonucleotide Ligation and Detection is a novel approach that does not rely
on sequencing by DNA polymerization. Each sequence is 50 bp and yields 30–50 Gbp per
run.

(Pinctada maxima) [49], California red abalone (Haliotis rufescens)
[50] and a marine annelid (Streblospio benedicti) [51]. Both EST and
454 sequencing (Table 2) yield relatively long reads compared
with the Illumina and ABI platforms (Table 2), making de novo
transcriptome assembly easier and potentially more robust. To
compare the utility of Roche 454 (which produces 300–500 bp
reads but only 0.5 Gbp/run [52]) and Illumina Genome Analyzer
(GAII, which only produces 100 bp reads, but >30 Gbp [52])
sequencing for SNP discovery, both platforms were used for SNP
discovery in European hake (Merluccius merluccius). Not surpris-
ingly, the average contig length (set of overlapping DNA se-
quences) was 3-fold greater for 454 than for Illumina, while the
average coverage depth was 7.5-fold greater for Illumina.
However, the two approaches yielded similar percentages of
polymorphic loci, showing that both platforms are suitable for
large-scale SNP discovery [53]. Other marine species using shorter
read sequencing platforms such as Illumina and ABI SOLiD plat-
forms for SNP discovery include mammals, fish, mollusks and
arthropods [31, 32, 54–60]. De Wit et al. outline a transcriptome
assembly approach followed by SNP marker development and
genotyping [61]. The idea is relatively straightforward: one
assembles a transcriptome from RNA-seq (RNA sequencing or
whole transcriptome sequencing) data, then aligns RNA-seq
reads to this assembly and finally identifies polymorphic loci.
The RNA-seq reads might come from individual transcripts or
from transcripts pooled from many individuals.
Focusing SNP discovery on expressed genes has the signifi-
cant advantage of targeting many functional portions of the
genome (i.e. all SNPs come from mRNAs, most of which encode
proteins and do not include non-coding genomic regions) [61].
Knowing the function of a locus containing a SNP of interest
gives further insight into the putative SNP phenotype and is
useful for downstream analyses, for instance gene functional
category enrichment analyses. This knowledge can be critically
important when looking for signatures of natural selection and
trying to determine the functional consequences of variation
that seems to be under selection or adaptive.
One shortcoming of targeting expressed genes results from
differences in allele-specific expression. Allele-specific expres-
sion, where one allele is preferentially expressed, can result
in inaccurately calling an individual a homozygote rather than
a heterozygote. Perhaps the biggest disadvantage of using

Figure 1. Frequency of expression. The number of gene-specific sequence tags
(as log based 10) versus their relative frequency for 5890 different transcripts.
Most genes have few sequence reads and a few genes have most reads.
RNA-seq to identify SNPs results from the widely varying mRNA
concentrations within a sample [62]. Figure 1 shows the fre-
quencies of genes and their relative expression level from an
RNA-seq experiment. Most mRNAs (4536 or 76%) have few reads
each, and the sum of all these reads represents only a fraction
(6%) of the total number of reads. A few mRNAs (136 or 2%) are
greatly overexpressed and thus have many reads each; the sum
of these reads represents >80% of all reads. This means that a
gene that is highly expressed will be sequenced many times at
the expense of genes that are less highly expressed, without
any additional SNP identification. More importantly, because
highly expressed genes use up the sequencing capacity, fewer
individuals can be sequenced at the same time, losing the ad-
vantage of simultaneously discovering SNPs and genotyping
hundreds of individuals. For example, using the RNA-seq ap-
proach in red abalone, approximately 22 000 SNPs were identi-
fied among 39 individuals using seven lanes of an Illumina
HiSeq 2000 [55]; in contrast, typically a similar number of ran-
dom, genome-wide SNPs can be identified in 100s of individuals
using a single lane [63].
Identifying SNPs in genomic DNA avoids the problem of
biased representation of the most highly expressed mRNAs, as
there are only two copies of each locus per cell and each locus
occurs at the same frequency (Table 2). However, as suggested
above, due to both sequencing and genome assembly and anno-
tation costs, a more practical strategy for population genomics
of natural marine populations is to only sequence a select small
portion of the genome. The most common strategy is ‘reduced
representation genome sequencing’ (RRGS, Table 2, reviewed in
[27, 64, 65]). RRGS defines genotypes by sequencing and is also
described as RAD-seq (restriction site-associated DNA sequenc-
ing) [58, 66] and GBS (genotyping by sequencing) [63]. In RRGS,
genomic DNA is cut with one or more restriction enzymes, and
only short DNA fragments are sequenced (reviewed in [27, 64,
65]), constituting approximately 0.1% to 1% of the genome. This
approach yields many thousandfold sequence coverage for the
same fragments. Many hundreds of individuals can be
sequenced simultaneously by using barcodes that identify each
individual. With 100s of millions of sequence reads of approxi-
mately 100 bp, there will be 1000s to 10 000s of the same loci
Ecological population genomics | 345
sequenced in 100s of individuals with >10-fold coverage per nu-
cleotide per individual. This sequencing depth for many indi-
viduals allows one to define SNPs, genotypes for all individuals
and allele frequencies across populations. With barcoding,
RRGS allows hundreds of individuals to be genotyped simultan-
eously and relatively cheaply and has opened the door for a
huge range of population genomics studies using any non-
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
model species. Due to the ease of these approaches, their re-
peatability [67–69] and their potential to produce thousands of
genetic markers, RRGS approaches to identify SNPs are fast
becoming the approach of choice for population genomic ana-
lyses in marine organisms.
Unlike SNPs developed using mRNAs, SNPs identified using
RRGS approaches with genomic DNA are most frequently from
non-functional, non-coding genomic DNA regions. While many
‘non-functional’ or neutral markers are useful for defining
population genetic parameters [12], RRGS utility to discover
SNPs that are evolving by natural selection may be less effective
than RNA-seq because so many of the SNPs should be associ-
ated with genomic DNA that has little if any function. Despite
this, we and others have used GBS or RAD tag sequencing
approaches to define SNPs evolving by natural selection [70, 71].
Thus, both sequencing cDNAs or reduced representative gen-
ome approaches have their strengths and weaknesses, and the
choice to use one or the other will depend on research goals.
Implications for marine ecology
Genomic approaches with marine organisms provide finer reso-
lution of population structure; for example several thousand
SNPs are able to define coral reef population structure along the
Florida coast where previous microsatellite or DNA sequences of
one or few loci suggest no differences along this coast [72].
Additional genomic approaches more readily identify loci with di-
vergence patterns that are unlikely to represent neutral diver-
gence but instead are most likely due to natural selection [32, 54,
55, 59, 60, 71, 73]. These genomic approaches that estimate the
shared nucleotide diversity among individuals and populations
provide thousands of loci, many of which are neutral. It is the
neutral loci that are used to estimate basic population genetic
parameters and demography estimates [12], and thus it is pos-
sible to more accurately define loci with non-neutral patterns
(potentially adaptive loci). An approach to defining adaptive loci
is to distinguish SNPs with FST values that are dissimilar to FST
values achieved when permutating the data (FST values are too
large relative to expected values [74, 75]). FST values [76, 77] pro-
vide measures of the genetic distance among populations and
range from 0 to 1.0. FST values measure the genetic variation be-
tween populations relative to the total variation [76, 77]. A simple
but inexact estimate of a FST value is: FST ¼ ((HT  HS)/HT), where
HT represents expected heterozygosity per locus of the total
population and HS represents expected heterozygosity of a sub-
population. Thus, a genomic approach that identifies 1000s of
SNPs is statistically more powerful in identifying adaptive SNPs
because SNPs with too large FST values are more readily defined
in comparison with many neutral loci [75, 78, 79].
One example of the strength of a genomic approach is in the
estuarine fish sailfin molly (Poecilia latipinna). An RRGS approach
was used to define approximately 1320 SNPs found in 80% of
120 individuals. Among three geographically close (within 10
km) populations in the Florida Keys (Figure 2A), a majority of
SNPs suggest little genetic distance with FST values <0.0125 (less
than 1.25% of the variance among populations relative to the
total) [71]. Although this FST value is significant, it is very small.
 346 | Crawford and Oleksiak
Figure 2. Sailfin molly GBS. A: Three geographically close Florida Keys populations (BP: Big Pine Key, NNK: No Name Key). B: Outlier test for 1.3K SNPs among 120
individuals; red highlights SNPs with extreme FST values unlikely to occur due to neutral processes. C. Structure analysis of Florida Keys populations using SNPs with
extreme FST values. Each vertical bar represents a multivariate summary of the ancestral genotype. (A colour version of this figure is available online at: http://
bfg.oxfordjournals.org)
The RRGS data for the genetic variation within and among these
and other populations are very similar to previous studies using
allozymes and microsatellites [80, 81], which suggests that there
is little if any difference among South Florida sailfin molly
populations. Yet, among the salt marsh flats in the Florida Keys
populations, there are SNPs with significant genetic diversity
among populations (Figure 2B [71]). Specifically, these three
similar populations have 18 SNPs with FST values that are un-
likely to occur relative to permutation of SNPs with similar het-
erozygosities (p-value < 0.01, false discovery rate ¼ 0.1; Figure
2B). Using these loci, there is a strong structure among popula-
tions (Figure 2C). The meaningful difference between RRGS and
previous measures of genetic diversity is the number of loci
examined, which allows more precise delineations of popula-
tion structure and, as well, facilitates identifying loci with ex-
cessive FST values that could indicate adaptive divergence.
Thus, the power of RRGS studies is that they enhance previous
studies and can also distinguish subtle but evolutionarily im-
portant genetic differences.
The resolution of genetic differences based on RRGS is found
in other studies. In 37 red abalone (Haliotis rufescens), the vast
majority of SNPs had little genetic differentiation among popu-
lations [55]. Yet, 691 from 22 000 loci had significantly higher
FST values that readily distinguish populations along the
California Coast [55]. Similarly, among Baltic and Atlantic her-
ring (Clupea harengus), most SNPs support genetic homogeneity,
reflecting the lack of obvious dispersal barriers in marine habi-
tats [32, 54]. These SNPs for many or a majority of loci yield
similar results to studies using allozymes or microsatellites
where population differentiation across large geographic areas
is very low (FST values 0.005). Yet, out of 5985 SNPs, there were
117 (2.0%) that showed evidence of substantial divergence, cor-
responding to FST values ¼ 0.128 [54]. In a similar analysis, when
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
investigating 440 817 SNPs, 3847 SNPs (approximately 1%)
showed significantly strong differences among populations,
with a few at or near fixation [32]. Similar data are found in tem-
poral isolation of pink salmon (Oncorhynchus gorbuscha)
and Atlantic salmon (Salmo salar), where a few out of 1000s of
SNPs are associated with migration and reproduction timing
[59, 60].
Data from these studies consistently detect a few percent-
ages of SNPs with elevated divergence between groups or popu-
lations that exceed neutral expectations. These studies
highlight the power of genomic approaches: discovery and char-
acterization of many thousands of polymorphic loci provide a
wide breadth of information where 1-5% of SNPs distinguish im-
portant population structure. Many of these informative SNPs
have FST values that are indicative of non-neutral processes and
thus suggest that adaptive evolution acting on many loci is
common.
Technical aspects: alignment, sequence depth,
HWE and LD
The big difference between traditional genetic markers and gen-
etic markers from genomic approaches is the number of
markers. Genomic approaches can yield hundreds of thousands
and even millions of genetic markers. However, except for num-
ber, genetic markers generated via genomic approaches are very
similar to those generated via less high-throughput approaches
and should be treated so [31, 71]. This entails filtering the data
so that only high-quality and high-confidence genetic markers
are used to address the relevant biological questions. An initial
filtering step begins with sequence coverage depth. The general
idea is that a polymorphism that occurs in multiple sequences
 Ecological population genomics | 347

Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019

Figure 3. Distribution of minor allele frequencies (MAFs) and Ho/He. SNPs from 120 individuals among five Florida sailfin molly populations [71]. A: Histogram of MAFs
for 3878 SNPs prior to filtering (i.e. without removing SNPs that do not occur in 90% of 120 individuals from five populations, and >1% MAF. MAFs of 5% and 1% are high-
lighted in red and black, respectively). B: Ratio of observed heterozygosity (Ho) over expected heterozygosity (He ¼ 2pq). SNPs with Ho > He and significantly different
from Hardy–Weinberg expectation (HWE, p < 0.01) highlighted in red. Notice that many SNPs have Ho  He, because the Wahlund effect among five populations would
reduce Ho relative to He [93]. These SNPs with low Ho are not removed. (A colour version of this figure is available online at: http://bfg.oxfordjournals.org)

and in multiple individuals is unlikely to be a technical artifact
(e.g. a sequencing error). Yet, there are potential technical prob-
lems specific to RRGS approaches [27, 64, 65].
Five technical aspects to be aware of are ascertainment bias,
alignment method, sequence depth, mis-alignment of paralogs
and linkage disequilibrium (LD). SNP discovery should use indi-
viduals from across the population range versus genotyping
one population and investigating these population-specific
polymorphisms. Examining SNPs discovered in only one popu-
lation leads to ascertainment bias [82]. This ascertainment bias
occurs because polymorphisms are population-specific, and
defining SNPs in one population will cause one to miss SNPs in
a second population, thus creating a false sense of lower nu-
cleotide diversity in the second population.
An important technical aspect of RRGS approaches starts with
alignment. In general, short sequence reads or tags are aligned to
a reference genome or, alternatively, are assembled into a set of
contiguous sequences, to which individual sequence reads are
aligned. The alignment method has a large effect on the resulting
SNP pool. For instance, the Tassel pipeline [83, 84] for species
with a reference genome and the related UNEAK pipeline for spe-
cies without a reference genome are used to identify SNPs from
RRGS data. Tassel and other bioinformatics approaches that rely
on aligning short sequence reads have the option of using Bowtie
[85] and BWA [86]. However, using the same raw sequencing
data, Bowtie and BWA result in different final SNP sets: in our
and others’ experience, only approximately 50% of reads pro-
duced shared alignments with these two tools [87–89]. Even after
filtering to remove SNPs with too few reads and excessive Ho,
this difference in which SNPs are identified is not eliminated, al-
though it does greatly reduce the frequencies of alternate SNP
identification [72]. It is difficult to suggest the cause or the conse-
quences of identifying different SNPs from the same sequences,
but one can compare these two approaches and inquire about
the frequencies of paralogs and sequencing depth to choose an
objective approach (see below).
 348 | Crawford and Oleksiak
Sequencing depth, the number of reads per locus or SNP, de-
fines the probability of identifying allelic variation, defining in-
dividual genotypes and avoiding sequencing errors. To identify
true allelic variants (versus sequencing errors) and genotypes
(where the identity of a SNP variant on both chromosomes is
estimated) requires sufficient read depth. If there were an equal
chance of sequencing both alleles, the probability of identifying
the second allele would be likely (>90%) with five reads.
Unfortunately, the distribution of reads is Poisson or negative
binomial [90], which means that one needs approximately 10
reads per locus per individual to accurately estimate genotypes
[91, 92]. Notice that genotype can be used to estimate popula-
tion allele frequencies from pooled samples of sufficient num-
bers of individuals. However, identifying both alleles in an
individual and thus its genotype is important for association
studies (relating genotypes to phenotypes) and for identifying
paralogs (different loci for similar genes, e.g. the three genes for
lactate dehydrogenase: Ldh-A, Ldh-B and Ldh-C). Identifying
genotypes is readily accomplished by ligating on a barcode for
each individual: a sequence-specific DNA adaptor that is pre-
sent in each sequence read and is used to identify every
individual.
Sequencing or PCR errors that can contribute to incorrect
SNP calls can be avoided by requiring a minimum number of
reads in more than one individual. Sequencing errors of 0.01%
to 0.1% can produce rare SNPs [92]. Similarly, in many RRGS
approaches, there is a PCR step that can introduce errors in
many copies of a sequence tag if the error occurs in the first few
cycles. These will create many sequence tags in an individual
with a novel SNP. Both errors can be avoided by requiring that
the rare SNP both has sufficient read depth and occurs in many
individuals. Thus, if one requires a minimum of five reads per
SNP with 1% minimum allele frequency (MAF) when sequencing
>100 individuals, all SNPs will be represented in more than one
individual multiple times. Removing SNPs with low MAF has a
small effect on the total number of SNPs (Figure 3A), and thus is
a simple way to avoid errors. Moreover, removing these SNPs
enhances downstream analyses that rely on permutation of the
data [94]. The downside of this approach of choosing MAFs is
that one will miss rare variations that may contribute to pheno-
typic variation [95].
RRGS approaches using short reads generated by next-gen-
eration sequencers can produce alignments among different
paralogs (similar genes that occur at different locations so they
do not recombine). A hallmark of SNPs that represent align-
ments among paralogs (variants between loci instead of true al-
lelic variants at a single locus) is that there will be excessive
heterozygosity. For example, imagine two genes that code for
similar proteins and thus share a high degree of similarity for
parts of their genes; further imagine that each gene has a fixed
different nucleotide at the same position. If a short read aligns
to both genes at this position, all individuals will appear to be
heterozygotes. The first step to avoiding this is removing tag se-
quences that align to more than one portion of the genome, a
default requirement for several alignment tools. This requires a
complete genome where all paralogs are identified. However,
not all RRGS projects in marine species will have a genome, and
not all genomes are assembled well-enough so that all paralogs
are identified. An additional approach is to identify all SNPs
with excessive observed heterozygosity (Ho) relative to the ex-
pected heterozygosity (He) using the Hardy–Weinberg equilib-
rium test (Figure 3B). Notice, only SNPs with large excessive Ho
are rejected because loci with heterozygote deficits (low Ho)
relative to He are likely among divergent populations (Wahlund
effect [93], Figure 3B), and these informative SNPs can be used
to describe the connectivity and differences within a species.
LD is the significant correlation among SNPs. This statistical
association among SNPs when physically close is due to limits
of recombination to create independence among sites. A poten-
tial advantage for genomic approaches that identify 1000s of
SNPs is a more accurate estimate of effective population size
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
using the LD method [96]. Many marine organisms are believed
to have very large effective population sizes, and so, current
methodologies using 10–20 loci are not very informative (i.e.
95% confidence intervals often include infinity). This arises be-
cause LD due to drift is low in large populations and difficult to
detect. Large numbers of markers are expected to overcome this
difficulty, allowing the upper bounds of Ne to be estimated [96].
While LD and the size of linkage groups can be a hallmark of
adaptive divergence [97], it also creates a bias in that SNPs in LD
are not independent. Thus, for example, many SNPs in LD that
have lower divergence among populations could inflate meas-
ures of connectivity. If the primary goal of an RRGS project is to
identify demographic processes, one can readily remove SNPs
in LD. This is best applied for SNPs with a genome, so that one
can examine a subset of SNPs that are physically close to one
another. It is more difficult with SNP discoveries without a gen-
ome because it would require defining the correlation among all
possible pairwise combinations, a computationally difficult task
with thousands of SNPs. There are other approaches. One is to
‘thin’ the data, so that no SNPs occur on the same tag sequence
or are within a fixed distance (e.g. only keeping SNPs that are
>100 bp apart). While these approaches are effective, one loses
informative SNPs because even though there may be a signifi-
cant correlation between loci, rarely are the correlations perfect,
and thus these sites have some information content. One sug-
gestion is to estimate population parameters by permuting the
data, using a subset of unlinked SNPs or defining the LD among
the most informative SNPs. The last suggestion is important for
defining the number of independent loci that explain the diver-
gence among populations and, because fewer SNPs are
involved, is more readily accomplished.
There are other concerns with GBS or RAD-seq approaches
[65, 98]; these include PCR-biased amplification that causes al-
lelic bias, allele drop off or loss of specific loci. These biases will
reduce the ability to identify genetic variation, making different
populations more similar because one allele is missing or
under-represented. Other problems are population-specific
polymorphisms in restriction sites such that the proper size
fragments are only produced in one or a few populations. If
SNPs are filtered such that they occur in a vast majority (>75%)
of individuals, these problems will only reduce the number of
informative SNPs, not create bias in divergence estimates
among populations.
In summary, most technical problems with RRGS
approaches are addressed by having sufficient read depth
among many individuals and eliminating SNPs with too large
Ho/He. These requirements may underestimate variation be-
cause rare or low-frequency alleles will not be considered.
Linkage is not necessarily a problem because it can be used to
estimate non-neutral patterns of divergence, but it will overesti-
mate the number of independent loci.
Conclusion
Marine ecological genomics provides the ability to better under-
stand population structure, genetic divergence among popula-
tions and selectively important genes. These data are important

because they inform us about the conservation genetics of iso-
lated populations, the genes affecting important phenotypes
(e.g. reproductive schedules) and the effectiveness of adaptive
change. Currently, many genomic studies suggest that marine
species have greater population structure than previously
appreciated. Additionally, many of these studies identify many
loci that appear to be evolving by natural selection. An exten-
sion of these observations is that natural selection is more ef-
fective than currently appreciated, resulting in marine
populations adapted to subtle differences in their local environ-
ments. If it is true that natural selection is more effectively
shaping population-specific genotypes, it suggests that current
climate changes will be mitigated by adaptive change in many
marine organisms with sufficient genetic variation.
Key Points
• Many marine organisms inhabit diverse environments
and have highly connected populations because of pe-
lagic life stages.
• Genomic approaches provide many thousands of poly-
morphic nucleotide sites.
• Most population analyses using genome-wide poly-
morphism data support previous population analyses
using fewer genetic markers.
• The strength of genomic approaches is that many
thousands of neutral loci can be used to define loci
that exhibit population-specific or extreme FST values.
• These non-neutral loci have greater power to define
population structure or explain genotype to phenotype
relationships.
Funding
M.F.O. and D.L.C. was provided by NSF grants from MCB
1434565, IOS 1147042 and DEB-1265282.
References
1. Cossins AR, Crawford DL. Fish as models for environmental
genomics. Nat Rev Genet 2005;6:324–33.
2. Grishkevich V, Yanai I. The genomic determinants of geno-
type  environment interactions in gene expression. Trends
Genet 2013;29:479–87.
3. Oleksiak M F., Crawford DL, Functional genomics in fishes, in-
sights into physiological complexity. In: Evan D., Claiborne J.
(eds). The Physiology of Fishes. Boca Raton: CRC Press, 2006,
523–50.
4. Oleksiak MF. Genomic approaches with natural fish popula-
tions. J Fish Biol 2010;76:1067–93.
5. Romero IG, Ruvinsky I, Gilad Y. Comparative studies of gene
expression and the evolution of gene regulation. Nat Rev
Genet 2012;13:505–16.
6. Stranger BE, Forrest MS, Clark AG, et al. Genome-Wide
Associations of Gene Expression Variation in Humans. PLoS
Genet 2005;1:e78.
7. Townsend JP, Cavalieri D, Hartl DL. Population genetic vari-
ation in genome-wide gene expression. Mol Biol Evol
2003;20:955–63.
8. Whitehead A, Crawford DL. Variation within and among spe-
cies in gene expression: raw material for evolution. Mol Ecol
2006;15:1197–211.
9. Williams RBH, Chan EKF, Cowley MJ, et al. The influence of gen-
etic variation on gene expression. Genome Res 2007;17:1707–16.
Ecological population genomics | 349
10. Wittkopp PJ. Variable gene expression in eukaryotes: a net-
work perspective. J Exp Biol 2007;210:1567–75.
11. Wray GA, Hahn MW, Abouheif E, et al. The evolution of tran-
scriptional regulation in eukaryotes. Mol Biol Evol 2003;20:
1377–419.
12. Allendorf FW, Hohenlohe PA, Luikart G. Genomics and the
future of conservation genetics. Nat Rev Genet 2010;11:
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
697–709.
13. Luikart G, England PR, Tallmon D, et al. The power and prom-
ise of population genomics: from genotyping to genome typ-
ing. Nat Rev Genet 2003;4:981–94.
14. Sick K. Haemoglobin polymorphism in fishes. Nature 1961;
192:894–6.
15. Lewontin RC, Hubby JL. A molecular approach to the study of
genic heterozygosity in natural populations. II Amount of
variation and degree of heterozygosity in natural populations
of Drosophila pseudoobscura. Genetics 1966;54:595–609.
16. Harris H. Enzyme polymorphisms in man. Proc R Soc Lond
1966;164:298–310.
17. Nevo E. Genetic variation in natural populations: patterns
and theory. Theor Popul Biol 1978;13:121–77.
18. Selander RK, Johnson WE. Genetic variation among verte-
brate species. Annu Rev Ecol Syst 1973;4:75–91.
19. Charlesworth B. Why we are not dead one hundred times
over. Evolution 2013;67:3354–61.
20. Haldane JBS. The cost of natural selection. J. Genetics
1957;55:511–24.
21. Kimura M. Evolutionary rate at the molecular level. Nature
1968;217:624–6.
22. Kreitman M. The neutral theory is dead. Long live the neutral
theory. Bioessays 1996;18:678–83; discussion 683.
23. Wayne ML, Simonsen KL. Statistical tests of neutrality in the
age of weak selection. Trends Ecol Evol 1998;13:236–40.
24. Schlötterer C. The evolution of molecular markers—just a
matter of fashion? Nat Rev Genet 2004;5:63–9.
25. Jarne P, Lagoda PJ. Microsatellites, from molecules to popula-
tions and back. Trends Ecol Evol 1996;11:424–9.
26. Seeb JE, Carvalho G, Hauser L, et al. Single-nucleotide poly-
morphism (SNP) discovery and applications of SNP geno-
typing in nonmodel organisms. Mol Ecol Resour 2011;11
(Suppl 1):1–8.
27. Davey JW, Hohenlohe PA, Etter PD, et al. Genome-wide gen-
etic marker discovery and genotyping using next-generation
sequencing. Nat Rev Genet 2011;12:499–510.
28. Jones FC, Grabherr MG, Chan YF, et al. The genomic basis of
adaptive evolution in threespine sticklebacks. Nature
2012;484:55–61.
29. Zhang G, Fang X, Guo X, et al. The oyster genome reveals
stress adaptation and complexity of shell formation. Nature
2012;490:49–54.
30. Star B, Nederbragt AJ, Jentoft S, et al. The genome sequence
of Atlantic cod reveals a unique immune system. Nature
2011;477:207–10.
31. Flight PA, Rand DM. Genetic variation in the acorn barnacle
from allozymes to population genomics. Integr Comp Biol
2012;52:418–29.
32. Lamichhaney S, Martinez Barrio A, Rafati N, et al. Population-
scale sequencing reveals genetic differentiation due to local
adaptation in Atlantic herring. Proc Natl Acad Sci USA 2012;
109:19345–50.
33. Montes I, Conklin D, Albaina A, et al. SNP discovery in
European anchovy (Engraulis encrasicolus, L) by high-
throughput transcriptome and genome sequencing. PLoS One
2013;8:e70051.
 350 | Crawford and Oleksiak
34. Narum SR, Buerkle CA, Davey JW, et al. Genotyping-by-
sequencing in ecological and conservation genomics. Mol Ecol
2013;22:2841–7.
35. Hodges E, Xuan Z, Balija V, et al. Genome-wide in situ
exon capture for selective resequencing. Nat Genet 2007;39:
1522–7.
36. Willette DA, Allendorf FW, Barber PH, et al. So, you want to
use next-generation sequencing in marine systems? Insight
from the Pan-Pacific advanced studies institute. Bull Mar Sci
2014;90:79–122.
37. Bowman S, Hubert S, Higgins B, et al. An integrated approach
to gene discovery and marker development in Atlantic cod
(Gadus morhua). Mar Biotechnol 2011;13:242–55.
38. Hubert S, Higgins B, Borza T, et al. Development of a SNP re-
source and a genetic linkage map for Atlantic cod (Gadus
morhua). BMC Genomics 2010;11:191.
39. Ciobanu DC, Bastiaansen JWM, Magrin J, et al. A major SNP re-
source for dissection of phenotypic and genetic variation in
Pacific white shrimp (Litopenaeus vannamei). Anim Genet
2010;41:39–47.
40. Du ZQ, Ciobanu D, Onteru S, et al. A gene-based SNP linkage
map for pacific white shrimp, Litopenaeus vannamei. Anim
Genet 2010;41:286–94.
41. Diopere E, Hellemans B, Volckaert FA, et al. Identification and
validation of single nucleotide polymorphisms in growth-
and maturation-related candidate genes in sole (Solea solea
L.). Mar Genomics 2013;9:33–8.
42. Williams LM, Ma X, Boyko AR, et al. SNP identification, verifi-
cation, and utility for population genetics in a non-model
genus. BMC Genet 2010;11:32.
43. Quilang J, Wang S, Li P, et al. Generation and analysis of ESTs
from the eastern oyster, Crassostrea virginica Gmelin and
identification of microsatellite and SNP markers. BMC
Genomics 2007;8:157.
44. Foley BR, Rose CG, Rundle DE, et al. A gene-based SNP re-
source and linkage map for the copepod Tigriopus californi-
cus. BMC Genomics 2011;12:568.
45. Galindo J, Grahame J, Butlin R. An EST-based genome scan
using 454 sequencing in the marine snail Littorina saxatilis.
J Evol Biol 2010;23:2004–16.
46. Gallardo-Esca  rate C, Nun~ ez-Acun~ a G, Valenzuela-Mun ~ oz V.
SNP discovery in the marine gastropod Concholepas concho-
lepas by high-throughput transcriptome sequencing. Conserv
Genet Resour 2013;5:1053–4.
47. Gallardo-Esca  rate C, Valenzuela-Mun ~ oz V, Nu
n~ ez-Acun ~ a G,
et al. SNP discovery and gene annotation in the surf clam
Mesodesma donacium. Aquac Res 2015;46:1175–87.
48. Helyar SJ, Limborg M, Bekkevold D, et al. SNP discovery using
next generation transcriptomic sequencing in Atlantic her-
ring (Clupea harengus). PloS One 2012;7:e42089.
49. Jones DB, Jerry DR, Fore ^t S, et al. Genome-wide SNP valid-
ation and mantle tissue transcriptome analysis in the silver-
lipped pearl oyster, Pinctada maxima. Mar Biotechnol
2013;15:647–58.
50. Valenzuela-Mun ~ oz V, Araya-Garay JM, Gallardo-Esca  rate C.
SNP discovery and High Resolution Melting Analysis from
massive transcriptome sequencing in the California red aba-
lone Haliotis rufescens. Mar Genomics 2013;10:11–6.
51. Zakas C, Schult N, McHugh D, et al. Transcriptome analysis
and SNP development can resolve population differentiation
of Streblospio benedicti, a developmentally dimorphic mar-
ine annelid. PloS One 2012;7:e31613.
52. Metzker ML. Sequencing technologies: the next generation.
Nat Rev Genet 2010;11:31–46.
53. Milano I, Babbucci M, Panitz F, et al. Novel tools for conserva-
tion genomics: comparing two high-throughput approaches
for SNP discovery in the transcriptome of the European hake.
PloS One 2011;6:e28008.
54. Corander J, Majander KK, Cheng L, et al. High degree of cryptic
population differentiation in the Baltic Sea herring Clupea
harengus. Mol Ecol 2013;22:2931–40.
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
55. De Wit P, Palumbi SR. Transcriptome-wide polymorphisms of
red abalone (Haliotis rufescens) reveal patterns of gene flow
and local adaptation. Mol Ecol 2013;22:2884–97.
56. Everett MV, Grau ED, Seeb JE. Short reads and nonmodel
species: exploring the complexities of next-generation
sequence assembly and SNP discovery in the absence of
a reference genome. Mol Ecol Resourc 2011;11 (Suppl 1):
93–108.
57. Fernandez R, Schubert M, Vargas-Velazquez AM, et al. A
genomewide catalogue of single nucleotide polymorphisms
in white-beaked and Atlantic white-sided dolphins. Mol Ecol
Resourc 2016;16:266–76.
58. Hohenlohe PA, Amish SJ, Catchen JM, et al. Next-generation
RAD sequencing identifies thousands of SNPs for assessing
hybridization between rainbow and westslope cutthroat
trout. Mol Ecol Resourc 2011;11:117–22.
59. Johnston SE, Orell P, Pritchard VL, et al. Genome-wide SNP
analysis reveals a genetic basis for sea-age variation in a wild
population of Atlantic salmon (Salmo salar). Mol Ecol 2014;
23:3452–68.
60. Seeb LW, Waples RK, Limborg MT, et al. Parallel signatures of
selection in temporally isolated lineages of pink salmon. Mol
Ecol 2014;23:2473–85.
61. De Wit P, Pespeni MH, Palumbi SR. SNP genotyping and popu-
lation genomics from expressed sequences–current advances
and future possibilities. Mol Ecol 2015;24:2310–23.
62. Bonaldo MF, Lennon G, Soares MB. Normalization and sub-
traction: two approaches to facilitate gene discovery. Genome
Res 1996;6:791–806.
63. Elshire RJ, Glaubitz JC, Sun Q, et al. A robust, simple genotyping-
by-sequencing (GBS) approach for high diversity species. PloS
One 2011;6:e19379.
64. Andrews KR, Luikart G. Recent novel approaches for popula-
tion genomics data analysis. Mol Ecol 2014;23:1661–7.
65. Puritz JB, Matz MV, Toonen RJ, et al. Demystifying the RAD
fad. Mol Ecol 2014;23:5937–42.
66. Baird N, Etter P, Atwood T, et al. Rapid SNP discovery and gen-
etic mapping using sequenced RAD markers. PloS One
2008;3:e3376.
67. DaCosta JM, Sorenson MD. Amplification biases and consist-
ent recovery of Loci in a double-digest RAD-seq protocol. PloS
One 2014;9:e106713.
68. Gonen S, Bishop SC, Houston RD. Exploring the utility of
cross-laboratory RAD-sequencing datasets for phylogenetic
analysis. BMC Res Notes 2015;8:299.
69. Gorjanc G, Cleveland MA, Houston RD, et al. Potential of geno-
typing-by-sequencing for genomic selection in livestock
populations. Genet Sel Evol 2015;47:12.
70. Hohenlohe PA, Bassham S, Etter PD, et al. Population gen-
omics of parallel adaptation in threespine stickleback using
sequenced RAD tags. PLoS Genet 2010;6:e1000862.
71. Nunez JC, Seale TP, Fraser MA, et al. Population genomics
of the euryhaline teleost Poecilia latipinna. PloS One 2015;10:
e0137077.
72. Drury C, Dale KE, Panlilio JM, et al. Genomic variation among
populations of threatened coral: acropora cervicornis. BMC
Genomics 2016, in press.

73. Milano I, Babbucci M, Cariani A, et al. Outlier SNP markers
reveal fine-scale genetic structuring across European
hake populations (Merluccius merluccius). Mol Ecol
2014;23:118–35.
74. Beaumont MA, Balding DJ. Identifying adaptive genetic diver-
gence among populations from genome scans. Mol Ecol
2004;13:969–80.
75. Lotterhos KE, Schaal SM. Genome scans for the contemporary
response to selection in quantitative traits. Mol Ecol 2014;23:
4435–7.
76. Holsinger KE, Weir BS. Genetics in geographically structured
populations: defining, estimating and interpreting FST. Nat
Rev Genet 2009;10:639–50.
77. Weir BS, Cockerham CC. Estimating F-statistics for the ana-
lysis of population-structure. Evolution 1984;38:1358–70.
78. Lotterhos KE, Whitlock MC. Evaluation of demographic his-
tory and neutral parameterization on the performance of FST
outlier tests. Mol Ecol 2014;23:2178–92.
79. Whitlock MC, Lotterhos KE. Editor: Judith LB. Reliable
Detection of Loci Responsible for Local Adaptation: Inference
of a Null Model through Trimming the Distribution of FST. Am
Nat 2015;186:S24–36.
80. Apodaca JJ, Trexler JC, Jue NK, et al. Large-scale natural dis-
turbance alters genetic population structure of the sailfin
molly, Poecilia latipinna. Am Nat 2013;181:254–63.
81. Trexler JC, Travis J, Dinep A. Variation among populations of
the sailfin molly in the rate of concurrent multiple paternity
and its implications for mating-system evolution. Behav Ecol
Sociobiol 1997;40:297–305.
82. Clark AG, Hubisz MJ, Bustamante CD, et al. Ascertainment
bias in studies of human genome-wide polymorphism.
Genome Res 2005;15:1496–502.
83. Bradbury PJ, Zhang Z, Kroon DE, et al. TASSEL: software for as-
sociation mapping of complex traits in diverse samples.
Bioinformatics 2007;23:2633–5.
84. Glaubitz JC, Casstevens TM, Lu F, et al. TASSEL-GBS: a high
capacity genotyping by sequencing analysis pipeline. PloS
One 2014;9:e90346.
85. Langmead B, Salzberg S. Fast gapped-read alignment with
Bowtie 2. Nat Methods 2012;9:357–9.
Ecological population genomics | 351
86. Li H, Durbin R. Fast and accurate short read alignment
with Burrows–Wheeler transform. Bioinformatics 2009;25:
1754–60.
87. Nielsen R, Mattila DK, Clapham PJ, et al. Statistical approaches
to paternity analysis in natural populations and applications
to the North Atlantic humpback whale. Genetics 2001;157:
1673–82.
Downloaded from https://academic.oup.com/bfg/article-abstract/15/5/342/1742085 by Universidade de São Paulo - IO user on 17 December 2019
88. O’Rawe J, Jiang T, Sun G, et al. Low concordance of multiple
variant-calling pipelines: practical implications for exome
and genome sequencing. Genome Med 2013;5:1–18.
89. Bayer T, Aranda M, Sunagawa S, et al. Symbiodinium tran-
scriptomes: genome insights into the dinoflagellate sym-
bionts of reef-building corals. PloS One 2012;7:e35269.
90. Gautier M, Foucaud J, Gharbi K, et al. Estimation of population
allele frequencies from next-generation sequencing data:
pool-versus individual-based genotyping. Mol Ecol 2013;22:
3766–79.
91. Lynch M. Estimation of allele frequencies from high-
coverage genome-sequencing projects. Genetics 2009;182:
295–301.
92. Maruki T, Lynch M. Genotype-frequency estimation from
high-throughput sequencing data. Genetics 2015;201:473–86.
93. Hartl DL, Clark AG. Principles of Population Genetics.
Sunderland, MA: Sinauer Associates, 2007.
94. Excoffier L, Laval G, Schneider S. Arlequin (version 3.0): an
integrated software package for population genetics data
analysis. Evol Bioinform 2005;1:47–50.
95. Nelson MR, Wegmann D, Ehm MG, et al. An abundance of rare
functional variants in 202 drug target genes sequenced in
14,002 people. Science 2012;337:100–4.
96. Waples RS, Do C. Linkage disequilibrium estimates of con-
temporary N e using highly variable genetic markers: a
largely untapped resource for applied conservation and evo-
lution. Evol Appl 2010;3:244–62.
97. Pritchard JK, Pickrell JK, Coop G. The genetics of human adap-
tation: hard sweeps, soft sweeps, and polygenic adaptation.
Curr Biol 2010;20:R208–15.
98. Andrews KR, Hohenlohe PA, Miller MR, et al. Trade-offs and
utility of alternative RADseq methods: reply to Puritz et al.
Mol Ecol 2014;23:5943–6.