Breaking RAD: an evaluation of the utility of restriction site‐associated DNA sequencing for genome scans of adaptation

Understanding how and why populations evolve is of fundamental importance to molecular ecology. Restriction site‐associated DNA sequencing (RADseq), a popular reduced representation method, has ushered in a new era of genome‐scale research for assessing population structure, hybridization, demographic history, phylogeography and migration. RADseq has also been widely used to conduct genome scans to detect loci involved in adaptive divergence among natural populations. Here, we examine the capacity of those RADseq‐based genome scan studies to detect loci involved in local adaptation. To understand what proportion of the genome is missed by RADseq studies, we developed a simple model using different numbers of RAD‐tags, genome sizes and extents of linkage disequilibrium (length of haplotype blocks). Under the best‐case modelling scenario, we found that RADseq using six‐ or eight‐base pair cutting restriction enzymes would fail to sample many regions of the genome, especially for species with short linkage disequilibrium. We then surveyed recent studies that have used RADseq for genome scans and found that the median density of markers across these studies was 4.08 RAD‐tag markers per megabase (one marker per 245 kb). The length of linkage disequilibrium for many species is one to three orders of magnitude less than density of the typical recent RADseq study. Thus, we conclude that genome scans based on RADseq data alone, while useful for studies of neutral genetic variation and genetic population structure, will likely miss many loci under selection in studies of local adaptation.     

Unbroken: RADseq remains a powerful tool for understanding the genetics of adaptation in natural populations

Recently, Lowry et al. addressed the ability of RADseq approaches to detect loci under selection in genome scans. While the authors raise important considerations, such as accounting for the extent of linkage disequilibrium in a study system, we strongly disagree with their overall view of the ability of RADseq to inform our understanding of the genetic basis of adaptation. The family of RADseq protocols has radically improved the field of population genomics, expanding by several orders of magnitude the number of markers available while substantially reducing the cost per marker. Researchers whose goal is to identify regions of the genome under selection must consider the LD of the experimental system; however, there is no magical LD cutoff below which researchers should refuse to use RADseq. Lowry et al. further made two major arguments: a theoretical argument that modeled the likelihood of detecting selective sweeps with RAD markers, and gross summaries based on an anecdotal collection of RAD studies. Unfortunately, their simulations were off by two orders of magnitude in the worst case, while their anecdotes merely showed that it is possible to get widely divergent densities of RAD tags for any particular experiment, either by design or due to experimental efficacy. We strongly argue that RADseq remains a powerful and efficient approach that provides sufficient marker density for studying selection in many natural populations. Given limited resources, we argue that researchers should consider a wide range of trade‐offs among genomic techniques, in light of their study question and the power of different techniques to answer it.     

Demystifying the RAD fad

We are writing in response to the population and phylogenomics meeting review by Andrews & Luikart ( 2014 ) entitled ‘Recent novel approaches for population genomics data analysis’. Restriction‐site‐associated DNA (RAD) sequencing has become a powerful and useful approach in molecular ecology, with several different published methods now available to molecular ecologists, none of which can be considered the best option in all situations. A&L report that the original RAD protocol of Miller et al. ( 2007 ) and Baird et al. ( 2008 ) is superior to all other RAD variants because putative PCR duplicates can be identified (see Baxter et al. 2011 ), thereby reducing the impact of PCR artefacts on allele frequency estimates (Andrews & Luikart 2014 ). In response, we (i) challenge the assertion that the original RAD protocol minimizes the impact of PCR artefacts relative to that of other RAD protocols, (ii) present additional biases in RADseq that are at least as important as PCR artefacts in selecting a RAD protocol and (iii) highlight the strengths and weaknesses of four different approaches to RADseq which are a representative sample of all RAD variants: the original RAD protocol (mbRAD, Miller et al. 2007 ; Baird et al. 2008 ), double digest RAD (ddRAD, Peterson et al. 2012 ), ezRAD (Toonen et al. 2013 ) and 2bRAD (Wang et al. 2012 ). With an understanding of the strengths and weaknesses of different RAD protocols, researchers can make a more informed decision when selecting a RAD protocol.     

Using RAD‐seq to recognize sex‐specific markers and sex chromosome systems

Next‐generation sequencing methods have initiated a revolution in molecular ecology and evolution (Tautz et al. 2010 ). Among the most impressive of these sequencing innovations is restriction site‐associated DNA sequencing or RAD‐seq (Baird et al. 2008 ; Andrews et al. 2016 ). RAD‐seq uses the Illumina sequencing platform to sequence fragments of DNA cut by a specific restriction enzyme and can generate tens of thousands of molecular genetic markers for analysis. One of the many uses of RAD‐seq data has been to identify sex‐specific genetic markers, markers found in one sex but not the other (Baxter et al. 2011 ; Gamble & Zarkower 2014 ). Sex‐specific markers are a powerful tool for biologists. At their most basic, they can be used to identify the sex of an individual via PCR. This is useful in cases where a species lacks obvious sexual dimorphism at some or all life history stages. For example, such tests have been important for studying sex differences in life history (Sheldon 1998 ; Mossman & Waser 1999 ), the management and breeding of endangered species (Taberlet et al. 1993 ; Griffiths & Tiwari 1995 ; Robertson et al. 2006 ) and sexing embryonic material (Hacker et al. 1995 ; Smith et al. 1999 ). Furthermore, sex‐specific markers allow recognition of the sex chromosome system in cases where standard cytogenetic methods fail (Charlesworth & Mank 2010 ; Gamble & Zarkower 2014 ). Thus, species with male‐specific markers have male heterogamety (XY) while species with female‐specific markers have female heterogamety (ZW). In this issue, Fowler & Buonaccorsi ( 2016 ) illustrate the ease by which RAD‐seq data can generate sex‐specific genetic markers in rockfish (Sebastes). Moreover, by examining RAD‐seq data from two closely related rockfish species, Sebastes chrysomelas and Sebastes carnatus (Fig.  1 ), Fowler & Buonaccorsi ( 2016 ) uncover shared sex‐specific markers and a conserved sex chromosome system.     

Don't throw the baby out with the bathwater: identifying and mapping paralogs in salmonids

Many eukaryotic genomes contain a large fraction of gene duplicates (or paralogs) as a result of ancient or recent whole‐genome duplications (Ohno 1970 ; Jaillon et al. 2004 ; Kellis et al. 2004 ). Identifying paralogs with NGS data is a pervasive problem in both ancient polyploids and neopolyploids. Likewise, paralogs are often treated as a nuisance that has to be detected and removed (Everett et al. 2012 ). In this issue of Molecular Ecology Resources, Waples et al. ( 2015 ) show that exclusion might not be necessary and how we may miss out on important genomic information in doing so. They present a novel statistical approach to detect paralogs based on the segregation of RAD loci in haploid offspring and test their method by constructing linkage maps with and without these duplicated loci in chum salmon, Oncorhynchus keta (Fig.  1 ). Their linkage map including the resolved paralogs shows that these are mostly located in the distal regions of several linkage groups. Particularly intriguing is their finding that these homoeologous regions appear impoverished in transposable elements (TE). Given the role that TE play in genome remodelling, it is noteworthy that these elements are of low abundance in regions showing residual tetrasomic inheritance. This raises the question whether re‐diploidization is constrained in these regions and whether they might have a role to play in salmonid speciation. This study provides an original approach to identifying duplicated loci in species with a pedigree, as well as providing a dense linkage map for chum salmon, and interesting insights into the retention of gene duplicates in an ancient polyploid.     

Trade‐offs and utility of alternative RADseq methods: Reply to Puritz et al.

Puritz et al. provide a review of several RADseq methodological approaches in response to our ‘Population Genomic Data Analysis’ workshop (Sept 2013) review (Andrews & Luikart 2014). We agree with Puritz et al. on the importance for researchers to thoroughly understand RADseq library preparation and data analysis when choosing an approach for answering their research questions. Some of us are currently using multiple RADseq protocols, and we agree that the different methods may offer advantages in different cases. Our workshop review did not intend to provide a thorough review of RADseq because the workshop covered a broad range of topics within the field of population genomics. Similarly, neither the response of Puritz et al. nor our comments here provide sufficient space to thoroughly review RADseq. Nonetheless, here we address some key points that we find unclear or potentially misleading in their evaluation of techniques.     

RADseq approaches and applications for forest tree genetics

As tree species vary extensively in genome size, complexity, and resource development, reduced representation methods have been increasingly employed for the generation of population genomic data. By allowing rapid marker discovery and genotyping for thousands of genomic regions in many individuals without requiring genomic resources, restriction site-associated DNA sequencing (RADseq) methods have dramatically improved our ability to bring population genomic perspectives to non-model trees. The rapid recent increase in studies of trees utilizing RADseq suggests that it is likely to become among the most common approaches for generating genome-wide data for a variety of applications. Here we provide a practical review of RADseq and its application to research areas of tree genetics. We briefly review RADseq laboratory methods and consider analytical approaches for assembly, variant calling, and bioinformatic processing. To guide considerations for study design, we use in silico analyses of eight available tree genomes to illustrate how expected marker number and density vary across laboratory approaches and genome sizes, and to consider the ability of RADseq designs to query coding regions. We review the empirical use of RADseq for different research objectives, considering its strengths and limitations. Many studies have used RADseq data to perform genome scans for selection, although limited marker density and linkage disequilibrium will often compromise its utility for such analyses. Regardless of this limitation, RADseq offers a powerful and inexpensive technique for generating genome-wide SNP data that can greatly contribute to research spanning phylogenetic and population genetic inference, linkage mapping, and quantitative genetic parameter estimation for tree genetics.     

Small‐ and large‐scale heterogeneity in genetic variation across the collard flycatcher genome: implications for estimating genetic diversity in nonmodel organisms

Population genetic studies in nonmodel organisms are often hampered by a lack of reference genomes that are essential for whole‐genome resequencing. In the light of this, genotyping methods have been developed to effectively eliminate the need for a reference genome, such as genotyping by sequencing or restriction site‐associated DNA sequencing (RAD‐seq). However, what remains relatively poorly studied is how accurately these methods capture both average and variation in genetic diversity across an organism's genome. In this issue of Molecular Ecology Resources, Dutoit et al. (2016) use whole‐genome resequencing data from the collard flycatcher to assess what factors drive heterogeneity in nucleotide diversity across the genome. Using these data, they then simulate how well different sequencing designs, including RAD sequencing, could capture most of the variation in genetic diversity. They conclude that for evolutionary and conservation‐related studies focused on the estimating genomic diversity, researchers should emphasize the number of loci analysed over the number of individuals sequenced.     

The discovery of Antarctic RNA viruses: a new game changer

Antarctic ecosystems are dominated by micro‐organisms, and viruses play particularly important roles in the food webs. Since the first report in 2009 (López‐Bueno et al. 2009 ), ‘omic’‐based studies have greatly enlightened our understanding of Antarctic aquatic microbial diversity and ecosystem function (Wilkins et al. 2013 ; Cavicchioli 2015 ). This has included the discovery of many new eukaryotic viruses (López‐Bueno et al. 2009 ), virophage predators of algal viruses (Yau et al. 2011 ), bacteria with resistance to phage (Lauro et al. 2011 ) and mechanisms of haloarchaeal evasion, defence and adaptation to viruses (Tschitschko et al. 2015 ). In this issue of Molecular Ecology, López‐Bueno et al. ( 2015 ) report the first discovery of RNA viruses from an Antarctic aquatic environment. High sequence coverage enabled genome variation to be assessed for four positive‐sense single‐stranded RNA viruses from the order Picornavirales. By examining the populations present in the water column and in the lake's catchment area, populations of ‘quasispecies’ were able to be linked to local environmental factors. In view of the importance of viruses in Antarctic ecosystems but lack of data describing them, this study represents a significant advance in the field.     

Homomorphic plant sex chromosomes are coming of age

Sex chromosomes are a very peculiar part of the genome that have evolved independently in many groups of animals and plants (Bull 1983 ). Major research efforts have so far been focused on large heteromorphic sex chromosomes in a few animal and plant species (Chibalina & Filatov 2011 ; Zhou & Bachtrog 2012 ; Bellott et al. 2014 ; Hough et al. 2014 ; Zhou et al. 2014 ), while homomorphic (cytologically indistinguishable) sex chromosomes have largely been neglected. However, this situation is starting to change. In this issue, Geraldes et al. ( 2015 ) describe a small (~100 kb long) sex‐determining region on the homomorphic sex chromosomes of poplars (Populus trichocarpa and related species, Fig.  1 ). All species in Populus and its sister genus Salix are dioecious, suggesting that dioecy and the sex chromosomes, if any, should be relatively old. Contrary to this expectation, Geraldes et al. ( 2015 ) demonstrate that the sex‐determining region in poplars is of very recent origin and probably evolved within the genus Populus only a few million years ago.     

A hot topic: the genetics of adaptation to geothermal vents in Mimulus guttatus

Identifying the individual loci and mutations that underlie adaptation to extreme environments has long been a goal of evolutionary biology. However, finding the genes that underlie adaptive traits is difficult for several reasons. First, because many traits and genes evolve simultaneously as populations diverge, it is difficult to disentangle adaptation from neutral demographic processes. Second, finding the individual loci involved in any trait is challenging given the respective limitations of quantitative and population genetic methods. In this issue of Molecular Ecology, Hendrick et al. (2016) overcome these difficulties and determine the genetic basis of microgeographic adaptation between geothermal vent and nonthermal populations of Mimulus guttatus in Yellowstone National Park. The authors accomplish this by combining population and quantitative genetic techniques, a powerful, but labour‐intensive, strategy for identifying individual causative adaptive loci that few studies have used (Stinchcombe & Hoekstra 2008 ). In a previous common garden experiment (Lekberg et al. 2012), thermal M. guttatus populations were found to differ from their closely related nonthermal neighbours in various adaptive phenotypes including trichome density. Hendrick et al. (2016) combine quantitative trait loci (QTL) mapping, population genomic scans for selection and admixture mapping to identify a single genetic locus underlying differences in trichome density between thermal and nonthermal M. guttatus. The candidate gene, R2R3 MYB, is homologous to genes involved in trichome development across flowering plants. The major trichome QTL, Tr14, is also involved in trichome density differences in an independent M. guttatus population comparison (Holeski et al. 2010) making this an example of parallel genetic evolution.     

Mock communities highlight the diversity of host‐associated eukaryotes

Host‐associated microbes are ubiquitous. Every multicellular eukaryote, and even many unicellular eukaryotes (protists), hosts a diverse community of microbes. High‐throughput sequencing (HTS) tools have illuminated the vast diversity of host‐associated microbes and shown that they have widespread influence on host biology, ecology and evolution (McFall‐Ngai et al. 2013 ). Bacteria receive most of the attention, but protists are also important components of microbial communities associated with humans (Parfrey et al. 2011 ) and other hosts. As HTS tools are increasingly used to study eukaryotes, the presence of numerous and diverse host‐associated eukaryotes is emerging as a common theme across ecosystems. Indeed, HTS studies demonstrate that host‐associated lineages account for between 2 and 12% of overall eukaryotic sequences detected in soil, marine and freshwater data sets, with much higher relative abundances observed in some samples (Ramirez et al. 2014 ; Simon et al. 2015 ; de Vargas et al. 2015 ). Previous studies in soil detected large numbers of predominantly parasitic lineages such as Apicomplexa, but did not delve into their origin [e.g. (Ramirez et al. 2014 )]. In this issue of Molecular Ecology, Geisen et al. ( 2015 ) use mock communities to show that many of the eukaryotic organisms detected by environmental sequencing in soils are potentially associated with animal hosts rather than free‐living. By isolating the host‐associated fraction of soil microbial communities, Geisen and colleagues help explain the surprisingly high diversity of parasitic eukaryotic lineages often detected in soil/terrestrial studies using high‐throughput sequencing (HTS) and reinforce the ubiquity of these host‐associated microbes. It is clear that we can no longer assume that organisms detected in bulk environmental sequencing are free‐living, but instead need to design studies that specifically enumerate the diversity and function of host‐associated eukaryotes. Doing so will allow the field to determine the role host‐associated eukaryotes play in soils and other environments and to evaluate hypotheses on assembly of host‐associated communities, disease ecology and more.     

Bacteriophage exclusion,a new defense system

The ability to withstand viral predation is critical for survival of most microbes. Accordingly, a plethora of phage resistance systems has been identified in bacterial genomes (Labrie et al, 2010 ), including restriction‐modification systems (R‐M) (Tock & Dryden, 2005 ), abortive infection (Abi) (Chopin et al, 2005 ), Argonaute‐based interference (Swarts et al, 2014 ), as well as clustered regularly interspaced short palindromic repeats (CRISPR) and associated protein (Cas) adaptive immune system (CRISPR‐Cas) (Barrangou & Marraffini, 2014 ; Van der Oost et al, 2014 ). Predictably, the dark matter of bacterial genomes contains a wealth of genetic gold. A study published in this issue of The EMBO Journal by Goldfarb et al ( 2015 ) unveils bacteriophage exclusion (BREX) as a novel, widespread bacteriophage resistance system that provides innate immunity against virulent and temperate phage in bacteria.     

Will reduced host connectivity curb the spread of a devastating epidemic?

The white‐nose syndrome (WNS), caused by the fungal pathogen Pseudogymnoascus destructans, is threatening the cave‐dwelling bat fauna of North America by killing individuals by the thousands in hibernacula each winter since its appearance in New York State less than ten years ago. Epidemiological models predict that WNS will reach the western coast of the USA by 2035, potentially eliminating most populations of susceptible bat species in its path (Frick et al. 2015; O'Regan et al. 2015). These models were built and validated using distributional data from the early years of the epidemic, which spread throughout eastern North America following a route driven by cave density and winter severity (Maher et al. 2012). In this issue of Molecular Ecology, Wilder et al. (2015) refine these findings by showing that connectivity among host populations, as assessed by population genetic markers, is crucial in determining the spread of the pathogen. Because host connectivity is much reduced in the hitherto disease free western half of North America, Wilder et al. make the reassuring prediction that the disease will spread more slowly west of the Great Plains.     

The genomics of adaptation,divergence and speciation: a congealing theory

In this issue, Flaxman et al. ( 2014 ) report the results of sophisticated whole‐genome simulations of speciation with gene flow, enhancing our understanding of the process by building on previous single‐locus, multilocus and analytical works. Their findings provide us with new insights about how genomes can diverge and the importance of statistical and chromosomal linkage in facilitating reproductive isolation. The authors characterize the conditions under which, even with high gene flow and weak divergent selection, reproductive isolation between populations can occur due to the emergent stochastic process of genomewide congealing, where numerous statistically or physically linked loci of small effect allow selection to limit effective migration rates. The initial congealing event can occur within a broad range conditions, and once initiated, the self‐reinforcing process leads to rapid divergence and ultimately two reproductively isolated populations. Flaxman et al.'s ( 2014 ) work is a valuable contribution to our understanding of speciation with gene flow and in making a more predictive field of evolutionary genomics and speciation.     

Host plant richness explains diversity of ectomycorrhizal fungi: Response to the comment of Tedersoo et al. (2014)

Exploring the relationships between the biodiversity of groups of interacting organisms yields insight into ecosystem stability and function (Hooper et al. 2000 ; Wardle 2006 ). We demonstrated positive relationships between host plant richness and ectomycorrhizal (EM) fungal diversity both in a field study in subtropical China (Gutianshan) and in a meta‐analysis of temperate and tropical studies (Gao et al. 2013 ). However, based on re‐evaluation of our data sets, Tedersoo et al. ( 2014 ) argue that the observed positive correlation between EM fungal richness and EM plant richness at Gutianshan and also in our metastudies was based mainly from (i) a sampling design with inconsistent species pool and (ii) poor data compilation for the meta‐analysis. Accordingly, we checked our data sets and repeated the analysis performed by Tedersoo et al. ( 2014 ). In contrast to Tedersoo et al. ( 2014 ), our re‐analysis still confirms a positive effect of plant richness on EM fungal diversity in Gutianshan, temperate and tropical ecosystems, respectively.     

Enalikter is not an annelid: homology,autapomorphies and the interpretation of problematic fossils

A megacheiran arthropod, Enalikter aphson, was recently described by Siveter et al. (2014) from the mid‐Silurian (late Wenlock) of Herefordshire. Previously, megacheirans had only been recognized from the Cambrian. Struck et al. (2015) considered the body plan of Enalikter to be incompatible with this affinity, arguing that many of the arthropod features were either not present or misinterpreted. Instead, they compared Enalikter to polychaete annelids, identifying characters from numerous polychaete lineages which they considered to be present in Enalikter. A reply to this critique by Siveter et al. (2015) reaffirmed arthropod affinities for Enalikter by presenting additional evidence for key arthropod features, such as arthropodized appendages. Here, we augment Siveter et al. by critically addressing the putative annelid characters of Enalikter presented by Struck et al. and additionally explore the morphological and phylogenetic implications of their hypothesis. We conclude that similarities between Enalikter and polychaetes are superficial and that character combinations proposed by Struck et al. are not present in any annelid, living or extinct. This taxon highlights the importance of using a phylogenetic framework for interpreting fossils that present unusual morphologies, such that proposed shared characters are hypotheses of homology rather than merely phenotypic similarities. Crucially, we argue that autapomorphic characters of subgroups of large taxa (like families or classes within phyla) should not be used to diagnose problematic fossils.     

Identifying homomorphic sex chromosomes from wild‐caught adults with limited genomic resources

We demonstrate a genotyping‐by‐sequencing approach to identify homomorphic sex chromosomes and their homolog in a distantly related reference genome, based on noninvasive sampling of wild‐caught individuals, in the moor frog Rana arvalis. Double‐digest RADseq libraries were generated using buccal swabs from 30 males and 21 females from the same population. Search for sex‐limited markers from the unfiltered data set (411 446 RAD tags) was more successful than searches from a filtered data set (33 073 RAD tags) for markers showing sex differences in heterozygosity or in allele frequencies. Altogether, we obtained 292 putatively sex‐linked RAD loci, 98% of which point to male heterogamety. We could map 15 of them to the Xenopus tropicalis genome, all but one on chromosome pair 1, which seems regularly co‐opted for sex determination among amphibians. The most efficient mapping strategy was a three‐step hierarchical approach, where R. arvalis reads were first mapped to a low‐coverage genome of Rana temporaria (17 My divergence), then the R. temporaria scaffolds to the Nanorana parkeri genome (90 My divergence), and finally the N. parkeri scaffolds to the X. tropicalis genome (210 My). We validated our conclusions with PCR primers amplifying part of Dmrt1, a candidate sex determination gene mapping to chromosome 1: a sex‐diagnostic allele was present in all 30 males but in none of the 21 females. Our approach is likely to be productive in many situations where biological samples and/or genomic resources are limited.     

Two are better than one: combining landscape genomics and common gardens for detecting local adaptation in forest trees

Predicting likely species responses to an alteration of their local environment is key to decision‐making in resource management, ecosystem restoration and biodiversity conservation practice in the face of global human‐induced habitat disturbance. This is especially true for forest trees which are a dominant life form on Earth and play a central role in supporting diverse communities and structuring a wide range of ecosystems. In Europe, it is expected that most forest tree species will not be able to migrate North fast enough to follow the estimated temperature isocline shift given current predictions for rapid climate warming. In this context, a topical question for forest genetics research is to quantify the ability for tree species to adapt locally to strongly altered environmental conditions (Kremer et al. 2012 ). Identifying environmental factors driving local adaptation is, however, a major challenge for evolutionary biology and ecology in general but is particularly difficult in trees given their large individual and population size and long generation time. Empirical evaluation of local adaptation in trees has traditionally relied on fastidious long‐term common garden experiments (provenance trials) now supplemented by reference genome sequence analysis for a handful of economically valuable species. However, such resources have been lacking for most tree species despite their ecological importance in supporting whole ecosystems. In this issue of Molecular Ecology, De Kort et al. ( 2014 ) provide original and convincing empirical evidence of local adaptation to temperature in black alder, Alnus glutinosa L. Gaertn, a surprisingly understudied keystone species supporting riparian ecosystems. Here, De Kort et al. ( 2014 ) use an innovative empirical approach complementing state‐of‐the‐art landscape genomics analysis of A. glutinosa populations sampled in natura across a regional climate gradient with phenotypic trait assessment in a common garden experiment (Fig. 1 ). By combining the two methods, De Kort et al. ( 2014 ) were able to detect unequivocal association between temperature and phenotypic traits such as leaf size as well as with genetic loci putatively under divergent selection for temperature. The research by De Kort et al. ( 2014 ) provides valuable insight into adaptive response to temperature variation for an ecologically important species and demonstrates the usefulness of an integrated approach for empirical evaluation of local adaptation in nonmodel species (Sork et al. 2013 ).     

Coalescing molecular evolution and DNA barcoding

The DNA barcoding concept (Woese et al. 1990 ; Hebert et al. 2003 ) has considerably boosted taxonomy research by facilitating the identification of specimens and discovery of new species. Used alone or in combination with DNA metabarcoding on environmental samples (Taberlet et al. 2012 ), the approach is becoming a standard for basic and applied research in ecology, evolution and conservation across taxa, communities and ecosystems (Scheffers et al. 2012 ; Kress et al. 2015 ). However, DNA barcoding suffers from several shortcomings that still remain overlooked, especially when it comes to species delineation (Collins & Cruickshank 2012 ). In this issue of Molecular Ecology, Barley & Thomson ( 2016 ) demonstrate that the choice of models of sequence evolution has substantial impacts on inferred genetic distances, with a propensity of the widely used Kimura 2‐parameter model to lead to underestimated species richness. While DNA barcoding has been and will continue to be a powerful tool for specimen identification and preliminary taxonomic sorting, this work calls for a systematic assessment of substitution models fit on barcoding data used for species delineation and reopens the debate on the limitation of this approach.