While there is a long history of controversy over what species are and how they should be delimited, the General Lineage Concept of species (de Queiroz 1998, 2007), defining species as the smallest metapopulation lineages on independent evolutionary trajectories, encapsulates what most systematists would regard as the essence of what a species is. As noted by de Queiroz (1998, 2007), most prior disputes about “species concepts” disagreed more about the criteria by which such lineages might be diagnosed (e.g., reproductive isolation vs. diagnosable character states) than about the actual concept of what a species is.

Hennig (1950, 1966) defined the point of speciation as the point where tokogenetic relationships, where individuals mate and their genes recombine within a lineage, are replaced by phylogenetic lineages, where there is no such exchange or recombination between lineages on an evolutionary timescale, and the lineages show bifurcating, hierarchical relationships amongst themselves. Within this framework, it becomes the central task of species delimitation to identify the point where tokogenetic patterns of genetic exchange cease and lineages become independent. Where potential species are in geographical contact, more or less complete reproductive isolation remains a key test of whether two lineages are indeed on independent evolutionary trajectories, and thus a fundamental criterion for species delimitation (Hillis 2019; Hillis et al. 2021; Jarrett et al. 2025), even though newer sequencing technologies have highlighted the persistence of varying degrees of introgression and hybridisation even among well-defined separate species (Schield et al. 2019; Edelman and Mallet 2021; Aguillon et al. 2022; Schöneberg et al. 2023; Myers et al. 2024).

Until the second half of the 20^th century, morphology and other phenotypic traits (e.g., behaviour, physiology), as well as crossbreeding experiments, were the only sources of evidence available to systematists to support their inferences. The molecular revolution in systematics gained momentum with increasing ease of access to DNA sequence information. In particular, mitochondrial DNA (mtDNA) provided a relatively accessible route to genetic information, which moreover evolves at a rate that makes it suitable for systematics at low taxonomic levels (Avise et al. 1979). Sequencing mtDNA fragments became the default approach following the development of the Polymerase Chain Reaction, the use of thermostable primers (Mullis and Faloona 1987; Saiki et al. 1988), and direct sequencing of PCR products using conserved priming sites (Kocher et al. 1989). Since then, despite multiple technological and conceptual developments, mitochondrial DNA has retained a perhaps disproportionate importance in herpetological systematics and species delimitation.

The properties that first made mitochondrial DNA popular in systematics include its high rate of evolution, making it useful at low taxonomic ranks, including for intraspecific phylogeography (Avise et al. 1987), and its clonal, matrilineal mode of inheritance. The lack of recombination means that the entire mtDNA molecule is inherited as a single locus, allowing multiple gene sequences to be concatenated into larger datasets to provide greater support for the mitochondrial gene tree, although care must be taken to avoid the accidental inclusion of nuclear integrations (Numts – Zhang and Hewitt 1996). The smaller effective population size of mitochondrial DNA (1/4 that of nuclear autosomal genes) results in more rapid coalescence times. As a result, other things being equal, a mitochondrial gene tree is more likely to track organismal phylogeny than any single autosomal gene fragment (Moore 1995). Moreover, as the gene content of the mtDNA molecule is largely conserved across all Metazoa, orthologous gene sequences can be obtained from a phylogenetically broad spectrum of animal life. Due to these unique properties and the insights they provide, mtDNA remains a key component of systematic research in animals (Blair 2023).

The role of mitochondrial DNA in species delimitation was further emphasised by the advent of the barcoding initiative (Hebert et al. 2003a, 2003b), whose central aim is to establish a standard mitochondrial marker for both specimen identification and biodiversity discovery, including species delimitation (Hebert et al. 2004b). The formulation of the concept of a “barcoding gap” (Hebert et al. 2003b) separating intraspecific from interspecific divergences (Hebert et al. 2004a) triggered the development of multiple approaches to species delimitation from single locus (usually mtDNA) data. Some of these methods are based on the detection of a barcode gap (e.g., ABGD, ASAP; Puillandre et al. 2012a, 2021), whereas others are tree based (e.g., GMYC, PTP; Pons et al. 2006; Zhang et al. 2013). The use of barcoding for species delimitation was taken to extremes in entomology, where some researchers described hundreds of species based on nothing but a consensus barcode sequence and a photograph (e.g., Sharkey et al. 2021), causing extensive controversy (Ahrens et al. 2021; Ahrens 2024). This extreme approach has so far not been extended to vertebrate taxonomy.

While its properties, mode of inheritance and rapid rate of sequence evolution make mtDNA logistically and economically attractive, these very features are responsible for the many limitations that make it unsuitable as a sole source of evidence in species delimitation (Miralles et al. 2024). Its solely matrilineal mode of inheritance, contrasting with the biparental inheritance of the nuclear genome, precludes the detection of admixture between lineages from mtDNA alone (Fig. 1) and can generate conflict between mitochondrial phylogeographic structure and genomic (nuclear) genetic structure in multiple ways. This phenomenon, variously referred to as mitonuclear or cytonuclear discordance, can potentially cause mitochondrial phylogeography to be grossly incongruent with patterns of genomic population structure. In the following paragraphs, I review the mechanisms underlying mitonuclear discordance in a number of herpetological examples.

Figure 1. 

Schematic illustration of the impact of the different modes of inheritance of nuclear and mitochondrial DNA on the ability to detect admixture between a hypothetical red and a blue species. A Nuclear DNA is inherited biparentally and reflects admixture in hybridization. In this case, the nuclear markers of the F2 offspring will reflect a genome originating to 75% from the red species. B Mitochondrial DNA is inherited matrilineally only. Therefore, the information on lines of descent involving male ancestors (indicated by dashed outlines) is lost from the mitogenome of the offspring. The mtDNA of the F2 offspring will reflect the matrilineal ancestry of the blue species only, without any indication of the 75% admixture from the red species.

Do phylogeographic breaks necessarily reflect population genomic breaks? A common assumption in phylogeographic and mtDNA-led systematic studies is that phylogeographic breaks correspond to past or present barriers to gene flow and reflect discontinuities in the genomic make-up of the phylogroups. However, this need not be the case. Irwin (2002) demonstrated that phylogeographic structure can evolve in situ in a continuously distributed taxon, without past or present range fragmentation. Contrasting population structures suggested by mitochondrial and nuclear markers have long been documented (Palumbi and Baker 1994): Female stasipatry combined with male vagility can result in strong mitochondrial phylogeographic structure within a single gene pool that misleadingly suggests population differentiation or speciation, potentially resulting in gross overestimates of species numbers (Eberle et al. 2019; Schultze et al. 2020).

Inability to assess introgression and hybridisation. Since the entire mtDNA molecule is (usually) inherited matrilineally and clonally, it follows that it cannot show admixture. Even in the presence of rampant hybridisation, each hybrid will carry the mtDNA haplotype inherited from its mother and her mother, etc. (Fig. 1). Consequently, a mtDNA phylogeny of two hybridising species will identify each specimen by its matrilineal ancestry only, and the topology of the mitochondrial gene tree alone will give no indication of the occurrence of hybridisation (e.g., Zancolli et al. 2016). Where mitochondrial haplotype clades overlap geographically, the mtDNA phylogeny alone cannot reveal whether these haplogroups represent independently evolving, sympatric organismal lineages or relictual mitochondrial haplotype clades within a single gene pool, e.g., after secondary contact (Castoe et al. 2007; Schield et al. 2015).

Since detecting admixture is key to distinguishing between tokogenetic and phylogenetic processes, the central task of species delimitation, it follows that on this basis alone, mtDNA cannot be used as the sole source of evidence for species delimitation. The North American ratsnake (Pantherophis obsoletus) complex is a prominent example where parapatric contact zones inferred from mtDNA led to the delimitation of four separate species (Burbrink 2001), whereas later genomic analyses (Burbrink et al. 2021a) showed broad zones of admixture between three of the lineages. Similarly, whereas the main mitochondrial genetic break in western rattlesnakes (Crotalus oreganus) along the North American Pacific coast coincides with the Transverse Ranges of southern California (Schield et al. 2019; Pook et al. 2000), a recent genomic study (Holding et al. 2021) placed the main genetic break in the species near the Sacramento and San Joaquin River Delta on San Francisco Bay.

Selection pressure versus population history. Especially in the presence of male-biased gene flow, changes in selection pressures, such as across ecotones, may drive population genomic breaks reflecting major barriers to gene flow that are entirely independent of mitochondrial phylogroups that reflect past lineage fragmentation. In some cases, patterns of genomic differentiation can cut orthogonally across the distribution of mtDNA haplotype clades (Smith et al. 1997; Ogden and Thorpe 2002). In these situations, the mitochondrial phylogeographic structure is almost completely unrepresentative of the genomic and phenotypic differentiation of these populations, and thus taxonomically irrelevant.

“Lost” mitochondrial species identity. Mitochondrial introgression can lead to the loss of species-specific mitochondrial haplotype clades. For instance, the Carpathian newt (Lissotriton montandoni) is a well-supported and universally recognised species of newt. However, due to mitochondrial introgression from L. vulgaris, it “disappears” into the latter species in a mtDNA-only phylogeny (Babik et al. 2005), whereas there is little evidence of matching nuclear gene flow (Zieliński et al. 2013).

Mitochondrial “ghost” species. Matrilineally inherited haplotype clades can persist in gene pools despite extensive gene flow with other populations, even when most of the genome has been replaced by that of the introgressing species. Vipera walser was described from northwestern Italy by Ghielmi et al. (2016), based on mtDNA sequences associating these populations with the V. kaznakovii complex from the Caucasus, rather than with the widespread Vipera berus to which they had previously been assigned. However, Doniol-Valcroze et al. (2021) and Dufresnes et al. (2024) found that the taxon instead shares most genomic similarity with the northern Italian clade of V. berus (V. berus marasso), and Dufresnes et al. (2024) recognised it as a subspecies of V. berus, V. b. walser. The origin of the Caucasian mtDNA haplotype clade of these populations remains an intriguing biogeographical enigma, but has little or no bearing on their overall affinities. In a similar vein, based on a mitochondrial phylogeny, Pyron and Burbrink (2009a, 2009b) recognised the black and speckled kingsnakes, previously subspecies of Lampropeltis getula, as separate species, L. nigra and L. holbrooki. However, a later genomic study showed both to be part of a single organismal lineage together with L. getula sensu stricto, leading to their synonymy with the latter (Harrington and Burbrink 2022).

The problematic role of mtDNA in species delimitation. While single-locus species delimitation algorithms claim increased precision in inferring species limits from mitochondrial sequence data (e.g., Pons et al. 2006; Puillandre et al. 2012a; Zhang et al. 2013), it must be remembered that all rely on the assumption that the mitochondrial tree reflects organismal phylogeny: No degree of sophistication of these methods can control for or overcome the problem of cytonuclear discordance (Miralles et al. 2024).

In summary, while mitochondrial DNA sequences remain invaluable as tools to generate species hypotheses, and may well reflect aspects of population history such as past range fragmentation, numerous potential confounding factors preclude their use as the sole or dominant source of evidence for species delimitation. Critical testing of mtDNA–derived species hypotheses using independent marker sets, which can be either nuclear genetic markers or phenotypic characters, and analytical approaches capable of detecting admixture between putative taxa must be considered a fundamental and critical part of any species delimitation attempt (Padial et al. 2010; Miralles et al. 2024; Vences et al. 2024).

The formulation of the General Lineage Concept of species (de Queiroz 1998, 2007) emphasised the multiplicity of criteria that can be used to delimit species, and spawned the framework of integrative taxonomy. Authors such as Dayrat (2005), Will et al. (2005) and Padial et al. (2010) advocated for an integrative approach to taxonomy, whereby different sources of evidence are used to test species hypotheses formulated on the basis of one marker. In particular, Padial et al. (2010) formulated a workflow where mitochondrial haplotype clades are treated as testable hypotheses termed Candidate Species (Padial et al. 2010) or Primary Species Hypotheses (Puillandre et al. 2012b; Miralles et al. 2024), which can then be critically tested with additional evidence (Padial et al. 2010; Miralles et al. 2024). Where such additional lines of evidence support species status for a Candidate Species or Primary Species Hypothesis, these achieve the status of Confirmed Candidate Species (Padial et al. 2010) or Secondary Species Hypotheses (Puillandre et al. 2012b), which form the basis for taxonomic decisions

Although de Queiroz (2025) re-emphasised that evolutionary lineages should be treated as species without recourse to additional properties, such as reproductive isolation, most treatises on species delimitation continue to place considerable emphasis on the need to demonstrate a considerable degree of reproductive isolation (Padial et al. 2010; Miralles et al. 2024). Where gene flow between putative lineages is high, it becomes increasingly difficult to support the argument that there are indeed multiple lineages in existence. Consequently, where candidate species occupy contiguous ranges in parapatry or sympatry, understanding the extent of reproductive exchange between them remains a critical test of these hypotheses, even though there remains ample legitimate disagreement in the taxonomic treatment of incipient lineages with incomplete reproductive isolation (e.g., Burbrink et al. 2021b; Burbrink and Ruane 2021; Hillis et al. 2021; Thiele et al. 2021; Hillis 2022).

Mitochondrial DNA sequences will unquestionably continue to be an important source of data in animal systematics in the foreseeable future. However, the main role of mtDNA should be seen as the crucial and often underrated function of hypothesis generation rather than that of hypothesis testing, or, worse, as a sole source of evidence in species delimitation (Blair 2023). However, framing the role of mitochondrial DNA in this manner then begs the question of what approaches would be appropriate to provide the critical tests of mitochondrially defined candidate species, as recommended by recent works (Padial et al. 2010; Miralles et al. 2024).

The key requirement for any delimitation analysis is that it must be designed to critically test the status of mitochondrial candidate species as independent evolutionary lineages. A critical test must be able to explicitly reject such a primary species hypothesis, not just fail to support it. For example, failure to find statistically significant differences in a phenotypic trait would fail to support a primary species hypothesis, but not reject it; in contrast, if a trait displays a pattern of clinal variation that crosses the putative boundary between the candidate species, this would constitute explicit evidence favouring rejection of the species hypothesis.

I therefore distinguish between two types of approaches: (i) critical approaches that can test and actively reject species hypotheses; and (ii) confirmatory approaches that can at best fail to actively support a species hypothesis, but cannot refute it. Broadly, methods that simply treat each of the candidate species as pre-defined units of comparison cannot critically test them, and are therefore confirmatory. To illustrate the importance of the distinction between critical tests and confirmatory approaches, Figure 2 shows a hypothetical scenario where two mitochondrial candidate species are distributed parapatrically along a geographic and environmental gradient. Genomic data and morphology both show continuous clinal variation across the gradient, including across the boundary between the mitochondrial lineages. These patterns should lead to rejection of the two candidate species hypotheses, as the mitochondrial haplotype clades clearly exist within a single panmictic gene pool. Any species delimitation analysis intended to offer a critical test must therefore be capable of revealing these patterns. Any analysis that simply compares the two pre-defined candidate species as single units of comparison would mask the clinal pattern of variation, and lead to the misinterpretation of the data as confirming the two candidate species. Consequently, for contiguously distributed candidate species, such approaches are essentially tautological.

Figure 2. 

Hypothetical scenario of two mitochondrial candidate species distributed along a geographic and ecological gradient. A observed patterns: 1) parapatrically distributed mitochondrial candidate species CS1 and CS2; 2) clinal phenotypic variation along geographical gradient; 3) genomic signature of isolation by distance along geographical gradient as revealed by software such as STRUCTURE (Pritchard et al. 2000); 4) continuous distribution through ecological gradient. Patterns 2–4 reject the candidate species hypothesis derived from 1. B consequences of analyses treating the candidate species CS1 and CS2 as a priori units of comparison: 1) misleading impression of confirmation of candidate species from phenotypic differences; 2) misleading impression of confirmation of candidate species from ecological niche differences.

This scenario is unlikely to be of relevance in situations where mitochondrial candidate species are allopatric and restricted to small, isolated ranges, such as oceanic islands or isolated mountains, where continued genetic exchange is highly improbable. However, structure within candidate species and continuing genetic exchanges between them must be critically evaluated in any situation where the candidate species are contiguously distributed, or likely to have been recently, prior to Pleistocene climatic and sea level fluctuations.

Population genetic approaches to the analysis of nuclear loci. The critical analysis of nuclear loci is a key component of species delimitation, especially where candidate species were first formulated from mtDNA sequence data. A multitude of approaches that do not require a priori grouping of samples into candidate species are available. Algorithms that can detect the number of distinct populations in a set of individual specimen data, such as STRUCTURE or STRUCTURAMA (Pritchard et al. 2000; Huelsenbeck et al. 2011), or analogous programs for next-generation sequencing data, such as ­NGSADMIX (Skotte et al. 2013), have clear potential for species delimitation and independent testing of mitochondrial candidate species, although even here, interpretation must be cautious (Wiens and Colella 2025). Haplotype networks generated by software such as NETWORK (Fluxus Technology Ltd. – www.fluxus-engineering.com) can visualise the extent of haplotype sharing between individuals, which can be sufficient to document reproductive isolation in some cases (e.g., Ratnarathorn et al. 2023). Multilocus genetic distances can be calculated through programmes such as POFAD (Joly and Bruneau 2006), and the resulting matrix of between-individual distances visualised through a principal coordinates analysis (Barlow et al. 2013; Gowri Shankar et al. 2021). Genetic distance matrices can also be tested for the relative degree of association with geographic distance (IBD) and candidate species membership using Mantel tests (Jensen et al. 2005). All these approaches can visualise or explicitly test the degree of reproductive isolation between candidate species without the biases often associated with a priori grouping of specimens.

Testing candidate species with phenotypic characters. Most phenotypic traits are likely to be under polygenic control, and the analysis of multiple traits can be used to provide a proxy for overall genomic variation, either in conjunction with or instead of nuclear genetic markers. Candidate species would be confirmed by fulfilling the prediction that they constitute phenotypically discrete entities distinct from other such entities. Similar considerations apply here as to nuclear markers: Only methods that consider variation both within and between candidate species can positively reject these primary species hypotheses. Suitable approaches include PCA and derivatives such as MFA and NMDS, as well as the use of Mantel tests to determine whether a given pattern of geographic variation is better explained by IBD or assignment to candidate species (Puorto et al. 2001).

Analyses of concatenated multilocus sequence data. The key reason for generating multilocus datasets should be to test for cytonuclear discordance, thereby challenging mitochondrial candidate species. However, phylogenetic analyses of concatenated multilocus datasets shoehorn all loci into a single tree topology, and thereby entirely negate their ability to reveal cytonuclear discordance. Concatenating multiple highly variable mitochondrial sequences with a few relatively conserved nuclear loci generates a mitochondrial phylogeny with added noise, not a multilocus phylogeny in any meaningful sense (Wüster et al. 2024). Concatenated analyses may be appropriate for inferring a species tree where there is no significant conflict between different markers but have no place in species delimitation.

Some multispecies coalescent (MSC) approaches. Bayesian species delimitation analyses such as those implemented in the popular programs BPP and BPP3 (Yang and Rannala 2010, 2014) have become popular in systematics (Leaché and Fujita 2010). Their ability to account for conflict between gene trees in both species tree reconstruction and species delimitation represents a major step forward compared to concatenated analyses of multilocus data. However, important limitations of the models as originally conceived include (i) the requirement for specimens to be grouped a priori into candidate species, (ii) assumption of lack of genetic structure within the candidate species, and (iii) assumption of a lack of migration (i.e., gene flow) between candidate species (Barley et al. 2018). Since the extent or lack of gene flow between candidate species is exactly what should be tested rather than assumed, this makes the use of such methods problematic in cases of candidate species with contiguous distributions (Chambers and Hillis 2020). Similarly, the methods are sensitive to isolation by distance, especially in conjunction with sparse geographic sampling, which is liable to lead to oversplitting (Sukumaran and Knowles 2017; Chambers and Hillis 2020; Mason et al. 2020). While MSC approaches are being refined to take gene flow into account (Jackson et al. 2016; Kornai et al. 2024), users must remain critically aware of the assumptions of the specific analyses they are using and how these may be violated by their data.

Analyses of phenotypic variation treating candidate species as units of comparison. As shown in Figure 2, it is essential to test critically for congruence between phenotypic variation and mitochondrial phylogeography. Any methods that require grouping individuals into a priori Operational Taxonomic Units (OTUs) cannot do so if each candidate species is treated as a single OTU. That includes simple tables of morphological comparisons between candidate species, t-tests or ANOVAs treating candidate species as units of comparison. Approaches that seek to maximise separation between a priori-defined groups, such as canonical variate or discriminant function analyses, are especially problematic if they treat the candidate species as single OTUs. These procedures effectively cherry-pick those traits that happen to approximately coincide with the candidate species while ignoring those that show different patterns of variation and could potentially reject these hypotheses. They are thus effectively tautological. A work-around for many of these approaches can be dividing each candidate species into multiple OTUs, which then allows testing of the prediction that multiple OTUs of each candidate species should form homogeneous groups distinct from the OTUs of other candidate species (Thorpe 1983).

It is also important to be aware of approaches that rely on a priori grouping of specimens as a preparatory step prior to further analyses, such as corrections for size or the estimation of missing data. These can be problematic even if the final analysis itself does not require a priori grouping. For instance, correcting for overall size to identify differences in shape requires grouping into OTUs (Thorpe 1975, 1976). At this point, grouping artefacts can be introduced inadvertently. For instance, if, as suggested for multispecies analyses in the description of the popular R package GroupStruct (Chan and Grismer 2022), dependent variables are regressed to a different adjusted SVL in each OTU, then the impacts of size differences between OTUs will be magnified, not controlled, which will eclipse the differences in shape and artificially inflate the distinctness of the OTUs in downstream analyses. Moreover, the use of different regression slopes ß in different OTUs, rather than the pooled within-OTU regression slope (Thorpe 1975, 1976), to adjust a dependent variable could lead to misleading downstream results (Reist 1986). While grouping specimens for preparatory analyses is often unavoidable, care must be taken to avoid grouping artefacts that will impede the usefulness of the data for critically testing species boundaries.

Ecological niche modelling. The notion of ecological divergence as an indicator of speciation has a long history in systematics (Van Valen 1976). The advent of ecological niche modelling (Guisan and Thuiller 2005) has allowed a much more rigorous circumscription of the ecological niches of a taxon, and has attracted the attention of systematists as an additional tool for species delimitation. However, Meik et al. (2015) cautioned against the use of climatic data as surrogates for physiological adaptations, and hence as indicators of speciation. It is important to understand that the observed distributions of taxa reflect their realised niches, not the fundamental niche within which they could exist without constraints imposed by biotic interactions, such as competition with close relatives. Consequently, the occupation of different niches by geographic segments of contiguously distributed candidate species could reflect either shifts in the fundamental niche (i.e., adaptive evolution of likely inherited traits), or simply the spatial division of a wider range by taxa with a single broad fundamental niche, perhaps through competition or as a result of historical contingency (Tingley et al. 2014; Meik et al. 2022). While tests like that proposed by Raxworthy et al. (2007) can help distinguish between these hypotheses, the simple comparison of realised niches of candidate species is likely to confound realised and fundamental niches and thus lead to an overestimate of species numbers. In particular, where candidate species occur along an environmental gradient, constructing niche models for each and projecting them back onto a map will inevitably result in the model suggesting different climate niches, and would be tautological (Fig. 2).

This paper was prompted by multiple recent cases where species limits in high-profile taxa were primarily inferred using an mtDNA-led approach, without critical independent testing of the putative taxa, and the results were later found to be misleading once a multilocus dataset was brought to bear on them. Further recent controversial species delimitation studies and the ensuing criticism (Dubois et al. 2024; Reyes-Velasco 2024; Vásquez-Restrepo et al. 2024; Wüster et al. 2024; Reynolds 2025) have again focussed attention on the problems of over-reliance on mtDNA for species delimitation. However, beyond these high-profile cases, it is clear that mtDNA-led species delimitation remains widespread in herpetology, despite the well documented intrinsic limitations of this marker. This paper is thus an expression of the desire to put the mtDNA molecule into perspective and to encourage more rigorous approaches to species delimitation that capitalise on the strengths of mtDNA while avoiding its pitfalls. I approach this by analysing the use of different marker systems and analytical approaches in recent species delimitation studies in non-avian reptiles, assessing to what extent different markers are used to critically test, rather than just confirm, mtDNA-based species limits, and how currently widespread approaches can be improved to achieve more convincing species delimitations, while remaining mindful of the resource limitations afflicting much of systematics.

Many or most of the problems highlighted above could be prevented through the use of next-generation sequencing methods, potentially coupled with increasing standardisation of the markers used (Eberle et al. 2020). However, their minimal uptake in herpetological species delimitation suggests that significant barriers continue to exclude most herpetologists from using these techniques (Carneiro et al. 2025). Consequently, my aim in the following paragraphs is to stimulate taxonomists from all resource settings to make the most of whatever means are at their disposal. My comments will therefore be based on the methodological status quo, where the overwhelming majority of taxonomic herpetology is restricted to morphological analyses and Sanger sequencing of a small number of loci.

Sequence smarter, not harder

The need for more genuinely multilocus analyses inevitably raises questions about the cost of additional sequencing, especially in resource-poor settings. However, the data collected here offer only very partial support for this concern: In many studies, researchers sequence multiple mitochondrial genes for numerous specimens, presumably to enhance the resolution of the resulting mitochondrial gene tree, but only a small subset of specimens is sequenced for fewer nuclear genes (Fig. 5). At least for the purposes of species delimitation, this is a misallocation of effort. Sequencing multiple mitochondrial genes is important if the main goal is an accurate mitochondrial phylogeny. However, the formulation of primary species hypotheses only requires the monophyly of the haplotypes of each candidate species to be strongly supported in the mtDNA tree, not so much the phylogenetic relationships within or between them. This is often achievable through the sequencing of a single mitochondrial gene. Indeed, that was the entire starting point of the barcoding initiative (Hebert et al. 2003a, 2003b, 2004b). While a second mitochondrial gene may be advisable for greater resolution and also to guard against misleading phylogenetic signals due to nuclear integrations (Numts - Zhang and Hewitt 1996), the use of further mitochondrial genes provides rapidly diminishing returns in species delimitation. Diverting the resources spent on additional mtDNA sequences towards sequencing nuclear genes would enable genuine multilocus analyses and more rigorous testing of barcode-inferred species boundaries for the same outlay (Fig. 5).

Figure 5. 

Allocation of sequencing effort between mitochondrial and single copy nuclear genes in reptile species descriptions. While the use of few loci overall (blue box) may be an inevitable consequence of resource limitation, the sequencing of numerous mitochondrial gene fragments but few or no nuclear markers (orange box) constitutes a suboptimal allocation of resources for the purposes of species delimitation.

As an example, one recent study later criticised for excessive reliance on mtDNA (Arteaga et al. 2024; Reyes-Velasco 2024) generated 206 new gene sequences for 70 specimens across 7 gene fragments. However, only 18 of these sequences, from 11 individuals of 5 candidate species (out of 10 recognised in the paper), were from two nuclear loci, while the remaining 188 sequences were distributed across 5 mitochondrial loci! An alternative strategy for the same outlay would have been to sequence a single mitochondrial gene for all specimens, and to allocate the remaining 136 sequences to the two nuclear markers for virtually all specimens. If an accurate mitochondrial phylogeny was deemed essential, the same sequencing effort would have allowed all 70 specimens to be sequenced for one mitochondrial gene, one specimen of each candidate species for the remaining four mitochondrial genes, and 48 specimens (4–5 for each species) for the nuclear NT3 and RAG1 genes. Both approaches would have allowed more rigorous tests of the evolutionary independence of the mitochondrial candidate species and forestalled later criticisms of the work.