There is little doubt that the central unit for taxonomy is the species, and that associating scientific names unequivocally to species is pivotal for a reliable reference system of biological information 1 . Since the advent of Linnaean nomenclature in 1758, taxonomists have been describing and naming thousands of species every year--currently around 15.000-20.000 among animals only 2 3 --numbers that rapidly increase for many groups of organisms due to the incorporation of new tools for discovery and the exploration of poorly known areas of the planet 4 5 6 7 8 . Indeed, this progress is being made possible despite important impediments 9 because species taxonomy is resurging as a solid scientific discipline 10 that incorporates technological advances, such as virtual access to museum collections 11 , high-throughput DNA sequencing 12 , computer tomography 13 , geographical information systems 14 , and multiple functions of the internet 15 . Also, taxonomic information is increasingly digitized and made available through several global initiatives, such as Species2000, The Encyclopaedia of Life (EOL), The Global Biodiversity Information Facility (GBIF), or ZooBank. The future has been envisioned to be an interactive "cybertaxonomy" with dynamic online description and publication of new species, and where updated taxonomic information would be accessible for almost everybody from everywhere 16 17 .

However, modern taxonomy still faces two major challenges. First, a qualitative challenge is to reach scientific consensus about the basic category around which taxonomy is built --the species-- and thus improve species delimitation. The second, a quantitative challenge, is the sheer number of species on earth that require discovery and description, estimated in at least 10 million, only considering eukaryotes, and of which a small fraction of less than 2 million have so far been named 18 . These challenges are closely tied to two deliverables that the scientific community and society expects from taxonomy. On one hand, to provide empirical rigor to species hypotheses and stability to their names, which requires a careful and often painstaking and time-consuming labor of species delimitation. On the other hand, an acceleration in the pace of species description, with the peril of erroneous species hypotheses and thus of unstable names.

We here review recent proposals for the development of taxonomy that have been launched to meet both challenges. We argue that recent conceptual advances will allow taxonomy to improve species delimitation through the integration of theory and methods from disciplines studying the origin and evolution of species. Also, we emphasize the importance of applying existing integrative protocols and of developing novel ones for proceeding toward a complete inventory of life on Earth in a reasonable time.

Most evolutionary biologists will now agree that species are separately evolving lineages of populations or metapopulations [sensu 19 ], with disagreements remaining only about where along the divergence continuum separate lineages should be recognized as distinct species 20 21 . This emerging consensus might appear as a minor advance, but it has led to a renewed discussion about species delimitation that is paramount for catalyzing the integration of new knowledge and methods of population biology, phylogenetics, and other evolutionary disciplines into taxonomy 22 23 24 25 26 27 28 29 30 31 32 . Taxonomists are realizing that what matters for the study of speciation matters for taxonomy as well, and that species will be better delimited if we know what caused their origin and determined their evolutionary trajectories. As illustrative examples, the discovery and description of three Californian species of trapdoor spiders 33 required inferences about the origin, genetic structure and degree of ecological interchangeability of divergent lineages; and a recent taxonomic revision of cardinal birds 34 involved the reconstruction of the populational, phylogenetic and biogeographic history of lineages.

One consequence of this trend is that after more than 250 years of predominance of comparative morphology in species discovery, new methods and data--mainly molecular--are conquering a great piece of the realm of taxonomy. Although many taxonomists have received with joy the prospects of this new taxonomy 11 12 16 35 36 , others are more skeptical 37 38 . Nonetheless, a general view has emerged that now is the time to construct a more "integrative taxonomy" that would accommodate new concepts and methods 36 37 39 40 41 , and a considerable number of studies have already echoed the new term "integrative taxonomy" 28 31 42 43 44 45 . A close look at this literature reveals, however, a lack of consensus about what an integrative taxonomy should be. For example, Dayrat 36 proposes to broadly integrate molecular methods and approaches of population genetics at the expense of classical taxonomic procedures, while others hold just the opposite idea 37 38 . Also, major disagreements concern the degree of congruence that different characters must show to consider a population or a group of populations as a separate species. For example, some taxonomists see congruence among molecular and morphological characters as a necessary requisite 36 40 43 while, for others 23 41 46 , the strength of integration lies in avoiding any a priori selection of character combinations. We here dub these two frameworks as "integration by congruence" and "integration by cumulation", respectively (Figure 1) and explain their advantages and limitations.

Congruence approaches have a long tradition in systematics. In phylogenetics, hypotheses of character homology can be examined following the principle of reciprocal illumination, in which each individual character hypothesis is evaluated by the extent to which it agrees with the overall favored phylogenetic hypothesis 47 . In contrast, the congruence approach for species discovery does not examine character hypotheses but lineage divergence hypotheses and can thus most convincingly be founded on population genetics theory [but see 37 38 . In this framework, phylogeographers introduced the genealogic concordance method of phylogenetic species recognition, GCPSR 48 . The GCPSR states that congruent identification of a population-level phylogenetic lineage by several unlinked genetic loci indicates that it is genetically isolated from other such lineages, and thus qualifies as a species, because only in such isolated lineages will the coalescent histories of the different markers agree. Integration by congruence follows the same rationale under the assumption that concordant patterns of divergence among several taxonomic characters indicate full lineage separation. Taxonomists also expect that species so discovered will more often correspond with distinct evolutionary units because it is highly improbable that a coherent pattern of character concordance will emerge by chance. As an example, DeSalle and collaborators 49 illustrated in a work diagram that congruence between two taxonomic characters is an important factor to reach a conclusion about species status (Figure 2a-2c). Different combinations of taxonomic characters may be deemed necessary by different investigators to propose and support species (Figure 3a), such as the congruence of molecular and morphological characters 36 43 , or even more restrictive combinations requiring evidence about reproductive isolation 50 51 .

The major advantage of the congruence approach is that it promotes taxonomic stability: most taxonomists will agree on the validity of a species supported by several character sets, as long as it is clear that they are unlinked and fixed (see below). The major limitation inherent in demanding congruence among taxonomic characters is the risk of underestimating species numbers (Figure 1) because the process of speciation is not always accompanied by character change at all levels [e.g. 52 , and the relative rates of character change during lineage divergence are heterogeneous (Figure 4). Indeed, several empirical studies support the view that lack of character congruence is a frequent situation [reviewed for arthropods by 41 ] resulting from the different modes and circumstances of speciation [e.g. 24 53 54 55 . A further risk of applying integration by congruence could be the bias toward uncovering older species. Species that diverged in the distant past will have an increased probability of showing complete gene lineage sorting and reciprocal monophyly for many loci 56 . The probability of new mutations--and thus character differences--arising and being fixed through adaptive or neutral processes also increases (Figure 4). Furthermore, extinction of relatives creates larger gaps among old species, whereas recent radiations may often be overlooked by a strict consensus approach 53 . For example, in Darwin's finches or some lineages of cichlid fishes, groups of closely related species show striking morphological differences that originated through fast divergent selection associated with ecological transitions, but show weak reproductive isolation, low genotypic clustering, and little neutral genetic differentiation [reviewed by 55 ].

The framework of integration by cumulation is based on the assumption that divergences in any of the organismal attributes that constitute taxonomic characters can provide evidence for the existence of a species 23 . This approach defends the view that because all taxonomic characters are contingent in existence, order of appearance, and magnitude of divergence during speciation, the only way for true integration is allowing any source of evidence--even a single one--to form the basis for species discovery. In this approach congruence is desired but not considered necessary 23 . In practice, evidence from all character sets is assembled cumulatively, concordances and discordances are explained from the evolutionary perspective of the populations under study, and a decision is made based on the available information, which can lead to recognition of a species on the basis of a single set of characters if these characters are considered good indicators of lineage divergence (Figure 3b 41 46 ).

A major advantage of this approach is that it does not bind species delimitation to the identification of any particular biological property. Taxonomists can thus select and focus on the most appropriate set of taxonomic characters for each group of organisms. Indeed, this has been the traditional approach of morphological taxonomy (Figure 3b) before the massive incorporation of other characters. Also, cumulation is probably most suitable to uncover recently diverged species in adaptive radiations 53 due to the stepwise process of speciation along ecological gradients 57 58 . The main limitation of the cumulative approach is that the uncritical use of a single line of evidence (e.g. a single locus of mtDNA) can lead to overestimation of species numbers (Figure 1). For example, because of genetic drift, small populations isolated only for a very short period could already become reciprocally monophyletic with respect to some character and be thus diagnosable. Such situations do not represent the type of diversity of interest to most ecologists and evolutionary biologists 59 and the question remains if these populations should be recognized and named as species or not. As an example, Meiri and Mace 50 disagreed with the recognition of Bornean and Sumatran populations of the clouded leopard Neofelis nebulosa diardi as a full species 60 for exactly that reason. Indeed, other cases of elevation of subspecies to species rank in several groups of organisms [review by 4 ], but especially in birds and primates 61 , have been criticized as an unjustified inflation of biodiversity with detrimental consequences for macroecology and conservation 61 . However, in many situations the steep increase of species numbers reflects a genuine discovery of previously unknown evolutionary lineages and thus valuable taxonomic progress 5 6 7 .

Further discussion about the alternatives for integration requires a closer look to what these approaches try to integrate--the characters. Taxonomic characters are organismal traits used as evidence for species discovery 62 . Characters can (i) be classified by the level of biological organization of the attributes to which they refer: biochemical, molecular, morphological, behavioral or ecological. They can (ii) be qualitative or quantitative, to describe variation that is discrete or continuous. Characters can (iii) have fixed states (that is, for each of the compared species or population there is a unique state for all individuals); or be polymorphic within species but with states distributed in different frequencies across species. Most important for taxonomy--and often-neglected--is the classification of characters (iv) by the evolutionary processes that shaped them (sexual or natural selection, or genetic drift), and by the role they play in the speciation process (Figure 4).

Taxonomists need to resort to different taxonomic characters to conform to the biological peculiarities of particular taxa. For example, behavioral characters--especially those genetically fixed without ontogenetic learning--such as call patterns of insects, bats and frogs, are routinely used by taxonomists working on those groups [e.g. 63 ], and their use has led to the discovery of many cryptic species in some groups of organisms 64 . Also, the ecology of organisms can be an important source of evidence in some cases. For example, the degree of ecological interchangeability may be a decisive taxonomic character to distinguish between closely related species 26 33 65 . In bacteria, the lack of a conspicuous morphology coupled with extensive gene transfer has forced taxonomists to develop a model-based strategy that combines data on ecology and on genetic diversity to delimit species 66 67 .

However, much of the discussion around integrative taxonomy deals with the merits of morphological versus molecular characters 36 37 38 39 41 49 . For practical and historical reasons most species have been primarily described based on morphology, including color. As a main advantage, morphological characters often serve to allocate individuals to species immediately by visual inspection, and are applicable to living as well as preserved specimens and fossils. Disadvantages are: (i) there is always a subjective component when defining and interpreting character states, (ii) demonstrating the fixation of a state requires large sample sizes 68 , (iii) the continuous rather than discrete nature of many characters on which taxonomists heavily rely, e.g., reptile scale counts 69 or mollusk shell size and shape 70 , and (iv) their unsuitability for some groups of organisms, either because speciation occurs without morphological change 52 , which leads to morphologically cryptic species 64 , or because morphological structures are labile or difficult to study--e.g. as in prokaryotes 66 67 .

Molecular characters used in taxonomy have historically been allozymic or chromosomal (number and structure of chromosomes), but today these are mostly sequences of mitochondrial (or chloroplast) DNA and, increasingly, of nuclear genes. While the analysis of allozymes has the advantage of simultaneously screening variation at several presumably unlinked nuclear loci, DNA sequences provide many more characters (nucleotide sites), can be amplified from much smaller samples, and can be obtained in unprecedented amounts through high-throughput sequencing from fresh samples, preserved historical collection material, and even from Pleistocene fossils 71 . DNA sequences can be examined using non-tree based methods to provide diagnostic differences among species 49 , but most frequently they are analyzed using tree-based methods to search for monophyletic groups that could represent species 22 . A limitation of tree-based methods is that it remains difficult to choose which among the multiple strongly supported clades detected represent species. Also, a growing body of evidence shows that discordance between species trees and gene trees is a common phenomenon caused by processes such as incomplete lineage sorting, hybridization, gene duplication, reticulated evolution, or recombination 54 . These situations greatly complicate the resolution of taxonomic problems 72 73 . Most promising are tree-based methods that rely on coalescent theory 54 74 , because they can identify signals of species divergence even under complex circumstances of gene tree incongruence and non-monophyly [e.g. 26 ]. A recent large-scale study on insects of Madagascar 75 shows that single-locus coalescent models perform well for both testing and discovering species from large sample sets even without prior hypotheses of population coherence, providing thus a potential empirical substitute for traditional tree-based methods for preliminary biodiversity screening and species identification (e.g., DNA barcoding approaches).

To be usable in an integrative taxonomy, characters should be evaluated taking into account the evolutionary forces driving the speciation process (Figure 4). This new perspective may help to overcome many of the long-standing discussions about characters in taxonomy. For example, as long as characters have a genetic basis, are unlinked and are not influenced by the same selective pressures, any character should be considered as an independent, equivalent and combinable unit. In other words, the potential of a character to clarify a taxonomic problem has to be carefully evaluated in every situation, and a particular nucleotide or morphological character state might be deemed more important than all other nucleotides and morphological characters, because it might be particularly informative to understand the process of lineage splitting and divergence. For instance, in those cases where the speciation process is driven by sexual and/or by natural selection, characters known to be under the influence of any of these forces in one species might be directly indicative of lineage divergence in the whole taxonomic group to which this species belongs and, thus, be more informative than those that are known to evolve neutrally.

Such inferences need to be based on careful evaluations to avoid circularity of arguments. As a very obvious example, in a group of animals where speciation has been demonstrated to be mainly driven by sexual selection of male colouration, differences in colour will be given a higher importance as taxonomic character than in a group of subterraneous and blind species. If speciation in a clade of phytophagous insects is known to be driven mainly by the switch to new host plants, then the discovery that a new population belonging to this clade feeds on a novel host plant might be a more relevant taxonomic character than it would be in a clade of host plant generalists. And in a group of microendemic species where specialization to narrow and distinct bioclimatic envelopes has been demonstrated to be the main force leading to speciation, the specialization of a newly discovered population to a bioclimatic niche distinct from all known species in the group might be a suitable argument to advocate its species status, while in groups where most species are known to be tolerant of a variety of bioclimates such data would be less relevant.

In general, sexually selected characters might be more likely to represent species-specific differences than naturally selected characters because they contribute to the reproductive cohesion and isolation of species while, at the same time, their rapid evolution contribute to create larger gaps between closely related species. Thus, the most important taxonomic characters would be those indicative of reproductive isolation or limited gene flow, such as crucial mutations in "speciation genes" 76 or any trait that directly mediates a premating or postmating reproductive barrier, such as wave form and frequency differences in advertisement calls of insects or frogs, or genital structures of arthropods and squamate reptiles. For example, a single amino acid substitution can suffice to produce striking plumage differences mediating species recognition and leading to speciation in birds 77 . Also, differences in a coding vision gene can affect female mating preferences and initiate lineage divergence in fish 78 . Some researchers further argue that molecular markers prone to high intraspecific gene flow might be less affected by interspecific gene flow, and be thus more effective for delimiting species 27 , because in cases of gene introgression among sister species substantial intraspecific gene flow will reduce the frequency of introgressed alleles. But, also, the first move in speciation can result of the accumulation of multiple neutral mutations in DNA sequences causing hybrid incompatibilities 55 , which make those mutations ideal diagnostic characters to separate species.

In short, future discussions in taxonomy should not be about morphology versus molecules but about how characters reflect lineage divergence or about the functional relevance of some characters in the speciation process. As a consequence, taxonomy will no longer be a science restricted to the description of patterns but will be tightly linked to the study of processes generating diversity.

In the practice of current (increasingly molecular) systematics, phylogeography and DNA barcoding studies of eukaryotes are revealing units that might represent potential new species at a faster pace than results can be followed up by taxonomists. This situation suggests a need for guidelines to order and classify this undescribed diversity. The bacteriological concept of candidate species 79 has recently been explored and applied to vertebrates for such units 8 63 80 81 . A further developed stepwise working protocol (Figure 2c) recognizes three subcategories of candidate species 8 . Groups of individuals within nominal species showing large genetic distances, but without further information, are considered unconfirmed candidate species (UCS) deserving further study. When additional data indicate that these genealogical units are not differentiated at the species level, they are flagged as deep conspecific lineages (DCL). The third category, confirmed candidate species (CCS), applies to those deep genealogical lineages that can be considered good species following standards of divergence for the group under study but that have not yet been formally described and named. For example, confirmed candidate species are sister lineages in syntopy showing no evidence of interbreeding, or allopatric lineages with distinct morphological or bioacoustical character divergences.

A more standardized nomenclatural system might help to communicate with precision about candidate species, inventory them and track their changing status. Murray and Schleifer 79 proposed a formal naming system for candidate prokaryotes that consists in placing the epithet Candidatus before a preliminary species name, as for example: "Candidatus Liberobater asiaticus". This system has since been broadly accepted and implemented in the Bacteriological Code. However, it is inapplicable for eukaryotes because the zoological and the botanical codes do not specify minimum scientific criteria for recognizing species names as valid. Thus, while "Candidatus Liberobater asiaticus" can be kept as an informal name until the species is proved to be valid by accepted standards of the discipline and becomes thus formally described, any informal name given to a candidate species of animal could qualify as valid name under the Zoological Code if the proposal is accompanied by a voucher and diagnostic differences (e.g. exclusive haplotypes).

For different groups of animals, naming schemes of candidate species have been established. As one example, catfishes of the family Loricariidae are provided with so-called L-numbers, a system of consecutive numbers introduced in 1988 by R. Stawikowski, A. Werner and U. Schliewen, in which each putative new species is referred by a unique number combination after the letter "L"--from Loricariids. These numbers are designated upon publication of photographs of unknown color variants of Loricariids in the German journal "Die Aquarien und Terrarien Zeitschrift" and are used beyond the realm of aquariologists.

A standardization of such schemes across taxonomic groups of eukaryotes would be clear progress for data retrieval systems. A naming scheme for candidate species should not be mistaken for a substitute or competition with the established Linnaean system of nomenclature but, rather, it is a mean to facilitate the assembly of data sets that could eventually lead to the description of new species under the Linnaean system. To avoid conflicts with the Codes of nomenclature, we propose to designate candidate species of eukaryotes through the combination of the binomial species name of the most similar or closely related nominal species, followed (in square brackets) by the abbreviation "Ca" (for candidate) with an attached numerical code referring to the particular candidate species (more than one candidate might be recognized under a valid species), and terminating with the author name and year of publication of the article in which the lineage was first discovered. The vouchers the candidate species cfould be the GenBank accession numbers of the sequences used to propose the candidate status, or any equivalent information (e.g. MorphoBank accession numbers for morphological candidate species, or a voucher specimen number from a public collection). As an example, Hirudo medicinalis [Ca3 Siddall et al. 2007], would be the exclusive name combination referring to a particular candidate species of European leech (Ca3), as defined in the corresponding reference 82 . This system should be flexible enough to accommodate situations in which no unambiguous candidate species definition has been proposed in a study--in this case, referring to a GenBank accession number should be possible: Hirudo medicinalis [Ca3 EF405599], where the number refers to a highly divergent sequence of H. medicinalis. When not even a tentative assignment of a candidate species to a most similar nominal species is possible, then it would be possible to assign a candidate species just to a genus or family, i.e., Hirudo sp. [Ca3 EF405599] or Hirudinidae sp. [Ca3 EF405599].

This system maintains the traditional structure of binomial species names and helps to list together both valid and candidate species in hierarchical and alphabetically ordered databases as GenBank, or repositories of morphological or geographical data. Adding a numerical code helps to avoid repetitive proposal of candidates; and GenBank accession numbers provide a direct link to source data. Candidate species should create a link between the activities of molecular biologists (e.g. ongoing DNA barcoding initiatives) and taxonomists to redirect taxonomic efforts and accelerate species descriptions.

Allele: one of two or more alternative forms of a gene that arise by mutation and are found at the same locus in a chromosome.

Allopatry: the condition of species or populations occurring in separate, non-overlapping geographical areas.

Candidate species: a set of organisms identified as a putative new species.

Coalescent theory: retrospective model of population genetics that employs a sample of individuals from a population to trace all alleles to the most recent common ancestor.

DNA barcoding: the use of short standardized DNA sequences to identify species.

Ecological niche: environmental conditions under which a species exist.

Gene lineage: ancestor-descendant series of alleles.

Locus (plural loci): a fixed position on a chromosome that may be occupied by one or more alleles of a gene.

Monophyletic group: a group consisting of an ancestor and all its descendants.

Neutral character: observable or quantifiable organismal trait whose evolution and variation can be explained by random processes.

Non-neutral character: observable or quantifiable organismal trait whose evolution and variation can be explained by natural or sexual selection.

Parapatry: the condition of species or populations occurring in contiguous geographical areas.

Paraphyletic group: a group consisting of an ancestor and some of its descendants.

Phylogenetics: biological discipline focused on reconstruction of the evolutionary relationships among organisms.

Phylogeography: study of historical processes responsible for intraspecific patterns (or patterns among closely-related species) of geographical distribution and diversity of gene lineages.

Species hypothesis: the hypothesis that a group of populations represents a separately evolving and divergent lineage.

Species lineage: ancestor-descendant series of metapopulations.

Speciation: the array of processes that result in the origination of new species.

Subspecies: infraspecific Linnaean category sometimes used to classify allopatric or parapatric populations showing some degree of divergence--traditionally in morphological traits--not considered large enough for the species rank.

Sympatry: the condition of species occurring in the same geographical area.

Syntopy: the condition of species occurring in the same locality at the same time.

Systematics: biological discipline that studies evolutionary patterns of biological diversity, including the fields of taxonomy and phylogenetics.

Taxonomy: biological discipline that identifies, describes, classifies and names extant and extinct living beings and deals with the theory of classification.

Taxonomic character: observable or quantifiable organismal trait used to separate species.