Contributions to Zoology, 79 (4) – 2010Michel Laurin: The subjective nature of Linnaean categories and its impact in evolutionary biology and biodiversity studies

To refer to this article use this url:

Biological criteria for rank assignment at a given level

Some criteria have been proposed to rank taxa at a single rank. For instance, many authors define species as reproductive communities (Lee, 2003), or lineages of such reproductive communities (De Queiroz, 1998); Dubois (1988) defines the genus as a set of potentially hybridizing species; phyla are often considered to be defined by a unique body plan (Erwin et al., 1997; McHugh, 1997; Valentine, 2004). Between the genus and the phylum levels, I am not aware of any proposed objective criterion to rank taxa. Of these three ranks, the species seems to be recognized by the least subjective criterion, since most biologists would recognize that reproductive communities should play a role in their delimitation, although there is considerable variability about the importance and way in which this criterion is used, and some authors view species as subjective entities (Ereshefsky, 2002). For phyla, uniqueness of body plan is only a vague criterion that has generally neither been used in a precise way nor quantified. The proposal to define the genus level using hybridization can illustrate the more general problem of making Linnaean objective, since this rank has received the most attention, after the species level (Dubois, 1988, 2007a: 32).

Using hybridization capability to objectively assign ranks to taxa could in theory be done, although this would require many nomenclatural changes, as Dubois (1988: 40) recognized. Dubois (1988: 40) also admitted that the geological age of the taxa thus delimited would be highly variable, about 2–3 Ma for placental mammals, but 20–23 Ma for birds and anurans. Thus, implementing this suggestion would make genera comparable only in some respects (possibility of hybridization) but not in others (geological age, and, possibly, phenotypic divergence), in addition to being very costly (since millions of experiments would need to be conducted). Such an implementation would be technically difficult because two distantly related species may retain their ancestral ability to reproduce, while more closely related species of the same clade may have lost this ability. Dubois (1988) suggested that the capacity to hybridize reflected overall genetic similarity (of structural and regulation genes), which is plausible, but it is also known that reproductive isolation can arise rapidly in hybrids or polyploids (Venditti and Pagel, 2010: 18). Application of this reproductive criterion to some taxa would be difficult, if not strictly impossible. For instance, neither reproductive communities nor sets of potentially hybridizing species can be recognized for extinct organisms represented only by fossils, nor for asexual taxa, such as eubacteria, archeans, and some eukaryotes. Finally, other authors disagree with the application of the potential for hybridization to delimit taxa because it is based on shared primitive features whose loss may not be particularly significant (Lherminier, 2009).

Dubois (1988: 31) suggested that the origin of new genera involved a special process (called ‘geniation’) involving a genetic revolution, and that this process was distinct from that of other speciations. Such a distinct process would help to delimit genera (if defined as sets of species capable of hybridization), although some of these delimitations would presumably yield paraphyletic taxa, and no theoretical or empirical justification for the existence of a geniation process can be found in modern genetics. Genetic theory predicts that, in some cases, reproductive isolation evolves gradually. Although genetic revolutions may play an important role in some speciations – according to some analyses, about 20% of the genetic divergence result from cladogeneses rather than anagenetic change (Venditti and Pagel, 2010: 15) –, some authors have expressed doubt about the general importance of genetic revolutions (Lherminier, 2009: 44). Genetic revolutions are involved in cases of hybridization and autopolyploidy (Turelli et al., 2001: 334), or when founder effects are important (Bush, 2007: 376), but these may consist of simple allelic frequency changes that do not result in reproductive isolation (Excoffier and Ray, 2008). Thus, the use genetic revolutions to delimit genera may not be easier than application of the hybridability criterion.

Consequences of the subjective nature of Linnaean categories

next section

The non-equivalence of taxa of a given rank has long been acknowledged as a major problem of rank‑based nomenclature (Dubois, 1988), and interesting proposals to assign ranks objectively (Dubois, 1988; Avise and Johns, 1999) or to bypass the problem caused by the lack of objectivity of Linnaean categories (Avise and Mitchell, 2007) have been devised by systematists. Despite this, Linnaean categories continue to be widely used in evolutionary biology and biodiversity studies (see below). This creates several problems that have already been discussed (e.g. Rowe and Gauthier, 1992; Laurin, 2005, 2008; Bertrand et al., 2006), so only a brief summary needs to be presented here. The main problems created by absolute ranks fall mostly into two categories: instability in delimitation of taxa, with the resulting imprecision in meaning of taxonomic names and the need to specify membership of taxa in each study (Rowe and Gauthier, 1992; Laurin, 2005, 2008), and problems in comparative studies or biodiversity assessments that rely on any supraspecific taxa (De Queiroz and Donoghue, 1988: 334; Bertrand et al., 2006). Other, smaller problems include the proliferation of redundant taxon names, and the unnecessary change in taxonomic composition created by the application of the principle of priority within ranks (or coordinated sets of ranks).

Imprecision in the meaning of taxonomic names

Given that the absolute rank assignment of taxa is subjective, ranking and delimitation of taxa can vary substantially between authors, or even between papers by a single author. This problem can affect any biological field, because biological knowledge is usually taxon‑specific; more than 1.5 million species have been described, and there are probably between 3.5 and 10.5 million extant species (Alroy, 2002). Thus, most biological knowledge is useful to the extent that the clade to which it applies is known with some precision. No objective reason for changing the rank allocation of a taxon, or for attributing a ranked taxon to another clade than specified in the original study is required by any of the rank‑based codes. Similarly, the rank‑based codes allow taxa to be put into synonymy (Fig. 1b) or for additional taxa to be erected within previously recognized taxa (Fig. 1c, d) without any justification or change in our objective knowledge about nature (i.e. discovery of new species or change in our understanding of the phylogeny). Thus, to take a simple hypothetical example consisting of four species (j-n) forming two genera (O and P) and one family (Oidae) under the originally proposed nomenclature (Fig. 1a), twelve alternative, equally, simultaneously and indefinitely valid nomenclatures can be proposed (Fig. 1b-d), resulting in thirteen nomenclatures. Under that system, the delimitation of taxa is ambiguous; hypothetical genus O can include species j and k, as in the original nomenclature, but it can also include only its type species j (Fig. 1c) or species j-n (Fig. 1b). The delimitation of family Oidae can fluctuate in the same way. The number of possible nomenclatures increases very fast with the number of taxa considered, so this number must be extremely high for the biodiversity that is already known. This problem occurs even in the case in which the phylogeny is stable and in which no new species is discovered, a combination of circumstances that should lead to maximal nomenclatural stability.

Much confusion arose in the discussions of nomenclatural stability in PN and RN because three kinds of nomenclatural stabilities (maximal, minimal, and realized) were conflated. Some proponents of RN have argued that the lack of precision in delimitation in RN is an advantage because it allows the limits of taxa to be adjusted when the phylogeny changes and when new taxa are discovered (Benton, 2007). This argument rests on the untested assumption that systematists spontaneously agree on the name and delimitation of taxa; if this were true, no system of biological nomenclature would be required. This is the maximal nomenclatural stability allowed by the system (Laurin, 2008), and it is undeniably greater than under PN, but it is seldom achieved, as empirical examples provided below demonstrate. The minimal nomenclatural stability provided by a system, if users abide by its rules, is probably more relevant, and it is in this respect that PN vastly outperforms RN because under a given phylogeny, only one delimitation of each taxon is generally possible (Fig. 1e, f). The debates between proponents of RN and PN can thus be reformulated in terms of the relative importance of minimal and maximal nomenclatural stability. Which one should be maximized? Minimal stability should be maximized if taxonomists generally fail to spontaneously agree on taxon delimitation, but maximal stability should be maximized if systematists generally agree on taxon delimitation. To determine which situation prevails, case studies of the realized nomenclatural stability are needed, and a few are provided below. However, various statements in the literature suggest that spontaneous agreement on nomenclatural matters is rare; after all, ‘It has been said that most scientists would rather use another scientist’s toothbrush than his terminology’ (McShea, 2000: 330).

Empirical studies show that the lack of delimitation provided by RN result poor realized nomenclatural stability, although this has been thoroughly investigated for few taxa. For instance, Rowe and Gauthier (1992) showed that the delimitation of Mammalia (ranked as a class, in rank‑based nomenclature) has varied greatly between authors, and even between various studies by a given author (Fig. 3a). The least inclusive clade called ‘Mammalia’ in the literature that they surveyed is usually called Theria (the smallest clade that includes placentals and marsupials), and the most inclusive Synapsida (the largest clade that includes mammals but not extant reptiles). The difference in composition between the least and most inclusive clades thus called Mammalia is modest if only extant forms are included because Monotremata was the only extant taxon that has been excluded by a small minority of studies. However, when extinct forms are considered (Fig. 3b-d), the difference is great, no matter which criterion is emphasized. For instance, the time of origin of the least inclusive clade (Theria) is no earlier than Jurassic according to some molecular dating studies (Bininda‑Emonds et al., 2007), and the paleontological evidence suggests an even later (Early Cretaceous) age of about 130 Ma (Benton and Donoghue, 2007). At the other extreme, Synapsida is known to have originated in the Carboniferous, at least about 315 Ma ago (Marjanović and Laurin, 2007). When looking at the morphology, the diversity of aspects encompassed by the first mammal is also impressive. At one extreme, the first therian was probably a moderately small (less than 25 cm snout‑vent length; Hu et al., 2005), possibly nocturnal, viviparous form with mammary glands and fur (Carroll, 1988). At the other extreme, the first synapsid was probably oviparous, devoid of mammary glands, diurnal, hairless, and larger, with a snout‑vent length of about 34 cm (Laurin, 2004).


Fig. 3. Delimitation of the taxon Mammalia under rank‑based (RN) and phylogenetic nomenclature (PN). Under RN, the name Mammalia (a) has been applied to several nested clades (some of which are identified by an asterisk), as shown by Rowe and Gauthier (1992). Under PN, the name could be defined using a crown-, apomorphy-, or total clade definition (numbers 1-3 in bold, blue type), but, once published in conformity with the PhyloCode, one definition would have priority and would not change. Most proponents of PN use a crown‑clade definition of Mammalia (as shown here), but many proponents of RN have advocated using the appearance of the dentary/squamosal joint to delimit Mammalia, although this proposal has not been consistently followed. The possible time of appearance of several other mammalian characters is shown. The vertical bar denotes the considerable uncertainty that surrounds the time of origin of most non‑skeletal ‘mammalian’ characters, which are known in the crown, but whose presence in other members of more inclusive taxa (e.g. Cynodontia, Eutheriodontia, Therapsida) cannot be assessed. A few taxa that have been occasionally considered part of Mammalia under RN are illustrated: Tetraceratops (b), Haptodus (c), and the dinocephalian Titanophoneus (d). Modified from Laurin and Cantino (2007) and Laurin and Reisz (1990). Scale bar (b-d) equals 2 cm. The geological time scale is from Gradstein et al. (2004). The two periods that could not be labeled on the figure because of lack of space are (from bottom to top) the Guadalupian and Luopingian (Middle and Late Permian). Abbreviations: D/Sq J, dentary‑squamosal joint; En, endothermy; Ha, hair; M Gl, mammary glands; Ma, million year ago; Mar, Marsupialia; Mo, Monotremata; Pl, Placentalia.

It could be argued that in the case of Mammalia, rank‑based nomenclature could not stabilize their delimitation only because, under the ICZN, taxa above the family-series have no types. However, two facts refute this argument. Firstly, the ICZN and the ICNB clearly state that rank‑based nomenclature does not delimitate taxa. Thus, according to Principle 2 in the introduction of the ICZN (1999), ‘[n]omenclature does not determine the inclusiveness or exclusiveness of any taxon, nor the rank to be accorded to any assemblage of animals, but, rather, provides the name that is to be used for a taxon whatever taxonomic limits and rank are given to it.’ General Consideration 4 of the ICNB likewise states that ‘[r]ules of nomenclature do not govern the delimitation of taxa…’ (Lapage et al., 1990). Thus, rank‑based nomenclature seems to have been designed specifically to avoid delimiting taxa. This is perplexing because at least some proponents of rank‑based nomenclature suggest that nomenclatural stability ‘would certainly be greatly appreciated by non‑systematists’ (Dubois, 1988: 31). Secondly, problems in delimitation similar to those evoked above for Mammalia also plague lower‑ranking taxa in the family- and genus-series. For instance, Keesey (in Laurin and Bryant, 2009) showed that names typified by Homo or Homo sapiens were associated with two (Hominoidea) to six (Hominidae) nested clades. Similar problems have affected names in the genus Rana (Hillis and Wilcox, 2005; Frost et al., 2006; Dubois, 2007b). In this case, the problem is exacerbated by the low number of Linnaean categories available to describe the diversity of the very speciose genus Rana (over 1000 species). Because of this, Hillis and Wilcox (2005) erected subgenera within subgenera, but, as pointed out by Dubois (2007b), this is contrary to the basic principles of rank‑based nomenclature (although the ICZN, contrary to the ICNB, does not state this clearly). Proponents of rank‑based nomenclature have long recognized that delimitation of taxa under that system is unstable, even at ranks at which taxa have types under the ICZN, such as the genus (Dubois, 1988). This brief review hopefully shows that the subjective nature of Linnaean categories contributes to vagueness in the meaning of taxon names, which is hardly surprising given that the authors of the rank‑based codes apparently think that taxa should be left undelimited.

Comparative studies that rely on supraspecific taxa

Using rank‑based nomenclature can lead to various problems that concern especially evolutionary biology. These include placing unnecessary constraints on the taxonomic sample and placing undue confidence in absolute or relative biodiversity indices derived from counts of higher taxa, an approach called taxonomic surrogacy. The first problem can occur when taxa of a given rank are selected to trace the evolution of a character. For instance, Smith et al. (2009: fig. 3) inferred the evolution of the cecal appendix in mammals using 79 taxa ranked as families, in addition to Amphibian and Reptilia (usually ranked as classes). While the use of taxa of a given rank (families) gives the reassuring appearance of a homogeneous sampling effort, this was probably not the best strategy because the appendix evolved much faster in some taxa than in others. Thus, in Laurasiatheria, a major clade of placental mammals that probably originated near the K/T boundary (about 65 Ma ago), the appendix never appeared. In Euarchontoglires (another large clade of placentals originating near the K/T boundary), the appendix evolved so fast that inferring its evolution using character optimization is difficult. In several of the ‘families’, the appendix was scored as ‘variable’ (usually termed ‘polymorphic’) when only some species in the taxon had an appendix (Smith et al., 2009: 1992). A more accurate estimate of the character history would have been obtained by replacing the polymorphic terminal taxa by smaller clades (typically ranked as genera) showing a single state. Thus, in this case, use of Linnaean categories gave unjustified confidence in a suboptimal taxonomic sampling scheme.

The problems raised by taxon surrogacy are even more acute. Prance (1994) showed that the neotropical region has a much greater biodiversity of embryophytes than equivalently sized paleotropical regions at the species level (Table 1). Looking at biodiversity at the genus or family level suggests that the flora of Africa and Malesia is about as diverse as that of the neotropics. It might be tempting to accept the conclusion that the neotropics support a greater biodiversity than the paleotropics based on Prance’s (1994) data, but this may not be justified. Perhaps systematists working in the neotropics were just more extreme ‘splitters’ when erecting species (or they may have erected a greater proportion of subjective synonyms) than those working in the paleotropics. Conversely, perhaps botanists working in the paleotropics preferred to include fewer species in each genus and family than botanists working in the neotropics, a question that has long divided systematists (Dubois, 1988); in this case, the neotropics might really have a greater biodiversity than the paleotropics. Another possibility is that the neotropics are home to a greater proportion of geologically recent species, which would explain that the number of genera and families is no higher in the neotropics than in the paleotropics. If Linnaean categories approximately reflected geological time, the three levels (species, genus, and family) would tell us something about the biodiversity of paleo- and neotropics. Unfortunately, this correlation, if present at all, seems to be very weak (Laurin, 2005; Bertrand et al., 2006), and rank assignment of taxa is too subjective to be informative. Bertrand et al. (2006: table 1) provided a compelling example of this using annelids.


Table 1. Biodiversity of embryophytes of the Neotropics and Paleotropics. Modified from Prance (1994).

Other studies have also found that using taxa of a given category to predict the biodiversity at a lower taxonomic level is problematic. For instance, Andersen (1995) showed that the number of ant genera in Australian localities was rather poorly correlated (r2 ≅ 0.5) with the number of species, and that the regression coefficient varied between habitats, size of area surveyed, and sampling effort, thus complicating the use of genus count to predict species counts. Grelle (2002) reported a fairly good correlation between the number of species and that of genera of neotropical mammals (r2 = 0.8) in various localities, but this varied strongly between taxa. Explained variance ranged from 0.45 for rodents to 0.96 for primates. Family and order number were generally not significantly correlated with species numbers. These patterns largely reflect the number of species per genus, family, and order, in taxonomies and in localities. As Grelle (2002) suggested, genus richness is often a good predictor of species richness when most genera are represented by one or two species in each locality. As that number increases, precision of the estimate of species number decreases. Thus, Terlizzi et al. (2003: 558) noted that ‘[t]he response of different taxonomic levels might change according to biogeographical features and internal diversity of taxa’; thus, ‘if the family richness is a good surrogate of species diversity in the North Sea, it might not be the same for the Mediterranean sea’. This conclusion is confirmed by the subjective nature of Linnaean categories, but the very fact that studies are still done on the efficiency of taxon surrogacy suggests that the subjective nature of absolute ranks is not fully understood by many biologists. Terlizzi et al. (2003: 559) concluded that taxonomic surrogacy ‘may be tolerated only when difficulties in sampling, data analysis and identification of some particular organisms make this procedure strictly necessary, not merely to save costs whatever the aim of the study is’, and appropriately noted that ‘[i]f species loss is the main concern of conservation biology (together with habitat loss), it is simply absurd to pretend to perform conservation studies without considering species.’

The lack of unified concepts of individual Linnaean categories precludes interpretation of patterns found through taxon surrogacy. For instance, Grelle (2002: 105) reported that, in most neotropical localities, primates and marsupials were generally represented by a single species in each genus (but several genera per locality); in contrast, several species of bats and rodents represented each genus. Grelle (2002) suggested that these differences reflected a higher turnover of primate and marsupial species than of bats and rodents. This is not likely because bats (1100 species) and rodents (2277 species) are the most speciose clades of placental mammals (5400 species, approximately; species counts from Wilson and Reeder, 2005), and since their antiquity is approximately equal with that of primates and less than that of marsupials (Wible et al., 2007), the diversification rate of bats and rodents has obviously been higher than that of most other mammalian clades (marsupials and primates have about 334 and between 190 and 350 species, respectively). Similarly, Grelle’s (2002: 105) suggestion that the difference in pattern between primates and marsupials on one hand, and rodents and chiropterans on the other, reflects ‘different assembly rules organizing (or not) these communities’ is not fully warranted. Such an interpretation might possibly be validated if it were shown that the minimal divergence time between primate or marsupial species co‑occurring in localities were significantly different from that of rodent or chiropteran species that co‑occur in similar localities. An alternative interpretation would be that ecological divergence allowing habitat sharing occurs faster in some of these taxa than in others. But such conclusions must rest on time‑calibrated trees, not on ‘ontologically empty’ (Ereshefsky, 2002: S309) Linnaean categories.

Even the species level is not completely objective in the absence of a universally agreed‑upon and universally applied species concept, and Ereshefsky (2002) has forcefully argued that this Linnaean category is as subjective as any other, even among sexually reproducing organisms. Furthermore, the equivalence between species of sexually reproducing organisms and asexual ones is even more problematic (Turelli et al., 2001: 336). Pleijel and Rouse (2003) even argued that we should drop the term ‘species’ and use the more ontologically neutral term LITU (least inclusive taxonomic unit) instead to denote the smallest recognizable clades. Thus, it could be argued that even counting species is not adequate to assess biodiversity, and that more objective indices based on phylogenies and evolutionary time should be used, such as Faith’s (1992) phylogenetic diversity index (PDI). The PDI is simply the sum of branch lengths linking all terminal taxa (of an area, or a clade, or of any group of taxa of interest). Computing the PDI is now feasible for many taxa since user‑friendly software such as Mesquite (Maddison and Maddison, 2009) can compute it, at least with some optional modules (Josse et al., 2006). The only remaining difficulty is obtaining a phylogeny of the relevant taxa with estimated divergence times, but recent progress in molecular (Thorne and Kishino, 2002; Sanderson, 2003; Brochu, 2004) and paleontological dating (Marjanović and Laurin, 2007, 2008) should facilitate this in the future. This is already shown by large compilations of time‑calibrated trees (e.g. Hedges and Kumar, 2009). Studying floral biodiversity of the paleo- and neotropics using the PDI would allow testing of the various hypotheses formulated above to explain the apparently greater biodiversity of the neotropics than the paleotropics.

All this raises serious questions about the validity of long‑admitted biodiversity patterns shown by paleontological and neontological examples. Many classical paleontological studies about the evolution of biodiversity through the Phanerozoic have been done at the family (e.g. Raup, 1979; Raup and Sepkoski, 1984) or genus level (Raup and Boyajian, 1988). Bertrand et al. (2006: 150) reviewed several such studies and pointed out that the most comprehensive paleontological databases that have been used to study the evolution of biodiversity are still compiled at the genus or family level. Various neontological studies about diversification have assessed dominance (the proportion of taxa at a given rank present in a single taxon of a higher rank, among a taxon of a still higher rank) and have argued that the ‘hollow curve distribution’ (HCD) implies that ‘one clade (or several clades) has had many more speciation events and/or fewer extinctions than other clades at the same taxonomic level’ (Dial and Marzluff, 1989: 26). That would be interesting (it would translate into a higher diversification rate) if taxa of a given rank were monophyletic and had equal geological ages, but neither is required by rank‑based nomenclature, and neither is true in the vast majority of taxonomies. Even though monophyly is increasingly enforced by authors and the lack of monophyly of taxa has been described as one of the greatest limitations of the use of taxonomies to assess biodiversity (Gaston and Williams, 1993: 5), some authors still claim that monophyly should not be required (e.g. Stuessy and König, 2008), and the commissions responsible for emending the rank‑based codes refuse to enforce monophyly as a taxonomic requirement (Laurin, 2008: 224). Yet, the adverse effects of using paraphyletic taxa were clearly shown, among other examples by Patterson and Smith (1987), in the context of assessing cyclicity in extinction patterns. Until dominance and HCD are reassessed on clades of equal geological age (regardless of absolute rank), the significance of these patterns will remain unknown.