Characters Reviewed and Sources of Difficulty

As mentioned in the Introduction, many of the fossil taxa are represented only by isolated teeth and jaws. The majority of the characters reviewed in this analysis (see appendix 1) were gathered primarily from the paleontological literature. In addition, some new characters and character states were added from personal observations made during the course of this analysis (appendix 1; see the following discussion).

All character states were compiled into a total of 246 transformation series. In many cases, the states within a given transformation series were so numerous and complex that it was more practical to handle them as a series of multiple binary transformations. Those cited transformation series that presented interpretive problems (e.g., relative size, relative position) are discussed below. It should be noted that the sequence in which the character states are listed in a given transformation series does not imply transformation additivity or polarity. Moreover, this phase of the analysis does not attempt to posit phylogenetic relationships: Outgroup comparison, and subsequent hypotheses of character polarity and directionality are addressed in the second phase (see Phylogenetic Analysis below).

Dental nomenclature follows that of Rich (1981); refer to figure 2. In general, because the nomenclature is fairly consistent across the Erinaceomorpha, the majority of the literature-based apomorphies are self-explanatory (refer to fig. 2) and need no discussion. The characters pertaining to the molars are illustrated in figures 2a, b, an idealization of the occlusal surfaces of upper and lower tribosphenic molars (following Salay, 1969 and Rich, 1981). Stereo photographs of occlusal surfaces of representatives of each genus reviewed in this analysis are presented in figures 3–7.

Interpretation difficulties are almost exclusively confined to those transformation series that attempt to characterize size and shape in a nominal (i.e., noncontinuous) fashion. For example, “the hypocone is larger than the protocone” (Storch and Qiu, 1991)—it is unclear whether the size “larger” refers to the height of the cusps, or the gross size (volume) of the cusps, or both. At first glance, this may seem trivial, but many fossil taxa, such as those that are represented only by dental material, are described and diagnosed based on such character states (appendix 1).

Herein I have tried to accurately define the size parameters to which I refer, however, there still remains the problem of visualizing size without the aid of controlled measurements (e.g., employing the use of calipers). This difficulty became apparent on many occasions when I, along with W. R. Downs, reviewed the same specimen with considerably different impressions. Orientation is partly responsible for the variant interpretations: When the tooth is in situ (i.e., in the jaw), it does not sit on a level plane, therefore the heights of the cusps appear different depending on whether they are viewed labially or lingually. Because of the sheer magnitude of the number of specimens reviewed in this analysis (227), along with the varying size and fragility of the specimens, taking measurements for every cusp was precluded. Instead, the specimens (skull and jaws) were placed so that the occlusal surfaces were on approximately the same plane (a natural position) and the relative heights of the cusps were recorded.

Other ambiguous characters concern cusp position illustrated, for example, in the statement “the paracone is lingual to the metacone” (Koerner, 1940). As with size, descriptions of cusp position depend on which part of the cusp is referred to: the base or the apex. In many cases, especially the protocone and protoconid, the cusp is somewhat crescentic in shape (see figs. 3–7), and thus the apex extends lingually beyond the base of the cusp. The apex of the protocone, however, becomes more aligned with its base with progressive wear (personal obs.). I have thus tried to standardize these relative-position characters by referring to only the base of the cusp. These revised positional definitions may not be in accord with the original intention of the author(s) who first observed and noted these characters (appendix 1), nevertheless, the base of the cusp is much less susceptible to wear, making its position less likely to be compromised.

Equally difficult to interpret is what constitutes a character or character state in the mind's eye of another investigator. Is an enamel “bead” on the labial side of the tooth equivalent to the presence of a labial cingulum? Or is a mediolateral crest extension of the protoconid on the p4 considered a distinct cusp (i.e., paraconid)? Fortunately, there are just a few examples where observer interpretation varies greatly. I have tried to adhere as closely as possible to the literal definitions of the characters reviewed (e.g., a labial enamel bead is not a cingulum, nor is a crest considered a distinct cusp). In some cases, I found it necessary to introduce new transformation series to accommodate commonly expressed character states (e.g., cuspules) that could not be accommodated comfortably in an already cited transformation series. I did not include those anomalous character states (e.g., double apex on the p4 paraconule) that were unlikely to have any potential for phylogenetic inference, since they are all individually specific, but it should be noted that such variation occurs.

The most pervasive difficulty in the analysis is apprehending the effects of wear on discrete dental characters. As discussed in the Introduction, wear is a consequence of many factors and processes, which ultimately results in the alteration of discrete dental characters at differing rates on an animal's full complement of teeth, both deciduous and permanent. These ontogenetic differences may be easy to apprehend as the effects of wear in a large sample. In those cases where only a few specimens are readily available (or even exist), however, this type of ambiguity could lead to unconscious character-state inference on the part of the investigator (Nixon, 1996).

Lastly, sample size itself may also pose problems: 25 individuals per taxon may not be a large enough sample to detect interspecific polymorphism, let alone the covariances with ontogenetic stages considered in the analysis.

Scoring of Characters

All individuals (see appendix 2) included in this analysis were personally reviewed and appropriately scored for the listed transformation series (appendix 1). The maximum number of multistates within a given transformation series is 5 (i.e., 0–4). Due to asymmetry, however, character coding is not as straightforward as 0→4. The fashion in which the asymmetrical data were recorded was designed to clearly indicate “morphographic distribution” of the polymorphism in a single individual. For example, states in the left and right teeth described as [2,1] would mean the right tooth has state 2, and the left tooth has state 1. Because most statistical packages cannot accommodate entries with commas, necessitating recoding for final analysis, coding of asymmetry starts with 5 and ends with 14. Coding is as follows:

[0, 1] or [1, 0] 5; [0, 2] or [2, 0] 6; [0, 3] or [3, 0] 7; [0, 4] or [4, 0] 8; [1, 2] or [2, 1] 9; [1, 3] or [3, 1] 10; [1, 4] or [4, 1] 11; [2, 3] or [3, 2] 12; [2, 4] or [4, 2] 13; [3, 4] or [4, 3] 14.

Although many of these combinations of asymmetrical polymorphism do not exist in the taxon-specific matrices, it was more efficient to recode all the possibilities through a linear editing function approach of the statistical package in which the data were collected (see below) than to accomplish this task by hand.

A result of this coding method is the acute loss of asymmetry distribution (e.g., [1, 3] or [3, 1] 10). Because this analysis seeks only to acknowledge that dental asymmetry exists among erinaceids, without exploring its nature, the loss of distributional information on asymmetry is considered insignificant. The raw data are preserved in Gould (1997; appendices 9–18).

As with virtually any other data set, missing values are present. In those cases where observation of a character state was unequivocally compromised by wear, the cell was left blank. Also, for those transformation series that were not applicable to the taxon being reviewed, the cells were also left blank. I did not code these data differently from other missing data, because operationally they are treated the same in a phylogenetic analysis that employs the parsimony principle (Maddison, 1993; Nixon et al., 1994; Nixon, 1996).

Taxa Reviewed

Of the 19 extant phylogenetic species currently recognized (fig. 1b), 10 were considered in this analysis (appendix 2). Many of the living erinaceids are surprisingly rare in North American collections, which constrained the sampling criteria accordingly. The optimal sampling parameters were as follows: (1) all specimens had to be accessible for personal review; (2) each taxon had to be represented by an ideal of 25 individuals and a minimum of 10, and the sampling constrained, if possible, to one geographic population; and finally, (3) each genus had to be represented by at least 2 species. Criterion no. 2 may violate classical sampling criteria in that the sample size may be insufficient to clearly apprehend variation (Sokal and Rohlf, 1995), either due to a paucity of specimens, or to a biased sampling within a population. Nevertheless, when dealing with the vertebrate fossil record, such parameters quickly become irrelevant in the face of inadequate sample sizes (e.g., one, or two specimens). Therefore, a small sample size of living taxa does not seem to be operationally any different than that of the fossil record, and in fact, a sample of 25 individuals per taxon is rare.

The limited sample size of this analysis also results from the need for personal review. As mentioned, North American collections of erinaceids are limited, the majority of which are located at the Smithsonian Natural History Museum and the American Museum of Natural History. Thus, taxa chosen were predetermined by their availability in these two collections. Of the 7 recognized living genera (following Frost et al., 1991; fig. 1b), 5 were sampled. Due to the rarity of Podogymnura (Hylomyinae, 2 species recognized) and Mesechinus dauuricus (Erinaceinae, monospecific genus) in these collections (fewer than 6 specimens per taxon were available), they were omitted from this analysis. Echinosorex (fig. 3), currently considered a monospecific taxon (E. gymnura) following Corbet (1988) and Frost et al. (1991), is the only taxon that is represented by more than 25 individuals (32 were reviewed). This exception to the maximum sample size exceeds the target sample size of 25 individuals from (presumably) one population from western Malaysia, and 5 individuals from the island of Borneo, which previously had been considered a separate (sub)species, E. gymnura albus (Corbet, 1988). Given the availability of these specimens and the nature of this project (inter- and intraspecific variation), the addition of these specimens seemed appropriate.

The following genera were analyzed: Hylomys (fig. 4), Atelerix (fig. 5), and Erinaceus (fig. 6) each represented by 2 species; and Hemiechinus (fig. 7), represented by 3 species. Three hemiechinines were analyzed to ensure the inclusion of at least one taxon that was previously considered to be Paraechinus (Rich and Rich, 1971; Rich, 1981; Corbet, 1988; Frost et al., 1991). Twenty-five individuals of the species Hylomys suillus, Erinaceus amurensis, Atelerix algirus and Hemiechinus hypomelas were not available for review, and therefore, smaller sample sizes had to be accepted: 16, 11, 24, and 24, respectively.

Other Data Collected

Data regarding the relative age, sex, and geographic localities of each individual (see appendix 2; see also Gould, 1997, appendices 9–18) have also been compiled; they constitute the variables against which discrete dental variation was tested for covariance.

The relative age categories—juvenile, mature adult, worn teeth—are representative of wear stages, as there is no question that wear is principally a function of the age of an individual (Brockie, 1959; Skoudlín, 1976, 1981; Gregory, 1976; Kahmann and Vesmanis, 1977; Vasilenko, 1988). The identification of juveniles is straightforward; it is based on the presence of deciduous teeth. The remaining two age categories are based on the following definitions: [category: worn teeth] those specimens identified as old run the gamut from having teeth worn to the roots to teeth worn just to the point where identification of certain discrete characters becomes murky (e.g., metaconule, cristae); [category: mature adult] all remaining individuals that do not have deciduous teeth or morphology that is clearly compromised by wear. These categories may seem arbitrary or imprecise, but to estimate the age of a hedgehog accurately is no simple task. Previous investigations regarding age estimation among erinaceines suggest that the only reliable methods are: (1) measuring the relative dry weight of eye lenses, which increases with age (Morris, 1969, 1970, 1971); (2) noting the stage of epiphyseal fusion (Morris, 1971; Reeve, 1981; Dickman, 1988); (3) determining the number of periosteal growth lines in the lower jaw (Kristoffersson, 1971; Kratochvíl, 1975; Dickman, 1988); and (4) observing dental wear stage (Brockie, 1959; Skoudlín, 1976, 1981; Gregory, 1976; Kahmann and Vesmanis, 1977; Vasilenko, 1988). Given the lack of access to fresh eye lenses and postcranial material for review of epiphyseal fusion and of permission to take thin sections from hundreds of specimens for age determination, wear stage was deemed acceptable for estimating age.

Admittedly, using a second age variable would have increased the rigor in this analysis. However, an early analysis of cranial suture closure (basioccipital and premaxillary-maxillary-palatine) demonstrated that these sutures close at approximately the same time very early on in ontogeny (personal obs.; Gould, 1997), thus, they would not have provided any additional information regarding the age of an individual.

The other variables—sex and geographic locality—were determined from specimen tags.

Analyses Conducted

Data were initially collected in MicroSoft Excel 4.0 for the Macintosh. It was then subsequently transposed and imported into both StatView 4.1, and a promotional version of (SAS) JMP for the Macintosh. The vast majority of all of the discrete dental data analyses (DDA) were conducted using StatView 4.1. All taxon data matrices (Gould, 1997, appendices 9–18) were first reviewed for intraspecific variation, as well as asymmetry within a transformation series (DDA 1). The data were analyzed by generating frequency tables for each transformation series across all the taxa (Gould, 1997, appendices 9–18). The results have been compiled in one table (appendix 3) for a global overview of variation. Identified interspecific polymorphism was then analyzed for covariance with sexual dimorphism (DDA 2), deciduous dentition and wear stage (DDA 3), and geographic locality (DDA 4). Bar chart cell plots were employed to visualize the distribution of the data, and their covariation with the variables noted (see appendix 4 for examples).

DDA 1: Polymorphism and Asymmetry

Polymorphism and asymmetry are prevalent in all of the taxa reviewed (table 1, appendix 3; see also Gould, 1997, appendices 9–18). For all the transformation series considered, overall polymorphism within a given taxon ranges from 25% to 44.5%, asymmetry being slightly more conservative, ranging from 4.1% to 39.8%.

The overall amount of polymorphism (and asymmetry) detected across and within 10 taxa and 246 transformation series does not seem too surprising considering the quantity of characters reviewed. What is surprising, however, are the characters that are consistently polymorphic across all the taxa—the number of upper canine roots, the number of P2 roots, the presence and absence of the P3 lingual lobe, and the shape of the P4 and condition of its lingual roots, to mention a few (appendix 3). These character states have all been cited in the literature as diagnostic for a taxon, either at the species level or higher (see appendix 1). Moreover, the polymorphic presence and absence of an entire tooth (I2, I3, P1, and P3) within a species is even more disconcerting (appendix 1). Without large series of individuals from a single population by which to detect such variation, these characters could be considered evidence of multiple species.

DDA 2: Covariation of Polymorphism and Sexual Dimorphism

The results of this analysis suggest that there is no expression of sexual dimorphism in the discrete dental characters among the reviewed taxa, and most probably throughout the living erinaceids as well.

DDA 3: Covariation of Polymorphism with Relative Age

All positively correlated results of polymorphism with the relative age of an individual (wear stage and deciduous vs. permanent teeth) are illustrated in table 2 (see also Gould, 1997, appendix 4).

In some taxa, the deciduous dentition can be quite different from the permanent teeth. Results of this analysis, however, indicate that among hedgehogs, the morphology of deciduous and permanent teeth is not easily distinguishable. Of the 10 taxa reviewed, only 3 distinctly demonstrate polymorphism in deciduous and adult teeth (see table 2): Hylomys suillus (upper canine size relative to postcanines); Atelerix algirus (presence of I3 posterior cingulum); and Hemiechinus auritus (P3 is reduced, dP3 protocone is present). (see also Gould, 1997, appendix 4, figs. 5, 18, and 34, respectively.)

Thirty-nine characters were found to be positively correlated with wear (table 2; see also Gould, 1997, appendix 4, figs. 2–4, 6–17, 19–33, 36–51); those that are consistently affected are: premolar cuspules, cingula, parastyle, and cristae. These wear-dependant characters are not tooth specific, they tend to be unfailingly distributed across almost all the teeth that exhibit that particular character state. For example, the parastyle is subject to wear on the P3, P4, M1 and M3; cingula are subject to wear on the upper canine through the M3, and the m2 (table 2). These wear patterns are directly correlated with occlusal surfaces of the parastyle: the P4 parastyle occludes with the posterior crest on the lower canine; the M1 parastyle occludes with m1 protoconule; the M2 parastyle occludes with m2 protoconule; and the M3 parastyle occludes with the m3.

The wear of the cingula is not as clear cut. Only two of the four cingula (on a premolar, upper molar, or lower molar) are occlusal surfaces: the anterior and posterior cingula, which occlude with the protocones and protoconids, respectively. Wear of the labial and lingual cingula among erinaceids must be a result of diet (or “bug wear,” sensu D. R. Frost). Hedgehogs have a varied diet, including: insects, snakes, eggs, small mammals, and small lizards (Lui, 1937; Krishna, 1956; Brockie, 1959; Burton, 1969; Herter, 1969; Campbell, 1973; Roberts, 1977; Merrit, 1981; Maheshwari, 1984; Corbet, 1988; also see Reeve, 1994 for a complete review), all of which can be abrasive to teeth. Hedgehogs are also known to dispatch relatively large invertebrates using their molars (Reeve, 1994; pers. obs.), instead of tearing with their incisors, or even canines. This observation would explain the wear of the labial cingulum recorded in this analysis.

DDA 4: Covariation of Polymorphism with Geographic Location

Only one taxon, Echinosorex (fig. 3), exhibited dental variation (presence/absence of I1) that conclusively covaried with geographic locality (table 2; see also Gould, 1997, appendix 4, fig. 1).

Echinosorex is the largest member of the Erinaceidae, and as well, the largest of the living lipotyphlans (Frost et al., 1991). Its known distribution extends throughout the Indonesian Peninsula and the Malayan Archipelago (Lim, 1967), to include the islands of Burma, Sumatra, Malaya, Thailand, and Borneo. The genus Echinosorex has been previously thought to contain at least three [sub]species: E. dealabatus, E. alba and E. gymnura (Corbet, 1988). I bracket the [sub] as these taxonomic designations have not been consistent. Recent revisions of the taxonomy of Erinaceidae considered Echinosorex to be a monospecific taxon (Corbet, 1988; Frost et al., 1991). In this analysis, the absence of the I1 seems to be apomorphic for the population in Borneo, although this is based on a review of only five specimens.

Brief Synopsis

Polymorphism and asymmetry were discovered to be quite common across all of the taxa reviewed in this analysis. Of the 246 transformation series considered, 204 (83%) were found to exhibit intraspecific variation. Of all the polymorphism exhibited, very little could be attributed unequivocally to any of the variables (age, sex, geography) considered in this analysis. This does not suggest that the discrete dental characters do not covary with these variables (except perhaps for sexual dimorphism); it simply suggests that it is very difficult to discern covariation from random individual variation.

Hylomys suillus (fig. 4) and Erinaceus amurensis (fig. 6) have the least amount of polymorphism (and asymmetry) relative to all of the taxa reviewed, with Atelerix albiventris (fig. 5) exhibiting the most. The relatively low frequencies of polymorphism in the two above mentioned taxa may be attributable to small sample sizes: 16 and 11 individuals, respectively.

Phylogenetic Analysis 1: Data Set 1

Analysis of the 100 transformation series from the discrete dental analysis and 10 taxa discovered six trees of 105 steps, with the following statistics (excluding uninformative characters): CI = 0.634; RI = 0.528; and RC = 0.357; the strict consensus and the Adams tree are depicted in figs. 9a and b, respectively. In all the trees discovered, every polytypic genus is rendered paraphyletic (except Erinaceus), and the monophyly of both extant subfamilies is challenged (compare with figs. 1a and b).

Phylogenetic Analysis 2: Data Set 2

Analysis of Frost et al.'s (1991) dental data across the 10 taxa reviewed in this investigation discovered 4 trees: length 38; CI = 0.816, and 0.80 (excluding uninformative characters); RI = 0.897; RC = 0.732. The strict consensus tree and the Adams tree are the same (fig. 9c). The only genus discovered to be monophyletic is Erinaceus both subfamilies are rendered paraphyletic.

Analysis of Frost et al.'s (1991) dental data and the 19 living taxa they considered discovered 9 trees: length: 41; CI = 0.756 and 0.744 (excluding uninformative characters); RI = 0.917; RC = 0.694. The strict consensus and Adams tree are illustrated in figs. 9d and e, respectively: hylomyine monophyly is challenged; Hylomys is never discovered to be a member of that group. Moreover, monophyly of all the living genera is suspect except for Erinaceus

Brief Synopsis

The recovered trees for all three analyses differ in overall topology both among themselves and with the hypotheses posited by Frost et al., (1991) and Gould (1995; figs. 1a and b). In each of the discovered topologies, all the taxa were rendered paraphyletic. Interestingly, the most parsimonious trees discovered in all three analyses were seemingly well supported, as evidenced by the high indices.

The apomorphy lists for two discovered trees are presented in appendices 5 and 6 with their respective data matrices. In all three analyses, Tree #1 was selected as the token topology from which to generate an apomorphy list (the strict consensus trees for the analyses are depicted in figs. 9a, c, and d). This arbitrary decision was based on the fact that not one of the discovered trees remotely approximates any of the previously posited phylogenetic hypotheses that are based on all available morphological data (see Corbet, 1988; Frost et al., 1991; Gould, 1995).

The purpose of this analysis is not to propose a phylogenetic hypothesis, but to explore the phylogenetic resolving power of the discrete dental characters. Given the incongruous results with the most recent hypotheses of extant erinaceid relationships (Corbet, 1988; Frost et al., 1991; Gould, 1995), detailed discussion of character support is foregone.