Species delimitation and recognition in the Pediomelum megalanthum complex (Fabaceae) via multivariate morphometrics

Abstract Pediomelum is a genus endemic to North America comprising about 26 species, including the megalanthum complex, which consists of Pediomelum megalanthum and its varieties retrorsum and megalanthum, Pediomelum mephiticum, and the recently described Pediomelum verdiense and Pediomelum pauperitense. Historically, species of the megalanthum complex have been variably recognized at the species or variety levels, dependent upon the relative importance of morphological characters as diagnostic of species. Ten quantitative morphological characters regarded as diagnostic at the species level were analyzed using multivariate morphometrics across these taxa in order to examine the discriminatory power of these characters to delineate species and to aid in species delimitation. The analyses support the recognition of Pediomelum megalanthum, Pediomelum mephiticum, and Pediomelum verdiense at the species level, Pediomelum retrorsum as a variety under Pediomelum megalanthum, and suggest the sinking of Pediomelum pauperitense into Pediomelum verdiense. The findings of the present study help quantify the power of certain characters at delimiting taxa and provide a basis for taxonomic revision of the Pediomelum megalanthum complex.


Species delimitation and recognition in the Pediomelum megalanthum complex (Fabaceae) via multivariate morphometrics
Ashley N. Egan 1

Introduction
Pediomelum Rydb. (Psoraleeae; Leguminosae) includes about 26 species, all native to North America (Rydberg 1919). Th e genus radiated recently and rapidly, with diversifi cation shifts taking place within the last 2 mya, likely due to the impacts of Pleistocene glaciations (Egan and Crandall 2008a). Th is recent radiation is illustrated in the shallow branch lengths within phylogenies, relative to related genera within the Psoraleeae tribe, even phylogenies based on a combined dataset of eight DNA markers (Egan and Crandall 2008b). Th is is also refl ected in the disparate taxonomic views among botanists, with several taxa variably recognized at specifi c or varietal levels. Th e most taxonomically contested group of species lie within subgenus Disarticulatum sensu Grimes (1990), which includes 10 species, with most variably restricted to areas of Texas and the deserts of the southwest U.S. An example of these contrasting taxonomic views can be found within the subspecifi c classifi cation of P. megalanthum (Wooton & Standl.) Rydb. Some botanists have recognized this as having three varieties (e.g. Grimes 1990;Isely 1998): var. megalanthum is found mostly in Uintah, Grand, and San Juan counties down the eastern border of Utah as well as neighboring counties in Colorado; var. epipsilum (Barneby) Grimes is endemic to the Dixie Corridor of Kane Co., UT and neighboring Coconino Co., AZ; var. retrorsum (Rydb.) Grimes is restricted to Nye, Lincoln, and Clark Co, NV, Mohave, Coconino, and disjunct in Graham Co AZ. Other botanists treat each as a separate species (e.g. Rydberg 1919;Welsh et al. 1993). Grimes (1990) favors varietal ranking based on overlapping qualitative and quantitative morphological characters, with no clear diagnostic features enabling species distinction. Isely (1998) even throws P. mephiticum (S. Watson) Rydb. into the 'megalanthum-mephiticum-epipsilum' complex. On the other hand, Welsh et al. (1993) utilized directionality of pedicel and peduncle hairs (P. retrorsum Rydb. vs. P. megalanthum) and absence of pubescence on the upper leaf surface (P. epipsilum (Barneby) S.L.Welsh) as diagnostic characters to maintain their distinctiveness.
Given the narrow distributions of these taxa and the disparate views on the usefulness of certain morphological characters for species delimitation in this group, many might see this as an example of the age old war between lumpers, those who tend to recognize fewer species, often allowing considerable breadth of morphological variation as inherent in species concepts, and splitters, those who split species based on more minute morphological diff erences, sometimes down to the population level. Hewitt C. Watson eloquently exemplifi ed this war in a letter to Charles Darwin dated 13 August 1855 wherein he wrote "Taking J. D. Hooker & Jordan as representative men for the opposite factions in botany,-'lumpers & splitters', the former would reduce the species of vascular plants to three score thousand, or perhaps much fewer;-while Jordan would raise them to three hundred thousand." (Darwin Correspondence Database, 25 Sept 2014).
Whilst preparing the treatment of Pediomelum for the Flora of North America, I was confronted with the question of where I lie on the spectrum of lumpers vs. split-ters, for P. megalanthum and its varieties, and in particular, in regards to two recent species described by Welsh and Licher (2010): P. pauperitense S.L.Welsh, Licher, & N.D. Atwood, described based on six collections, and P. verdiense S.L.Welsh & Licher, described from four specimens. Th ese species are said to diff er from P. mephiticum in the directionality of pedicel and peduncle hairs as well as the shape of the lateral and upper calyx teeth. Th e species are said to diff er distinctly from each other in the size of pedicels, bracts, seeds, and fl ower lengths, as well as fl ower color and peduncle length. Th ey join the ranks of what I refer to as the P. megalanthum complex in subgenus Disarticulatum.
Th e rapid radiation of the genus coupled with the variable recognition of specifi c or varietal ranking and the contrasting opinions of the relative discriminatory power and usefulness of several morphological characters invites an approach using multivariate morphometric analysis as a means of delimiting species or varieties among contested taxa. Th is is particularly so in the case of the newly described P. pauperitense and P. verdiense. Th is work will additionally examine the relative power of certain quantitative morphological characters historically used to delimit species in this group. Here, I aim to objectively delimit species within the P. megalanthum complex using a multivariate morphometric approach, incorporating the morphological diversity of P. megalanthum (vars. megalanthum and retrorsum), P. mephiticum, P. verdiense, and P. pauperitense. I have chosen to recognize P. epipsilum at the specifi c level (discussed below) and so it is not included in these analyses.

Methods
Plant material -Twenty-seven herbarium specimens from Utah and Arizona (deposited at ARIZ, ASC, BRY, and US) were chosen for study and tentatively identifi ed as P. mephiticum, P. megalanthum var. megalanthum, P. m. var. retrorsum, P. pauperitense, and P. verdiense. A list of the specimens included in the morphometric analyses, with voucher information and origin, are given in Table 1. Eff ort was made to obtain all identifi ed specimens of P. verdiense and P. pauperitense through a request to ARIZ, ASC, and BRY for such, especially those listed in Welsh and Licher (2010). However, some specimens were not available for loan. Because of this, comparatively few specimens for these taxa were available. So as not to over-weigh the analyses with specimens from P. mephiticum and P. megalanthum and its varieties, and thereby introducing sampling bias, a comparable number of these specimens were used as well, resulting in a set of specimens smaller than that used in some other studies. However, trends were still very visible in these data. Most specimens were collected or annotated by botanists having authority on the genus, including S.L. Welsh, J.W. Grimes, A.N. Egan, M. Licher, and N.D. Atwood, and taxonomic identifi cations were initially accepted according to the most recent annotation on the specimens. Th ose recognized at the specifi c vs. subspecies levels with the same epithet were analyzed together, i.e. P. retrorsum was grouped in analyses with P. megalanthum var. retrorsum.  Characters scored -Ten quantitative morphological characters (Table 2) were scored. Characters were chosen based on those quantitative traits widely used in fl ora and monographic works to distinguish species, particularly fl ower and calyx characters, as well as those characters used by Welsh and Licher (2010) as diagnostic of species. Other studies have used a similar number of characters for morphometric analysis within species complexes with positive information content (e.g. Kaplan and Marhold 2012;Lihová et al. 2004). Characters were scored for between two and fi ve sites per specimen using digital calipers. While there are a few qualitative characters -such as direction of pedicel and peduncle hairs -that are used by some to diff erentiate taxa (i.e. P. megalanthum varieties), these were not included here. Diff erent researchers have debated the relative utility and importance of directionality of hairs as diagnostic of species or lower-level taxa (see discussion above). I chose explicitly not to include these characters because my focus is on the use of quantitative (continuous) morphological traits with the aim to determine whether quantitative traits can separate taxa along similar lines as some previous researchers do for species vs. subspecies based on the qualitative character of vestiture (e.g. P. megalanthum varieties).
Multivariate morphometric analyses -A combination of multivariate analyses and hierarchical clustering were employed to investigate species limits in this group. All statistics were computed in the statistical package JMP v. 11.1.1 (SAS Institute Inc., Cary, NC). As an initial step, correlation coeffi cients were computed on the total dataset and on each species' dataset to reveal any highly correlated character pairs that may distort downstream analyses. In addition, departure from a normal distribution for each character within each species was tested using the Shapiro-Wilk goodness of fi t test (Shapiro and Wilk 1965).
Morphometric multivariate analyses were conducted on values of individual measurements without averaging across multiple observations per specimen. Th is was done because of the limited number of specimens available for P. pauperitense and P. verdiense. Use of all observations both within and across specimens will likely provide a better view of the intraspecifi c variation within a character. Th is is akin to Pedersen's (2010) justifi cation for using values measured on each individual plant as opposed to using a population mean.
A hierarchical cluster analysis (HCA) was performed to investigate how specimens would group based on overall morphological similarity using Ward's minimum variance method with the data standardized by standard deviation (Ward Jr 1963). Th is method groups specimens by minimizing the increase in the error sum of squares upon each addition of a cluster. Because the use of multiple methods is recommended to ascertain the robustness of clusters (Marhold 2011), UPGMA (unweighted pair-group method using arithmetic averages) with data standardized by standard deviation was also employed (Sokal and Michener 1958).
Principal component analysis (PCA; Sneath and Sokal 1973) was used to delineate patterns of morphological variation across the P. megalanthum complex. Th is method is a good fi rst tool for investigating overall patterns in morphology as each character is weighted the same. Th is was fi rst applied to the complete dataset with all characters and species included. For greater resolution among the main groups, PCA was then conducted on two subgroups: i) P. mephiticum, P. verdiense, and P. pauperitense; ii) P. m. var. megalanthum and P. m. var. retrorsum.
Canonical discriminant analyses (CAN) were employed to investigate the spread of means across each species group and determine how well the characters (Y) predicted the separation of species based on means. Th is method measures the distance of each point from the centroid, or multivariate mean, of its group as defi ned previously by species or subspecies. Th e distance measure is based on the Mahalanobis distance, which incorporates the variances and covariances between variables. CAN is classically implemented using a linear method, which assumes that Y variables are normally distributed with the same variances and covariances, or a quadratic method in which covariances can be diff erent across groups. Because not all of the character distributions were normal, I employed a regularized, compromise method, which is a mixture between the linear and quadratic methods (Friedman 1989). Th e regularized method incorporates two parameters: lambda deals with the shrinkage to a common covariance and ranges from 0 = quadratic to 1 = linear; gamma deals with the shrinkage to diagonal and ranges from 0 = no shrinkage to 1 = diagonals only. A low gamma is suggested when variables are correlated. Here I used a lambda proportional to the number of non-normal character distributions by species (λ = 0.8) and a gamma of 0.
Among the multivariate analyses employed here, several have been used by other researchers to determine those variables causally impacting the separation of species. As a variable reduction technique, PCA helps to discern which characters are responsible for grouping individuals. However, it does not assume an underlying causal model. For determining those characters responsible for delineating species, factor analysis may be a more appropriate method as this technique makes the explicit assumption of an underlying causal model (Jolliff e 2005;Suhr 2005). Factor analysis was performed using principal components and the varimax rotation (FAPC) as well as the maximum likelihood framework with varimax rotation (FAML). Similar to PCA, CAN reduces the variable space to those discriminants that are responsible for assigning individuals to previously defi ned groups, and is often employed to determine those variables with the most discriminatory power (e.g. Koch et al. 2013;Semple and Chmielewski 1987). Stepwise discriminant analysis (SDA) is a method designed to order variables by discriminatory power, adding variables in a stepwise fashion according to the amount of correlation of the variable to the reduced eigenvectors. SDA was compared to results from PCA, CAN, and factor analyses to provide a robust investigation into which of the morphological characters are most informative at delineating species or taxa.
Each analysis was conducted on a series of three data sets or levels: (1) the fi rst dataset included all species, (2) the second dataset includes only data from P. mephiticum, P. verdiense, and P. pauperitense (the MVP group), (3) the third data set includes the P. megalanthum varieties, P. m. var. megalanthum and P. m. var. retrorsum. Datasets are thus notated by the type of analysis and the dataset used such as CAN1, HCA2, PCA3, and so on.

Results
Following Kaplan and Marhold (2012), a cutoff value of 0.90 for the correlation between characters was used to ascertain exclusion of characters from the data analyses. Correlation coeffi cients did not exceed 0.90 for any character pair within any species, and so all characters were used in subsequent analyses. Only three character pairs across all species had correlation coeffi cients above 0.8 in either the positive or negative direction: P. verdiense had a correlation coeffi cient of 0.8775 for calyx:lower calyx tooth; P. m. var. retrorsum exhibited a negative correlation of -0.8108 for pedicel:bract; P. pauperitense exhibited a correlation of 0.8783 for fl ower:bract. For all species, the correlation between calyx:lower calyx tooth was between 0.6 and 0.8, an expected result considering the overlapping nature of these characters. However, inclusion of both incorporates the plasticity of calyx morphology into the overall analysis -a key character used for species delimitation -and thus all characters are kept for further analyses.
Ward's cluster analysis of all specimens (HCA1) produced a dendrogram with two main groups: one comprised of entirely P. mephiticum, P. verdiense, and P. pauperitense with the exception of a single P. m. var. megalanthum data point, and the other group comprised of three subgroups, two comprising mixtures of the two P. megalanthum varieties and one comprised mainly of P. verdiense (Fig. 1). UPGMA cluster analysis (data not shown) produced three main clusters, two comprised of a mixture of P. megalanthum varieties, one with a few P. verdiense or P. mephiticum specimens included, and one comprised wholly of the P. mephiticum-verdiense-pauperitense complex (hereafter denoted as the MVP group), again with exception of the single P. m. var. megalanthum data point. Considering the strong support for the MVP group, a separate hierarchical analysis was conducted on the MVP group only (HCA2). Th is analysis suggests two main clusters, one cluster almost entirely of P. mephiticum data points with two P. verdiense data points included therein, and a second main cluster mostly comprised of P. verdiense and P. pauperitense with three P. mephiticum data points scattered throughout (Fig. 2). A cluster analysis of the P. megalanthum varieties (HCA3) showed no clear division between taxa (data not shown).
Th e ordination diagram from the principal component analysis based on all specimens (PCA1; Fig. 3) also suggested two main groups. Specimens of P. mephiticum, P. verdiense, and P. pauperitense were separated from the P. megalanthum varieties along the fi rst axis with all characters contributing to this division. Th e second axis separated P. mephiticum from a mixture of P. verdiense and P. pauperitense but did not separate the P. megalanthum varieties from each other. Floral characters (fl ower, calyx, calyx tube, lower calyx tooth, pedicel) vs. vegetative characters (peduncle, petiole, bracts, stipules, and leafl ets) separated P. verdiense and P. pauperitense from P. mephiticum along the second axis, with vegetative characters contributing more to diff erentiation along the second component ( Fig. 3; Table 5).
Independent principal component analyses were also conducted on the two main subgroups defi ned by PCA1. PCA2, comprising the MVP group, showed good separation along the fi rst axis of P. mephiticum from P. verdiense and P. pauperitense, with fl oral vs. vegetative characters strongly affi liated with this break (Fig. 4A). However, P. verdiense and P. pauperitense created a mixed group with most of the P. pauperitense data points clustering below the second axis (capturing 20% of the variation in the data) amidst P. verdiense, suggestive of a lack of diff erentiation normally found between species. Th e same result is found in PCA3, the analysis of the diff erentiation between P. m. var. megalanthum and P. m. var. retrorsum, illustrating a mixture of P. m. var. megalanthum and P. m. var. retrorsum data points (Fig. 4B). Contributions of characters to each multivariate axis for each of the three PCA analyses are listed in Table 4.    Table 5.
Canonical and classifi catory discriminant analyses were also conducted to investigate the spread of means per species group. In accordance with HCA and PCA analyses, CAN1 shows P. mephiticum largely distinct from the others, with the two P. megalanthum taxa creating one group while P. verdiense and P. pauperitense another (Fig.  5). Th e inner circles of each species represent the 95% confi dence region for the true mean (of all characters taken together) of the species. Th e 95% confi dence region for P. verdiense and P. pauperitense are overlapping, as are those of P. m. var. megalanthum and P. m. var. retrorsum, suggesting that the overall mean for these species based on all characters is not statistically diff erent.
Factor analyses and PCA were taken together to help elucidate those characters most infl uential in separating predefi ned groups based on expert identifi cation (Table 5). For  Table 5. . Th e second factor or component axis largely separates P. mephiticum from P. verdiense and P. pauperitense and is most strongly associated with various vegetative characters (Table 5). Th is same association between discriminatory power, axes, and characters follows into analyses conducted on the MVP group only (FAPC2, FAML2, and PCA2),   megalanthum, + P. megalanthum var. retrorsum. Inner circles by group are the 95% confi dence region for containing the true overall mean of the group; the outer circles by group are the normal 50% contours, the normal ellipse region that contains 50% of the population for each group. Rays show the coordinate directions in canonical space. Table 6. Rank order of relative discriminatory power of characters for distinguishing species as ascertained by stepwise discriminatory analysis. F-ratio and probability are calculated based on the stepwise inclusion into the set of characters ranked previously. that diff ered across analytical method (Fig. 5).

All specimens MVP only P. megalanthum varieties rank Character F-ratio Prob>F Character F-ratio Prob>F Character F-ratio Prob>F
Stepwise discriminatory analysis showed fl ower length to be the only signifi cant trait off ering some measure of discriminatory power between P. m. var. megalanthum and P. m. var. retrorsum (SDA3; Table 6) with all other characters not signifi cant. For MVP only, SDA2 determined that bracts were the most important factor in discriminating between species, followed by fl ower length, pedicel length, peduncle, calyx tube, and stipules (Table 6). Scatterplots showing the fi t of peduncle × calyx tube lengths (Fig. 6A) and bract × fl ower lengths (Fig. 6B) show the range and distinguishing power of these traits for MVP. For all species analyzed together, SDA1 calyx tube length was strongly correlated with separation of groups, followed by bract and fl ower lengths; leafl ets, pedicel, peduncle, and stipules were also associated with discriminatory power across these groups, but less so (Table 6).

Discussion
Species recognition often relies on deciphering nonoverlapping patterns in morphology between biological entities (Davis and Heywood 1963;Mayr 1942), with these gaps in morphology used as a means of delimiting species or lower taxa. Recognition of taxa at the specifi c vs. varietal or subspecifi c level can often be a diffi cult choice to make, especially for plants. Ideally, subspecies or varieties should be characterized by some cohesive trait along side morphology, such as geography, ecology, or phylogenetic traits (Hamilton and Reichard 1992). Often this is exacerbated by a disagreement concerning the relative importance of various morphological characters as being diagnostic of species or varieties or the lack of a clear supporting character not of the morphological type. Th is battle is evident in the genus Pediomelum, particularly for species of the intermountain west.
Relationships among the species or varieties of the Pediomelum megalanthum complex have been debated among botanists, largely due to diff ering opinions as to which morphological characters are most important for distinguishing species. In his key to species of Pediomelum, Rydberg (1919) emphasized the pubescence of the peduncle (hairs appressed vs. hairs spreading or retrorse) as diagnostic between P. megalanthum and P. mephiticum or P. retrorsum. Welsh et al. (1993) followed Rydberg (1919). Ockendon (1965) commented extensively on the P. megalanthum complex, suggesting that this group of species required an extensive review coupled with reproductive studies to truly decipher amongst species. He considered the pubescence character to be useful, but recognized that use of pubescence on the peduncle could be problematic, favoring that of the petioles as being more diagnostic and leading to his recognition of Psoralea megalantha Wooton & Standl. and Ps. mephitica S.Watson, with specimens relegated to retrorsum being subsumed under Ps. mephitica. Ockendon (1965) also recognized the potential of fl ower size as a distinguishing character and suggested that new species might be recognized in the future. Ultimately, Ockendon (1965) concluded that "Rather than altering the taxonomy of this group in a piecemeal fashion, it seems best to wait until a more unifi ed treatment is possible." Grimes (1990) took up Ockendon's challenge and produced the most recent treatment of the genus, taking a more quantitative approach emphasizing fl ower and calyx lengths. Th is relative increase in the importance of fl ower size as delimiting between taxa in this species group admitted less variability in both P. mephiticum and P. megalanthum, these being distinguished by those plants having calyx tubes 4.5 mm or shorter vs. those 6 mm or longer, respectively. Th is shifted specimens previously referred by Ockendon (1965) to Ps. mephitica being now recognized as a variety of P. megalanthum due to fl ower size and other overlapping quantitative characters (see Table 5 of Grimes 1990).
Th e morphometric analyses conducted herein largely support the use of fl ower and calyx sizes as being useful characters for species delimitation. All level 1 morphometric analyses involving all species in the complex, P. mephiticum, P. verdiense, P. pauperitense, and the two varieties of P. megalanthum, illustrated a clean break in overall morphological variation between P. mephiticum, P. verdiense, and P. pauperitense (the MVP group) and the P. megalanthum varieties (HCA1, PCA1; Figs. 1, 3). Canonical discriminant analysis (CAN1; Fig. 5) also illustrated a strong break between the MVP group and the P. megalanthum varieties, with no overlap in overall species means. Flower size, calyx, and calyx tube were most strongly associated with the fi rst canonical axis, while calyx tube and bracts showed the strongest association with the second canonical axis.
In most analyses, fl oral characters separated species with the greatest discriminatory power along the fi rst component or axis of the analysis, whereas the suite of vegetative characters contributed more to the second component divisions ( Fig. 3; Table  5). Th ese results are not unexpected, as most current species determinations were based on Grimes ' (1990) criteria of fl ower and calyx morphology as being most distinguishing of species. Th e stepwise discriminatory analysis for all taxa (SDA1) suggested that the strongest discriminatory character amongst taxa was calyx tube length (Table 6), a fi nding in line with most researchers who use this character in dichotomous keys to separate species (Grimes 1990;Isely 1998;Welsh et al. 1993) and is here largely responsible for the separation of the megalanthum varieties from the MVP group. Bract size is the second most discriminatory character (Table 6), responsible largely for the separation of P. mephiticum from P. verdiense + P. pauperitense. Flower size is the third most discriminating character for distinguishing amongst the previously determined species groups.
Th e distinction of large and small fl owered forms within the megalanthum complex has been recognized by previous researchers (Grimes 1990;Isely 1998). Th e larger-fl owered group comprising the P. megalanthum varieties has been another battleground for taxonomists in this group. Pediomelum m. var. retrorsum has variably been recognized at the varietal level (Grimes 1990;Isely 1998), at the specifi c level as P. retrorsum (Rydberg 1919;Welsh et al. 1993), or included as either Psoralea megalantha or Ps. mephitica (Rydberg 1919), depending on the relative importance of fl ower size vs. pubescence type and direction of peduncular and petiolar hairs deemed by the researcher. Th at said, the analyses herein do not support a clear distinction between species designated as P. retrorsum vs. P. megalanthum based on the quantitative characters employed, by either principal component analyses (PCA1,PCA3;Figs. 3,4B) or canonical discriminant analysis (CAN1; Fig. 5). In fact, all of the ten characters employed here present overlapping character ranges (Table 3), a fi nding illustrated best by the canonical discriminant analysis that shows strong overlap in the 50% normal contours, the normal ellipse region (outer circle) that contains 50% of the species' quantitative morphological diversity based on the 10 characters included here (Fig. 5). In addition, there is an overlap, albeit slight, in the 95% confi dence region between megalanthum and retrorsum, suggesting that the overall morphological means are not statistically diff erent. Grimes (1990) presented a comparison of 14 quantitative characters across the three varieties of P. megalanthum and concluded something very similar to these fi ndings: "…the varieties overlap in most qualitative and quantitative characters, and the diagnostic characters for all three varieties overlap so much as to make specifi c status untenable." (see Table V of Grimes 1990: 81).
Th at said, there is some quantitative morphological and geographic separation evident between P. m. var. megalanthum and P. m. var. retrorsum. Th e varieties are fairly distinct geographically, with megalanthum primarily of eastern Utah, western Colorado, and northwest New Mexico and retrorsum of southern Nevada, northwestern Arizona, and sporadically along the Gila River drainage elsewhere in Arizona. Flower length was the only character having signifi cant discriminatory power in the stepwise discriminant analysis (SDA3; Table 6), suggesting that fl ower length may be the only causal quantitative character off setting the separation of the population normal contours in the canonical discriminant analysis. Th is suggests that perhaps there is ongoing diff erentiation among the megalanthum varieties, perhaps spurred by geographic separation that may in time lead to species diff erentiation. Th e lack of quantitative separation from each other argues against recognizing these taxa as separate varieties and instead lumping them under P. megalanthum. However, some qualitative diff er-ence in the directionality of prevailing hair types and geographic separation exists, providing some justifi cation for recognition and separation at the varietal level. In an eff ort to favor tradition and lessen the upset to prevailing taxonomic concepts, I am in favor of recognizing these taxa at the varietal level as P. megalanthum var. megalanthum and P. megalanthum var. retrorsum, largely following the concept of Grimes (1990).
I agree with Grimes' conclusions to a point with the recognition of P. m. var. megalanthum and P. m. var. retrorsum. However, after careful comparison of the character ranges of P. m. var. epipsilum with the others as ascertained by Grimes, I fi nd suffi cient distinguishing characters that separate P. epipsilum from the other varieties, including having leafl ets smaller and glabrate above or sparingly strigose along veins (vs. leafl ets larger and pubescent above and below in vars. megalanthum and retrorsum), and bracts larger and caudate (vs. smaller, acuminate, acute, or shortly caudate in the others). Indeed, Grimes' quantitative character comparison shows non-overlapping ranges for leafl et and bract size, separating P. epipsilum from the others. Barneby (1943), in his description of the species as Psoralea epipsila Barneby, states that it differs from Ps. mephitica by its caulescent nature and conspicuous bicolored leaves and that it is intermediate between Ps. mephitica and its variety Ps. m. var. retrorsa (Rydb.) Kearney & Peebles in fl ower size. He also states that the banner is barely exserted from the calyx, whereas the others are well exserted. Th is preponderance of both quantitative and qualitative diff erences, coupled with the fact that P. epipsilum is set apart in phylogenetic analyses (Egan and Crandall 2008b), lead me to recognize P. epipsilum at the species level, as others before have also done (Barneby 1943;Welsh et al. 1993).
During an examination of Pediomelum in Arizona, Welsh and Licher (2010) discovered some collections that did not key out well based on fl ower size and peduncular pubescence. All these off ending specimens had small fl owers and ascending hairs on pedicels and peduncles, a character combination at odds with the prevailing concepts of P. megalanthum and P. mephiticum. Th ose plants having fl owers with a banner that is purple or white suff used with purple that is not strongly contrasting with the wings or keel in color and found mainly from the Verde Limestone Formation were described as P. verdiense whilst those plants from near Poverty Mountain having a banner and wings of white or cream that contrasts strongly to the purple color of the keel, with leaves that tend to exceed infl orescence in height were described as P. pauperitense. For comparison, P. mephiticum has a white or cream banner with wings and keel purple or white suff used with purple and is present in the extreme northwest corner of Arizona and adjacent areas in Utah and Nevada.
Furthermore, Welsh and Licher (2010) stated that P. verdiense corresponded to those plants with pedicels 3-3.5(-5) mm long, bracts 5-8 mm long; fl owers 10-11.3 mm long whereas P. pauperitense corresponded to those plants with pedicels 1.5-2.5(-3) mm long, bracts 3-5 mm long; and fl owers 7.3-10 mm long. Th ese ranges suggest a clean break between these species and were used to key out specimens. However, my examination and measuring of specimens used in this study, many of which were also cited by Welsh and Licher or were paratypes thereto, gave a very diff erent picture (Table 3). In fact, of these characters, my measurements for P. pauperitense pedicel and fl ower length were completely non-overlapping with the ranges suggested by Welsh and Licher (pedicels 3-4 mm and fl owers 10.7-12.2, as compared to those above). Th is made me question the authenticity of these species.
Th e multivariate morphometric analyses on the MVP group only (level 2 analyses) were very telling. Th ere seems to be a distinct separation between P. mephiticum and the other two species, as evidenced by all methods applied herein, but strong overlap between P. verdiense and P. pauperitense. Th is is perhaps best illustrated by the hierarchical cluster analysis (HCA2) which shows two main clusters, one cluster almost entirely of P. mephiticum and a second main cluster mostly comprised of P. verdiense and P. pauperitense (Fig. 2).
As in the case with the P. megalanthum varieties, canonical discriminant analysis (Fig. 5) shows strong overlap in the 50% normal contours as well as a large overlap in the 95% confi dence region between P. verdiense and P. pauperitense, suggesting that the overall morphological means are not statistically diff erent. Lastly, principal component analyses (Figs. 3,4A) showed separation of P. mephiticum from the others along the fi rst component. Factor analysis suggested that fl ower, calyx tube, peduncle and bracts contributed most to the separation along the fi rst axis. A linear regression of peduncle vs. calyx tube (Fig. 6A) and of bracts vs. fl ower (Fig. 6B) supports the distinction between P. mephiticum and P. verdiense+P. pauperitense and illustrates the lack of distinction between the latter two taxa. Th is is evident in the overlapping 95% confi dence bands between P. verdiense and P. pauperitense in both linear regressions, with no overlap with P. mephiticum. SDA2 suggests that bracts are the most distinguishing character in the MVP group, followed by fl ower, pedicel, peduncle and calyx tube in rank order (Table 6).
Given the overlap in continuous character distributions between several species or taxa in this study, some researchers may invoke hybridization as one reason behind overlapping morphology. Traditionally, hybridization is said to create morphological intermediacy (Anderson 1949). However, several studies have shown that hybridization does not always result in morphological intermediacy, but that it can, in fact, produce parental and even novel morphological characters or combinations (e.g. Rieseberg 1995;Rieseberg and Ellstrand 1993). Furthermore, the use of multivariate morphometric techniques for detecting hybridization has been called into question, as these methods cannot distinguish between divergence and hybridization (Wilson 1992). With these caveats in mind, the presence of hybridization within or between taxa in this complex cannot be proven nor ruled out. Indeed, it is possible that P. verdiense or P. pauperitense, or any of the species in this group, could be hybrids involving one or more of the other species in the complex. However, this study cannot address this at this time. Th e role of hybridization in this complex may best be addressed using molecular or genomic methods spanning the species and population levels.
Taken together, the results of this study argue for the recognition of P. mephiticum and P. verdiense at the specifi c level, but do not support P. pauperitense as its own species. Some researchers might suggest that P. pauperitense be recognized as a variety of P. verdiense based on geographic separation, diff erences in peduncle length relative to petiole length, or fl ower color. However, given the few numbers of populations and specimens relegated to P. verdiense and P. pauperitense, I deem it premature to make this distinction, especially considering the lack of any non-overlapping quantitative morphological character to justify this separation. Now, with all this said and done, I revisit the initial question posed to myself: where do I lie on the spectrum of lumpers vs. splitters? Considering my conclusions in the paragraph above, I think me a lumper -at least in the case of Pediomelum. And yet, my initial inclination -prior to this analytical undertaking -was to synonymize both P. verdiense and P. pauperitense under P. mephiticum. Th is exercise convinced me to do otherwise -to recognize P. verdiense at the species level. Th is is more leaning towards a splitter mentality. Th e problem? Not knowing the dimensions of the spectrum! I guess I lie somewhere in the middle…

Conclusions
Given the conglomeration of past research with current fi ndings shown herein, I support the recognition of P. megalanthum as having varieties megalanthum and retrorsum. I also recognize the specifi c status of P. mephiticum. As per the sinking of P. pauperitense under P. verdiense, a new description of P. verdiense is given below, along with a key to the taxa investigated or discussed herein.
Key to the species