PHYLOGENETIC, ECOLOGICAL, AND ALLOMETRIC CORRELATES OF CRANIAL SHAPE IN MALAGASY LEMURIFORMS

Authors


Abstract

Adaptive radiations provide important insights into many aspects of evolution, including the relationship between ecology and morphological diversification as well as between ecology and speciation. Many such radiations include divergence along a dietary axis, although other ecological variables may also drive diversification, including differences in diel activity patterns. This study examines the role of two key ecological variables, diet and activity patterns, in shaping the radiation of a diverse clade of primates, the Malagasy lemurs. When phylogeny was ignored, activity pattern and several dietary variables predicted a significant proportion of cranial shape variation. However, when phylogeny was taken into account, only typical diet accounted for a significant proportion of shape variation. One possible explanation for this discrepancy is that this radiation was characterized by a relatively small number of dietary shifts (and possibly changes in body size) that occurred in conjunction with the divergence of major clades. This pattern may be difficult to detect with the phylogenetic comparative methods used here, but may characterize not just lemurs but other mammals.

Adaptive radiations provide a window into adaptation, phenotypic and taxonomic diversification, as well as the role of ecology and divergent selection in shaping evolution (Guyer and Slowinski 1993; Schluter 1996, 2000; Givnish and Sytsma 1997; Grant and Grant 2008; Losos 2010). Among the best-studied adaptive radiations are the Galapagós finches, Hawaiian silverswords, East African cichlids, and Caribbean anoles. The lemuriforms, a monophyletic clade of strepsirrhine primates endemic to Madagascar, are often thought of as representing an adaptive radiation in an insular environment (Martin 1972; Jungers 1980; Godfrey 1988; Fleagle 1999; Godfrey and Jungers 2002). Malagasy lemurs could therefore provide insights into evolution on islands and adaptation in a morphologically and ecologically variable clade of mammals that range in size from 30 g to ∼160 kg. This study uses phylogenetic comparative methods to investigate how size, activity pattern, and three subtlety different aspects of diet influence lemur cranial form. We particularly focus on diet because correlated ecological and morphological diversification along dietary axes is a feature of numerous vertebrate adaptive radiations; examples include the Galápagos finches (Grant and Grant 2008), New World phyllostomid bats (Freeman 2000), multituberculate mammals (Wilson et al. 2012), and modern whales (Slater et al. 2010).

Investigating how skull shape relates to ecology will also contribute to our understanding of the complex underpinnings of vertebrate skull diversity. Among the important functional roles played by the cranium are food acquisition and processing. Highlighting the ways in which the skull has adapted to different dietary demands is not straightforward, however. For one, the cranium has additional roles to fulfill related to housing sensory organs, respiration, vocalization, display, and defense. Moreover, integration among regions of the skull is likely to result in compromises in morphology that muffle the dietary signal (Lieberman 2011). However, neither the presence of competing functional demands nor the presence of integration negates a priori the possibility that the cranium is influenced by diet.

Empirical evidence supporting the association between diet, biomechanical function, and cranial shape is mixed. A positive relationship between diet and cranial form has been demonstrated for some vertebrates (e.g., Dumont 1997; Freeman 2000; Caumul and Polly 2005; Metzger and Herrel 2005; Van Valkenburgh 2007; Wroe and Milne 2007; Nogueira et al. 2009; Samuels 2009; Goswami et al. 2011; Dumont et al. 2012; McCoy and Norris 2012), including primates (e.g., Ravosa 1990; Anapol and Lee 1994; Koyabu and Endo 2009, 2010). However, some of these studies failed to explicitly incorporate phylogeny into their analyses. Moreover, other studies failed to recover or recovered only a weak relationship between ecological variation, including differences in diet, and cranial form (e.g., McBrayer 2004; Cardini and Elton 2008; Jones and Goswami 2010; Perez et al. 2011).

This work also contributes to a growing body of literature that integrates geometric morphometric data into phylogenetic comparative methods (e.g., Klingenberg and Ekau 1996; Rüber and Adams 2001; Clabaut et al. 2007; Meloro et al. 2008; Morgan 2009; Figueirido et al. 2010; Raia et al. 2010; Klingenberg and Marugán-Lobón 2013).

STUDY SYSTEM

The numerous species of extant and subfossil lemurs present an impressive array of morphological, behavioral, and ecological specializations that contributed to their occupation of the many varied habitats of Madagascar (Tattersall 1982; Godfrey et al. 2006; Gould and Sauther 2006; Mittermeier et al. 2010). The last evaluation of the infraorder Lemuriformes as an adaptive radiation sensu stricto was more than three decades ago (Martin 1972). This study is part of a broader effort aimed at reevaluating the status of lemurs as an adaptive radiation in light of the four criteria for an adaptive radiation outlined by Schluter (2000): (1) monophyly, (2) rapid speciation (particularly early in the radiation), (3) a correlation between environment and phenotype, and (4) demonstrated performance advantage of phenotypic traits in their environment. Although Schluter's (2000) emphasis was on radiations at low taxonomic levels, adaptive radiations can also characterize taxa at higher levels.

Lemur monophyly is well-documented (Yoder et al. 1996; Horvath et al. 2008; Perelman et al. 2011), and lemurs may have undergone a rapid cladogenesis early in their evolutionary history, followed by decreasing speciation rates (Pybus and Harvey 2000), although this pattern should be reevaluated using more recent phylogenetic and taxonomic data as well as information about divergence dates (e.g., Mittermeier et al. 2010; Perelman et al. 2011). Lemurs first populated Madagascar in the late Cretaceous/early Paleocene (Perelman et al. 2011) or early Eocene (Springer et al. 2012), and is one of only six or seven terrestrial mammal orders (depending on whether the “Malagasy aardvark” belongs to Tenrecoidea or its own order Bibymalgasia; MacPhee 1994; Buckley 2013) to occupy the island. The fossil record of lemurs is very shallow, having been recovered from terminal Pleistocene and Holocene deposits spanning only ∼29,000–510 years ± 80 B.P. (Simons et al. 1995; Burney et al. 2004; Crowley 2010). This is presumably because Tertiary deposits are almost entirely absent on Madagascar (Godfrey and Jungers 2002), and lemurs are not represented in the late Cretaceous mammalian fauna of the island (Krause et al. 2006). This study addresses the third criterion (correlation between environment and phenotype), while the fourth criterion (establishing whether particular morphological traits provide a performance advantage) is typically addressed using experimental approaches that relate morphology to fitness.

Within lemurs, there is limited evidence that aspects of the face track dietary variation (Ravosa 1992; Vinyard et al. 2003; Ravosa and Daniel 2010), and the study with the greatest taxonomic breadth suggested a strong relationship between lemur cranial shape (as captured by 2D landmarks) and diet (Viguier 2004). However, none of these studies incorporated phylogeny into the analyses despite the fact that neither diet nor cranial morphology is independent of phylogeny in lemurs. The present study is the first investigation of whether lemur cranial anatomy reflects ecological variation within a phylogenetic framework, and to test three distinct models that relate specific attributes of food to skull morphology.

RELATIONSHIP BETWEEN ECOLOGY AND MORPHOLOGY

Activity pattern

Although orbit size is not a perfect predictor of eye size (Kirk 2006a), there is a tight positive relationship between the two variables. Larger eyes may allow for greater light sensitivity as a larger cornea and pupil increase the amount of light falling on the retina (Kirk 2004, 2006a). This provides a physiological explanation for the empirical observation that nocturnal primates have larger orbits (scaled to cranial length) than cathemeral (activity throughout the diel cycle; Tattersall 1979; Donati et al. 2013) or diurnal primates (Walker 1967; Kay and Cartmill 1977). This relationship also holds within lemurs as indicated by the fact that cathemeral and diurnal lemurs have orbits smaller than predicted by the regression of orbit size on cranial length based on only the nocturnal lemur sample (Jungers et al. 2002). However, when orbit size is instead regressed on body mass, the extant cathemeral and diurnal lemurs straddle the regression line extrapolated from the nocturnal sample. These conflicting results suggest that cranial size is an imperfect, and possibly biased, proxy for body mass (Jungers et al. 2002). Thus, there should be a relationship between activity pattern and cranial shape (related to relative orbit size) in the extant taxa when centroid size (CS) of the cranium is used as a covariate (CS predicts cranial length almost perfectly in our current dataset: R2 = 1.0; P < 0.001), but not when body mass is used as a covariate. Because the subfossil lemur sample fell below the nocturnal regression line regardless of the size proxy used, we expect a relationship between cranial shape and activity pattern when the subfossil lemurs are included regardless of which size proxy is used.

Diet

Diet is a complex concept that can be challenging to characterize as it is influenced by both ecological factors (e.g., habitat and season) and social factors (e.g., group size and intergroup competition). For another, some foods in a diet might influence cranial shape more than others. This study evaluates three distinct hypotheses that implicate subtly different aspects of diet as the primary selective force operating in molding cranial anatomy: (1) typical diet, (2) the presence of mechanically challenging foods in the diet even at low proportions, and (3) selection for great mechanical advantage versus for wide gape.

Regarding the first dietary hypothesis, cranial adaptations may reflect the structural and material properties of the most common elements in a diet such that some measure of the “typical diet” will explain a large proportion of morphological variation. Specifically, taxa that commonly consume hard or tough foods such as seeds, leaves, or bamboo will exhibit adaptations related to (1) greater jaw-muscle activity, (2) higher bite forces, and (3) resisting the higher strains that result from elevated bite forces or repetitive loading. These adaptations may include larger muscle attachment areas, taller and more flaring zygomatic arches, and an anteroposteriorly shorter face. Alternatively, the second hypothesis predicts that the most resistant food item processed/consumed, even if consumed rarely (e.g., “fallback” foods eaten in seasons of low food variety), could be the prime driver of morphological adaptation to diet in primates (Rosenberger and Kinzey 1976; Yamashita et al. 2009). Morphological adaptations in species consuming hard/tough foods even at low frequencies would be similar to those just discussed.

The final dietary hypothesis rests on the idea that great mechanical advantage versus large gape (distance or angle between the jaws in an open mouth) may represent a performance trade-off because morphologies that favor mechanical advantage tend to decrease the potential for achieving wide gapes (Herring and Herring 1974; Dumont and Herrel 2003; Williams et al. 2009; Ravosa et al. 2010). Different diets will likely produce selection for one or the other of these strategies (e.g., small resistant seeds would require great mechanical advantage while large ripe fruits would favor wide gapes), although in a few cases there may be selection for both (e.g., large resistant seeds). Cranial adaptations to a wider gape include a temporomandibular joint (TMJ) closer to the occlusal plane, longer jaws, and an anteroposteriorly elongated glenoid fossa (Herring and Herring 1974; Vinyard et al. 2003).

Materials and Methods

DATA COLLECTION

Landmark acquisition

Thirty-nine three-dimensional (3D) landmarks were acquired with a Microscribe 3DX digitizer (Fig. 1 and Table 1) from all families and most genera of extant and extinct lemurs with the exception of the smallest species (because these were below the resolution of the Microscribe digitizer; Table 2). Most of the subfossil data were collected by WLJ, whereas data for a few subfossil lemurs and the extant comparative sample were collected by KLB. Both authors (WLJ and KLB) collected data from three of the same species. Interobserver error is not expected to impact this analysis as the average intraspecific Procrustes distances for those species including specimens digitized by KLB and WLJ was 0.12 (range: 0.08–0.16). These values are similar to both the intraspecific (average: 0.10; range: 0.07–0.13) and interspecific (average: 0.11; range: 0.8–0.19) comparisons for the subfossil taxa where only one worker recorded data. Importantly, members of the same species overlap and a nested ANOVA found no significant observer effect for the principal components discussed below (the first four components). Small sample sizes within individual species are not expected to directly affect statistical power in comparative analyses such as this because species means are the units of analysis. Although the small sample sizes (which are unavoidable in many cases due to limited subfossil samples available for study) have likely decreased the accuracy of the species mean estimates, we do not expect the degree of error introduced will impact our results greatly (Supporting Information Methods 1).

Table 1. Names, abbreviations, and definitions of the 3D cranial landmarks used to quantify cranial shape and size in this study (also illustrated in Fig. 1)
LandmarkAbbreviationDefinition
NasionNasJunction of frontonasal and internasal sutures
RhinionRhiAnterior-most midline point on nasal bones
NasospinaleNspTip of anterior nasal spine
ProsthionProMidline of palate in same plane as incisors (may be extrapolated)
IncisivionIncJunction of palate midline and the plane of the posterior margin of incisive foramina
MaxillopalatineMpaJunction of intermaxillary and maxillopalatine sutures on inferior surface of palate
StaphylionStaPosterior-most point in midline of palate
BasionBasAnterior-most point of foramen magnum
Superior orbitSorbSuperior orbital margin as defined by a chord perpendicular to Frankfort horizontal
Inferior orbitIorbInferior orbital margin as defined by a chord perpendicular to Frankfort horizontal
FrontomalareorbitaleFmoWhere frontomalare suture passes the inner orbital rim
FrontomalaretemporaleFmtWhere frontomalare suture passes the temporal line (or outer orbital rim)
Inferior lacrimalLacInferoanterior point on the lacrimal fossa
Zygotemporale superiorZtsWhere zygotemporale suture crosses superior margin of zygomatic arch
Zygotemporale inferiorZtiWhere zygotemporale suture crosses inferior margin of zygomatic arch
ZygomalareZgmInferior-most point of zygomaxillary suture
Posterior caninePcanPosterior-most point on canine tooth, projected onto buccal alveolar margin
P4-M1 contactP4M1Point of P4-M1 contact, projected onto buccal alveolar margin
Distal M3DM3Point of distal M3, projected onto buccal alveolar margin
Pterygoid intersectionPterInferior-most point of contact between medial and lateral pterygoid plates
Lateral glenoidLglenLateral-most point on edge of temporal glenoid fossa
PorionPorSuperior-most point on margin of the external auditory meatus
BregmaBrJunction of sagittal and coronal sutures
OpisthionOpPosterior-most point of foramen magnum
InionInPosterior-most point in midline of occipital bone (junction of superior nuchal lines)
Table 2. Variables used in this study organized by species: sample size, body mass, brain mass, cranial centroid size, activity pattern, average diet, biomechanical strategy, and food resistance score
  Size variables  Biomechanical strategy 
TaxonSample sizeBody mass (kg2)Cranial centroid sizeActivity pattern scoreAverage diet categoryGapeMechanical advantageFood resistance Score
  1. a

    These are extinct subfossil species.

  2. b

    H1 refers to the traditional reconstruction of Hadropithecus as a hard-object feeder, whereas H2 is consistent with more recent reconstructions of a diet composed of bulbs/corms of grasses and sedges or leaves/roots of succulent.

Cheirogaleus medius30.1686.510Frugivore100
Mirza coquereli40.3199.970Liquid consumers100
Phaner furcifer30.34102.730Liquid consumers110
Lepilemur leucopus100.61103.070Folivore011
Lepilemur dorsalis10.50108.100Folivore011
Cheirogaleus major30.40110.560Frugivore100
Avahi laniger91.18114.030Folivore011
Lepilemur microdon10.97116.330Folivore011
Lepilemur ruficaudatus20.75116.840Folivore011
Lepilemur mustelinus41.02118.080Folivore011
Hapalemur griseus101.08130.821Folivore011
Eulemur mongoz31.49154.950.5Frugivore101
Eulemur coronatus41.18156.670.5Frugivore101
Lemur catta122.21161.531Frugivore111
Propithecus verreauxi183.49168.721Folivore111
Eulemur fulvus212.13175.890.5Frugivore101
Eulemur macaco31.82176.370.5Frugivore101
Daubentonia madagascariensis42.56186.900Granivore112
Propithecus diadema65.89187.311Folivore112
Mesopropithecus globicepsa211.30188.921Folivore012
Varecia variegata/rubra103.58203.711Frugivore101
Indri indri96.19207.291Folivore012
Pachylemur insignisa311.50216.421Frugivore101
Pachylemur jullyia313.40237.071Frugivore101
Babakotia radofilaia120.70238.351Folivore012
Archaeolemur majoria818.20263.741Frugivore112
Archaeolemur edwardsia1126.50295.471Frugivore112
Hadropithecus stenognathusa(H1)b135.40305.851Granivore112
Hadropithecus stenognathusa(H2)b 35.40305.851Folivore011
Palaeopropithecus kelyusa135.00331.041Folivore012
Palaeopropithecus ingensa141.50360.671Folivore012
Megaladapis madagascariensisa246.50456.261Folivore012
Archaeoindris fontoynontiia1161.20519.371Folivore012
Megaladapis edwardsia385.10566.551Folivore012
Figure 1.

Line drawing of Lemur catta cranium with 3D landmarks. The abbreviations refer to the landmark definitions in Table 1. Although shown only unilaterally, bilateral landmarks were taken on both right and left sides of the cranium.

Landmark configurations for all specimens were superimposed by generalized Procrustes analysis (GPA) in Morpheus (Slice 1998), which removes variation in the data due to differences in translation, orientation, and scale (Rohlf and Slice 1990). The original CSs (the square root of the sum of squared distances from the centroid to all landmarks in a landmark configuration) were retained for subsequent analyses. Aligned specimens are usually projected onto a Euclidean tangent space. However, the correlation between distances among specimens in the “GPA” shape and the tangent space was r = 1.00 (calculation performed in tpsSmall; Rohlf 2003), so this step was not performed. The position of each missing bilateral landmark was estimated from its antimere (if present) through reflected relabeling (Baab and McNulty 2009; Gunz et al. 2009). A minimal number of midline landmarks or bilateral landmarks missing on both the right and left sides were reconstructed for five individuals (mostly from extinct taxa) based on the position of that landmark in specimens that were close phylogenetic relatives and that shared similar morphologies in the localized region containing the missing landmark (Supporting Information Methods 2). Analyses were performed on the symmetric component of shape only, as calculated in the MorphoJ software package (Klingenberg, 2011).

Centroid size and body mass

Log-transformed CS calculated from the cranial landmarks, a proxy for overall cranial size, was used as the scaling variable in most analyses. Both log-transformed CS and log-transformed body mass were used as covariates in separate tests of the relationship between activity pattern and cranial shape based on the analysis of Jungers et al. (2002) (size data are available in Table 2).

Ecological variables

In general, small extant lemurs are nocturnal, larger lemurs (with the exception of Daubentonia) are diurnal (Mittermeier et al. 2010), and a few medium-sized lemurs (e.g., brown lemurs) are cathemeral. Diurnality is inferred for all subfossil lemur species examined here based on the absolutely and relatively small size of their orbits (Walker 1967; Jungers et al. 2002; Godfrey et al. 2006). For activity pattern, animals were scored as an ordinal variable with “0,” “1,” or “2” if they were nocturnal, cathemeral, or diurnal, respectively, based on Mittermeier et al. (2010) (Table 2). This coding assumes that cathemerality will require adaptations that are intermediate between the other two states (e.g., Kirk 2006b).

Extant lemurs include folivorous, frugivorous, omnivorous, insectivorous, gummivorous, and even bamboo-specializing species (see Supporting Information for references). The extinct subfossil lemurs are suggested to range from folivorous, to frugivorous, to mixed feeders (Table 2). Quantitative dietary data for extant lemurs were tabulated from published field observations and proportions of each food type in the diet were averaged for each species (Table S1). These data are, by their nature, incomplete because they rarely sample all populations across all seasons, but represent the most accurate source of ecological information as they derive from direct observation in natural settings. Estimates can be updated as new information becomes available. Dietary inferences in the subfossil taxa were based on a “total evidence” approach using relatively recent analyses of coprolites, dental morphology, dental microwear and topography, finite-element modeling, and isotopic signatures (see Table S1 for references). There was generally broad agreement about dietary estimates for the subfossil taxa. However, two very distinct dietary estimates exist in the literature for Hadropithecus stenognathus. For this reason, two hypotheses (H1 and H2) will be tested separately for this group (Supporting Information Methods 3). Dietary variables designed to capture the key element for each hypothesis were formulated based on these proportional data.

To create categories for “typical diet,” we performed a cluster analysis based on unweighted pair group means analysis of Euclidean distances among taxa based on the proportion of the diet in each of five main food types. The five food types were fruit, foliage (leaves/buds/flowers/bamboo), fauna, liquids (exudates, insect secretions), and hard objects (seeds/cankers/galls/USOs [underground storage organs]). The analysis identified four main clusters that also make intuitive sense. These categories correspond roughly to species that are primarily (1) frugivorous, (2) folivorous, (3) granivorous (seed-eating), or (4) liquid consumers (Table 2). Most taxa fall into one of the first two categories, with only a few specialized species classified in the latter two.

The second dietary categorization reflects difficult-to-process food items in the diet, even when consumed rarely (Table 2). Bamboo, leaves, seeds, and USOs are particularly mechanically resistant (Williams et al. 2005) and challenging to process (Hylander 1979; Kinzey and Norconk 1993; Ravosa 1996; Yamashita 1996; Lucas 2004; Williams et al. 2005; Dominy et al. 2008; Norconk et al. 2009; Yamashita et al. 2009). The “resistance” variable is an ordinal variable where a species was scored as “1” if leaves or bamboo comprised >5% of its diet, “2” if seeds or USOs were present in the diet at proportions of >5%, and “0” if none of these foods was present at levels above 5%. If foods were present from both categories, then the species received a “2.”

Finally, we hypothesize that selection will favor large gape, great mechanical advantage, or both, based on the physical and mechanical properties of the foods consumed (Dumont and Herrel 2003; Vinyard et al. 2003; Taylor and Vinyard 2009; Perry and Hartstone-Rose 2010; Ravosa et al. 2010; Perry et al. 2011; Table 2). Selection will favor a large gape in species consuming large ripe fruit (because relatively large foods require a larger gape) and gums (Vinyard et al. 2003), whereas it will favor increased mechanical advantage in taxa whose members eat leaves, USOs, and insects with tough exoskeletons in primates weighing more than 1 kg (because if all else remains equal, harder and tougher foods require greater mechanical advantage). If ≥60% of a species’ diet was judged to require a large gape, then it was given a score of “1” for gape; if ≥60% of a species’ diet favors great mechanical advantage, then it received a “1” for mechanical advantage. A few species may experience selection for both wide gape and great mechanical advantage if they eat foods for which both are favorable (e.g., large seeds or unripe fruit, bamboo culm pith, and insects with tough exoskeletons when eaten by small lemur taxa) or a combination of foods that each require different strategies. Species received a “1” in both categories (signifying selection for both traits) if <60% but ≥20% of the diet fell into each of these categories or if ≥60% of their diet required both strategies (e.g., seeds). The aye-aye received a “1” for both variables because it gnaws trees to access insects, an activity that requires both a wide gape and great mechanical advantage.

Phylogeny

To use phylogenetic comparative methods, we compiled a single composite phylogeny whose basic structure (topology and branch lengths) was based on nuclear DNA sequence data (Perelman et al. 2011; Fig. 2). Extant taxa not included in that study were added based on the position of close relatives in the Perelman et al. study or other genetic analyses (Roos et al. 2004; Andriaholinirina et al. 2006). Subfossil taxa were added to the tree based on morphological (especially dental) and genetic evidence (ancient DNA has been recovered for Pachylemur, Megaladapis, Palaeopropithecus, Archaeolemur, and Hadropithecus; Crovella et al. 1994; Karanth et al. 2005; Orlando et al. 2008). Branch lengths represent time since divergence. The basic framework for these dates is based on Perelman et al. (2011). The remaining branch lengths were reconstructed by the authors. For extant taxa (e.g., Cheirogaleus major and C. medius), branch lengths were based roughly on amount of genetic divergence relative to other taxa with known branch lengths. For the extinct taxa, we choose a date of 6.25 Ma for the most recent subfossil species splitting events, a date similar to that seen for many extant species divergences. Branch lengths between neighboring nodes within the palaeopropithecid clade were then each set to 6.25 Ma, resulting in an even distribution of divergence events through time.

Figure 2.

Composite phylogenetic tree used for phylogenetic comparative methods. Both the branching pattern and divergence dates are based primarily on the Perelman et al. (2011) molecular phylogeny (nodes with black stars). Other taxa were added to the tree based on other genetic data (mtDNA or SINE insertions) when available (nodes with gray stars). The position of Lepilemur mustelinus was based on that of its sister taxon, L. jamesi. Subfossil taxa for which ancient DNA is either not available or has not been analyzed were inserted based on morphological considerations (nodes without stars). Divergence dates that are italicized were estimated by the authors; all other divergence dates were based on genetic data.

STATISTICAL ANALYSIS

We present results from nonphylogenetic analyses and analyses that take the expected lack of independence due to phylogeny into account. We used standard (nonphylogenetic) and phylogenetic principal components analyses (PCA and PPCA, respectively) of our species mean shapes to investigate the main patterns of cranial shape variability in lemurs (Revell 2009; Polly et al. 2013). We used ordinary least squares (OLS) and phylogenetic generalized least squares (PGLS) regressions to evaluate the relationship between diet and the size/shape of the skull (Grafen 1989; Rohlf 2001, 2006; Martins et al. 2002). We assessed phylogenetic signal by estimating an additional parameter, Pagel's λ (Pagel 1999), during calculation of the PGLS (Revell 2010). By generalizing the calculations presented in the appendix of Revell (2010) to the multivariate case (>1 dependent variable), we found that λ was very close to 1.0 for the full model using all species in each of the four ecomorphological models. A value of λ = 1.0 suggests that our data are compatible with a Brownian motion model (Pagel 1999), and this was therefore the model assumed in the phylogenetic analyses.

The first four PCs (77.3% of the total variance) were used as the dependent variables for the OLS analyses and the first four PPCs (64.9% of the total variance) were used as the dependent variables in the PGLS analyses. The result is many fewer variables than observations (species means), which is important in a multivariate analysis. We regressed (P)PC 1–4 scores separately on size and the relevant ecological variables in separate multiple regression analyses for each of the ecological hypotheses (activity pattern, typical diet, food resistance, and gape/mechanical advantage; Table 2). However, individual (P)PC axes are not expected to be biologically meaningful. Therefore, we also performed multivariate multiple regressions using the (P)PC 1–4 scores as dependent variables simultaneously for each of the hypothesized relationships between ecology and cranial shape (and pairwise comparisons for the MANCOVA/phylogenetic MANOVA). For the multivariate multiple regression, we examined the influence of all independent variables simultaneously (full model), all ecological variables holding size constant, and each independent variable separately with all other variables held constant.

Statistical significance was assessed at the α ≤ 0.05 level, and a sequential Bonferroni correction for multiple tests was applied (Holm 1979). All tests were run twice, with Hadropithecus scored differently to reflect the two main dietary hypotheses present in the literature (Table 2 and Supporting Information Methods 3). The regression analyses were also performed using the extant taxa only (based on the same (P)PC scores as above) because we have more confidence in the accuracy of the dietary data and phylogenetic relationships of these species. The PGLS calculations were programmed in SAS-IML and Matlab based on the phylogenetic transformation of the x and y values described in Garland and Ives (eq. 4; 2000) and Rohlf (eqs. 20–21; 2001), modified slightly based on the GLS equations found in Section 4.2 of Rao and Toutenburg (1999) (Matlab code available from authors upon request).

We also ran the same set of regression/MANOVA analyses using the coordinate data (from the symmetric component of the shape variation) directly as dependent variables (in place of PC scores) using the geomorph package (Adams and Otárola-Castillo 2013) for the R software environment (R Core Team 2013). We do not report the results of these analyses here as size was the only variable significantly associated with cranial shape.

Results

CRANIAL SHAPE VARIATION

The subspaces spanned by the first two components of the nonphylogenetic and phylogenetic PCAs are very similar (Fig. 3), and results are described for the latter only. Two of the major lemuriform clades separate along PPC 1: members of the Lemuridae (except Hapalemur) + Megaladapidae clade occupy the negative end of PPC 1, whereas the Indriidae + Palaeopropithecidae + Archaeolemuridae clade dominate the positive end of PPC 1 (Fig. 3B). The two small-bodied families (Cheirogaleidae and Lepilemuridae), which together form the sister group of this latter clade, are positioned centrally on the first axis. Cheirogaleids score more negatively than all Lepilemur species except L. microdon. Daubentonia (the aye-aye) is positioned just outside the envelope of extant lemur variation. Its position closest to the indriids is consistent with previous observations of (superficial) similarities in cranial morphology between these groups (Tattersall and Schwartz 1974).

Figure 3.

Ordinations of the first and second components of the (A) nonphylogenetic and (B) phylogenetic PCAs in shape space, conducted at the level of species means. Although not identical, the two ordinations are very similar. Minimum convex hulls around lemuriform families emphasize the strong phylogenetic signal along these components.

This first component reflects differences between prognathic species with relatively low, narrow neurocrania and orthognathic taxa with relatively tall and broad cranial vaults (Fig. 4A). In the latter, the facial skeleton and zygomatic arch are both much taller superoinferiorly. Both elongation and greater anterior protrusion of the palate result in the very long lateral profile of specimens that score low on PPC 1. The root of the zygomatic arch (estimated by the inferior zygomaxillary suture landmark) is positioned posterior to the tooth row in long-faced forms, and closer to the premolar-molar contact in more orthognathic species. These shape differences are not related to overall size; two large-bodied taxa occupy opposite ends of this axis (Megaladapis and Hadropithecus), and log-transformed CS accounts for little variation (R2 = 0.08) and is not significant (P = 0.12).

Figure 4.

Shape variation associated with the first two axes of the phylogenetic PCA in shape space. The shape differences along PPC 1 (A) and PPC 2 (B) are illustrated in right lateral and superior views. Lines have been added between points to create “wireframes” for easier visualization. In all cases, the shape changes are shown at the extreme negative (left) and positive (right) species positions along the axes. Although only the symmetric component of shape was analyzed, the left side was mirrored in the superior view for easier visualization.

The second component captures some allometric variation: all of the lowest scoring specimens are large subfossil taxa, and CS accounts for 69% of variation in PPC 1 scores (P < 0.01). Larger taxa have relatively small orbits and superiorly deflected neurocrania (the negative end of PPC 2), whereas smaller ones have larger orbits and more coplanar neurocrania and rostra (positive end of PPC 2; Fig. 4B). Additional shape differences along this component relate to palatal shape, the orientation of the infraorbital region, the height of the zygomatic arch, and the position of the lateral orbital margin.

The third and fourth components again contain a phylogenetic signal, but are not clearly related to allometric or ecological factors. The subfossil palaeopropithecids (especially Babakotia radofilai and Palaeopropithecus kelyus) are separated from the other taxa on PPC 3, whereas the aye-aye (Daubentonia) and the extinct megaladapids (particularly M. madagascariensis/grandidieri) are distinguished from the other taxa on PPC 4 (not shown). Twenty-two percent of the variance on PPC 3 is accounted for by log-transformed CS (P = 0.01).

RELATIONSHIP BETWEEN SHAPE AND SIZE AND ECOLOGY

In the multivariate multiple regression using (P)PCs 1–4 as dependent variables, cranial size is highly correlated with cranial shape independent of the ecological variables in nearly all of our analyses (Tables 36). Broadly speaking, there are more statistically significant effects in the OLS than the PGLS analyses, but the pattern of which independent variables account for the most shape variance are similar in both analyses. The fewer statistically significant results in the PGLS analyses may be due in part to the fact that PGLS recognizes that the effective sample size is actually smaller due to the presence of correlated observations, thereby decreasing power to reject a false null hypothesis.

Table 3. Results of OLS and PGLS regression analyses of cranial shape on activity pattern and log (centroid size)/log (body mass) based on all species (including subfossil taxa) and extant species only
 Activity pattern—log (centroid size)a
 OLSPGLS
 All species All species 
 Pr2 Pr2 
  1. a

    An asterisk denotes statistical significance at the α = 0.05 level after a sequential Bonferroni correction for multiple tests. The eta-squared (η2) is similar to an r2 value and is calculated as 1-Wilks’ lambda (Λ).

PC 10.53160.0412 0.18810.1453 
PC 2<0.00010.8742 <0.00010.5217 
PC 30.43770.0536 0.87360.0226 
PC 40.65980.0273 0.89370.0199 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.08210.9179<0.0001*0.31360.6864
Size<0.0001*0.22030.77970.0010*0.51770.4823
Activity pattern0.63120.91200.08800.71490.92720.0728
 Extant species only Extant species only 
 Pr2 Pr2 
PC 10.89090.0127 0.32550.1709 
PC 2<0.00010.7570 0.00070.6041 
PC 30.51650.0708 0.56190.105 
PC 4<0.00010.7202 0.71910.0700 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.04200.95800.06130.40340.5966
Size<0.0001*0.17270.82730.08910.60250.3975
Activity pattern0.0169*0.46840.53160.75790.88890.1111
 Activity pattern—log (body mass)a
 OLSPGLS
 All species All species 
 Pr2 Pr2 
PC 10.92290.0053 0.48990.0763 
PC 2<0.00010.8757 <0.00010.5676 
PC 30.49510.0458 0.89930.0191 
PC 40.77890.0165 0.8930.0200 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.09910.9009<0.0001*0.32540.6746
Size<0.0001*0.26610.73390.0016*0.53720.4628
Activity pattern0.94650.81960.18040.88110.95860.0414
 Extant species only Extant species only 
 Pr2 Pr2 
PC 10.21250.1581 0.41290.1436 
PC 2<0.00010.7194 0.00100.5875 
PC 30.07850.2463 0.17980.2326 
PC 40.00210.4972 0.83170.0462 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.09740.90260.08630.43050.5695
Size0.00570.40030.59970.13390.64290.3571
Activity pattern0.0063*0.40610.59000.58970.83840.1616

Based on Jungers et al. (2002), we predicted a relationship between cranial shape and activity pattern when cranial size (log-transformed CS) was used as a covariate, but not when log-transformed body mass was used as a covariate for the extant taxa, and a relationship regardless of the size covariate used when the full sample (including the subfossil taxa) was analyzed. However, activity pattern only accounted for a statistically significant proportion of the size-independent shape variation in the OLS analysis restricted to the extant taxa (η2 = 0.53 when cranial size was the covariate; 0.59 when body mass was the covariate; Table 3). Activity pattern did not account for a significant proportion of size-independent cranial shape variation in any of the other analyses, including both phylogenetic analyses.

When size was held constant, the four dietary categories together accounted for a significant proportion of cranial shape variation in the extant-only PGLS analysis prior to the sequential Bonferroni correction (Table 4). The relationship between shape and granivory bordered on significance in both the H1 and extant-only PLGS analyses (η2 = 0.28 and 0.48, respectively), and a diet dominated by foliage also explained cranial shape variation independent of the other variables in the extant-only PGLS analysis (none of these relationships remained significant after the multiple comparisons were taken into account using a sequential Bonferroni correction). The OLS results were similar: the four dietary categories together predicted a statistically significant amount of shape variance in the H2 and extant-only analyses (and nearly so in the H1 analysis), granivory was a significant or nearly significant predictor in all three analyses, and the relationship between shape and liquid consumption was significant before the Bonferroni correction in the H2-OLS analysis.

Table 4. Results of MANCOVA and P-MANCOVA analyses of cranial shape and average diet (dietary categories) and size based on all species (including subfossil taxa) and extant species only
 Average dieta
 OLSPGLS
 All species (H1) All species (H1) 
 Pr2 Pr2 
  1. a

    An asterisk denotes statistical significance at the α=0.05 level after a sequential Bonferroni correction for multiple tests. The eta-squared (η2) is similar to an r2 value and is calculated as 1-Wilks’ lambda (Λ).

PC 10.0420.2898 0.00800.4124 
PC 2<0.00010.8734 0.00020.5683 
PC 30.04900.2806 0.58970.1188 
PC 40.13760.2138 0.99790.0096 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.03450.96550.0002*0.19750.8025
Size<0.0001*0.10020.8998<0.0001*0.34300.6570
Dietary category0.01140.3830.61700.27050.58400.4160
Frugivore versus rest0.49960.87870.12130.89130.98790.0121
Folivore versus rest0.42410.86160.13840.38000.85080.1492
Granivore versus rest0.01370.61650.38350.07600.72170.2783
Liquid consumer versus rest0.12830.75930.24070.27960.82240.1776
 All species (H2) All species (H2) 
 Pr2 Pr2 
PC 10.38430.1339 0.08810.2778 
PC 2<0.00010.8812 0.00060.5224 
PC 30.00410.4104 0.65740.1053 
PC 40.01360.3521 0.82770.0705 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.02630.97370.0008*0.23270.7673
Size<0.0001*0.10000.9000<0.0001*0.33760.6624
Dietary category0.0009*0.29240.70760.60920.68800.3120
Frugivore versus rest0.55160.88960.11040.90170.96030.0397
Folivore versus rest0.35890.84530.15470.33590.83900.1610
Granivore versus rest0.0006*0.46990.53010.51130.88120.1188
Liquid consumer versus rest0.02430.64920.35080.51060.88100.1190
 Extant species only Extant species only 
 Pr2 Pr2 
PC 10.00070.6780 0.00300.6446 
PC 20.00060.6852 0.00050.7185 
PC 30.00840.5545 0.45250.2369 
PC 40.00060.6846 0.38970.2590 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.00680.99320.0029*0.09510.9049
Size<0.0001*0.07000.93000.03550.47640.5236
Dietary category0.0001*0.07580.92420.02590.20950.7905
Frugivore versus rest0.14460.61190.38810.96070.95650.0435
Folivore versus rest0.19090.64470.35530.05810.51900.4810
Granivore versus rest0.0073*0.36440.63560.06030.52240.4776
Liquid consumer versus rest0.05090.50720.49280.08410.55420.4458

There was no significant relationship found between the presence of mechanically resistant foods and cranial shape in the PGLS analyses, but this variable did explain a significant proportion of the cranial shape variation independent of cranial size in all three of the OLS analyses (Table 5).

Table 5. Results of OLS and PGLS regression analyses of cranial shape on size dietary strategy variables based on all species (including subfossil taxa) and extant species only
 Food resistancea
 OLSPGLS
 All species (H1) All species (H1) 
 Pr2 Pr2 
  1. a

    An asterisk denotes statistical significance at the α=0.05 level after a sequential Bonferroni correction for multiple tests. The eta-squared (η2) is similar to an r2 value and is calculated as 1-Wilks’ lambda (Λ).

PC 10.00600.2891 0.16440.1541 
PC 2<0.00010.8725 <0.00010.5075 
PC 30.22410.0949 0.38490.0950 
PC 40.48630.0469 0.96120.0096 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.05460.9454<0.0001*0.30090.6991
Size<0.0001*0.10150.8985<0.0001*0.33500.6650
Resistance0.0074*0.60650.39350.51430.88980.1102
 All species (H2) All species (H2) 
 Pr2 Pr2 
PC 10.05960.1714 0.18970.1447 
PC 2<0.00010.8719 <0.00010.5154 
PC 30.08960.1485 0.34850.1024 
PC 40.44790.0521 0.96470.0090 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.05990.9401<0.0001*0.29820.7018
Size<0.0001*0.09920.9008<0.0001*0.33300.6670
Resistance0.0227*0.66570.33430.47460.88170.1183
 Extant species only Extant species only 
 Pr2 Pr2 
PC 10.11580.2130 0.33260.1685 
PC 2<0.00010.6915 0.00050.6194 
PC 30.05780.2715 0.60590.0948 
PC 40.00240.4894 0.96830.0137 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.03760.96240.04010.37320.6268
Size<0.0001*0.05570.94430.0056*0.39930.6007
Resistance0.0080*0.41970.58030.53780.82230.1777

Similarly, selection favoring wide gape versus great mechanical advantage successfully predicted shape variation in the nonphylogenetic analyses only. The two variables together (i.e., selection for gape and selection for great mechanical advantage) explained a large proportion of shape variation in the three OLS analyses (η2 = 0.54–0.70; Table 6). Conditions favoring selection for mechanical advantage also explained variation in shape independent of the other variables in all three nonphylogenetic analyses.

Table 6. Results of OLS and PGLS regression analyses of cranial shape on food resistance scores and size based on all species (including subfossil taxa) and extant species only
 Dietary strategya
 OLSPGLS
 All species (H1) All species (H1) 
 Pr2 Pr2 
  1. a

    An asterisk denotes statistical significance at the α=0.05 level after a sequential Bonferroni correction for multiple tests. The eta-squared (η2) is similar to an r2 value and is calculated as 1-Wilks’ lambda (Λ).

PC 10.00420.3607 0.13660.2080 
PC 2<0.00010.8733 0.00020.5138 
PC 30.05870.2235 0.58400.0906 
PC 40.61500.0592 0.76060.0603 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.03690.96310.0002*0.26770.7323
Size<0.0001*0.08720.9128<0.0001*0.33740.6626
Gape + mechanical advantage0.0019*0.40990.59010.60040.79160.2084
Gape0.17740.79140.20860.71890.92540.0746
Mechanical advantage0.0003*0.45900.54100.43720.86940.1306
 All species (H2) All species (H2) 
 Pr2 Pr2 
PC 10.02910.2635 0.11590.2190 
PC 2<0.00010.8751 0.00030.5128 
PC 30.04490.2391 0.64700.0796 
PC 40.59720.0618 0.63820.0811 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.04120.95880.0002*0.26230.7377
Size<0.0001*0.08930.9107<0.0001*0.33670.6633
Gape + mechanical advantage0.0061*0.45810.54190.53940.77570.2243
Gape0.50750.88450.11550.62030.90690.0931
Mechanical advantage0.0010*0.50560.49440.53020.88920.1108
 Extant species only Extant species only 
 Pr2 Pr2 
PC 10.01360.4577 0.14960.3136 
PC 2<0.00010.7022 0.00170.6184 
PC 30.02270.4217 0.41530.1967 
PC 40.00710.4992 0.35360.2177 
 PWilks' Λη2PWilks' Λη2
Full model<0.0001*0.02660.97340.02400.22770.7723
Size<0.0001*0.07580.92420.02930.48570.5143
Gape + mechanical advantage0.0166*0.29660.70340.22360.50180.4982
Gape0.33560.88450.11550.13640.62600.3740
Mechanical advantage0.0163*0.44270.55730.63080.84170.1583

Discussion

Our PPCA revealed that the primary pattern of variation in cranial shape across Malagasy lemurs reflects variation in prognathism and the relative width and height of the neurocranium. The second PPC axis highlighted an allometric pattern related to relative orbit size and the orientation of the cranial vault relative to the rostrum. A strong phylogenetic pattern was recovered, and differentiation among clades along PPC 1 and PPC 2 suggests that taxonomic divergence has occurred along both allometric and nonallometric axes of shape differentiation, a pattern also described for the papionin Old World monkeys where taxa have diverged along both allometric and nonallometric lines (Singleton 2002; Frost et al. 2003). This contrasts with the pattern reported for New World monkeys and the guenon tribe of Old World monkeys, where most differentiation among groups has occurred along allometric lines (Marroig and Cheverud 2005; Cardini and Elton 2008). In lemurs, the larger taxa occupy more disparate regions of morphospace than smaller ones, perhaps reflecting greater niche divergence or reduced evolutionary constraints at larger body sizes.

Are lemurs a true adaptive radiation? Lemur monophyly is well-established (Yoder 1997; Perelman et al. 2011; Springer et al. 2012) and if previous suggestions of rapid speciation early in the radiation are correct (Fleagle and Reed 1999), then what remains to be evaluated is whether phenotypic variation reflects underlying environmental diversification, and, finally, whether these variations are adaptive in their environments (Schluter 2000). There are a number of documented phenotype-environment correlations in lemurs related to vision and light levels, postcranial anatomy and locomotion, and body size and niche productivity. Therefore, establishing a correlation between feeding ecology and cranial form in lemurs is not a prerequisite to recognizing them as an adaptive radiation. However, a relationship between cranial form and diet is widespread across other putative vertebrate adaptive radiations, and many instances of convergent evolution for this relationship have been documented (e.g., Anapol and Lee 1994; Freeman 2000; Nogueira et al. 2005; Van Valkenburgh 2007; Wroe and Milne 2007; Samuels 2009; Goswami et al. 2011).

Of the previous studies linking lemur craniofacial variation to ecological factors, only a few took a broad taxonomic view. Instead, most focused on comparisons among a few closely related species or a few subfossil taxa (Jolly 1970; Tattersall 1973; Ravosa 1991, 1992; Vinyard et al. 2003; Ravosa and Daniel 2010; Ravosa et al. 2010; Dumont et al. 2011). The broadest view of lemur cranial diversity was provided by Viguier (2004), who argued that craniofacial structure in lemurs primarily reflects functional adaptation to diet based on an analysis of 2D craniofacial landmarks using a broad taxonomic sample (in agreement with the nonphylogenetic analyses presented here). The most important difference between the Viguier study and the present one is probably the use of phylogenetic methods here. In the Viguier (2004) study, five of six dietary categories were composed of members of a single lemuriform family, so the similarities found could just as easily be attributed to phylogeny as to diet. The only category that contained taxa from different families was the “folivore” category, the same category that showed a wide distribution along PC 1 and PC 2. It is also notable that there was a high degree of overlap between the “gummivore” and “omnivore” groups, both of which included only cheirogaleids.

When phylogeny is ignored (implying a star phylogeny), there are several significant correlations between cranial shape variation and ecological variables, as described above. However, when phylogenetic methods are applied (PGLS), cranial shape is only related to membership in some of the dietary categories (particularly granivory), and even then, the relationship is only of borderline statistical significance (see also Metzger and Herrel 2005). The justification for applying phylogenetic comparative methods has been laid out clearly elsewhere (e.g., Nunn 2011), and will not be reiterated here. As discussed by Rohlf (2001), both OLS and PGLS provide unbiased estimates of the same model parameters and it is not possible to determine which estimates are correct when they differ in a particular dataset. However, if the phylogenetic tree is correct, then the PGLS estimates are expected to be closer to the true parameters more often than those estimated using OLS. Even if this phylogenetic tree used here differs from the true phylogeny, it is likely to be much closer to the true one than is a “star phylogeny.”

Relying on the PGLS analyses, we find support only for the hypothesis that skull shape reflects the most commonly consumed food items, and even then, support is weak. Highly granivorous taxa scored high on PPC 1, indicating a more orthognathic face with a higher position for the TMJ, consistent with expectations for a higher mechanical advantage. However, even this result should be viewed with caution as this positive result was only apparent in the H1 and extant-only analyses. The H1 analysis reflects a dietary reconstruction of hard-object feeding for Hadropithecus, which is controversial (e.g., Dumont et al. 2011). A similarly weak relationship between cranial shape and ecology when phylogeny is taken into account was also documented for guenon monkeys (Cardini and Elton 2008) and New World monkeys (Perez et al. 2011). The weak positive associations between ecology and cranial shape in the phylogenetic analyses can be interpreted in several ways.

The stronger relationships recovered by OLS indicate that the correlations between cranial shape and ecology are due mostly to relationships among closely related species. Biologically, this could indicate that much of the cranial shape and dietary variation in lemurs is partitioned among several major clades that diverged early in lemur evolutionary history (perhaps reflecting niche conservatism). If the morphological divergence that occurred at these bifurcations was driven by ecological changes, it may be difficult to recover this pattern as phylogenetic methods treat closely related taxa as providing partially redundant information, and therefore downweights their similarities relative to similarities evinced by more distantly related taxa (see also Polly et al. 2013). This interpretation is supported by the close correspondence of the observed shape differences between granivores and the morphological predictions for this type of diet based on biomechanical models. Further support is found in morphological convergence among taxa that exhibit similar diets in different clades. For example, Hapalemur (a bamboo specialist within the generally frugivorous Lemuridae family) is positioned within the morphospace of the more folivorous indriids as predicted by functional morphology, and the folivorous lepilemurs are also positioned closer to the indriid centroid than the lemurid centroid (Fig. 3). However, there are also exceptions to this pattern: the subfossil genus Megaladapis does not converge on the cranial architecture of any other folivorous taxa. Similarly, if changes in size accompanied dietary shifts, this would be masked by including size as a covariate in the model. In fact, both of these patterns have been suggested to apply to New World monkeys: major dietary shifts occurred only four times during NWM evolution, and these were accompanied by changes in size (Marroig and Cheverud 2005). Macroevolutionary models designed to assess the presence of niche conservatism may be helpful in assessing this possibility (Cooper et al. 2010).

Other explanations for the overall weak ecology-morphology relationship include an overprinting of the ecological signal by other functional demands on the cranium, inadequacies of one or more of the datasets used here (e.g., the data about lemur diets are of insufficient resolution or dietary reconstructions for the extinct species are inaccurate), and a mismatch between current diet and morphology related to ecological retreat (Crowley et al. 2012). Ecological retreat refers to a postulated shift to wetter riparian (and more protected) habitats by those lemurs that survived colonization of Madagascar by humans. If this relatively recent retreat was characterized by changes in diet, the resulting mismatch between diet and morphological adaptations would lead to a weak correlation between ecology and cranial shape. However, the generally stronger relationship between ecology and morphology in the extant species counters the idea of a recent mismatch.

Overall, lemur cranial morphology retains a strong phylogenetic signal, and lineages have diverged along both allometric and nonallometric axes of shape variation. The correlation between diet and cranial form is weak when the underlying phylogeny is taken into account, a pattern also documented in other vertebrate lineages (e.g., McBrayer 2004; Cardini and Elton 2008; Jones and Goswami 2010; Perez et al. 2011). This may mean that diet has not strongly impacted cranial form, but may also be the result of an evolutionary history characterized by a relatively small number of dietary shifts that occurred in conjunction with the divergence of major clades (and possibly shifts in body size) and few instances of dietary convergences between these clades. This latter pattern could be difficult to detect with the type of phylogenetic comparative methods used here, but may characterize not just lemurs but other mammalian taxa, including South American primates (Rosenberger et al. 2009). Detailed assessment of shape differences as they relate to biomechanical predictions derived from theory and experimental analysis, alongside the use of macroevolutionary models that evaluate the likelihood of niche conservatism, are necessary to assess this possibility.

ACKNOWLEDGMENTS

We gratefully acknowledge the institutions that permitted us to collect data on the extant and extinct lemuriform samples: American Museum of Natural History, Duke Lemur Center Division of Fossil Primates, Académie Malgache, the University of Madagascar at Antananarivo, Museum National d’Histoire Naturelle, and Naturhistorisches Museum Basel. We also thank E. Otárola-Castillo for his analytical help and assistance with the geomorph program for R. We thank the Associate Editor (C. Klingenberg), A. Cardini, and two anonymous reviewers for their thoughtful comments. The authors have no conflict of interest to declare.

DATA ARCHIVING

The doi for our data is 10.5061/dryad.ts866.

Ancillary