How to make a flying squirrel: Glaucomys Anatomy in Phylogenetic Perspective

Abstract Molecular evidence strongly supports the derivation of flying squirrels from tree squirrels, with the sister group being the Holarctic tree squirrels (Sciurus) and their close relatives. We present a rationale for the hypothesis that the transition occurred among small squirrels, and we propose using the northern flying squirrel (Glaucomys sabrinus) and the southern flying squirrel (Glaucomys volans) as models. Thus, we compare their morphologies with similar-sized tree squirrels (the Central American dwarf squirrel [Microsciurus alfari] and the western dwarf squirrel [M. mimulus]). We compare body proportions of Glaucomys with those of Microsciurus, considering differences as potential adaptations for gliding associated with wing loading, aspect ratio, and parasitic drag. We use the following measurements: lengths of the centra of the thoracic, lumbar, sacral, and caudal vertebrae; and lengths of the long bones (humerus, radius, femur, and tibia), metacarpals, metatarsals, and proximal phalanges of the hands and feet. We then compare these proportions of Microsciurus with those of other species of Sciurini, and the proportions of Glaucomys with other species of Pteromyini, to determine if each is representative or derived within its clade. Members of the genus Glaucomys exhibit relative elongation of the lumbar vertebrae and the forearm, relevant to wing loading and aspect ratio, relative lengthening of the midcaudal vertebrae, and relative shortening of the more distal caudal vertebrae, perhaps of importance for stability and control. Members of the genus Glaucomys also have shorter hands and feet, but these appeared to be elongated in Microsciurus, rather than shortened in Glaucomys, leaving the issue of parasitic drag moot. Finally, we speculate on the genetic changes that have produced these morphological modifications and may facilitate the evolution of gliding flight.

Strong evidence supports the thesis that flying squirrels (Pteromyini) are derived from a common ancestor shared with the Holarctic tree squirrels (Sciurini). Mercer and Roth (2003), Steppan et al. (2004), and Herron et al. (2004) reached this conclusion based on a total of 6 genes (respectively, a combination of nuclear and mitochondrial genes: IRbp, 12S, and 16S, 2,659 base pairs [bp]; 2 nuclear genes: c-Myc and Rag1, 4,477 bp; and 1 mitochondrial gene: Cytb, 1,140 bp), and it is concordant with morphological features that are probably synapomorphies. We accept the common ancestry of the Pteromyini and the Sciurini (Thorington and Hoffmann 2005), figured in Arbogast (2007). This phylogeny is contradicted by some interpretations of paleontological evidence (de Bruijn and Ü nay 1989;Major 1893;Mein 1970), and our study was initiated as a challenge to these interpretations. In a previous paper,  showed that the dental characters previously used to identify flying squirrels in the fossil record are seen in a number of nonflying squirrels as well, calling into question the identity of pre-Miocene ''flying squirrels.'' They also suggested morphological features of the proximal and distal ends of long bones that are present or likely to be present in the fossil record of the Oligocene and Miocene and that would enable paleontologists to distinguish between flying squirrels and others. In the current study, we consider other morphological changes, particularly changes in proportions, required to transmute tree squirrel morphology into the gliding adaptations of flying squirrels.
The common ancestor of the Sciurini and the Pteromyini was most likely an animal with tree squirrel morphology. Douglassciurus of the late Eocene and Palaeosciurus of the early Oligocene, the 2 earliest squirrels in the fossil record, both with surprisingly complete skeletons, had postcranial anatomy that was remarkably like that of modern-day tree squirrels in proportions and details of their anatomy (Emry and Thorington 1982;Vianey-Liaud 1974). Douglassciurus was compared directly with the fox squirrel (Sciurus niger) and exhibited very similar proportions of limb bones, hind foot bones, and proximal caudal vertebrae (Emry and Thorington 1982). Palaeosciurus goti is suggested to be a ground squirrel, based on the relative lengths of its humerus and radius and of its femur and tibia. The illustrations accompanying the description of P. goti show that the hind limb and upper arm bones closely resemble those of Recent tree squirrels (Vianey-Liaud 1974). Palaeosciurus feignouxi, of the very early Miocene (Savage and Russell 1983), has long-bone ratios like those of tree squirrels. Accordingly, it is highly likely that the common arboreal ancestor of flying squirrels and tree squirrels, also of the early Miocene, was very similar morphologically to modern-day tree squirrels.
The polarity that we propose-that tree squirrel morphology is primitive and that flying squirrel morphology is derivedalso is supported by comparative data from other sciurids. First, we propose the obvious, that flying squirrels were not derived from terrestrial squirrels. It does not seem logical that a means of getting from 1 tree to another without coming to the ground would evolve in a terrestrial squirrel. Second, all tree squirrels are remarkably similar in body proportions, and where they differ, none are similar to flying squirrels. Most relevant, the Callosciurinae tree squirrels of southern Asia and the Sciurini tree squirrels of northern Eurasia, North America, and South America are very similar (Thorington and Heaney 1981). Either this is due to a common ancestral morphology or to convergent evolution. If the former, then the morphology must be primitive for the divergence of the Sciurini and the Pteromyini, which occurred slightly later (Mercer and Roth 2003). The early Miocene fossil P. feignouxi is a candidate for that ancestor or a close relative of it. If convergent evolution is responsible for the similarities between the Callosciurinae and the Sciurini tree squirrels, then the convergence would be expected in the arboreal ancestor of the Pteromyini. Other cases that can be argued to result from convergent morphology strengthen this argument. They are the African tree squirrels and perhaps the Asian giant tree squirrels (Ratufa) and the South American pygmy squirrel (Sciurillus), whose limb and trunk proportions are reported by Thorington and Heaney (1981) and Thorington and Thorington (1989). It seems that arboreal squirrels have repeatedly converged on similar morphologies or retained ancestral tree squirrel proportions. Accordingly, we submit that the common ancestor of the Sciurini and Pteromyini most likely had the limb and trunk proportions of the former.
We suggest that flying squirrels evolved from tree squirrels that were small for the following reasons. First, very few Recent Sciurini tree squirrels weigh more than 1 kg, most species are 500 g or less, and many are 250 g or less. Early Miocene sciurines are likely to have exhibited the same size distribution, although there is little evidence for judgment. Second, tree squirrels are excellent leapers and easily transfer from 1 perch or tree to another by leaping, but maximum leaping distances scale with body size. A small tree squirrel cannot leap as far as a large one; thus the advantage of gliding increases with decreasing size. Third, smaller animals are less likely to be hurt in falling because they reach lower terminal velocities and have less momentum; small inept gliders face fewer risks than large ones. Fourth, wing loading increases with body size. Other factors being equal, mass increases with the cube of linear dimensions and wing area increases with the square; thus wing loading should be lower in small squirrels. Minimal gliding speed is expected to vary directly with wing loading, so smaller, more lightly loaded squirrels are expected to be able to glide more slowly-hence more safely (Thorington and Heaney 1981). Fifth, large flying squirrels are more specialized than small flying squirrels. Large squirrels eat more vegetable matter and have more complex teeth than do small squirrels, including all the Holarctic tree squirrels (Bryant 1945;McKenna 1962: with particular reference to groups B II and B III). Among features relevant to the aerodynamics of gliding, discussed below, the large flying squirrels have an additional patagial region, the uropatagium, stretching between the hind limbs and the tail; they have a hind limb muscle (the semitendinosis III) with a more derived morphology than that found in small flying squirrels, including its origin from the caudal vertebrae, its course along the edge of the uropatagium, and its insertion at the ankle; and they also have a more derived origin of the tibiocarpalis muscle on metatarsal III, not found in the small Glaucomys or Petinomys (Johnson-Murray 1977; Thorington et al. 2002).
Accordingly, we submit that the 2 North American flying squirrels (Glaucomys sabrinus and G. volans) are good models for comparison with similar-sized Sciurini tree squirrels for the study of basic gliding adaptations in the Pteromyini. We consider Microsciurus to be an appropriate model for the ancestral sciurine that gave rise to the flying squirrels because of its small size and because it is closely related to Sciurus niger, which has been called a living fossil because of its similarities to the Eocene Douglassciurus (Emry and Thorington 1984). Therefore, we compare these flying squirrels (G. sabrinus and G. volans) with similar-sized members of the Sciurini-2 species of dwarf squirrels of the genus Microsciurus (M. alfari and M. mimulus)-thereby minimizing allometric differences.
The appropriateness of these 2 genera as models for the evolution of flying squirrels can be questioned on the basis of their phylogeny. Mercer and Roth (2003) estimated that the Pteromyini and Sciurini diverged 23 million years ago, that the genus Glaucomys diverged from other flying squirrels 14 million years ago, and that Microsciurus evolved less than 3 million years ago. However, models that would appear to be more appropriate, such as Eoglaucomys and Petaurista, are not small. It is not possible to support our choice of models with evidence from the fossil record because most purported flying squirrels in the fossil record are represented by teeth only. However, our hypothesis would be strongly challenged by the finding of a large flying squirrel in the early Miocene (23-18 million years ago) based on a fossil bone or bone fragment identified by the anatomical details described by . We are able to test whether Glaucomys and Microsciurus are similar to other members of their clade in morphological features that distinguish the 2 genera from each other. This enables us to determine to what extent either or both genera are derived within their clade for these characteristics, which hence may be irrelevant to our more general thesis. Thus, we use the comparison of the 2 genera as a heuristic hypothesis to highlight the major differences in body proportions that may be elements of the flying squirrel adaptation, subject to further test.
Our hypotheses are based on several aerodynamic features, such as wing loading, parasite drag, and aspect ratio. Wing loading is mass per unit area, so to reduce wing loading, one must increase wing area or decrease body mass, or both. Thorington and Heaney (1981) showed that the ponderal index, a measure of mass versus body length, is inversely related to size in flying squirrels, but not in tree squirrels. This caused us to look at vertebral lengths and to question whether some vertebrae are more elongated than others in flying squirrels relative to tree squirrels. Exploratory analysis led to the hypothesis tested, that an elongation of the lumbar region in flying squirrels relative to the thoracic region characterizes Glaucomys versus Microsciurus.
We examined the relative lengths of the caudal vertebrae, knowing that tails and tail lengths vary greatly among flying squirrels. Small flying squirrels have shorter tails relative to body length than large flying squirrels and the tails are distichously haired (feather-tailed) as opposed to the teretely haired (round-tailed) large flying squirrels. Because tree squirrel tails also are distichous, we expected no difference in the relative lengths of the caudal vertebrae between Glaucomys and Microsciurus. Exploratory analysis suggested that this hypothesis was wrong and that the middle caudal vertebrae were relatively longer in Glaucomys, so we tested this as the alternative hypothesis. The presence of the uropatagium in large flying squirrels suggests that there are some major differences in tail function between large and small flying squirrels, so we expected the relative lengths of their caudal vertebrae to differ, but we had no clear concept of how they would differ.
Aspect ratio is the ratio of wingspan to wing area, and it is complexly associated with glide distance and issues of control. The higher the aspect ratio is, the further a glider can go per meter of altitude loss; but the higher the aspect ratio is, the more sensitive the wing is to stalling at higher angles of attack. A high-aspect wing will permit long glides but it will not permit landing on vertical tree trunks the way flying squirrels do. Wingspan is a function of forelimb length, so we tested the hypothesis that the forelimb (humerus þ radius) is disproportionately lengthened in Glaucomys relative to Microsciurus. From previous work (Thorington and Heaney 1981), we know that the forearm (radius) is elongated in flying squirrels relative to callosciurine tree squirrels, so we focused on this relationship in our study of the more phylogenetically appropriate Sciurini-Pteromyini comparison. We then tested whether the radius was disproportionately lengthened relative to the humerus, as would be expected in order to keep muscle mass proximal on the limb, and lengthened disproportionate to the tibia, as hypothesized if the aspect ratio is a critical factor.
During glides, flying squirrels hold their hands and feet in positions that appear to reduce drag. Externally, Glaucomys appear to have shorter toes than Microsciurus. Using X-rays of skins, we measured the lengths of hand and foot bones to test this, including the metacarpal 4 and proximal phalanx of the 4th digit of the hand, and the metatarsal 4 and proximal phalanx of the 4th digit of the foot. The 4th is the longest digit of the hand and foot. The hypothesis is that flying squirrels reduce the length of the hand and foot to reduce parasite drag.
Gliding adaptations have evolved in Recent mammals at least 6 times: in the Pteromyini (the true flying squirrels, 15 genera including Glaucomys), the Asian Dermoptera (the flying lemurs, Cynocephalus and Galeopterus), the African Anomaluridae (the scaly-tailed ''flying squirrels,'' Anomalurus, Idiurus, and Zenkerella), and 3 times in the Australian fauna (the feather-tailed gliders, Acrobates; the lesser gliding possums, Petaurus; and the greater glider, Petauroides). The fossil record documents 2 or 3 additional lineages of gliding mammals (Meng et al. 2006;Stafford et al. 2002). Among these, flying squirrels probably present the best opportunity for comparison with living species that retain the ancestral morphology. As genetic analyses become less expensive and we approach the era of $1,000 genomes (Service 2006), it may become possible to document the molecular basis for the adaptations of flying squirrels and hence for the evolutionary novelty of gliding flight, which provides an additional motivation for thorough anatomical study and comparison of the Pteromyini with the Sciurini.

MATERIALS AND METHODS
We studied and measured skulls, postcranial skeletons, X-rays of hands and feet of museum skins, and scanned images of caudal vertebrae of Sciurini and Pteromyini squirrels. We also utilized measurements taken by collectors and recorded on the skin tags. Specimens used are listed in Appendix I. Skull lengths, lengths of limb bones, and lengths of segments of trunk skeletons were measured with digital calipers (Fowler Max-Cal, Newton, Massachusetts). X-rays and scanned images were measured from digital images on a computer using analy-SIS Soft Imaging System (Münster, Germany). All specimen measurements were made by the junior author.
Knowing that there is considerable geographic variation in size in the northern flying squirrel, we measured the greatest length of the skull in G. sabrinus from different parts of its range (Arbogast 2007;Hall 1981). We combined our measurements with those of Howell (1918), after measuring a number of specimens that he also measured, comparing our measurements with his, and concluding that differences appeared to be random and never exceeded 0.1 mm (his measurements were recorded to the nearest 0.1 mm). Because of the size variation within the species, we treated G. sabrinus in 2 different samples: the smaller eastern subspecies (G. s. macrotis, G. s. fuscus, and G. s. coloratus) versus the other larger western subspecies (G. s. sabrinus, G. s. fuliginosus, G. s. gouldi, G. s. klamathensis, G. s. lascivius, and G. s. oregonensis).
We used external measurements taken by the collectors and recorded on the skin labels for an initial test of differences between Glaucomys and Microsciurus in the relative lengths of the tail and hind foot. Tail length was subtracted from total length to provide a measure of the length of head and body, which we used as an estimate of size.
Vertebral measurements were taken ventrally on dried, straightened backbones. The vertebral columns were straightened by moistening the specimens with water, pinning them on a foam block, and air drying them. Measurements of thoracic vertebrae T1 through T10 and T1 through T12, thoracic vertebra T11 through lumbar vertebra L7, lumbar vertebrae L1 through L7, and sacral vertebrae S1 through S3 include the lengths of the centra of all the vertebrae in each segment. For Microsciurus, these vertebral measurements were taken from Thorington and Thorington (1989). The ratio of lengths of lumbar vertebrae divided by lengths of thoracic vertebrae is used to test whether there is an elongation of the lumbar region in flying squirrels. The measurements were summed (thoracic vertebrae T1 through T12, plus lumbar vertebrae L1 through L7, plus sacral vertebrae S1 through S3) and these are the trunk lengths used to determine relative lengths of the limbs and the individual long bones. For these, the lengths of the limbs and long bones of each specimen were divided by the trunk length of the same specimen, and these ratios were averaged. We scanned the trunk vertebrae of 14 specimens and all the caudal vertebrae of 127 specimens using a digital scanner. We measured the vertebrae individually using the imaging system after calibrating each image with a scale on the scanned image. For each vertebra, we obtained the average. We summed these averages to obtain trunk length (thoracic 1-sacral 3) and tail length (caudal 1-21), and we obtained the relative lengths of the vertebrae by dividing the average length of each by the appropriate total, either trunk length or tail length. Sample sizes in Tables 1 and 2 are the numbers of vertebral columns that were complete (e.g., all caudal vertebrae 1-21 were present) or for which the relevant measurements could be obtained (e.g., lengths of both forelimb and trunk).
The following limb measurements were taken: the greatest length of the humerus, from the head to the medial margin of the trochlea; the greatest length of the radius from the medial edge of the head to the tip of the styloid process; the greatest medial length of the femur, from the head to the medial condyle; the greatest length of the tibia, from the proximal articular surface to the tip of the medial malleolus; and the medial width of the femur. All limb-bone measurements were taken using digital calipers. The length of the forelimb is defined as the summed lengths of the humerus and radius. The length of the hind limb is defined as the summed lengths of the femur and tibia. Southern flying squirrels (G. volans) have been reported to be sexually dimorphic (Robins et al. 2000). We combined sexes in our study because of the small sample sizes of Microsciurus available and because the 1-2% sexual dimorphism Robins et al. (2000) reported was expected to be small compared with the intergeneric differences.
The hands and feet of museum specimens were X-rayed, using a digital X-ray unit, and measured using the imaging system. Total lengths of metacarpals 2-5, metatarsals 1-5, and corresponding basal phalanges were measured, when their images were visible, that is, not blocked by other bones or the supporting wire within the skin. Only the data for the 4th digits, the longest of the hand and foot, are presented here. The lengths of the foot bones were averaged and the relative lengths were computed by dividing this average and its standard deviation by an average trunk length. This was necessitated by the lack of axial skeletons for the specimens X-rayed. In most cases the average trunk lengths are listed in Table 1. For Hylopetes spadiceus and Iomys horsfeldii the trunk lengths were estimated from the measured trunk lengths of Glaucomys and the relative lengths of the head and body of Hylopetes, Iomys, and Glaucomys, as listed in Thorington and Heaney (1981). These computations should provide good estimates of the averages of relative lengths of the foot bones but are expected to overestimate their standard deviations if trunk and foot bone lengths are correlated.
Ratios sometimes have abnormal distributions. As in previous papers (Thorington and Heaney 1981;Thorington and Thorington 1989), we have found that the distributions of the ratios of body measurements do not differ greatly from normal. A slight skewness of the distributions probably causes the analysis of variance to underestimate the F-value and consequently the significance of the difference.
Analyses of the data all were conducted with Microsoft Excel (Microsoft, Redmond, Washington). In comparing 2 genera that probably have not shared a common ancestor for 20 million years, it is not surprising to find statistically significant differences. Confounding statistical analysis is the complexity of making Bonferroni adjustments on multiply correlated measurements and the 33% probability that both species of Glaucomys will be larger than both species of Microsciurus, or vice versa, in any particular comparison. Our solution to this problem is to consider only differences with extremely large significance values (e.g., cases in which F-values for analysis of variance [ANOVA] are much higher than those for P ¼ 0.001, and usually probabilities are much much lower than 0.00001), in the hope that these modifications of tree squirrel anatomy are the meaningful ones for the evolution of gliding. We report the actual F-values for the ANOVA with 1 degree of freedom in the numerator (Glaucomys species combined versus Microsciurus species combined) and probability levels as low as 0.00001.

RESULTS
Geographic variation in G. sabrinus in skull length caused us to divide the species into 2 samples (see ''Materials and Methods''). We report the values for both samples, except for the hand measurements, in which we included only the eastern subspecies. The eastern subspecies (G. s. coloratus, G. s. fuscus, and G. s. macrotis) include the smallest animals. The largest animals are those of northern Idaho and western Montana (G. s. latipes). For more details, see Appendix II.
The relative lengths of the thoracic to lumbar vertebrae differ between flying squirrels and tree squirrels (Table 1) (Table 1). Therefore, we interpret the relative differences between Microsciurus and Glaucomys to result from elongation of the lumbar region in the flying  (Fig. 1), it is clear that the thoracic vertebrae of Glaucomys are slightly shorter than those of Microsciurus, lumbar vertebrae are longer, that the transition occurs in the vicinity of thoracic vertebrae 10-12, and that there is a graded change in the relative lengths of the lumbar vertebrae with the greatest differences in lumbar vertebrae 4 and 5. The elongation of the lumbar vertebrae causes approximately a 10% elongation of the length of the trunk vertebrae in Glaucomys versus Microsciurus and should account for approximately a 10% increase in the area of the gliding membrane. If one distinguishes between vertebrae with lumbar-type articulations versus those with thoracic-type articulations, a ratio of vertebrae T11-L7/ T1-T10 also emphasizes the elongation of the lumbar vertebrae in the flying squirrels (F ¼ 121.6, d.f. ¼ 1, 113, P , 0.00001).
In the illustration of the trunks of the 4 species (Fig. 2), it is obvious that differences in lengths of the lumbar vertebrae are accompanied by diminution of their transverse processes. The transverse processes of Glaucomys are much smaller than those of Microsciurus, implying that the dorsal musculature is much reduced in the former and suggests that the mass of the animal does not increase as much as might be expected from the elongation of the vertebrae.
In an analysis of field-taken measurements (data not shown), relative tail length (tail length/length of head and body) was not consistently different between Microsciurus and Glaucomys (F ¼ 1.71, d.f. ¼ 1, 97, P . 0.05). These measurements taken by diverse collectors are variable and have large standard deviations, but the ranking of relative tail length (from largest to smallest)-eastern G. sabrinus, M. alfari, western G. sabrinus, However, measurements of the caudal vertebrae (Table 2) demonstrate significant differences between the tails of Glaucomys and Microsciurus. The tails of Glaucomys have slightly fewer vertebrae than those of Microsciurus, and the relative lengths of the individual vertebrae differ systematically between the flying squirrels and tree squirrels. The reduced number of vertebrae is difficult to quantify because of the frequency with which the distal vertebrae are lost during the preparation of specimens, but our samples of skeletons included Glaucomys with 21-23 vertebrae and Microsciurus with 25-27 vertebrae. The proximal 4 caudal vertebrae do not differ, but caudal vertebrae 5-13 are relatively longer in Glaucomys than in Microsciurus, and caudal vertebrae 14-21 are shorter (Fig. 3). The 9 caudal vertebrae C5-C13 average 60% of the length of C1-C21 in Glaucomys but only 48% in Microsciurus. A ratio of the lengths of the elongate middle vertebrae to the others (Table 1) emphasizes the difference between the flying squirrels and tree squirrels (F ¼ 119, d.f. ¼ 1, 58, P , 0.00001). Microsciurus does not differ from other Sciurini in this ratio, implying that Glaucomys is derived in this feature. Hylopetes and Eoglaucomys are similar to Glaucomys in this ratio, suggesting that it is characteristic of small to medium-sized flying squirrels, but a single large flying squirrel (Aeromys) is different.
The relative lengths of the limbs are presented in Table 1. Glaucomys has a significantly larger ratio of forelimb (humerus þ radius) to trunk (thoracic þ lumbar þ sacral vertebrae) than Microsciurus, but not a significantly larger ratio of hind limb (femur þ tibia) to trunk. In this case, the forelimb to trunk ratio of Microsciurus exceeds that of larger members of the Sciurini, being matched only by the ratio of another species of dwarf squirrel, ''M.'' flaviventris. The forelimb to trunk ratio of Glaucomys also exceeds that of larger species of Pteromyini. This suggests that there are 2 contributing factors to the elongation of the forelimb in Glaucomys-a ''small squirrel factor'' as well as a ''flying squirrel factor,'' both increasing the aspect ratio of the wing. The lengths of the long bones are given in Table 3 and their proportions relative to trunk length are given in Table 4. There is little difference between Glaucomys and Microsciurus in the relative lengths of the humerus (humerus/ trunk, F ¼ 7.36, d.f. ¼ 1, 116, P , 0.01), so most of  30.6 6 0.6 (12) 41.8 6 0.9 (13) 46.6 6 0.7 (14) S. deppei 34.6 6 0.4 (10) 32.9 6 0.9 (9) 43.8 6 1. strating that elongation of the forearm is a general feature of flying squirrels. The long bones of flying squirrels are visibly more gracile than those of tree squirrels. The femora of Glaucomys have smaller midshaft diameters than those of Microsciurus (F ¼ 75.6, d.f. ¼ 1, 117, P , 0.00001) and the relative width (diameter/length) is much less (F ¼ 191.7, d.f. ¼ 1, 116, P , 0.00001). Because robustness of long bones is expected to be proportional to body mass, this implies that mass relative to linear measurements is reduced in Glaucomys compared with Microsciurus.
Field measurements of Glaucomys and Microsciurus suggest that the hind foot of the flying squirrels is relatively shorter (hind foot length/head and body length, means and standard deviations) than in the tree squirrels: G. sabrinus, eastern (0.25 6 0.02); G. sabrinus, western (0.23 6 0.03); G. volans (0.24 6 0.02); M. alfari (0.28 6 0.02); M. mimulus (0.26 6 0.01; F ¼ 41.1, d.f. ¼ 1, 97, P , 0.001). Skeletal measurements support this observation (Table 6). The lengths of metatarsal 4 and proximal phalanx 4, relative to trunk length, are significantly shorter in Glaucomys than in Microsciurus. The sum of the 2, which comprise most of the length of the foot, is 22-23% of trunk lengths in Glaucomys and 26% in Microsciurus (F ¼ 117, d.f. ¼ 1, 43, P , 0.00001). In the hands, metacarpal 4 and proximal phalanx 4 also are significantly shorter in Glaucomys than in Microsciurus. As measured by the summed lengths of metacarpal 4 and proximal phalanx 4, the hands exhibit a greater difference than the feet-12-14% of trunk length in Glaucomys and 18% in Microsciurus (F ¼ 182, d.f. ¼ 1, 24, P , 0.00001). This is approximately a 15% difference in relative foot length, and 25% difference in relative hand length. However, a comparison of the relative hand and foot lengths in Microsciurus with other tree squirrels of the Sciurini suggests that Microsciurus has unusually long hands and feet for a tree squirrel, and that it, not Glaucomys, exhibits the derived morphology. Examination of our data for other Sciurini and other Pteromyini suggests that flying squirrels, in general, have shorter hands and feet, relative to trunk length, than tree squirrels (Table 6). However, these data are very dependent on the estimates of trunk length, and we are not confident of these estimates for flying squirrels other than Glaucomys because of small sample sizes (Table 1).
In summary, compared with Microsciurus, G. sabrinus and G. volans exhibit elongation of the lumbar vertebrae, the middle caudal vertebrae, and the long bones of the forearm. They exhibit shortening of the distal caudal vertebrae, the hands, and the feet, relative to Microsciurus (Fig. 4). However, the hands and feet are probably better considered to be elongated in Microsciurus, not shortened in Glaucomys.

DISCUSSION
Contrasting the anatomy, behavior, and ecology of flying squirrels and tree squirrels provides us with concepts of the evolutionary changes required of and resulting from gliding flight. For example, all tree squirrels are diurnal, although crepuscular activity is not uncommon. In contrast, all flying squirrels are nocturnal, although also commonly crepuscular. We presume this behavioral change is in response to predation during glides by diurnal aerial predators in the Miocene, like present-day Cooper's hawks (Accipiter cooperii) and northern goshawks (Accipiter gentilis), which fly much more rapidly than flying squirrels glide. However, testing hypotheses about causation in the Miocene is very difficult. In turn, nocturnality has driven a number of anatomical changes, including relative eye size, a retina comprised largely of rods, and correlated neurological changes (Walls 1942). The patagium itself is the most obvious contrast between flying squirrels and tree squirrels. The leading edge extends from the cheek to the wrist and is lined by a muscle, the platysma II. The importance of the wing tip in flight-for stability, control, and reduction in drag-is reflected in a constellation of anatomical changes exhibited by flying squirrels (Thorington and Darrow 2000;Thorington et al. 1997Thorington et al. , 1998. The major portion of the patagium extends from the body to the wrist and ankle and includes a number of thin muscles, including the tibiocarpalis muscle reaching from the wrist to the ankle (Bryant 1945;Johnson-Murray 1977). The uropatagium between the tail and the ankle is absent from Glaucomys, but very conspicuous in larger flying squirrels. The trailing edge of the uropatagium is supported by a 3rd head of the semitendinosis muscle of the thigh (Johnson-Murray 1977; Thorington et al. 2002).
The shape and size of the patagium determine important aerodynamic features, such as wing loading (body mass/wing area) and aspect ratio (wingspan/average wing chord), and we  The importance of wing loading to flying squirrels is unknown, but the morphological modifications of Glaucomys suggest that it is very important. Aerodynamically, wing loading is expected to affect the minimal speed of gliding, but not the distance (Thorington and Heaney 1981). In a study of G. volans, Bishop (2006) found a significant positive correlation between speed and coefficient of lift, which is anomalous, as she noted, because minimal speed is expected to be inversely proportional to the coefficient of lift, according to aerodynamic theory. This may result from the different gliding behaviors of her 2 animals-the more heavily loaded animal averaged a lower glide angle with a higher coefficient of lift, as well as a higher average velocity than the lighter animal. There was a significant correlation between wing loading and speed in Bishop's analyses of gliding in G. volans, but the correlation between speed and coefficient of lift was slightly larger, causing wing loading to be eliminated in her stepwise multiple regression analysis and leaving only coefficient of lift as a factor contributing to velocity. It seems unlikely to us that flying squirrels defy aerodynamic principles, so we still expect minimal airspeed to be directly proportional to the square root of wing loading divided by the coefficient of lift. Speed affects critical issues such as landing and control in flight. Rapid    (6) Other Sciurini versus Other Pteromyini gliding requires rapid reactions; slow gliding increases vulnerability to gusts of wind. In nature, flying squirrel body mass varies with age, season, and pregnancy, so we know that flying squirrels cope with differing wing loadings. We do not know if there are critical issues of stability and maneuverability among young animals learning to glide or among gravid females, but the importance of wing loading for such animals could be tested in flight chambers. Robins et al. (2000) suggested that the sexual dimorphism in G. volans is important because it reduces the wing loading of the females when pregnant. (Examination of their data shows that females average larger than males in long-bone lengths by 1.0-1.4%.) Their hypothesis seems counterintuitive because wing loading (mass/wing area) would be expected to be slightly larger in the larger females. In order to have lower wing loadings than males, nonpregnant females would need to be leaner. Unfortunately, we were unable to test this hypothesis with our data because mass is seldom recorded for older museum specimens.
We have considered other hypotheses involving the elongation of lumbar vertebrae. The elongation of vertebrae necessarily reduces flexibility by moving intervertebral joints further apart, so flying squirrels should be expected to have stiffer, less flexible lumbar portions of their spines. However, most of the flexibility of tree and flying squirrel spines is accomplished between the 10th thoracic vertebra and the 2nd lumbar vertebra (personal examination of X-rays), and there is very little elongation in this region in flying squirrels. Most of the elongation occurs in lumbar vertebrae 3-7. Accordingly, we think elongation of the lumbar vertebrae has minimal effects on the flexibility of the spine of flying squirrels.
In contrast, the relative elongation of caudal vertebrae 5-13 may well function to increase the stiffness of the tails of flying squirrels. The shorter vertebrae of Microsciurus result in more intervertebral joints per centimeter in this portion of the tail than are found in Glaucomys. We do not understand why this should be important, but we do suspect that flying squirrels while gliding and tree squirrels while leaping use their tails differently for balance and control (Scheibe et al. 2007, and videos 4 and 5, available online at http://dx.doi.org/10.1644/06-MAMM-S-331.s4 and http:// dx.doi.org/10.1644/06-MAMM-S-331.s5).
The elongation of the forelimb has 2 consequences. One is that it will increase the area of the wing, and hence decrease the wing loading. Second, and perhaps more important, it will increase the wingspan and hence the aspect ratio (AR) of the wing (AR ¼ (wingspan) 2 /(wing area) or (wingspan)/average chord length of wing). We hypothesized that evolution would favor elongation of the forelimb, because the aspect ratio is directly proportional to glide ratio (glide distance/vertical drop), so that the elongation should increase the distance of the squirrel's glides. We have pondered why flying squirrels do not have higher aspect ratios that would enable them to glide further yet. We suspect there are 2 trade-offs that account for this. First is the trade-off between the advantage of a higher glide ratio and the disadvantage of reduced agility in tree-trunk locomotion. The latter would be particularly important during mating chases, if these mating chases are similar to those of Holarctic tree squirrels (as seems likely, but is poorly documented). Second, there may be a trade-off between the advantage of a higher glide ratio and the disadvantageous attributes of higher aspect ratios on landing characteristics. Wings with low aspect ratios do not stall until they reach high angles of attack, allowing a flying squirrel to be nearly vertical when it lands on a tree trunk. This enables it to land on its feet, rather than on its nose. The 2nd hypothesis seems less probable, but it is more testable (with models in a wind tunnel) than is the 1st.
Because flying squirrels hold their hands and feet in positions that appear to minimize parasite drag, we hypothesized that a reduction in foot and hand length might also occur. We found that Glaucomys do have shorter hands and feet than Microsciurus, but this seems to be due to the fact that Microsciurus has especially long hands and feet, not that those of Glaucomys are short. Examination of our data suggests that other species of Pteromyini also have shorter hands and feet than the other species of Sciurini. However, because the data are scanty, we considered the issue moot.
In a perspective titled ''How to make a limb?' ' Duboule (1994) described the development of vertebrate limbs in terms of what was then known about Hox genes. The understanding of biochemical pathways of vertebrate development has increased greatly since then (Forlani et al. 2003) and should in the future enable us to determine what modifications in DNA have led to specific morphological changes during evolution. A selective review of the burgeoning literature leads us to suggest some biochemical pathways that may be responsible for the differences we have observed between flying squirrels and tree squirrels. These pathways are the links between the mutations of DNA and the adaptive morphological features of the phenotype that permit or facilitate the evolutionary novelty of gliding flight.
During development, presomitic mesoderm gives rise to the somites, which then give rise to the vertebrae and skeletal musculature. The vertebral modifications we observe in the evolution of flying squirrels from tree squirrels are possibly driven by changes in the molecular pathways during these transitions, in particular Hoxa, Hoxc, and Hoxd, paralogs 10, 11, and perhaps 13 (Deschamps and van Nes 2005). Wellik and Capecchi (2003) demonstrated that Hoxa, Hoxc, and Hoxd 10 are active in the presomitic mesoderm of the lumbar region and determine that the lumbar vertebrae have no ribs, and that the paralogs Hoxa, Hoxc, and Hoxd 11 are active in the presomitic mesoderm of the sacral vertebrae and determine their morphology. Carapuco et al. (2005) suggest that Hoxa11 is active in the somites of the proximal caudal vertebrae as well.
As the presomitic mesoderm gives rise to the somites, the boundaries between the somites appear to be controlled by the Notch pathway in a clocklike manner, under the coordination of the Wnt signaling pathway (Dequeant et al. 2006). The fibroblast growth factor (Fgf8 signaling pathway), coupled with the receptor for the fibroblast growth factor (Hajihosseini et al. 2004), is involved in determining the size of the somites (Aulehla and Herrmann 2004). A gradient of Fgf8 mRNA is established from the tail bud, which establishes a gradient of Fgf8, strongest toward the caudal end, weakest toward the rostral end (Dubrulle and Pourquie 2004), which determines the rate of inclusion of presomitic mesoderm in the somites as they are formed. This gradient, coupled with the timing of the somite boundaries by the Notch pathway, determines the size of the somites and could determine the size of the subsequently formed vertebrae. Another possibility is that the size differential between flying squirrel and tree squirrel vertebrae could be determined later in their embryology by different growth rates of the vertebrae, as described by Sears et al. (2006) for the phalanges in bat wings (discussed below). Because the embryology of tree and flying squirrels has not been studied, we cannot distinguish between these 2 possibilities. Boulet and Capecchi (2004) demonstrated that Hoxa11 and Hoxd11 are active in the forearm during the development of the radius and ulna, and that lack of function of these 2 leads to decreased mesenchymal condensation through reduced expression of Fgf8 and Fgf10 and delayed limb development. However, the major effect was the malformation of the growth plates of both bones. In the growth plates, the chondrocytes failed to mature and rarely formed hypertrophic cells. Thus, modifications of the biochemical pathways controlled by these 2 homeobox genes in the development of the forearm are likely the cause of differing adult forearm morphology in tree squirrels and flying squirrels.
The Hox genes, Hoxa and Hoxd 11, 12, and 13 are similarly important in the development of the hands and feet, particularly through the expression of Fgf8 and Fgf10 (Boulet and Capecchi 2004;Zakany et al. 1997). Knosp et al. (2004) demonstrated that Hoxa13 is essential for the development of the digits of the hands and functions through the expression of the bone morphogenetic proteins Bmp2 and Bmp7. Subsequently, Sears et al. (2006) demonstrated that the elongation of the digits of bats is affected by the up-regulation of Bmp2, which increases the rate of chondrogenesis during development. Bmp2 causes great enlargement of the hypertrophic zone of the growth plate and elongation of the digits, whereas its antagonist, Noggin, reduces the rate of chondrogenesis and causes the digits to be shorter. Upstream regulation of levels of Bmp2 or Noggin could therefore account for the different foot and hand lengths in Glaucomys and Microsciurus. As noted by Sears et al. (2006), this manner of control of the size of skeletal elements could be widespread in vertebrate development. If true, it also could be the basis of the lengthening and the shortening in the flying squirrel vertebral column and forearm.
Delineation of the DNA changes in the evolution of flying squirrels, the molecular pathways affected, and how these have led to their morphology and made gliding flight possible, will ultimately tell us how nature has built flying squirrels. The evolutionary novelty of gliding has evolved 6 different times among Recent mammals, but squirrels probably present the only opportunity to study these details of the evolutionary process, because of the persistence of the ancestral morphology among the American tree squirrels.