The homeosis hypothesis postulates that in sloths the number of vertebrae with a cervical identity is changed compared to mammals with seven cervical vertebrae, i.e., the number of anterior vertebrae without ribs, as well as the shape of the vertebrae around the cervico-thoracic boundary is changed (Figure 2). Hence, the prediction for Bradypus is that more vertebrae will have cervical shape characteristics, such as the presence of foramina transversaria and more vertebrae will not have ribs (Figure 2). In addition, it is expected that the anterior tubercles, which normally are largest on the sixth vertebra, will be largest on a more caudal vertebra. The reverse is predicted for Choloepus specimens, i.e. they will have fewer vertebrae with cervical shape characteristics, coinciding with the absence of ribs (Figure 2). This is in agreement with shape and rib changes in homeotic mice mutants with fewer or more than seven cervical vertebrae (Figure 3A, [e.g. 39 66 67 68 69 70 ].

The primaxial/abaxial hypothesis postulates that in Bradypus and Choloepus the first seven vertebrae have a cervical identity, which then should be visible in their shape characteristics (Figure 2). In addition, for Bradypus it predicts that the ribs of the 8^th and possibly the 9^th vertebra will only possess proximal and medial (vertebral) rib parts and no sternal parts. Sternal rib parts should be absent because of the hypothesized caudal shift of the abaxial domain, which provides necessary and permissive signalling for the sternal rib parts (Figure 2). For Choloepus, the prediction is that the first seven vertebrae will not have ribs, like almost all other mammals. The hypothesized forward shift of the abaxial domain cannot lead to the induction of ribs or parts of ribs in cervical vertebrae, as abaxial signalling for sternal rib parts is only permissive and not instructive (as discussed above). Hence, the prediction for Choloepus is of a normal mammalian pattern of seven cervical vertebrae.

The homeosis hypothesis predicts that vertebrae at boundaries may have a transitional identity regarding shape and the presence of ribs as a result of incomplete homeotic transformations. This follows from results on homeotic mice mutants and is supported by skeletal patterns in other mammals with homeotic transformations of vertebrae, including humans. Homeotic transformations induced by mutations of Hox genes, or genes upstream of Hox usually appear to be incomplete, resulting in transitional vertebral identities (Figure 3) 39 66 67 68 69 70 71 72 73 74 75 76 ). For instance, a transitional cervico-thoracic vertebra is characterised by an intermediate cervico-thoracic shape and the possession of rudimentary ribs. Rudimentary ribs can consist of only proximal parts, or proximal and medial parts or even of only proximal and sternal parts, possibly with the medial parts as fibrous bands (Figure 3A,C,D,G and 4D) 4 77 78 79 80 . It is important to note that fibrous bands cannot be seen on radiograph, nor in cleared and stained specimens and are also rarely preserved on museum skeletons. Rudimentary ribs are often fused to the adjacent rib (Figure 3D and 4F,H) or, when very short to the transverse process, leading to an enlarged transverse processes (Figure 4A and 4D) 43 81 . It is thought that incomplete homeotic transformation of vertebrae and ribs are common, because the identities of vertebrae are influenced by a set of partially redundant Hox genes [e.g. 13 14 16 18 82 ]. Hence, a mutation in only one gene may not lead to a complete homeotic transformation, and generally further mutations are expected to be necessary for complete transformations.

The primaxial/abaxial hypothesis does not predict intermediate identities of vertebrae, other than the presence of proximal and medial rib parts without sternal rib parts on the 8^th and 9^th vertebrae in Bradypus, due to a caudal shift of the abaxial domain. No rudimentary ribs are predicted for Choloepus. No intermediate shapes of vertebrae are expected, because the mechanism is supposed not to be centered on the vertebrae, but to impact rib formation from a distance and, hence a more gradational mechanism is expected to have but little influence at the vertebral shape, which is supposedly determined by a different mechanism.

The homeosis hypothesis predicts that homeotic changes at the cervico-thoracic boundary may be accompanied by homeotic changes at other boundaries. This follows from results on homeotic mice mutants, in particular mutations in genes upstream of Hox and compound Hox mutations (Figure 3F) 14 18 39 69 73 74 76 83 . A shift at several boundaries is also seen in skeletal patterns in other mammals with homeotic transformations of vertebrae, including humans (Figure 4A,C) [e.g. 10 81 84 85 ]. Hence, the homeosis hypothesis suggests the possibility of transitional and asymmetric vertebrae at other vertebral boundaries, e.g. lumbar vertebrae with rudimentary ribs. A homeotic shift that includes the lumbo-sacral boundary implies a shift of the sacrum, a change of the number of presacral vertebrae and the possibility of sacral vertebrae that are incompletely fused to the sacrum (Figure 3F).

The primaxial/abaxial hypothesis predicts no homeotic transformations of vertebrae at other vertebral boundaries. It also does not predict a shift in the presence of ribs at the thoraco-lumbar boundary. The primaxial/abaxial hypothesis predicts a shift of the sacrum and sites of sacral fusion along the vertebral column, leading to a change in the presacral number of vertebrae, just as the homeosis hypothesis.

The experiments with mutant mice indicate that vertebrae with a transitional identity are usually asymmetric, e.g. with asymmetric shape, with one rudimentary rib larger than the other, or with a unilateral rib (Figure 3A,G). This squares with the asymmetry in transitional vertebrae in other mammals, including humans (Figure 4C). The asymmetry in vertebrae with a transitional identity is presumably due to a coupling between A-P patterning of presomitic mesoderm and the preservation of left-right symmetry of somite formation. Both processes are to an important extent determined by the same A-P gradients of Retinoic acid, Fgfs and Wnts during early organogenesis (Figure 5) [ 20 25 26 see also ref. 10 on unilateral cervical ribs in humans]. Hence, a modification of these gradients is expected to affect both the A-P patterning of the presomitic mesoderm and the left-right symmetry of somites. The homeosis hypothesis, thus, predicts that transitional vertebrae will often be asymmetric regarding shape and the possession and size of ribs.

The primaxial-abaxial hypothesis does not predict left-right asymmetry in the shape of the vertebrae, however if the abaxial domains are not shifted to the same extent on the left and right side, asymmetry in the possession of ribs is expected and asymmetry in the fusion of sacral vertebrae to the sacrum.

The homeosis hypothesis predicts that changes in the number of cervical vertebrae will almost always be associated with pleiotropic effects as earlier documented in humans [ 10 30 , see also 9 43 47 48 86 87 88 89 90 91 ]. The unavoidability of pleiotropic effects is assumed to be due to the strong interactivity between the patterning of the A-P axis (vertebra identity), the dorso-ventral axis and the left-right axis during early organogenesis (Figure 5). By extension, the homeosis hypothesis predicts a wide variety of pleiotropic effects to be associated with homeotic transformations of cervical vertebrae in sloths.

Primaxial/abaxial patterning of the migrating sternal rib cells occurs after the highly interactive early organogenesis stages (in mice after E.D. 11, see 92 ), when development has become more modular and compartmentalized and mutational changes with an effect on these stages are associated with fewer pleiotropic effects 27 31 . Hence, changes affecting primaxial/abaxial signaling may be expected to be primarily associated with only local pleiotropic effects in neighbouring structures. In agreement with this, Buchholtz and Stepien 56 argue that potentially deleterious pleiotropic effects, such as found in humans with cervical ribs, are not expected in sloths.

Comparison of the shape of the most rostral vertebrae of wild-caught Choloepus and Bradypus specimens with those of other mammals confirms the traditional view that sloths have an abnormal number of cervical vertebrae. In the investigated Choloepus hoffmanni specimens, we found that the four or five most anterior vertebrae have no ribs and a cervical shape, with bilaterally a foramen transversarium, spinous processes with a dorsal and slightly caudal orientation and transverse processes with a lateral and somewhat ventro-caudal orientation (Figure 1C). The anterior tuberculi, that normally characterize the sixth vertebra in mammals were present on the fourth or fifth vertebra, indicating a homeotic transformation. The sixth vertebra had a transitional cervico-thoracic shape and rudimentary ribs. The seventh vertebra had a completely thoracic shape (Figure 1C) and full ribs that were fused to the sternum (Table 1). Thoracic shape characteristics included dorsolaterally oriented transverse processes, posteriorly oriented spinous processes and rib articulation facets. In the investigated Choloepus didactylus specimens, we found that the five or six most anterior vertebrae had cervical shape characteristics and no ribs (figure 1A and 1B). The seventh vertebrae always had a transient cervico-thoracic shape and rudimentary ribs (figure 1A and 1B, Table 1) and never a completely thoracic shape as in Choloepus hoffmanni. In most of the investigated Bradypus specimens, the first eight vertebrae had a fully cervical shape with bilaterally a foramen transversarium and the largest tuberculi anterior on the eighth vertebra instead of on the sixth as in other mammals (figure 1C). In two specimens, the eighth vertebra had a transitional cervico-thoracic shape (table). In all investigated specimens the ninth vertebrae had transitional cervico-thoracic shape characteristics and rudimentary ribs (Figure 1C,D, Table 1). The tenth vertebrae were in all cases the first fully thoracic vertebrae with complete ribs that were fused to the sternum, comparable to the eighth vertebrae in other mammals.

Hence, the shapes of the vertebrae and the presence or absence of ribs support that homeotic transformations of vertebral identity have taken place in sloths, resulting in a larger number of cervical vertebrae in Bradypus and a smaller number of cervical vertebrae in Choloepus (cf. Figures 1 and 2).

Clearly, our findings do not support the primaxial/abaxial hypothesis. This hypothesis can only explain the absence of sternal rib parts on the 8^th and 9^th vertebrae in Bradypus, i.e. rudimentary ribs that consist of the proximal and medial rib parts. However, the completely missing ribs on the eighth cervical vertebrae in Bradypus cannot be explained, nor the rudimentary ribs on the ninth that only consist of the proximal part of the rib (Figure 1C,D cf Figure 2). Similarly, the presence of rudimentary ribs on the sixth and full or rudimentary ribs on the seventh cervical vertebrae in Choloepus are not predicted by the primaxial/abaxial hypothesis (Figure 1A,B cf Figure 2). Buchholtz and Stepien 56 assume that such rudimentary ribs can be explained by the primaxial/abaxial hypothesis, but this would not only require instructive abaxial patterning of these rib parts and the development of isolated sternal rib parts, but also the unlikely articulation or fusion of sternal rib parts to the vertebrae, far away from the sternum. The latter would require the migration of future sternal rib cells away from the vertebrae towards the developing sternum to receive abaxial signalling, followed by the migration away from the sternum back towards the vertebrae. Finally, none of the shape changes of cervical and thoracic vertebrae in Bradypus and Choloepus was in agreement with the hypothesis.

As already mentioned above, the vertebrae at the cervico-thoracic boundary often had a transitional identity in Bradypus and Choloepus. The vertebrae at other vertebral boundaries also often had a transitional identity. At the thoraco-lumbar boundary transitional vertebrae had a transitional shape and rudimentary ribs (Figure 6A). At the lumbo-sacral and sacro-coccygeal boundaries, incomplete sacral fusions indicate transitional vertebral identities (Figure 6B,C). These transitional identities are in agreement with the predictions of the homeosis hypothesis.

The primaxial/abaxial hypothesis does not predict the observed rudimentary lumbar ribs in Bradypus. The rudimentary lumbar ribs in Choloepus are also not in agreement with this hypothesis, because they are not only missing the sternal parts, but also the medial parts of the ribs. The primaxial/abaxial hypothesis cannot explain the transitional shape characteristics of vertebrae at the lumbo-sacral and sacro-coccygeal boundaries which rather point to incomplete homeotic transformations. Incomplete and asymmetrical fusions at these boundaries are not in agreement with this hypothesis, because they were associated with transitional shape characteristics of the involved vertebrae.

Vertebrae with a transitional identity usually had a strong left-right asymmetry in Bradypus and Choloepus (figures 1C, 6A,C). This was true for vertebrae at all vertebral boundaries and could be apparent both in the shape of the vertebrae, the size and shape of the ribs and the extent of the fusion with the sacrum. The asymmetry is predicted by the homeosis hypothesis and only for the presence of ribs for the primaxial/abaxial hypothesis.

We found many abnormalities in the skeletons of Choloepus and Bradypus specimens, affecting vertebrae, ribs, sternum, cranium, pelvic girdle and limb bones (Figure 7, Table 1). Fusions of cervical vertebrae were found to be quite common, in particular of the second and third vertebrae, but also of other cervical vertebrae (Figure 7A). Fusions of vertebrae can be caused by segmentation defects during early development due to the link between segmentation and A-P patterning (Figure 5) or by abnormal fusion, which can be due to increased ossification at a later stage. Furthermore, we found defective chondrification and ossification of the sternum and pelvic girdle in three Choloepus specimens (Figures 7B,C, Table 1). Abnormal fibrous bands were found in two Choloepus specimens that connect rudimentary ribs with the sternum and apparently had developed from the anlagen of medial and distal rib parts instead of the normal bony parts [see 93 ]. Other abnormalities were asymmetric vertebrae and ribs, presumably caused by a disturbance of the preservation of left-right symmetry of developing somites 25 26 . These abnormalities in Choloepus and Bradypus specimens are in agreement with the predictions of the homeosis hypothesis In contrast, the primaxial/abaxial hypothesis does not predict this wide- array of skeletal abnormalities in different parts of the body, especially not the observed disturbances in chondrification and ossification and possibly segmentation.

Manatees, like Choloepus, have a reduced number of cervical vertebrae, based on shape characteristics and the number of anterior vertebrae without ribs (Figure 1E,F). Comparison of skeletons of wild-caught manatees (Trichechus manatus and T. senegalensis) with those of sloths showed a surprisingly high similarity. Manatees also typically displayed incomplete homeotic transformations at the cervico-thoracic and other vertebral boundaries: they usually had rudimentary ribs anterior to the first full ribs and the shape of vertebrae at the cervico-thoracic boundaries was often transitional (Figure 1E, Table 2). At the thoraco/caudal boundary vertebrae also often displayed a transitional character in shape and in presence of ribs (Figure 6D). More caudally it was difficult to distinguish clear vertebral boundaries, due to the reduction of the pelvic girdle and the absence of sacral fusions. Another similarity with sloths was the strong left-right asymmetry of transitional vertebrae, e.g. regarding size and presence of foramina transversaria (cervico-thoracic boundary), size of transverse processes (thoraco-caudal boundary) and size of rudimentary ribs (cervico-thoracic and thoraco-caudal boundary). The seventh vertebra of the investigated manatee specimens always had a full thoracic identity (Figure 1E,F, Table 2). The sixth cervical vertebrae were transitional with rudimentary ribs and often missed foramina transversaria (1E). Sometimes the fifth cervical vertebra also had rudimentary ribs (always smaller than those on the sixth).

We found many abnormalities in the skeletons of manatees (Table 2, Figure 7). Particularly frequent, like in Choloepus were fusions of the second and third cervical vertebrae (Figure 7D, Table 1). In Choloepus and Bradypus we found abnormally shaped and asymmetric vertebrae, ribs and sterna (Figure 7D). Midline fusion defects of the sternum were particularly common (Figure 7E), as well as reduced and unfused spinous processes of cervical vertebrae (Figure 7D). Finally, we also found large abnormal fibrous bands, including fibrous bands that connected rudimentary ribs to the sternum (Figure 7F). Hence, several skeletal abnormalities were remarkably similar to those found in sloths.

We investigated the vertebral pattern in 27 wild-caught armadillos (mainly Euphractus sexcinctus and Dasypus novemcinctus, see Table 4) and 8 wild-caught collared anteaters (Tamandua tetradactyla), sister taxa of the sloths within the Xenarthra. In armadillos and all but one collared anteater, the number of cervical vertebrae was found to be seven, as is typical for mammals (Table 3 and 4). We did find several transitional vertebrae at the thoraco-lumbar boundary with rudimentary ribs and at the lumbo-sacral boundary with incomplete sacral fusion. Transitional sacro-coccygeal vertebrae were quite common.

We did not find skeletal abnormalities in these specimens, other than the above-mentioned transitional vertebrae and irregulaties in tail vertebrae in some collared anteaters (Table 3). The transitional vertebrae commonly displayed left-right asymmetry, as in sloths and other mammals. Postnatal fusions of cervical vertebrae are typical for armadillos and supposedly adaptations to their digging habits 94 95

One anteater specimen was found to have rudimentary ribs on the seventh cervical vertebra and it had unilaterally one tuberculum anterior on the fifth cervical vertebra and one on the sixth vertebra, indicating incomplete homeotic transformations of C5, C6 and C7 (Table 3). This specimen was found to have incomplete ossification of the sternum and a phalanx missing in digits 1 and 2 of the left hindlimb (oligodactyly).

We investigated 11 wild-caught specimens of dugongs (Dugong dugon), a sister taxon of manatees within the order Sirenia. Unexpectedly, we found in approximately half of the specimens a reduced number of cervical vertebrae (6 of 11, Table 5). These speciments all had 6 cervical vertebrae and rudimentary ribs on the seventh vertebra, signifying a transitional cervico-thoracic identity, Figure 8B). All of these specimens had skeletal abnormalities, including irregularly shaped and asymmetric vertebrae, ribs and sternums, fusions of cervical vertebrae and ribs, abnormalities in limbs and scapula, and midline fusion defects in the skull (Figure 8C,F,G). We did not find skeletal abnormalities in those specimens that had the normal number of seven cervical vertebrae.

Furthermore, we investigated 16 wild-caught specimens of hyraxes (Procaviidae), a sister taxon of the Sirenia within the Paenungulata (Dendrohyrax arboreus, D. dorsalis and Procavia capensis). We found one D. arboreus and one P. capensis with 6 cervical vertebrae and rudimentary or complete ribs on the seventh cervical vertebrae (Figure 8A, Table 6). Both specimens had transitional thoraco-lumbar vertebrae and the D. dorsalis specimen also had an asymmetric sixth cervical vertebra with on the right a C6 identity and on the left a C7 identity (Figure 8A). Furthermore, both specimens had skeletal abnormalities, including irregularly shaped vertebrae and ribs, malformed scapulae and oligodactyly, Figure 8D,E, Table 6). The remaining 14 specimens had the normal number of seven cervical vertebrae with sometimes transitional vertebrae at more caudal boundaries, Table 6). We did not find any further skeletal anomalies in these specimens.

We found that the number of cervical vertebrae in Choloepus and Bradypus has changed from the standard seven of most mammals due to homeotic transformations of vertebrae, as first suggested by Bateson 34 . We conclude this based both on the shapes of the vertebrae and the absence or presence of ribs (cf. Figure 1 and 2). There was generally good agreement between the degree of transformation of vertebrae and the size of the ribs, e.g. vertebrae with full ribs displayed fully thoracic shape characteristics, whereas vertebrae with one or more rudimentary ribs displayed a transitional cervico-thoracic shape. Homeotic transformations also occurred at more caudal boundaries, as demonstrated by the presence of vertebrae with a transitional identity at these boundaries (Figure 6). The vertebral pattern with homeotic shifts along a large part of the vertebral column mostly resemble the vertebral patterns of mice with mutations in genes upstream of Hox, which affect multiple Hox genes [e.g. 69 74 76 83 96 97 , see also 23 ]. This is in agreement with preliminary findings of us that mutations in genes upstream of Hox are involved in the development of abnormal number of cervical vertebrae in humans (Bakker et al., unpublished data).

All other predictions of the homeosis hypothesis are also supported by our findings on the derived skeletal patterns of sloths. Incomplete homeotic transformations of vertebrae were common at all vertebral boundaries and transitional vertebrae often displayed strong left-right asymmetry. Furthermore, we found many skeletal and fibrous band abnormalities in Choloepus and Bradypus, similar to those found in humans and mice with a changed number of cervical vertebrae (cf. Figures 4D and 7F; 10 43 81 86 87 88 98 for humans, 19 66 67 72 75 76 99 for transgenic mice). This supports the hypothesis that homeotic changes of the number of cervical vertebrae in mammals are unavoidably associated with pleiotropic effects 10 .

Buchholtz and Stepien 56 argue instead that homeotic transformations of cervical and thoracic vertebrae have not taken place in sloths. However, their illustrations of vertebrae of Bradypus and Choloepus clearly show similar signatures of a homeotic transformation as did our specimens, i.e. a cervical shape of the eighth vertebra of Bradypus with bilateral foramina transversaria, no articulation facets and a cervical orientation of the processes (their Figure 1D) and a thoracic shape of the sixth and seventh vertebrae of Choloepus with no foramina transversaria, the presence of articulations facets and a thoracic orientation of processes (their Figures 2C and 2F). In addition, in their Figures 2C and F the shapes of the sixth and seventh vertebrae in Choloepus indicate respectively a transitional cervico-thoracic identity and a completely thoracic identity. Hence, their figures strongly support homeotic transformations of the vertebrae.

Most predictions of the primaxial/abasxial hypothesis are not supported by the vertebral anatomy of sloths. Only the absence of distal parts of ribs on the eighth and ninth vertebra in Bradypus can be explained by the primaxial/abaxial hypothesis, but not the absence of the medial rib parts or the complete absence of ribs (cf. Figure 1C,D with Figure 2). In addition, no other rudimentary ribs in Choloepus and Bradypus can be explained by this hypothesis, i.e. rudimentary ribs on cervical vertebrae in Choloepus and on lumbar vertebrae in Bradypus. Not only that, the cervical shape characteristics of the eighth vertebra and the cervical or cervico-thoracic characteristics of the ninth vertebrae in Bradypus, are in contradiction with this hypothesis and clearly demonstrate a homeotic transformation. Similarly, the cervico-thoracic or completely thoracic shape characteristics of the sixth and seventh vertebrae in Choloepus cannot be explained by this hypothesis nor other homeotic transformations of shapes, such as the shifted position of the largest anterior tuberculi. Finally, the primaxial/abaxial hypothesis does not predict the common left-right asymmetry of vertebrae and sternum, nor the associated skeletal abnormalities (Figures 6 and 7).

Manatees also have an exceptional number of cervical vertebrae and like in sloths this appears to be due to homeotic transformations. As in sloths, we find incomplete and complete homeotic transformations of vertebrae, as apparent from the shape of vertebrae and the absence or presence of ribs. Furthermore, we find that transitional vertebrae are usually asymmetric and that homeotic transformations occur at different vertebral boundaries. Hence, we find striking similarities between the vertebral patterns in sloths and manatees. In addition, we also find that the changes of the number of cervical vertebrae in manatees co-occur with skeletal and fibrous band abnormalities.

Our results, thus, strongly support the traditional view that the number of cervical vertebrae in manatees is changed due to homeotic transformations as in sloths 6 8 41 . The results also confirm that homeotic changes of cervical vertebrae generally co-occur with pleiotropic effects in other parts of the body.

Most investigated specimens of armadillos, anteaters and hyraxes, have the normal number of seven cervical vertebrae. Transitional vertebrae at more caudal boundaries were found to be common, as is more generally the case in mammals (Figure 3F,G,I and Figure 4A,C) [see further 74 ]. The number of vertebrae in caudal regions shows more interspecific variation in mammals 6 and, hence, caudal homeotic transformations appear to be less constrained, in agreement with an observed weaker selection against such changes in humans 10 . The transitional vertebrae were usually asymmetric, which appears to be a more general phenomenon, as asymmetric transitional vertebrae also appear to be common in Anolis lizards (at the thoraco-lumbar and more caudal boundaries, Andre Pires da Silva, pers. comm.). Unexpectedly, we found that 6 of 11 investigated specimens of Dugongs had a changed number of cervical vertebrae and rudimentary ribs on the seventh vertebrae (Figure 7B, Table 5). Thus, although this species is thought to have the normal number of vertebrae for mammals, it seems that the constraint on changes of the number of cervical vertebrae has also been broken in this sisterspecies of manatees, albeit to a lesser extent. In addition, we found that two out of 17 hyracoid specimens and one out of 8 anteater specimens had an abnormal number of cervical vertebrae, with rudimentary or full ribs on the seventh vertebra (Figure 8A, Table 4 and 6, Lin and Asher also found cervical ribs in some P. capensis specimens, pers. comm.). Hence, it is possible that the evolutionary constraint on changes of the cervical vertebral number is also weaker in these sister taxa. It will be interesting to investigate the strength of the evolutionary constraint in xenarthrans and paenungulates and possibly other afrotherians further. A first datum already is that the constraint on the number of presacral vertebrae also appears to be relaxed in afrotherians and xenarthrans 6 132 . One reason may be that biomechanical constraints on both transitional cervico-thoracic and lumbo-sacral vertebrae are weak due to relatively low activity rates of members of these taxa.

Importantly, none of the anteater, armadillo, dugong and hyrax specimens with a normal number of seven cervical vertebrae had conspicuous skeletal abnormalities, other than transitional vertebrae at caudal boundaries and minor anomalies such as irregular vertebrae in the long tail of anteaters. In contrast, all Dugong, anteater and hyracoid specimens with 6 cervical vertebrae had conspicuous skeletal abnormalities (Figure 8B to 8G), as we also find in manatees and sloths.

The skeletons of sloths and manatees that we investigated had a surprisingly high number of abnormalities, similar to those found in mice and humans with an abnormal number of cervical vertebrae. In addition, the dugong, anteater and hyrax specimens with a changed number of cervical vertebrae, also had similar skeletal abnormalities. This strongly supports the hypothesis that changes in the number of cervical vertebrae in mammals are unavoidably accompanied by pleiotropic effects 8 9 10 . These abnormalities were shown to lead to strong, indirect, selection against these changes in humans and we expect similarly strong selection against such abnormalities in most other mammalian species, especially in highly active mammals. However, it is plausible that skeletal abnormalities are less harmful in sloths, manatees and dugongs, because of their low activity and low metabolic rates. We cannot exclude that some of the abnormalities are also harmful for sloths, manatees and dugongs, for instance the virtually incomplete ossification of the pelvic girdle and sternum in two Choloepus specimens. Nevertheless, the high frequency of the abnormalities in wild-caught specimens supports the notion that fitness effects of the abnormalities appear to be limited.

We only investigated skeletons, hence we have no information on potential abnormalities in other tissues. Data on humans with a changed number of cervical vertebrae show that many other deleterious pleiotropic effects contribute to a lower fitness of these individuals and lead to strong prenatal and neonatal selection. In particular, in those individuals with a changed number of cervical vertebrae that survive the prenatal and neonatal period, there is a strongly increased chance for childhood cancers (120-fold, 8 ). The extremely low metabolic rates of manatees and sloths are expected to lower cancer rates 9 . A similar effect would apply to dugongs, anteaters and hyraxes as these also have low metabolic rates, albeit not as extreme as in sloths and manatees 133 134 135 136 and, hence, they may also have reduced cancer rates.