This large bone has a diameter of 8 cm at the shaft. The bone cortex is rather thin (maximal thickness: 1.5 cm, locally far less). Extensive perimedullar erosion bays spread into the cortex, which explains its local thinness. There is some amount of medullar bone trabeculae. On a thick section observed with the naked eye, 5 to 6 “cycles” of growth may be deciphered in the cortex (Fig. 1A). Examination of the whole cortex at low magnification ( Fig. 2 ) reveals that it is entirely built by primary periosteal bone tissue. The “cycles” of growth are far less conspicuous at this magnification (actually only one is obvious). The primary bone fits well with the general concept of fibrolamellar bone complex. It is here a very densely vascularized tissue with longitudinal, circular and radial vascular canals. It can be mostly described as plexiform because numerous radial canals occur. However the tissue locally modulates into laminar or longitudinal, depending on the lack of radial, and of radial and circular canals, respectively. As usual, the vascular canals form the lumen of primary osteons, themselves embedded into the “fibrous” component of the complex. The lumen of vascular canals varies locally in diameter, depending on the amount of centripetal lamellar deposition of the primary osteons. Around the marrow cavity and some erosion bays, a thin deposit of endosteal tissue occurs.

Along the caudal transect, the tissue follows the above description, with typical plexiform tissue close to the periphery (Fig. 1B) in the middle of the cortex, where the “fibrous” and “lamellar” components of the complex can be precisely observed (Fig. 1C) and close to the marrow cavity with a thin endosteal deposition (Fig. 1D). Along the medial transect, the primary tissue is generally similar but is locally modified by an extensive amount of Sharpey's fibres that pervade most of the cortex (Fig. 1E). All Sharpey's fibres spread from the periosteal (fibrous) component of the complex, not from the osteonal component. Along the rostral transect, the general structure is the same but there are no Sharpey's fibres and the vascular canals are predominantly circular, giving to the tissue a distinct laminar pattern (Fig. 1F). No Haversian reconstruction is observed, even in the deepest perimedullar regions of the cortex.

The bone has a diameter of 8–9 cm at the shaft. The cortex is relatively thin, with a maximal thickness of 1 cm. There is no active perimedullar erosion, hence the contrast between the cortex and the free marrow cavity is obvious. Only some thin endosteal bone trabeculae protrude from place to place in the marrow cavity (Fig. 3A). Histologically, the cortex has been entirely reworked by Haversian reconstruction into superimposed generations of secondary osteons (Fig. 3B, C).

The bone shaft cross section is oval and irregular, about 3 × 6 cm. The cortex is relatively thick, from 1 to 2 cm. There is a strong contrast between the compact deep cortex and the well-delineated marrow cavity. The marrow cavity contains a high amount of well-developed trabeculae (Fig. 3D), some lining round or oval marrow spaces, others having the shape of crane-like “X struts” with a likely biomechanical significance. As in femur MAD 364, the cortex has been entirely reworked by Haversian reconstruction into superimposed generations of secondary osteons (dense Haversian bone) (Fig. 3E, F).

The bone has a diameter of 6 to 7 cm at the shaft. The cortex is relatively thick, with a maximal thickness of 2 cm. Some perimedullar erosion bay locally invades the deep cortex, while relatively thick extrusions, cone- or dome-shaped, of the deep cortex protrude in the marrow cavity. The external cortex is entirely built by a compact primary tissue with a circular organization. Close observation shows that this tissue is actually the laminar modulation of the fibrolamellar complex. However, the primary osteons of the tissue have experienced a full maturation, such that extensive lamellar bone build-ups of osteonal origin took place while the lumen of the vascular canals became reduced. Scattered Haversian substitution has taken place, and fully “mature” secondary osteons (e.g. with a much reduced Haversian canal) are set even close to the external surface of the cortex (Fig. 4A). The mid- and deep cortices have a structure basically similar to that of the outer cortex, however the amount of secondary osteons is higher (Fig. 4B), and one (perhaps two) lines of arrested growth (LAGs) have interrupted the centrifugal deposition of the laminar tissue (Fig. 4C).

The inner cortex is quite peculiar, the bone is organized as a compacta but has an endosteal origin and is formed of very finely lamellar tissue, crossed by a few simple primary vascular canals (not primary osteons). Locally the bone forms compact, undulating thick “waves” of lamellae. There is much evidence of erosion-reconstruction cycles throughout (Fig. 4D), especially near the marrow limit, where highly remodelled bone trabeculae locally protrude inwards (Fig. 4E). Other evidence of erosion-reconstruction is given by enormous, “endosteal secondary osteons” with an incomplete redeposition (Fig. 4F). The observation of all three complete orthogonal transects in the cortex does not bring evidence of obvious qualitative histological differences between them.

The bone diameter is about 5 cm at the shaft. The cortex is relatively thick, up to 2 cm. The marrow cavity is well delineated from the deep cortex but is peculiar as there is apparently no “free” marrow cavity, most of its space being invaded by numerous thin bony trabeculae forming an extensive spongiosa; only a small, kidney-shaped space remains free of bony trabeculae.

The histology of this bone is closely similar to that of MAD 8814. The primary cortex is built of the laminar tissue pattern of the fibrolamellar complex, namely with extensive circular and longitudinal vascular canals. Here, the woven (periosteal) component of the complex is very much reduced and most of the bone tissue is thus formed by the lamellar (osteonal) component of the complex (Fig. 5A-B). Because most of the bony material, as observed on cross section, is lamellar with a general circular orientation (primary osteons around circular canals) the tissue superficially looks like a simple centripetal deposition of “lamellar-zonal” tissue laid down by the periosteum. This similarity is enforced by the very complete deposition of the osteonal material around many almost obliterated circular canals. This well matured primary tissue is extensively pervaded by Haversian substitution, from the deep cortex to the free bone surface (Fig. 5C-D). Many open cavities represent the early phase of the substitution, by local osteoclastic erosion of the primary cortex. More mature secondary osteons are observed diffusely throughout the cortex, but generally do not form extensive superposed generations of osteons.

The innermost cortex, again, shows similarity with the one in MAD8814 and is formed by endosteal finely lamellar tissue. In some places this endosteal deposit is regular and rather thin, in other regions it shows an extensive development. On the medullar side it turns into highly remodelled, complex endosteal trabeculae (Fig. 5E-F); on the external side, i.e. towards the deepest (oldest) periosteal cortex, there are a high number of large erosion bays with highly remodelled walls formed by endosteal lamellae. There are complete and extensive transitions, through large “endosteal osteons” from those large erosion bays to more typical intracortical secondary osteons (Fig. 5E-F). There are no obvious differences among the three orthogonal transects observed, apart from the different developments of the complex endosteal region.

The putative presence of a phylogenetic signal in bone histological data has been a matter of discussion in recent years. Cubo et al. (2005) concluded that phylogenetic signal was highly significant at the microstructural level for some histological traits, but not for all of them. de Ricqlès et al. (2008) argued that ‘the histological level of organization by itself may reflect at best a weak signal’. Legendre et al. (2013) performed supplementary analyses in a sample including fossils and concluded that most osteohistological features show a significant phylogenetic signal, a conclusion confirmed by supplementary analyses performed on an exhaustive sample of extant ratites (Legendre et al., 2014).

The observations show that most primary bone tissues laid down in the shaft cortex of at least the femur and tibiotarsus of Aepyornis are of the fibrolamellar general type. Tissues modulate locally, being perhaps more plexiform in the femur, more laminar in the tibiotarsus. These tissues can receive moderate to extensive Haversian replacement, Haversian bone sometimes entirely obliterating the primary tissues previously laid down during earlier life. Similar characteristics are often observed among non-avian dinosaurs of “moderate” body sizes, actually among archosaurs generally. Does that mean that one is dealing here with a “dinosaurian” (even archosaurian) histological “signature”, in other words a strong “phylogenetic signal” carried by bone histology? This may well be the case but alternative explanations should not be neglected. It seems unlikely that all the ancestral taxa of Aepyornis, down to the theropodian ancestor of birds, have always and continuously expressed those character-states in their bone tissues phenotypes (de Ricqlès, 2000). Because immediate bird ancestor and early birds were small (Turner et al., 2007), it is unlikely that their bone histological phenotypes were similar to those of most dinosaurs and large ratites. If so, the occurrence of such tissues among ratites would be a secondary development. It may be that the genetic capacity to produce fibrolamellar tissues and dense Haversian bone – given certain ontogenetic requisites – has always been available among archosaurs as a plesiomorphic trait (de Ricqlès, 1993) and retained even without actual phenotypic expression. If so, rather than a simple homoplastic development, the histological situation among large ratites would express “deep homology”. Nevertheless, the control of actual phenotypic tissue characters in a given taxon would be, first and foremost adaptive-ontogenetic autapomorphies, linked with species-specific life history traits (de Ricqlès et al., 2008).

In the context of the hypothesis suggesting that the presence (or the genetic basis) of a fibrolamellar complex may be the primitive condition among amniotes, we can hypothesize that each expression of a fibrolamellar complex in less inclusive clades is homologous among themselves. However, secondary homology tests (according to which only a character state that occurs once on a tree is taken to be homologous; de Pinna, 1991) refute this last hypothesis (for instance, the last common ancestor of sauropsids and that of diapsids does not show a fibrolamellar complex, because neither testudines nor lepidosaurs show this character state, whereas the ancestral condition for archosaurs involves the presence of fibrolamellar complex). Secondary homology tests through parsimony show that each expression of a fibrolamellar complex in less inclusive clades (e.g. ratite birds, if this group is indeed a clade, see introduction) constitutes an independent synapomorphy. This last view is congruent with the facts that: •

each historical event is by definition unique;

This view is also congruent with Dollo's (1893) law of irreversibility, according to which a structure or organ that has been lost or discarded through the process of evolution (e.g. absence of fibrolamellar complex at the nodes Sauropsida and Diapsida) will not reappear in the exact same form in that line of organisms (e.g. presence of a fibrolamellar complex at the node Archosauria). In conclusion, the fibrolamellar complex observed in synapsids and in archosaurs probably share the same genetic basis (“deep homology”), but each expression of this character state in less inclusive clades may constitute independent synapomorphies at the level of these clades.

In the case of Aepyornis maximus, the large body size may not have been reached as fast as in long-legged ratites (i.e. ostrich, rheas, emu and cassowaries), in spite of the high growth rates suggested by the occurrence of plexiform bone tissue. In ornithurine birds including most modern birds, Hesperornis and Ichthyornis, skeletal development is achieved in less than a year, so that bone growth marks are either absent or restricted to the outer cortical layer of bone cortices (Chinsamy, 2005). The evidence for several LAGs in adult-sized femur – if confirmed by further studies – suggests that Aepyornis maximus took several years to reach skeletal maturity. Interestingly, proctracted osteogenesis interrupted by several LAGs is found in New Zealand ratites, including the subfossil moa (Dinornithiformes; Turvey et al., 2005) and the extant kiwi (Apteryx; Bourdon et al., 2009b). It has been suggested that prolonged cyclical growth in New Zealand ratites evolved in response to insular environment with low predation pressure (Bourdon et al., 2009b and Turvey et al., 2005). Protracted cyclical growth also occurs in the putative bird Gargantuavis, which is known only from the Ibero-Armorican island of the Late Cretaceous European archipelago (Chinsamy et al., 2014). Chinsamy et al. (2014) proposed that the growth pattern of Gargantuavis may be linked to the prevailing environmental conditions in an insular setting. About 88 million years ago, Madagascar became isolated from other Gondwanian landmasses, which led to the development of a unique ecosystem with a high level of endemic species (Goodman and Jungers, 2014). This evidence suggests that Aepyornis probably evolved cyclical growth in response to the insular environment of Madagascar. The ontogeny of Aepyornis undoubtedly followed a K-selected evolutionary pathway, as suggested by the enormous size of the eggs. As for the moa (Turvey et al., 2005), it is clear that such strategies would have proved lethal as soon as these bird populations were confronted to human predation, although the extinction of elephant birds seems to have a complex causality (Goodman and Jungers, 2014). The total number of LAGs in Aepyornis cannot be ascertained in this preliminary study, and a precise assessment of its ontogeny will have to await more extensive investigations. On the issue of “growth lines”, it may be useful to pinpoint again that the optimal conditions to observe these may be quite different from the optimal conditions to observe fine histology or cells sizes/shapes.

Comparison of the two femora of roughly the same size (MAD 378 and 364) but with very distinct histologies shows that at least a part of the ontogenic trajectories are faithfully recorded by bone tissue. While MAD 378 strongly suggests a subadult condition with growth still ongoing, MAD 364 histology would record a fully adult or perhaps even an old individual.

Because there is no control in our material of different bones pertaining to the same individual or not, it is difficult to assess the significance of bone-specific differences as observed. Obviously there are microanatomical and histological differences between femora, tibiotarsi and tarsometatarsus. What do they mean? Are they caused by species-specific, organ-specific, ontogenetic or biomechanical or sexual differences? Or express all of them simultaneously?

Along a proximodistal axis in the limb bones, there seems to exist a “gradient” in which the histological “maturation” of cortical bone tissues differ. The femur would show a more “juvenile” histology than the tibiotarsus, and the tibiotarsus more “juvenile” than the other, more distal (and often smaller) bones. This has been fairly well documented in ostriches (Amprino and Godina, 1947) and dinosaurs (Horner et al., 1999) and is not restricted to cursorial limbs. The data available for Aepyornis, although weak, seem to concur: the tarsometatarsus is entirely substituted into secondary bone, an adult-sized femur is still mostly formed by primary tissues, and the tibiotarsi fall in between. This situation may be explained by the bone-specific growth dynamics of different bones within an individual skeleton. For example if a small metatarsal takes the same amount of time to reach its adult size than the much larger tibiotarsus and femur in the same individual (growth isochrony), it makes sense than different tissue types adapted to different growth rates will be laid down in the different bones. Poorly vascularized “lamellar-zonal” tissues would prevail in the smallest bones, and well vascularized “fibrolamellar” ones in the largest. In turn, the difference in initial vascularization of the primary tissues laid down may control the rate and amount of their secondary reconstruction, hence relatively more Haversian bone in the smaller, most distal bones and a more “mature” histology.

There is a distinct difference in the cortico-medullary ratios among the different bones studied. While the tarsometatarsus (especially) and tibiotarsus have a relatively thick cortex, the femur has a relatively thin one. Also, the amount of cancellous trabeculae in the marrow cavity seems to be very different from bone to bone, minimal in the femur, maximal in the tibiotarsus (MAD 8826). Assuming that our sample brings significant data, such differences may express in part the differences in biomechanical loading experienced among the leg bones. In cursorial birds, the femur, during locomotion, may assume an almost horizontal orientation with little actual motions while the more distal limb bones may experience intensive loading changes during each locomotory cycle (Gatesy, 1990). This could be expressed in the cortico-medullary ratios, development of Haversian substitution in the cortex, and amount of the perimedullary spongiosa. In the tarsometatarsus, the bone section departs strongly from a circular shape: this may be linked to the obvious “Warren struts” or X-shaped trabeculae observed in the marrow cavity.

In the tibiotarsus of both species, one has noted the peculiar development of endosteal deposition at the periphery of the marrow cavity. The endosteal bone tissue takes several aspects, from a dense compacta with undulating lamellae to highly remodelled trabeculae forming a spongiosa. The bone is apparently prone to intensive erosion/reconstruction phases, with production of very large “endosteal osteons”, etc. This situation may suggest that this endosteal deposit is linked to the reproductive physiology of the bird, especially to the storage/releasing cycles of calcium involved in yolk and especially eggshell production. Similar situations have been well documented among extant birds (e.g. Meister, 1951, Schweitzer et al., 2005 and Taylor and Moore, 1953), and have been suggested in several dinosaurs as well (Chinsamy, 2005, Chinsamy et al., 2013, Hubner, 2012 and Schweitzer et al., 2005).

A clue may be the comparison of the tibiotarsi in A. maximus (MAD 8814) and A. medius (MAD 8826). As stated above, the taxonomy of elephant birds is controversial, and some or all aepyornithid species may represent just one variable or sexually dimorphic species within each genus (Hume and Walters, 2012). Pronounced sexual size dimorphism occurs in other large ratites that evolved in an insular environment, namely the New Zealand Moa (Bunce et al., 2003). Assuming one is really dealing with two different species, with specific body-size differences, the histology of MAD 8826 would indeed suggest a lower growth rate than in MAD 8814, and also a more mature “older” bone, as suggested by the more generalized Haversian replacement. Alternatively, the data could simply express interindividual differences within a population. Be it as it may, the tibiotarsi histology is more congruent with each other than with any other bones of the sample, suggesting again a strong organ-specific (autapomorphic) signal, even if the samples pertain to different (but closely related) species.