The development of Ensatina differs in many respects from that of most other amphibians. We propose that large egg size is the major reason for most of these differences.
Egg size is a formal evolutionary constraint
An egg can become only so large before a shift from holoblastic to meroblastic cleavage occurs [1, 13, 55, 56]. Such an evolutionary event has occurred at least five times within vertebrates, in lineages leading to hagfishes, elasmobranchs, teleosts, coelacanths and amniotes . The egg size at which the shift from holoblastic to meroblastic cleavages occurs varies between animal taxa; knowing egg size alone does not necessarily predict the type of cleavage. Teleosts and reptiles have meroblastic cleavages, even though some species have eggs the same size as those of amphibians that cleave holoblastically. The identification of a new tissue type in the amphibian embryo - the nutritional endoderm - may provide an intermediate state towards the evolution of amniote meroblastic cleavage . In the large-egged frog, Eleutherodactylus coqui, the nutritional endoderm consists of yolk-rich cells that do not contribute to differentiated tissues. It is possible that Ensatina also has nutritional endoderm, but further study is necessary. By comparing different groups of amphibians (for example, frogs, salamanders and caecilians) it might be determined if these taxa respond similarly to increases in egg size with regard to early development and gastrulation.
We suggest that the asymmetries and asynchronies in early cleavage are influenced by egg size and represent a formal constraint on evolution . This developmental pattern has been reported in previous studies of other salamanders with relatively large yolky eggs: Eurycea bislineata , Desmognathus fuscus  and Cryptobranchus alleganiensis  (diameters of 2.5, 2.8 and 4.0 mm, respectively). The better studied salamander taxa such as Pleurodeles, Triturus and the axolotl (Ambystoma mexicanum) have smaller eggs (1.7, 2.1 and 1.9 mm diameter, respectively) . E. bislineata, D. fuscus and Ensatina represent two of the three major clades of plethodontids, whereas Cryptobranchus is not closely related to plethodontids [42, 43]. The fact that the asymmetries and asynchronies in early cleavage seen in these four species are qualitatively similar and that these species represent two disparate salamander families suggests that large egg size and not phylogenetic relationship accounts for the differences in development from amphibians with smaller eggs. Ensatina, with the largest egg size of any salamander studied to date, offers the most extreme example of asymmetry and asynchrony. Meroblastic-like cleavages of the sort reported in this study were described in E. bislineata, D. fuscus and Cryptobranchus, but in these species a cleavage plane splits the vegetal pole by about the eight-cell stage, earlier than in Ensatina.
Large egg size is clearly a factor in the asymmetrical and asynchronous cleavages seen in Ensatina. There may be biomechanical interference with positioning of the mitotic spindles and the actin contractions associated with cytokinesis of the developing embryo . The cellular mechanisms that can cause atypical cleavage patterns have been the subject of both theoretical studies [36, 38] and reviews. Most of the theoretical work has concentrated on invertebrate patterns. Experimental studies in sea urchin eggs, using centrifugation to concentrate the yolk and alter the plane of cleavage has provided supporting data for the theoretical work . Centrifugation experiments have also been performed on Xenopus to demonstrate the importance of cytoplasmic factors on pattern formation [61, 62]. This technique could be used in studies of early cleavage patterns as a means of testing the effect of a concentrated region of yolk on cytokinesis in amphibian embryos.
There is reason to believe that the mechanisms that govern cleavage in invertebrates may have some relevance to the understanding of amphibian cleavage . Work on invertebrates can help direct research that investigates mechanisms for the production of atypical cleavage patterns in amphibians. For example, a potential factor causing asymmetrical cleavage, not directly related to yolk, is an attachment of the mitotic spindle to the animal cortex of the dorsal region as seen in the mollusc Spisula solidissima . Studies have demonstrated that cortical complexes are important for pattern formation in Xenopus [64–66].
Blastula formation in Ensatina appears to be a modification of the typical amphibian pattern and again the differences can probably be attributed to the large amount of yolk. The blastocoel, compared with the diameter of the egg, is much smaller in Ensatina than in smaller-egged amphibians. Large and yolky endodermal cells make up most of the egg, which means that the animal pole cells represent a smaller proportion of the egg, creating a relatively small blastocoel. A one-cell-thick blastocoel roof means that the cells that form tight junctions to the outer environment of the embryo are also the cells secreting an extracellular matrix on which involuting cells migrate . In Xenopus, the blastocoel roof is several layers thick and consists of two different, stratified populations in terms of protein secretion . This implies that the epithelial sheet of the blastocoel roof in Ensatina has a single cell population that exhibits the role of two cell populations in other species, assuming similar secretory behavior to that in Xenopus. The thin (one cell thick) blastocoel roof is a characteristic of other species of plethodontids, some with relatively small eggs .
Convergent extension versus cell migration
Convergent extension of the involuting cells during gastrulation is the major mechanism moving cells along the blastocoel roof in Xenopus [26, 28, 68, 69]. Migration (the crawling of the cells along the blastocoel roof) is another possible cellular mechanism and is the major, although not exclusive, mechanism of involution in the salamander Pleurodeles waltl [31, 32]. The observations in these two species have been used to argue that cell migration is the more important morphogenetic process during gastrulation in salamanders, whereas convergent extension is the more important morphogenetic process in frogs [32, 70]. Interestingly, more recent analyses of other species of Anurans suggests that there are frog species without convergent extension . The importance of the role of convergent extension during salamander neurulation is accepted [33, 34]; the controversy here is its role during gastrulation. In the salamander A. mexicanum, both migration and convergent extension are important mechanisms [2, 16]. Convergent extension may be occurring during Ensatina gastrulation because Anuran species without convergent extension have symmetrically closing blastopores, whereas Ensatina has an asymmetrically closing blastopore with more involution on the dorsal side. Of course, in other species of salamanders, subduction, involving apical constrictions and ingression of mesoderm, seems to be a major morphogenetic process during gastrulation and the same may well be true for Ensatina [2, 19]. This study could not distinguish which mechanism of involution was more important or even estimate the extent of ingression in this species, but it did demonstrate that cell migration occurs along the blastocoel roof of Ensatina.
Blastopore closure and archenteron formation
Blastopore closure is a biomechanical problem when the egg is as large as in Ensatina [4, 26]. The slowness of the gastrulation process is one consequence. The differences in the rate of blastopore closure between Ensatina and other salamanders are probably due to the larger absolute egg size (including amount of yolk) of Ensatina. Epiboly of the noninvoluting cells presumably helps surround the yolky endoderm cells. The vitelline membrane is very elastic and tightly stretched around the embryo, and it provides support for the flaccid Ensatina embryo. Without this support the embryo could not develop, much less gastrulate. These differences do not entail any new cellular mechanisms.
Keller predicted that, unless there is some fundamental change in gastrulation mechanisms, the dorsal lip of a large-egged amphibian must form below the equator in order for blastopore closure to occur, which seems to borne out in the case of Ensatina . By contrast, the yolky sturgeon egg appears to avoid this difficulty by reproportioning and retiming these movements so that extension occurs first, without convergence, moving the marginal zone below the equator and then converging to close the blastopore [26, 71]. Despite their apparent versatility, convergent extension movements are not universally used in gastrulation, even for relatively small eggs, where they offer few apparent disadvantages. For example, P. waltl, which does not appear to use convergent extension until later stages, has eggs not much larger (1.7 mm)  than those of Xenopus (1.4-1.5 mm) , and their marginal zone lies near the vegetal pole. Subduction, the morphogenetic mechanism for blastopore closure seen in many salamander species (including P. waltl), can also result in asymmetrical blastopore closure [2, 19]. Even other species of frogs, some with larger eggs, do not seem to use convergent extension during gastrulation, as shown by the lack of notochord elongation as visualized by Brachyury staining [2, 8]. It is clear that there is still much to understand about the effect of egg size on morphogenesis.
The archenteron that is formed at an advanced stage of blastopore closure is small in Ensatina (Figure 5). The formation of such a small archenteron is indicative of the formation of an embryonic disk, as described in the tropical tree frog Gastrotheca. riobambae (Anura: Hylidae) [8, 73]. However, the archenteron of Ensatina is larger than that of G. riobambae. Furthermore, it is not symmetrically located around the blastopore but is skewed almost totally towards the dorsal side. Interestingly, this aspect of archenteron formation is more similar to that observed in Xenopus than to that observed in Dendrobatid frogs and G. riobambae . Archenteron formation in Ensatina is similar to that in another large-egged salamander, Andrias japonicus .
Comparisons with other amphibian clades
Early development of large amphibian eggs has rarely been examined in more than a cursory manner, and few reports are available for comparison with our own observations. The third order of amphibians, the Gymnophiona, has some species with large eggs and many of the developmental characteristics seen in Ensatina: initially incomplete cleavage, and presence of a thin blastocoel roof and an embryonic disk [11, 75]. However, early developmental data on gymnophiones are very scarce.
Early development of anurans is known in more detail and in more species. G. riobambae shows the three patterns of early development mentioned for gymnophiones and these patterns are similar in appearance to those observed in this study: initially incomplete, asynchronous and asymmetrical cleavage; a one-cell-thick, translucent blastocoel roof; and an embryo forming from a superficial embryonic disk [9, 53]. In G. riobambae, large egg size has been implicated as a possible explanation for these developmental patterns .
The above discussion of the possible effects of large egg size on early development has been cast in a causal light, concentrating on mechanistic explanations. In discussing the effects of large egg size on early development in an evolutionary context, it must be understood what is meant by large egg size. There are at least two possible criteria for defining what is a large egg size for amphibians: (i) an absolute measure based on the range of egg sizes (generally diameter) in all species of a taxon; or (ii) a relative measure based on a ratio of egg size (such as indicated by cytoplasmic volume) to some developmentally crucial cell parameter such as cytoskeletal elements and/or nuclear volume (typically represented by genome size) [76, 77]. A large egg size as defined by the first criterion will have implications to related quantitative characters such as volume, surface area and mass. We believe it is not unreasonable to assume some correlation between these measurements and another quantitative measure, the amount of yolk. A large egg size as defined by the second criterion is more complicated and probably more realistic. However, there are probably interspecies comparisons where the first criterion is a close approximation of the underlying cellular differences. Any discussion on the effects of large egg size on early development should take these criteria into account.
Both G. riobambae and Ensatina could be considered to have a large egg size by the usage of either criterion. Ensatina has a large egg size in an absolute sense because the egg diameter is close to the maximum observed across all amphibians (this maximum is 10 mm for urodeles and anurans , slightly more for gymnophiones). The absolute egg diameter of G. riobambae is not small, but it is smaller (4-5 mm diameter) than that of Ensatina. However, by taking into account genome size (the only relative and crucial cell parameter for which data are broadly available), the eggs of G. riobambae have a larger effective egg size because of their much smaller genome size [9, 78]. If both species are considered to have a large egg size, because they fulfill either criterion, then the unusual patterns of early development observed in both species (initially incomplete, asynchronous and asymmetrical cleavage; a one-cell-thick translucent blastocoel roof; and an embryo forming from a superficial embryonic disk) correlate with large egg size.
Data on egg size and early development of some other species of amphibians contradict this simple correlation, and suggest a more complicated evolutionary interaction between large egg size (defined by either of the two criteria mentioned) and morphogenetic patterns of early development. Studies on the development of a frog from a different family (Eleutherodactylus coqui; Leptodactylidae) supports this conclusion [55, 79]. The egg diameter of this species is the same as G. riobambae, yet its development is not nearly as unusual as that of G. riobambae; for example, E. coqui does not develop a superficial embryonic disk [55, 79]. Because genome sizes are similar for these two genera, determination of which has a larger egg size based on the relative criteria does not alleviate this apparent contradiction. Embryos of E. coqui do have one unusual feature in their early development, a translucent blastocoel roof, which indicates thinness .
Using the terminology of Gould, we propose that the differences between these two species of anurans suggest that the effect of egg size on two of the three developmental patterns discussed (one-cell-thick blastocoel roof and an embryo forming from a superficial embryonic disk) are not a formal but possibly a historical constraint in frogs . This is in contrast to effect of egg size on early cleavage patterns, which, as stated above, appears to be a formal constraint. If large egg size always correlated with unusual patterns of early development, then egg size could be considered a formal constraint, but this is not the case because frogs with similar sized eggs can have very different developmental patterns.
Comparisons with other species of plethodontids
A related plethodontid, Batrachoseps, shows two of the three patterns of early development seen in Ensatina and G. riobambae (initially incomplete, asynchronous and asymmetrical cleavage and an embryo forming from a superficial embryonic disk), but it differs in having a blastocoel roof two to three cells thick . Other species of plethodontids, such as Desmognathus quadramaculatus and D. wrighti, show all three patterns of early development [40, 41, 44]. Batrachoseps has a genome size similar to that of Ensatina  but an egg diameter (3.4 mm) just over half that of Ensatina; nevertheless, Batrachoseps is considered to have a large, yolky egg relative to that of other amphibians. The two species of Desmognathus have the same genome size , which is less than half that of Ensatina, and differ in egg size, with D. quadramaculatus having a larger egg diameter (3.9 mm) than that of D. wrighti (2.3 mm) . It is possible that the two species of Desmognathus have a large egg size based on a relative criteria using a ratio of genome size to egg diameter, explaining their unusual early development [41, 44]. Further studies need to be carried out on the early development of other species of plethodontids to better understand the extent of variation in early development
In interspecies comparisons, both between anuran families and within plethodontids, the effects of increased egg size on patterns of early development appear complex. The species comparisons that do not support a simple correlation between large egg size and patterns of early development are those between distantly related taxa, suggesting that large egg size may be a useful predictive factor for early development only within an evolutionary lineage.
Factors influencing the rate of development
In most species of salamanders studied, as in this one, development is slow, but in salamanders this may be due not only to egg size but also to the large amount of nuclear DNA, which is known to slow cell cycle time and regeneration [82, 83]. Comparative studies between anurans have led researchers to conclude that large egg size does not necessarily slow the rate of development . Based on comparative studies of plethodontid development, genome and egg sizes seem to affect the rate of early development only when they reach extremely large sizes (Collazo and Wake, unpublished data). The differences in genome sizes necessary to affect the rate of early development may need to be larger than those seen in the four species of plethodontids discussed above. However, the relatively large size of the genome in Ensatina together with its large egg size may be acting synergistically to slow down the rate of development. We believe that the most important single factor for the slow rate of development in Ensatina is probably egg size, simply because of the mechanics of cytokinesis through such a large egg volume and morphogenesis across such a large surface area. The same situation seems to be true in anuran lineages. Even though the genome size of G. riobambae is larger than that of most other hylids (6 pg per haploid genome), it is not much larger, and is in fact small relative to that of salamanders and many other anurans [9, 78, 84]. We see no obvious correlation between large genome size and the unusual patterns of development in the amphibians we have discussed. Further studies of the potential role of large genome sizes on early development are warranted.