The volume and variability of lacuna volume in extant actinopterygians and sarcopterygians are consistent with expectations based on the size of their moderate to small genomes (Table 1). When these data are added to the existing extant vertebrate dataset described above, we find substantial evidence for the relationship between genome and cell (lacuna) volume (P-val < 0.0001). In MCMC, we find that 99% of the posterior distribution of the first slope parameter deviates from zero (posterior β₂, μ = 0.60, σ = 0.21) and 94% of the dummy variable interaction slope deviated from zero (posterior β₄, μ = −0.35, σ = 0.22). We also find relatively large values for both the coefficient of determination (posterior r², μ = 0.59, σ = 0.06) and phylogenetic signal (posterior λ, μ = 0.79, σ = 0.11).

Our retrodictions for genome size in the extinct taxa are shown on Fig. 1. We estimate that Marmorerpeton had a genome size within the range of modern urodeles (Fig. 1A), consistent with our previous work (Laurin et al., 2016). Note that our predictions in log space were identical to Laurin et al. (2016), but our correction for anti-logging (Hayes and Shonkwiler, 2006 and Smith, 1993) resulted in estimates slightly larger than, but consistent with our conclusions in Laurin et al. (2016). The remaining extinct species show no evidence for outliers, with genome size estimates similar to actinopterygians and mammals (Fig. 1B). Genome size in lepidosaurs and non-avian archosaurs are generally smaller than estimates for the fossil non-amniote sarcopterygians, but not significantly so.

The evolutionary transition between an aquatic lifestyle of Devonian sarcopterygians and the branch of sarcopterygians, which became terrestrial and diversified into modern tetrapods is a well-studied evolutionary event in the history of vertebrates (Laurin, 2010). Many records of this transition exist, including insights from comparative genomics (Amemiya et al., 2013, Finn et al., 2014 and Nikaido et al., 2013), preserved fossils that track morphological modifications (Pierce et al., 2012, Pierce et al., 2013 and Shubin et al., 2014), impressions of footprints which, along with limb proportions, reveal gait and motion (Legreneur et al., 2013, Marsicano et al., 2014 and Niedźwiedzki et al., 2010), the locality of fossilized remains, and the rocks in which they are preserved (Bendix-Almgreen et al., 1990 and Ultsch, 1996) which help us, along with faunal associations (Laurin and Soler-Gijón, 2010), to understand the environments and conditions in which this group evolved. But in spite of these records, there are many aspects of this transition that are not preserved directly. We therefore do not understand fully what biological changes occurred as early limbed vertebrates evolved in the Devonian, nor what pressures the environment placed on this new land-going group – a prime example being the subsequent evolution of an amphibious lifestyle.

Studying the macroevolution of genome size is especially problematic. The large genomes of many lissamphibians and lungfish, and the notably small genomes of many teleosts, birds, bats, and some turtles, make an estimation of the ancestral tetrapod genome size difficult to determine with any degree of precision. There may be reasons to hypothesize that genome size changed during the water-to-land transition, but they are poorly supported. For example, some evidence suggests that freshwater species tend to have larger genomes than marine counterparts (Hardie and Hebert, 2004 and Rees et al., 2007). To address this problem, we included data from Eusthenopteron and Ichthyostega, two Late Devonian genera very close to the origin of the tetrapods and that were both primarily or exclusively aquatic, as well as more recent (Permo-Carboniferous) taxa that were more terrestrial. Eusthenopteron bears paired, dorsal, anal and caudal fins, and a pectoral girdle attached to the skull along with other anatomical features, which characterize it as fully aquatic (Laurin et al., 2007 and Long et al., 2006). Ichthyostega, on the other hand, possesses limbs with digits, a pectoral girdle unattached to the skull, a much larger pelvic girdle, and a more structurally established ribcage (Pierce et al., 2013). These features, along with other traits, constitute exaptations of Ichthyostega to terrestrial habitats, because other characters, such as a well-developed cephalic portion of the lateral-line organ and a caudal fin suggest that it was likely aquatic, perhaps with a lifestyle not unlike a mudskipper (Callier et al., 2009 and Pierce et al., 2013). The sampled Permo-Carboniferous taxa include some amphibious or terrestrial taxa, such as Acheloma, the unidentified dissorophoid, and Chroniosaurus (Laurin et al., 2004, McHugh, 2015 and Quemeneur et al., 2013) as well as more aquatic taxa, like Trimerorhachis and Diplocaulus (some of which may be secondarily aquatic). Despite the clear transition in habitat for the adult forms of these extinct species, we find evidence suggesting that the genomes in both habitats and before and after the transition to land were roughly equivalent in size.

The retrodicted C-values for Eusthenopteron and Ichthyostega have important implications for understanding how genome size evolved in extant finned sarcopterygians. The only two groups of extant finned sarcopterygians that persist today are the several species of lungfish (with massive genomes) and coelacanths (with a moderate-sized genome). The primitive C-value condition of sarcopterygians has long been unclear, despite work documenting long-term gradual increase in genome size within lungfish (Thomson, 1972). The additional information from the two Devonian taxa (Eusthenopteron in particular) presented here implies that the large genomes of lungfish are derived and that primitive sarcopterygians had genome sizes much more equitable to the coelacanth L. chalumnae. This finding is interesting in light of the coelacanth exome, which apparently also evolved slowly (Amemiya et al., 2013). It is also consistent with early work on dipnoan genome size evolution (Thomson, 1972).

Lungfish and urodeles have enormous genome sizes compared with other vertebrates (Fig. 2). It has been hypothesized that the simplification of the developmental program in these groups (paedomorphosis) relaxed constraints on genome size, causing them to expand over time (Gregory, 2002b). Despite having massive genomes, lungfish and urodeles appear to lack whole-genome duplications (Kaiya et al., 2014, Panopoulou and Poustka, 2005 and Smith et al., 2009). Such events in plants have been well researched (Mühlhausen and Kollmar, 2013, Song et al., 1995, Vanneste et al., 2014a and Vanneste et al., 2014b) and recent studies have looked towards animals, and specifically vertebrates, in search of evidence of similar whole-genome duplications (Berthelot et al., 2014 and Hufton et al., 2008). There is good evidence that two whole-genome duplications (WGD) events occurred between early chordates and the origin of the osteichthyan stem (Dehal and Boore, 2005, Kasahara, 2007 and Laurin, 2011). Such events were likely crucial for structuring important adaptations of vertebrates, such as developmental genes or the immune system (Otto, 2007), but do not appear to be related to the long-term evolution of genome size (Fig. 2). For instance, the major histocompatibility complex (MHC) in modern vertebrates exists as four distinct paralogs on four chromosomes (Kasahara, 2007). The cephalochordate Branchiostoma has a single MHC region, which likely represents the ancestral version of the four MHC paralogs found within jawed vertebrates. HOX genes, responsible for the major body layout designs of animals, are also found in four clusters on four different chromosomes in most vertebrates. Only one cluster exists in Branchiostoma (Holland et al., 2008) and HOX genes are simplified in the closest relative of vertebrates, the urochordates (Delsuc et al., 2006 and Singh et al., 2009).

Our approach has the potential to detect WGD in extinct lineages, but we find no evidence suggesting they occurred in temnospondyls. The Permo-Carboniferous limbed vertebrates Acheloma, Eryops, Trimerorhachis, and the Permo-Triassic Wetlugasaurus and Chroniosaurus, are all found to have similar, modestly sized genomes within the same range as Eusthenopteron and Ichthyostega. These values support the suggestion that genome sizes did not change dramatically in limbed vertebrates during their early evolution and were very similar to those of other sarcopterygians of the time. The evolution of genome size in lissamphibians is more complex (note that we do not include caecilians in our study owing to a lack of samples, their genome sizes fall within the range of anurans). In this clade, data presented here and elsewhere (Laurin et al., 2016 and Organ et al., 2011) suggest that the Permian lepospondyls Cardiocephalus and Diplocaulus had moderately-sized genomes. Considering the elevated size of genomes in anurans and caecilians, these data imply that genome size expansion in the lissamphibian stem began in the Permian, before the development of distinct urodele and anuran clades. Our analysis also suggests that in the Early Triassic to Middle Jurassic, urodeles underwent a second period of genomic expansion (Fig. 2). However, these lineages are poorly sampled in terms of paleogenomic data and new studies would certainly help clarify their patterns of genome evolution. Analysis of transposable elements (TE) in salamanders suggests that while possessing typical vertebrate families of TEs, their genomes contain larger amounts of long terminal repeat (LTR) retrotransposons, which strongly contribute to the massive genomes of extant urodeles (Herrick and Sclavi, 2014, Sun and Mueller, 2014 and Sun et al., 2012) in addition to expansive intronic sequences (Smith et al., 2009).

The moderately-sized genomes we infer for basal limbed vertebrates suggest that these taxa lacked the repetitive content and enlarged introns characteristic of urodeles or lungfish. By sampling fossil data, our method has the potential to reconstruct very large genomes owing to WGD in basal members of extant groups. The present analysis finds no such evidence, but our paleogenomic approach could be used in the future to better understand the tempo and mode at which genomes shrink following WGDs.