(Non)Parallel developmental mechanisms in vertebrate appendage reduction and loss

Abstract Appendages have been reduced or lost hundreds of times during vertebrate evolution. This phenotypic convergence may be underlain by shared or different molecular mechanisms in distantly related vertebrate clades. To investigate, we reviewed the developmental and evolutionary literature of appendage reduction and loss in more than a dozen vertebrate genera from fish to mammals. We found that appendage reduction and loss was nearly always driven by modified gene expression as opposed to changes in coding sequences. Moreover, expression of the same genes was repeatedly modified across vertebrate taxa. However, the specific mechanisms by which expression was modified were rarely shared. The multiple routes to appendage reduction and loss suggest that adaptive loss of function phenotypes might arise routinely through changes in expression of key developmental genes.


| INTRODUC TI ON
The vertebrate appendage demonstrates substantial diversity in form and function, having evolved into fins, wings, flippers, claws, hooves, and myriad other structures. Appendage reduction and loss is also a significant component of vertebrate appendage evolution. Repeated, independent instances of appendage reduction and loss offer an opportunity to investigate the extent to which the developmental bases of phenotypic evolution are shared and unique (i.e., (non)parallel) across vertebrate lineages (Bolnick et al., 2018).
Here, we review molecular pathways involved in appendage development to ask whether shared or unique genetic and developmental mechanisms are involved in independent instances of vertebrate appendage reduction and loss. For consistency, we chose to use the nomenclature rules usually reserved for mouse and rat (Gene and PROTEIN) throughout our review. Because there are no established guidelines for the discussion of regulatory elements, enhancer symbols will be capitalized and italicized (ENHANCER) ( Table 1).
Comparing the molecular drivers of appendage reduction and loss across vertebrate clades required that we find taxa that (a) show appendage loss or reduction and (b) have data on the molecular and developmental components driving reduction. Though there are hundreds of independent instances of lost or reduced appendage elements reported for vertebrates, we found only a handful of taxa for which the molecular pathways involved are described even in part, likely limited by the difficulty of studying development in nonmodel organisms.
The cases we did find span 450 million years of vertebrate evolution, from teleost fish to mammals (López et al., 2016). To address generality in appendage loss and reduction across vertebrates, we therefore must discuss homology between teleost fins and tetrapod limbs.
Teleost fins and tetrapod limbs arose by modifications to the paired fins of their last common ancestor and are superficially similar in position and function (Hall, 2007). Ancestral gnathostome fins were composed of long-bone segments arranged into three structures along the anteroposterior axis: the propterygium, the mesopterygium, and the metapterygium (Coates, 1994;Don et al., 2013;Hawkins et al., 2021) (Figure 1). In teleosts, the propterygium and mesopterygium form the fins, whereas the metapterygium is lost (Coates, 1994;Don et al., 2013;Hawkins et al., 2021) (Figure 1).
In contrast, only a modified metapterygium is retained in tetrapod limbs (Coates, 1994;Don et al., 2013;Hawkins et al., 2021). Thus, the teleost fin and the tetrapod limb are derived from distinct tissues. However, despite originating from different tissues, a sort of "deep homology" underlies fin and limb development (Shubin et al., 1997(Shubin et al., , 2009. That is, much of the genetic architecture controlling appendage development is shared between teleosts and tetrapods (Hall, 2007). For example, the Hedgehog pathway plays a role in anteroposterior appendage patterning and maintaining downstream gene expression in both fish and tetrapods (Chiang et al., 2001;Lettice et al., 2003;Ros et al., 2003;Sagai et al., 2005). Alterations to this signaling pathway result in aberrant appendage development and morphology in both clades: experimental loss of Shh expression resulted in truncated limbs in mice and in fin absence in the teleost medaka (Oryzias latipes) (Chiang et al., 1996;Letelier et al., 2018;Sagai et al., 2005). Similarly, the expression and function of Gli3, a Shh antagonist, is conserved from fish to tetrapods (Letelier et al., 2020). Gli3-knockout medaka grow extra fin elements; Gli3-deficient mice develop a similar polydactyl phenotype (Letelier et al., 2020;Litingtung et al., 2002;Lopez-Rios et al., 2012;te Welscher, Zuniga, et al., 2002).
Regulation of Hox genes, a gene family important for embryo patterning in most animals, is also shared in teleost fins and tetrapod limbs (Ahn & Ho, 2008;Cohn & Tickle, 1999;DuBuc et al., 2018;Hall, 2007;Parrish et al., 2009;Ramos et al., 2012;Ryan et al., 2007;Scott, 1993;Tanaka et al., 2005). For example, Hox genes are expressed in three phases in the pectoral appendage of zebrafish and chick; orthologous genes are expressed in similar regions of the appendage during each phase (Ahn & Ho, 2008).
Having supported homology between fins and limbs, we now define appendage reduction and loss, the main criteria for taxon inclusion for this review. Defining "loss" is straightforward: the absence of one or more bones from the appendage, from pelvic or pectoral girdles to fin rays or digits. "Reduction" has had a more varied definition over its study (Bickley & Logan, 2014;Brandley et al., 2008;Chiang et al., 2001;Greer, 1991;Klepaker et al., 2013;Kragesteen et al., 2018;Thompson et al., 2018;Wiens et al., 2006). For our review, we consider "reduction" to be a diminishment in the relative length or width of at least one bone in the appendage.
We now divide the rest of our review by clade, appendage, and modification type to allow for comparisons between taxa and establish if the same molecular mechanisms are used for appendage reduction or loss by distantly related vertebrates. While the complex gene regulatory networks dictating appendage development may

Gene symbol
Gene name

Sonic Hedgehog
Gli3 GLI Family Zinc Finger 3 Hox a gene family comprising a subset of homeobox genes

Hand2
Heart and neural crest derivatives expressed 2 offer numerous routes to reduction and loss, we found that these phenotypes most often resulted from modified regulation of the same of key developmental genes (Table 2).
The stickleback pelvic appendage is a modified pelvic fin comprised of two articulated spines and a bony girdle that extends along the belly and up the sides of the fish. Over 100 geographically distinct freshwater populations of G. aculeatus have evolved reduction and/or loss of the pelvic spines and girdle (Bell et al., 1993;Chan et al., 2010;Coyle et al., 2007;Klepaker et al., 2013;Shapiro et al., 2006Shapiro et al., , 2009Shikano et al., 2013). Because these freshwater populations were independently colonized by marine ancestors at the end of the last glacial maximum (Schluter & McPhail, 1992), they represent repeated instances of evolution

F I G U R E 1
The teleost pectoral fin is based on zebrafish fin morphology, while the tetrapod forelimb is based on human anatomy. Elements of the ancestral pectoral fin are retained and modified in extant vertebrates: Appendage structures are colored to reflect their evolutionary origins. The propterygium (yellow) and mesopterygium (red) were retained and modified in teleost evolution while the metapterygium (dark and light blues) makes up the tetrapod limb. The proximal portion of the metapterygium (dark blue) likely forms the stylopod, while the more distal elements (light blue) were likely elaborated into the distal limb structures (Ahn & Ho, 2008;Don et al., 2013;Freitas et al., 2007;Hawkins et al., 2021) F I G U R E 2 A simplified gene regulatory network implicated in vertebrate appendage development. Genes symbols coded in magenta are unique to the hindlimb, while those in blue are unique to the forelimb (Butterfield et al., 2009;Charité et al., 2000;Delgado et al., 2021;Delgado & Torres, 2015;Fernandez-Teran et al., 2000;Hockman et al., 2008;Jin et al., 2019;Lafage-Proust, 2015;McQueen & Towers, 2020;Minguillon et al., 2012;Ng et al., 2002;Nishimoto et al., 2015;Tanaka et al., 2005;te Welscher, Fernandez-Teran, et al., 2002;Xu & Wellik, 2011;Zúñiga, 2015) and provide a good system for the study of genetic parallelism of appendage reduction and loss (Bolnick et al., 2018).
Pelvic expression of Pitx1 in G. aculeatus is regulated by two pelvic-specific enhancers-PELA and PELB (Chan et al., 2010;Thompson et al., 2018;Xie et al., 2019). Pelvic-reduced sticklebacks have mutations in one or both enhancers and demonstrated reduced Pitx1 expression in pelvic tissue (Chan et al., 2010;Kragesteen et al., 2018;Thompson et al., 2018;Xie et al., 2019). Genomic studies have shown that mutations to PELA arise de novo, likely because the enhancer is in a chromosomal region prone to double-strand breakages (Xie et al., 2019). The PELA enhancer is subject to strong positive selection that drives the null allele to fixation (Chan et al., 2010;Xie et al., 2019). The strong selection for modified Pitx1 regulation suggests a potential route to appendage reduction in other taxa if the lack of constraint is shared (Chan et al., 2010;Xie et al., 2019).

TA B L E 2 Summary of molecular mechanisms of reduction and loss
Indeed, more than thirty populations of the ninespine stickleback (Pungitius pungitius) have pelvic reduction and loss and show no differences in the PITX1 amino acid sequence between pelvic-complete and pelvic-absent fish (Klepaker et al., 2013;Shapiro et al., 2004Shapiro et al., , 2006. Instead, Pitx1 expression is missing from the pelvic region of pelvic-absent ninespines, as in threespine stickleback (Shapiro et al., 2004(Shapiro et al., , 2006. Hybrids of threespine and ninespine stickleback with one pelvic-complete parent and one pelvic-reduced parent have a full pelvis, while hybrids with two pelvic-reduced parents demonstrate pelvic spine and girdle reduction (Shapiro et al., 2006). These results indicate that pelvic reduction is controlled by regulation of the same locus, Pitx1, in threespine and ninespine sticklebacks, despite their 26-million-year divergence (Shapiro et al., 2006;Varadharajan et al., 2019). Moreover, modified Pitx1 expression has been implicated in pelvic reduction of G. doryssus, a 10-million-year-old threespine stickleback species from the Miocene (Stuart et al., 2020). This inference stemmed from an observation of pelvic asymmetry in which left side vestiges were larger than right side vestiges in G. doryssus fossils-a similar phenotype to that found in extant pelvic-reduced stickleback (Nelson, 1971;Shapiro et al., 2004Shapiro et al., , 2006Stuart et al., 2020). As such, it appears that pelvic reduction and loss in more than 100 populations across at least three stickleback species shares a genetic cause.
However, modified Pitx1 expression does not drive pelvic appendage loss in a different teleost, the fugu (or pufferfish) Takifugu rubripes. Pelvic loss in fugu results instead from the absence of positional signaling by HoxD9a in the pelvic region. HoxD9, an orthologous gene, is important for appendage positioning and initiation in vertebrates ( Figure 3) (Cohn et al., 1997;Tanaka et al., 2005). For example, in stickleback embryos, HoxD9 expressed in pectoral and pelvic fin buds (Tanaka et al., 2005). In embryonic fugu, however, HoxD9a is expressed in the pectoral region but is absent from the pelvic region (Tanaka et al., 2005). Therefore, the absence of Hoxd9a expression in the pelvic region of fugu prevents fin and girdle formation.
Shh expression in the tetrapod limb is controlled by an enhancer called the ZRS (Galli et al., 2010;Lettice et al., 2003;Park et al., 2008;Riddle et al., 1993;Young & Tabin, 2017). The P. regius ZRS has three large deletion mutations relative to Anolis sagrei, a lizard with fully developed hindlimbs (Leal & Cohn, 2016). These mutations result in Shh expression that is reduced and terminates early (Leal & Cohn, 2016). Loss of SHH signaling is followed by a decrease in Fgf8 expression, preventing limb and girdle growth in P. regius (Figure 4) (Leal & Cohn, 2016). Notably, ZRS sequences are even more poorly conserved in advanced snakes, likely driving complete loss of the hindlimb and pelvis (Kvon et al., 2016;Leal & Cohn, 2016).

| SQUAMATE D I G IT LOSS
While less striking than the complete limb loss of snakes, digit loss in the fore-and hindlimbs of other nonsnake squamates has evolved over twenty separate times (Brandley et al., 2008;Greer, 1991).
Scincidae, a squamate family with over 1700 described species, accounts for nearly half of these instances of digit loss (Brandley et al., 2008;Uetz et al., n.d.). For example, fore and hindlimb digit number varies between the seven species of the Australian genus Hemiergis (Shapiro et al., 2003;Uetz et al., n.d.). Hemiergis initialis retains five digits on each limb, whereas H. peronii has lost 2 digits on every limb and H. quadrilineata has lost three digits on every limb (Shapiro et al., 2003). Variation in Hemiergis digit number correlates with the duration of Shh expression in the limb bud: shorter expression corresponds to fewer digits ( Figure 5) (Shapiro et al., 2003).
Reduction or loss of Pitx1 in the posterior appendage unmasks the asymmetrical expression of Pitx2, one of only six genes known to generate left-larger directional asymmetry in limb bud (Palmer, 2004 implicating Pitx1 in hindlimb loss and pelvic reduction in manatee ( Figure 6) (Shapiro et al., 2006).

| MAMMAL D I G IT REDUC TI ON
The number and size of digits is variable among mammals; more than half of mammalian orders demonstrate some form of digit reduction (Sears et al., 2011). The first digit in all adult even-toed ungulates (order Artiodactyla) is absent, and digits II and V are re-  Zúñiga, 2015).
Flight evolved early in the bat lineage and was facilitated by substantial changes to forelimb and pectoral girdle structure, including reduction in bone size (Hockman et al., 2008;Simmons et al., 2008).
Specifically, the length and width of the ulna are reduced relative to the radius, with the distal tip of the ulna fused to the radius (Sears, 2008;Sears et al., 2007). Ulnar reduction decreases wing weight without compromising its function (Sears, 2008).
Ulnar width reduction in bats results from differential growth rates between the radius and the ulna (Sears et al., 2007). In the short-tailed fruit bat (Carollia perspicillata) and the little brown bat (Myotis lucifugus), the cartilage condensations that will form the radius and ulna are initially similar in width (Adams, 1992;Sears et al., 2007). The relative width of the ulna begins to decrease with the onset of ossification, and it continues to narrow as the distal tip fuses to the radius (Adams, 1992;Sears et al., 2007). Two processes have been suggested to cause ulnar width reduction: (a) abnormal morphology of differentiating cartilage cells or (b) a lower rate of bone deposition (appositional growth) (Biga et al., n.d.;Sears et al., 2007).
Ulnar length reduction likely results from modified regulation of essential limb patterning genes. In C. perspicillata, M. lucifugus, and Miniopterus schreibersii (the common bent-wing bat), posterior HoxD gene expression is upregulated and prolonged in the developing wing relative to the hindlimb or mouse limbs (Chen et al., 2005;Ray & Capecchi, 2008;Wang et al., 2014). Additionally, the anterior edge of HoxD13 expression is shifted distally, and the posterior edge is shifted proximally in the bat forelimb bud (Chen et al., 2005;Ray & Capecchi, 2008). HoxD cis-regulatory elements have bat-specific changes that are not shared with other mammals (Booker et al., 2016;Ray & Capecchi, 2008). For example, the GCR is a regulatory region that drives HoxD gene expression in the mammalian forelimb (Ray & Capecchi, 2008). Compared to mouse or human GCRs, the Chiropteran GCR has several lineage-specific sequences and drives altered expression of HoxD genes when compared to mouse or human GCRs (Ray & Capecchi, 2008). Altered expression of HoxD genes results in aberrations in ulnar length Chen et al., 2005;Hérault et al., 1997;Peichel et al., 1997;Ray & Capecchi, 2008;Sears, 2008).
In the Natal long-fingered bat (Miniopterus natalensis), Shh expression is delayed but spatially expanded in the forelimb bud, relative to mouse (Hockman et al., 2008). In experimental studies, Shh-knockout mice showed reduced cell proliferation and increased cell death in forelimb buds, resulting in a mutant phenotype similar to the batwing-a normal radius and a reduced ulna (Ahn & Joyner, 2004;Chiang et al., 2001;Hockman et al., 2008;Sears, 2008). This change in Shh expression might also contribute to the expanded Hox gene expression that shrinks the ulna, discussed above (Chiang et al., 2001;Hockman et al., 2008).

The variable forelimb reduction and digit loss suggests that emu
wing morphology is not constrained (

F I G U R E 6
The gene regulatory network modified in the reduction of hindlimb and pelvic elements in cetaceans † and sirenians*. Gene symbols in red are not expressed in the hindlimb bud. Alternating red and orange lettering indicates that gene expression is reduced and terminates earlier than in typical limb development. The alternating red and orange arrow indicates that the interaction between diminished Shh and Fgf8 results in arrested limb development. Modified expression of Pitx1, written in orange, is suspected to underlie hindlimb loss and pelvic reduction in manatee Expression of Tbx5 in the emu wing bud is delayed relative to chick, reducing recruitment of progenitor cells in sternal and forelimb tissues (Bickley & Logan, 2014;Minguillon et al., 2005;contra Farlie et al., 2017). With fewer progenitor cells, rates of proliferation and outgrowth are reduced, and the emu wings grow 64% slower than chicken wings (Bickley & Logan, 2014;Farlie et al., 2017;Faux & Field, 2017;Smith et al., 2016). Notably, the emu wing bud emerges after and develops more slowly, than the hindlimb bud (Ahn & Joyner, 2004;Bickley & Logan, 2014;Butterfield et al., 2009).
The flightless Galápagos cormorant (Phalacrocorax harrisi) has a short radius and ulna relative to its humerus (Bickley & Logan, 2014;Burga et al., 2017). Compared to flying cormorant species, the Galápagos cormorant has a deletion of four amino acids in the CUX1 coding sequence (Burga et al., 2017). In experiments with mouse cell lines, the resultant protein was less effective in activating Ihh, a gene important for the proliferation and differentiation of cartilage cells (Burga et al., 2017;Kronenberg, 2003;Peckham et al., 2003).

| B IRD HINDLIMB REDUC TION
In all extant birds and their recent ancestors, the fibula is splinterlike and reduced, usually around 2/3 length of the tibia (Botelho et al., 2016;Paese et al., 2021). Initially, the two cartilaginous elements that form the tibia and fibula are approximately equal in size (Botelho et al., 2016;Paese et al., 2021). In one possible explanation, the fibula is reduced because it lacks a distal growth plate (Botelho et al., 2016). Without the growth plate, the fibula does not maintain a population of immature, proliferating cartilage cells that drive distal growth because the feedback loop between IHH and PTHrP is disrupted. Indian Hedgehog encourages the formation of bone from cartilage and the production of PTHrP (Botelho et al., 2016). Conversely, PTHrP delays cartilage maturation and inhibits IHH production (Botelho et al., 2016). The distal portion of the fibula does not maintain PTHrP expression, but the fibulare acts as a surrogate growth plate early in bone development. While the fibulare is appressed to the fibula, it provides PTHrP signaling that inhibits IHH production and allows for continued cartilage growth (Botelho et al., 2016). Over the course of bone development, the fibulare separates from the fibula and PTHrP signaling no longer reaches distal cartilage of the fibula (Botelho et al., 2016). Without PTHrP to maintain the feedback loop with IHH, the growth of the fibula is slow and terminates early, resulting in a short, splinter-like bone (Botelho et al., 2016).
Another explanation is that altered Hedgehog signaling disrupts anteroposterior polarity in the developing bird hindlimb (Paese et al., 2021). The talpid 2 mutant chicks, a 19-bp deletion in C2cd3 prevents formation of the repressive form of GLI3 (Paese et al., 2021). This mutation leads to ectopic SHH signaling, polydactyly, degradation of digit identity and autopod asymmetry, and fibular extension (Paese et al., 2021). That is, in talpid 2 chicks, the lengths of the tibia and fibula remain similar throughout development, while the tibia extends significantly relative to the wild-type fibula (Botelho et al., 2016;Paese et al., 2021 The mechanism most often shared among taxa was modulation of Shh expression and signal transduction, which was associated with limb reduction in squamates, cetaceans, artiodactylans, bat, and emu. The central role of SHH in limb patterning and outgrowth likely influences in its parallel modification in distantly related vertebrate clades. However, the specific molecular mechanisms by which SHH Perhaps it is unsurprising that vertebrate appendage reduction and loss is underlain by both shared and unique molecular mechanisms. Appendage development is controlled by spatially and temporally regulated expression of dozens of interacting genes-a complexity that creates potential for numerous routes to appendage reduction and loss. However, many key developmental genes have pleiotropic effects across the body plan, so evolution could be constrained to only a handful of pathways. Such constraint may explain the most salient finding of our review: in all cases but one, appendage reduction and loss resulted not from changes in protein coding DNA but from changes to enhancer sequences and limb-specific gene expression patterns. The evolutionary importance of regulatory mutations is contentious, especially for gain-of-function adaptations (Hoekstra & Coyne, 2007). However, our findings support the assertion that regulatory changes represent a major mode of evolution because of their repeated role in loss-of-function phenotypes that are likely adaptive (Chan et al., 2010;Hoekstra & Coyne, 2007).

ACK N OWLED G M ENTS
We thank M. Coates

CO N FLI C T O F I NTE R E S T
Authors involved in preparation of this manuscript have no conflicts of interest to declare.