Asymmetric cell convergence-driven zebrafish fin bud initiation and pre-pattern requires Tbx5a control of a mesenchymal Fgf signal

The T-box transcription factor Tbx5 plays a pivotal role in vertebrate forelimb initiation (Agarwal et al., 2003; Ahn et al., 2002; Garrity et al., 2002; Ng et al., 2002; Rallis et al., 2003; Takeuchi et al., 2003). For example, human TBX5 haploinsufficiency is associated with Holt–Oram syndrome (Basson et al., 1997; Li et al., 1997), in which affected individuals exhibit forelimb defects ranging from a reduction in the radius and the anterior-most digit (pre-axial hemimelia) to a complete loss of arm (phocomelia) (Basson et al., 1994; Holt and Oram, 1960; McDermott et al., 2005). The anteriorly biased reduction of the forelimb skeleton in Holt–Oram syndrome suggests a developmental patterning defect; however, the cellular behaviors affected by Tbx5 during limb bud initiation are unknown. Here, we have used zebrafish to investigate Tbx5a-dependent cell movements and the downstream genetic cascade that results in outgrowth and patterning of the early vertebrate limb bud.

Knowledge of the cellular processes underlying limb bud initiation offers unique insights into the genetic mechanisms underlying normal or pathological limb development. Early histological studies suggested that on-site cell proliferation is the sole mechanism for supplying limb bud mesenchymal cells (Harrison, 1918; Rallis et al., 2003; Searls and Janners, 1971; Swett, 1923). However, subsequent investigations suggested a cell movement-based mechanism, in which limb-field cells segregate from flanking regions because of higher tissue cohesivity (Damon et al., 2008; Foty et al., 1996; Heintzelman et al., 1978). In zebrafish, lateral plate mesoderm (LPM) cells appear to migrate in a Tbx5a-dependent manner into the future pectoral fin bud region prior to overt fin bud formation (Ahn et al., 2002; Garrity et al., 2002). Moreover, oriented cell motility and divisions of limb bud cells have been reported during the early outgrowth of mouse and chick limbs and zebrafish fins (Wyngaarden et al., 2010). Recently, Tbx5-dependent epithelial-to-mesenchymal transition was observed within the presumptive limb-field in chick embryos (Gros and Tabin, 2014). Together, these data suggest a conserved function for Tbx5a in the promotion of oriented cell motility during limb bud initiation.

Tbx5 expression initiates a cascade of various Fibroblast growth factor (Fgf) signals in the limb mesenchyme (Agarwal et al., 2003; Ng et al., 2002; Takeuchi et al., 2003), which eventually induces Fgf8 in the apical ectodermal ridge, a structure essential for the maintenance of limb cell proliferation (Bénazet et al., 2009; Boulet et al., 2004; Lewandoski et al., 2000; Ohuchi et al., 1997). Although the importance of ectodermal Fgf signaling has been well documented (Boulet et al., 2004; Crossley et al., 1996; Lewandoski et al., 2000; Moon and Capecchi, 2000; Vogel et al., 1996), much less is known about the roles of the earlier mesodermal Fgf signals. In chicks, epithelial-to-mesenchymal transition of limb-field LPM cells requires mesodermal expression of the Tbx5 target Fgf10 (Gros and Tabin, 2014), and in zebrafish, formation of the fin bud from fin-field LPM cells requires mesodermal expression of the Tbx5a target Fgf24 (Fischer et al., 2003). Together, these findings suggest a role for early mesodermal Fgf signals in regulating LPM cell motility during limb bud formation. However, it remains unclear how or if mesenchymal Fgf signals promote oriented cell behaviors observed in the LPM during limb bud initiation.

In this study, we establish the mechanistic connection between Tbx5a, mesodermal Fgf signals and motility patterns in the zebrafish pectoral fin-field. We combine single-cell-resolution fate mapping and three-dimensional cell tracking to demonstrate definitively that zebrafish pectoral fin-field cells converge along the anteroposterior (AP) axis while maintaining their relative initial AP positions topologically to form the fin bud. Here, we use the term ‘topological’ to refer to an ordered process that maintains the initial organization of a group of precursor cells, or field of cells, within the final tissue or organ. Our analyses show that LPM cells exhibit differences in cell motility that result in the asymmetric positioning of fin-field cells towards a defined location in the posterior fin-field. We further show that fin-field LPM cells can migrate towards an ectopic source of Fgf, supporting a role for Fgf24 in the posterior fin-field LPM as the convergence cue for fin bud precursors. Our study also uncovers an early AP pre-pattern within the pectoral fin-field. Partial knockdown of either Tbx5a or Fgf24 leads to a biased perturbation of the anterior fin-field cells, resulting in their failure to migrate into the fin bud. This preferential loss of anterior convergence results in anteriorly biased truncations of the pectoral fin disc in larval zebrafish that resemble aspects of the forelimb skeletal defects that define individuals with Holt–Oram syndrome and suggests that early fin-field cells contain positional identities prior to convergence.

We define the zebrafish pectoral fin-field as the area within the LPM from which the mesenchymal cells of the pectoral fin bud are derived. To map the territory of the fin-field at single-cell resolution, single blastomeres in 64- to 128-cell-stage embryos were injected with Kaede mRNA (Ando et al., 2002). We then photo-converted from green to red fluorescence in individual LPM cells at 14 hours postfertilization (hpf) and followed the labeled progeny up to 48 hpf. Using flanking inter-somitic borders as landmarks, we constructed a fate map of the LPM region at 14 hpf consisting of 105 labeled single cells between somite levels 1 and 6 (Fig. 1E). We classified later progeny clones into four fate groups according to the location and morphology of the labeled progeny cells at 28-48 hpf. All clones of cells derived from single cells labeled at 14 hpf gave rise to progeny of only one fate type, as follows: (1) cardiac (Fig. 1A, n=6), consisting of one or two labeled cells located anterior to the common cardinal vein within either the myocardial or the pericardial tissue; (2) fin (Fig. 1B; n=72), labeled cells consisting of two to four progeny of irregular cuboidal morphology within the mesenchyme of the fin bud proper beneath the fin epidermis; (3) peritoneum (Fig. 1C; n=14), typically consisting of four progeny spreading laterally over the yolk sac; and (4) medial (Fig. 1D, n=13), consisting of one or two labeled cells located more medially than other labeled clones and within the position of the intermediate mesoderm. Thus, the mapped LPM region is roughly segregated into cardiac, fin and peritoneum fields from anterior to posterior (Fig. 1E). The fin-field at 14 hpf occupies a region within the LPM roughly 200 µm along the AP body axis (flanking somites 1-4; Fig. 1E, between red dashed lines), approximately 1.6 times the AP length of the pectoral fin bud at 28 hpf (∼120 µm along the AP axis, flanking somites 2-3; Fig. 1E). This result is consistent with a previous report (Wyngaarden et al., 2010) and suggests that fin-field cells converge along the AP axis to become positioned within the fin bud.

We found that the relative AP positions of the cardiac, fin and peritoneum fields in the LPM are maintained after tissue differentiation from each precursor pool. Next, we asked whether the cells of the fin-field also exhibit a topological arrangement; that is, are relative AP positions of the cells maintained during convergence or does extensive mixing occur? We compared the initial AP position of labeled LPM cells in the fin-field with the averaged position of each resulting clone in the fin bud. This analysis revealed a correlation (Fig. S4B; P<2.2×10⁻¹⁶, n=72), indicating a topological transfer of the anteroposterior arrangement of cells from the LPM fin-field at 14 hpf to the fin bud at 28 hpf.

In order to characterize the spatial dynamics of fin-field cells further, we used a transgenic line Et(hand2:eGFP)ch2, in which all LPM cells are marked with GFP (Fig. 2Ai-Di; Fig. S1; supplementary materials and methods) allowing three-dimensional cell tracking (Fig. 2Ai′-Di′; Fig. S2A-D; supplementary materials and methods). The transformation from the fin-field to the fin bud involves movements of ∼120 fin-field cells on each side of the embryo from 18 (onset of AP convergence) to 28 hpf (condensation of fin bud primordia). We tracked 60 cells unilaterally from somite levels 1-5+ in each of eight wild-type embryos from 18 to 23 hpf, when most fin-field LPM cell movements take place (Fig. S2E). Fig. 2Ai-Di′ and Movie 1 show time-lapse images in a double transgenic but otherwise wild-type embryo, in which Et(hand2:eGFP)ch2 labels all LPM cells and Tg(h2afx:h2afv-mCherry)mw3 labels all nuclei (McMahon et al., 2009). Registered cell tracks from eight wild-type embryos are collectively shown in Fig. 3Ai, with the origin of the axes at the intersection between the midline of the embryo and the anterior border of somite 1. Clustering these cell tracks into five color-coded groups corresponding to the inter-somitic borders flanking the track origins, we observed a general trend of caudally and laterally oriented net motility relative to the body axis (Fig. 3Ai). Subtraction of the averaged cell tracks of each LPM group from the averaged track of all LPM cells further revealed a net convergence of track groups 1-4, which separates them from the more posteriorly located track group 5+ (Fig. 3Bi,Ci). In order to identify and distinguish the converging and non-converging cell groups quantitatively, we defined the measurement ‘scatter’ to describe the variation of individual AP positions within a delineated group of cells (supplementary materials and methods). An increasing or decreasing scatter value over time indicates cell dispersion or convergence, respectively. A sliding maximal detection limit along the AP axis allowed us to identify the lowest scatter value point to define the border between converging and the non-converging LPM cell groups; this point roughly aligns with the border between track groups 4 and 5 (Fig. S2F,G), in agreement with the posterior limit of the fin-field revealed by our fate map (Fig. 1E). Henceforth, we refer to the definitive ‘fin-field’ as those LPM cells originating between somites 1 and 4 along the body axis of the 14 hpf zebrafish embryo, with the peritoneum field residing in more posterior LPM regions. As shown in Fig. 3Di, fin-field cells converge along the AP axis, in contrast to adjacent peritoneum progenitors that disperse over time.

A tight correlation exists between the starting and ending AP positions of individual cell tracks (Fig. 3Ei), attesting to the preservation of relative AP positions between fin-field cells and their progeny in the fin bud. Together, the fate map and cell-tracking data reveal that the pectoral fin-field initially occupies a broader area along the primary body axis than the final AP dimensions of the fin bud. The LPM fin-field cells then converge topologically to form the fin bud mesenchyme.

The most anterior groups of LPM cells, initially residing next to somites 1 and 2 (‘anterior fin-field’; Fig. 3Ai,Bi, red and orange groups; Fig. 3Fi, AP positions <100 µm), migrate further relative to those at more posterior initial locations next to somites 3 and 4 (‘posterior fin-field’; Fig. 3Ai,Bi, blue and green groups; Fig. 3Fi, AP positions ∼100-200 µm), resulting in an anterior-to-posterior gradient in displacement (Fig. 3Fi). The displacement measure can be affected by two factors, namely speed and persistence of motion in a given direction. After subtraction of the overall LPM tissue movement, anterior fin-field cells migrate at higher average speeds along the AP axis (Fig. S3A) and exhibit elevated persistence (the ratio of displacement to total trajectory length; supplementary materials and methods; Fig. S3B) and mean square displacement (MSD) values (Fig. S3F), relative to posterior fin-field cell groups. The elevated levels of net directional migration of anterior fin-field cells in the anterior-to-posterior direction relative to those of posterior fin-field cells in the posterior-to-anterior direction results in a convergence point positioned within the posterior fin-field (center of mass at roughly 180 μm; Fig. 6L and supplementary materials and methods).

Tbx5 orthologs are required for forelimb bud formation in multiple vertebrates (Agarwal et al., 2003; Ahn et al., 2002; Garrity et al., 2002; Ng et al., 2002; Rallis et al., 2003; Takeuchi et al., 2003). Although we and others previously suggested the disruption of migration movements in Tbx5a-deficient embryos (Ahn et al., 2002; Garrity et al., 2002), it was not previously known how or when limb bud formation fails. In order to investigate the behavior of individual Tbx5a-deficient cells in the LPM further, we injected 1 ng Tbx5a antisense morpholino (a dose phenocopying the null Tbx5a mutant heartstrings) into Et(hand2:eGFP)ch2; Tg(h2afx:h2afv-mCherry)mw3 embryos and performed live tracking analyses (Fig. 2Aii-Dii; Fig. S2H; Movie 2; n=60 tracks from each of eight embryos). Tbx5a-deficient fin-field cells fail to converge along the AP axis and migrate in near-parallel tracks laterally from the dorsal midline towards the ventral side of the body (Fig. 2Aii- Dii′; Fig. 3Aii-Cii). Notably, the caudalward migration of anterior fin-field cells observed in wild-type embryos (Fig. 3Bi,Ci,Fi) is diminished in magnitude in Tbx5a morphants, (Fig. 3Bii,Cii,Fii). The scatter value in the Tbx5a morphant fin-field remains constant over time, as opposed to the decreasing value for wild-type fin-field cells as they converge (Fig. 3Dii, black versus gray solid arrowhead). Furthermore, the descending gradients in displacement (Fig. 3Fii, black arrowhead compared with gray arrowhead), average speed, persistence and MSD (Fig. S3A,B,F, second column compared with first column) along the AP axis are diminished in the Tbx5a-deficient fin-field. Collectively, these data demonstrate that fin-field LPM cells, although still able to migrate, fail to converge along the AP axis in Tbx5a morphants. These LPM cells eventually scatter over the yolk and adopt a peritoneum-like morphology.

Interestingly, although cell convergence fails, Tbx5a-deficient fin-field cells maintain their topological migration pattern (Fig. 3Eii). Given that in the peritoneum field, where Tbx5a expression is normally absent, LPM cells also exhibit topological migration (Fig. 3Ei, purple dots), we hypothesize that the maintenance of nearest neighbor relationships during migration might be an intrinsic property of all LPM cells independent of Tbx5a function.

Given that Tbx5a is expressed throughout the AP length of the fin-field (Fig. 4A-D), its expression alone cannot account for the asymmetry in fin-field cell motility. Rather, Tbx5a might regulate a convergence cue located within the posterior half of the fin-field. One candidate is fgf24, a Tbx5a target that is required for the coalescence of Tbx5a⁺ cells and formation of the pectoral fin (Draper et al., 2003; Fischer et al., 2003; Mercader et al., 2006). fgf24 is expressed in a more restricted domain within the posterior position in the tbx5a⁺ LPM (Fischer et al., 2003), roughly aligned with the future position of the prospective fin bud (Fig. 4E-H; note that the somitic landmarks have shifted posteriorly by one somite compared with 14 hpf).

To investigate whether Fgf24 is required as a convergence cue for fin-field LPM cells, we injected 1 ng of Fgf24 antisense morpholino into Et(hand2:eGFP)ch2; Tg(h2afx:h2afv-mCherry)mw3 embryos and performed live cell tracking (Fig. 3Aiii-Diii; Fig. S2I; Movie 3; n=60 tracks from each of eight embryos). Fgf24 morphant cell motility along the AP axis essentially phenocopies that of the Tbx5a morphants; fin-field cells no longer converge along the AP axis (Fig. 3Aiii-Ciii) and instead migrate in parallel with a near-constant scatter value (Fig. 3Diii, solid black compared with solid gray arrowhead) and diminished displacement (Fig. 3Fiii, black compared with gray arrowhead), average speed, persistence and MSD along the AP axis (Fig. S3A,B,F, third compared with first and second columns). As in wild-type embryos and Tbx5a morphants, the topological relationship of LPM cells is maintained in Fgf24 morphant fin-fields (Fig. 3Eiii).

Fgf24 expression is temporally and spatially correlated with its potential role as a convergence cue and is functionally required in fin bud initiation (Fig. 4; Fischer et al., 2003). To test whether fin-field LPM cells can migrate towards a localized Fgf source, we implanted polystyrene beads coated with recombinant human-mouse Fgf8b (a paralog of Fgf24; Jovelin et al., 2007) into the fin-field LPM in Fgf24 morphants and performed cell tracking (n=20 cells in each of eight embryos; Fig. 5; Fig. S5; Movie 4A,B).

We found that Fgf8b-coated beads can induce a convergence-like movement of fin-field LPM cells (Fig. 5E-H; Movie 4B; Fig. S5I-P) compared with bovine serum albumin (BSA)-coated control beads (Fig. 5A-D; Movie 4A; Fig. S5A-H), as indicated by the declining average distances between tracked cells and beads (Fig. 5I, orange compared with blue line). The presence of an Fgf8b-coated bead alters nearby fin-field cell behaviors in Fgf24 morphants, such that these cells migrate with higher speed (Fig. 5J; Wilcoxon test P=7.3×10⁻¹⁶) and persistence (Fig. 5K; Wilcoxon test P=2.8×10⁻¹⁵) compared with those adjacent to control BSA-coated beads.

Fgf24 loss of function perturbs the activation of pea3 (etv4 – Zebrafish Information Network) and fgf10 and the maintenance of tbx5a in the fin initiation gene network (Fischer et al., 2003; Fig. 5M,Q,U compared with Fig. 5 L,P,T). Although control beads failed to rescue expression of pea3, fgf10 and tbx5a in the fin-field (Fig. 5N,R,V), Fgf8b-coated beads induced their expression around the implantation sites within the fin-field (Fig. 5O,S,W). This suggests that ectopic Fgf8, when introduced in a comparable spatiotemporal pattern to endogenous Fgf24, can rescue aspects of the fin initiation gene network in Fgf24 morphants. Taken together, these results demonstrate that fin-field LPM cells can migrate towards a localized Fgf source in a chemoattractive fashion during fin initiation.

We have previously shown that suboptimal doses of Tbx5a morpholino can reduce the size of larval pectoral fins and cause tissue gaps preferentially on the anterior side of the endoskeletal disc (Ahn et al., 2002; Fig. 6B, asterisks). The anteriorly biased truncations are reminiscent of the pre-axial hemimelia observed in TBX5-haploinsufficient individuals with Holt–Oram syndrome (Basson et al., 1997; Brassington et al., 2003; Li et al., 1997). The intrinsic asymmetry in fin-field cell convergence pattern led us to hypothesize that partial reduction in Tbx5a or Fgf24 levels might have differential effects upon anterior versus posterior subpopulations within the fin-field.

To test whether the AP axis in the fin bud of Tbx5a or Fgf24 partial morphant embryos is disrupted, we injected Et(hand2:eGFP)ch2; Tg(h2afx:h2afv-mCherry)mw3 embryos with suboptimal doses of Tbx5a or Fgf24 morpholino (0.6 ng for Tbx5a, 0.4 ng for Fgf24; Fig. S2J,K; n=60 tracks in each of eight embryos per treatment). Tbx5a or Fgf24 partial morphant fin-field cells display an overall intermediate level of AP convergence (Fig. 6C,D; Fig. S6C,D; black solid arrowhead in Fig. 6E or Fig. S6E) and displacement gradient (black arrowhead in Fig. 6K or Fig. S6K) compared with wild-type embryos and Tbx5a or Fgf24 full morphants. The average speed, persistence and MSD of the anterior fin-field cells in partial Tbx5a or Fgf24 morphants more closely resemble those of Tbx5a or Fgf24 full morphants rather than wild-type embryos (Fig. S3A,B,F,G). We further quantified the proportions (relative to the total number initially present) of anterior or posterior fin-field cells that successfully became incorporated into the fin bud (Fig. 6L,M; Fig. S6L,M; supplementary materials and methods). In Tbx5a partial morphants, only 18% of anterior fin-field cells contributed to the fin bud, compared with a 47% incorporation rate of posterior fin-field cells (Fig. 6M; Wilcoxon test P=0.0027). Therefore, whereas cell migration across the entire fin-field is affected in partial Tbx5a morphants, the anterior population is more affected by decreased Tbx5a levels. A similar conclusion can be drawn with Fgf24 partial morphants (Fig. S6M; Wilcoxon test P=0.028).

To test whether the differential convergence defects between anterior versus posterior fin-fields observed in partial morphants are correlated with later patterning defects, we quantified the areas occupied by hoxc6a⁺ or hoxd12a⁺ cells, which mark the anterior and posterior fin bud mesenchyme, respectively (Fig. 6F,G; Gibert et al., 2006; Molven et al., 1990; Neumann et al., 1999). The expression domains of hoxc6a and hoxd12a are both reduced in Tbx5a or Fgf24 partial morphants (Fig. 6H,I; Fig. S6H,I) in comparison to wild-type embryos (Fig. 6F,G; Fig. S6F,G); however, the hoxc6a expression domain is reduced ∼40% more than that of hoxd12a (Fig. 6J, Tbx5a partial morphants; Fig. S6J, Fgf24 partial morphants).

Together, these results suggest that partial reduction in Tbx5a or Fgf24 levels results in preferential disruption of anterior fate in the early fin bud. Given that this is correlated with a greater reduction in the convergence of the anterior subpopulation of fin-field cells, we propose that an early mechanism might pre-pattern the vertebrate fin AP axis prior to or during the fin-field convergence. We propose that a greater sensitivity to the levels of Tbx5a or Fgf expression in anterior fin-field cells versus posterior fin-field cells could contribute to the anteriorly biased defects observed in individuals with Holt–Oram syndrome.

A quantitative model of how early LPM cells become organized prior to limb bud formation, especially with regard to cell migration, does not exist for any vertebrate embryo. Here, we have characterized the dynamics of LPM migration to form the fin bud by combining a single-cell-resolution fate map with live cell tracking. We established that the definitive fin-field occupies the area of the zebrafish LPM adjacent to somites 1-4. This broad extent of the fin-field, relative to the initial fin bud, which only covers the area between somites 2 and 3, suggests that LPM cells must migrate and converge to form the fin bud proper. Time-lapse analyses confirm that convergence movements of the fin-field LPM supply the majority of the mesenchymal complement of the early fin bud. Migration movements of the LPM fin-field cells are coordinated and coherent, maintaining a topological relationship between the initial AP organization of the LPM along the body axis with the final AP location of these cells within the forming fin bud.

Our studies show that fin-field cells do not simply converge symmetrically towards a center point, contrary to what had been proposed by previous studies using somites as landmarks (Ahn et al., 2002; Wyngaarden et al., 2010). In zebrafish, the LPM and somites are known to shift relative to one another at this stage (Devoto et al., 1996). By tracking the movements of individual cells within the fin-field, we have shown that a cell motility gradient along the AP axis results in the final position of the fin bud being located posteriorly within the fin-field. We hypothesize that the differential fin-field cell movements indicate either a pre-patterning of different movement characteristics or an active migration towards an asymmetrically localized signaling center. Together, these data define a model in which asymmetric but topological cell convergence transforms the initially broad LPM fin-field into the compact early fin bud.

The overall extent and subsequent shape changes of the fin-field closely mirror the dynamic expression pattern of zebrafish tbx5a gene during development (Fig. 4A-D). Previously, it was shown that Tbx5a deficiency causes a complete loss of pectoral fins (Ahn et al., 2002; Garrity et al., 2002). Here, we show that loss of Tbx5a function abolishes the differences both in distance and in angle of migration amongst the different groups of fin-field cells. As a result, fin-field cells migrate along parallel tracks relative to each other along the dorsoventral embryonic axis and display little to no relative movement in the AP direction. Therefore, loss of Tbx5a does not affect the ability of the cells to migrate per se, but does affect their ability to track collectively towards a convergence target area.

Although the importance of cell migration in limb bud formation is increasingly acknowledged, much less is known about the genetic and molecular functions of signals within the converging pre-limb bud mesenchyme. Several Fgfs are expressed in the intermediate and lateral plate mesoderm preceding formation of the limb bud. We and others (Gros and Tabin, 2014) hypothesize that mesenchymal Fgf signals play a key role in the transition from limb/fin-field to limb/fin bud. Fgf24 is a teleost-specific member of the fgf8/fgf17/18 family (Jovelin et al., 2007) and is the earliest Fgf member to be expressed in the fin-field region. Previously, knockdown of Fgf24 was shown to cause absence of pectoral fins, similar to Tbx5a deficiency (Draper et al., 2003; Fischer et al., 2003). Fgf24 is a known target of Tbx5a (Fisher et al., 2003). Here, we show that disruptions in LPM cell migrations are similar in Fgf24- and Tbx5a-deficient specimens, with one notable difference. In Tbx5a-deficient embryos, the differences in both directionality and distance displayed by different wild-type AP populations of fin-field cells are lost (compare Fig. 3Bi with Bii). In Fgf24-deficient embryos, although AP convergence movements are similarly abrogated, the anterior population of fin-field cells still retains the ability to migrate further and faster than the posterior fin-field cells along the ML axis (Fig. 3Biii; Fig. S3C-E,I). This difference reveals that anterior fin-field cells might have an intrinsically higher motility that requires Tbx5a function but is independent of Fgf24 function.

At present, the mechanism that produces a restricted fgf24 expression domain within a broader Tbx5 expression domain, despite Tbx5 being the only known activator, is not yet explained. Early signals, such as Wnt2b, which is required for expression of both tbx5a and fgf24 (Fischer et al., 2003), could potentially introduce a bias in Fgf24 activation within the tbx5a⁺ fin-field. Given that Wnt2b is expressed at an earlier time point in a restricted domain in the intermediate mesoderm immediately adjacent to the Tbx5⁺ LPM, it is a potential candidate to fulfill the role of a local activator of fgf24 expression; however, more in depth studies are warranted.

To test the hypothesis that Fgf24 acts as a non-cell-autonomous convergence cue to redirect the migration of anterior fin-field cells towards the posterior fin-field, we implanted beads coated with recombined Fgf8b into the fin-field of the zebrafish embryo. Although this approach did not elicit a full rescue of fin bud formation in Fgf24-deficient embryos, fin-field cells did reorient their migration pathways to converge upon the Fgf-coated beads. In chick embryos, Fgf-coated beads cause ectopic limb development, but the mechanism is controversial (Lewandoski et al., 2000; Moon and Capecchi, 2000). Recently, Fgf8-soaked beads placed into later-stage formed chick limb buds were shown to attract mesenchyme cells (Gros et al., 2010); however, this finding does not directly address a potential earlier role for Fgf as a chemoattractant within the tetrapod mesenchyme before limb bud formation. It has been postulated that an Fgf8 signal from the intermediate mesoderm could function as a limb inducer (Agarwal et al., 2003; Crossley et al., 1996; Moon and Capecchi, 2000; Vogel et al., 1996). Interestingly, conditional knockout of Fgf8 before the onset of expression in the apical ectodermal ridge reduces the size of the hindlimb bud in an apoptosis-independent manner (Sun et al., 2002). This suggests that a previously unsuspected Fgf mechanism might regulate the number of precursor cells that ultimately contribute to the limb bud. We speculate that an Fgf molecule might play a similar role in tetrapods as in zebrafish to induce cell convergence into the nascent limb bud position. Alternatively, as Fgf signaling has also been shown to promote cell motility locally in a non-directional manner (Bénazéraf et al., 2010; Gros et al., 2010), Fgfs might permissively promote cell motility in the limb-field, thus enabling limb-field cells to respond to an instructive signal such as Wnt5a, which has been shown to promote oriented cell behaviors during tetrapod limb bud development (Wyngaarden et al., 2010).

Preliminary studies (Q.M. and H.K.S., unpublished data) suggest that Tbx5a morphant fin-field cells do not appear to alter their migration direction, speed or persistence as robustly as Fgf24 morphant cells when presented with an Fgf8-coated bead. Although more studies are needed, these initial results suggest that, beyond inducing Fgf24 expression, Tbx5a might have additional roles in controlling fin-field LPM cell convergence. One possibility that we are currently investigating is whether Tbx5a is also required for the expression of an Fgf receptor (Fgfr) in the fin-field. If indeed the difference in fin-field responsiveness to an Fgf signal between Tbx5a and Fgf24 morphants is attributable to an Fgfr downstream to Tbx5a, we would hypothesize the existence of an Fgfr whose expression is directly regulated by Tbx5a but perhaps not by Fgf24.

We have demonstrated that partial reduction in Tbx5a or Fgf24 levels leads to an anteriorly biased perturbation of LPM cell convergence. Interestingly, loss of anterior cell migration often results in anteriorly distorted truncation of the AP axis in the pectoral fin cartilaginous disc, and this bias is also reflected in the expression of Hox (AP identity) genes. We suggest that the AP identity of fin-field cells (i.e. pre-pattern) might be specified early, preserved throughout the orderly fin-field cell convergence and eventually translated into the AP identity of the fin bud and fin skeleton. This can explain the correlated spatial bias between mis-migrating fin-field cells, reduction in fin bud fate markers, and skeletal defects in partial Tbx5a or Fgf24 morphants. It is possible that the pre-pattern can be shaped as early as gastrulation, when the LPM is exposed to multiple morphogen gradients, such as Retinoic acid along the AP axis (Keegan et al., 2005).

We propose that the pre-axial hemimelia displayed by individuals with Holt–Oram syndrome might be attributable in part to the preferential mis-migration of anterior limb-field cells during limb bud initiation. In multiple vertebrate species (mouse, chicken and zebrafish), the migration of LPM cells has now been shown to be involved during limb bud development (Wyngaarden et al., 2010). Here, we show that an early biased disruption in zebrafish fin-field cell convergence generates a distorted fin bud with proportionally fewer anterior fin progenitors than posterior ones. Interestingly, AP distortion in mouse limb bud fates by ectopic activation of Shh (Knezevic et al., 1997; Krawchuk et al., 2010; Zhang et al., 2010) recapitulates the pre-axial hemimelia condition observed in Holt–Oram syndrome. These findings argue that pre-axial hemimelia results from distortion of the AP axis, which can result from either early migration errors by fin-field cells or later patterning defects within the fin bud proper. Most importantly, our findings show that a balanced ratio of anterior to posterior fin progenitors is crucial for proper fin development. Given that many similar processes are involved in de novo development and in regeneration, we believe that the ratio of anterior versus posterior cells might be an important factor in the ability of a limb to regenerate. Finally, most congenital limb malformations possess a Mendelian underpinning in a defective developmental pathway (de Graaff and Kozin, 2009). We propose that our work provides a reference model for an examination of early cellular events during limb bud initiation in other models of human diseases, which might shed light on common root causes of limb birth defects.

Embryonic and adult zebrafish were raised and maintained in standard laboratory conditions (Westerfield, 2007). Et(hand2:eGFP)ch2 was generated with an eGFP-containing Tol2 transgenic construct (D. G. Ahn; supplementary materials and methods). Tg(h2afx:h2afv-mCherry)mw3 was a gift from B. Link (McMahon et al., 2009). Tg(b-act2:Brainbow1.0L)pd49 was provided by K. Poss (Gupta and Poss, 2012). Morpholino (Gene Tools, LLC) injections were performed as described by Nasevicius and Ekker (2000) (see supplementary materials and methods). Chromogenic mRNA in situ hybridizations were performed as described by Ahn et al. (2002) (see supplementary materials and methods). Quantification of gene expression area was performed by manual color thresholding with the RGB Threshold function in ImageJ (National Institutes of Health) for 100% incorporation of NBT/BCIP signals into the thresholded area.

A single marginal blastomere in the 64- or 128-cell-stage embryo was injected with 25 ng/µl Kaede mRNA. A single LPM cell from the labeled blastomere clone at 14 hpf was illuminated for 20-30 s through a pinhole in combination with a 63× dipping objective mounted on a Zeiss Axioplan microscope with light from a 100 W HBO mercury bulb passing through a DAPI filter set. Dorsal and lateral views of the embryos were photographed using a Nikon D5000 camera immediately after photo-conversion, at 28 and 48 hpf. The positions of the labeled progeny were analyzed further in ImageJ.

Bead implantation procedures followed those described by Picker et al. (2009) with the following modifications. Recombinant human-mouse Fgf8b protein (R&D Systems; cat. no. 423-F8-025) was diluted to 125 ng/μl in 0.1% BSA in 1× PBS and incubated with Polystyrene 45.0 μm microspheres (Polysciences; cat. no. 07314) overnight at 4°C. Embryos at 14 hpf were stabilized on 3% methylcellulose covered by Ringer solution before a mineral oil droplet was administrated through a micropipette to remove the epidermis over the targeted LPM region. A tungsten needle was used to position a loaded bead into the LPM. Embryos were then left undisturbed in E3 medium for at least 1 h before embedding into low-melting-point agarose for live imaging or fixing in 4% PFA for RNA in situ hybridization.

Live imaging of the fin-field was performed using Et(hand2:eGFP)ch2; Tg(h2afx:h2afv-mCherry)mw3 double hemizygous embryos. Dechorionated embryos were mounted in 0.7% low-melting-point agarose (Sigma; A9414) in a glass-bottomed Petri dish containing E3 medium with 0.16% tricaine and kept at 27°C on a heated stage. z-stacks were collected every 8 min with a 20×/0.7 water-immersion objective on an inverted Olympus DSU spinning disk microscope. Cell tracking was performed semi-automatically. Maximal intensity projections were created from the mCherry channel z-stacks and were used for generating two-dimensional tracks. Sixty cell tracks were generated per embryo with the FIJI Manual Tracking plug-in by randomly selecting nuclei from one side of the LPM. Two-dimensional tracks were then imported into a custom FIJI macro to generate z-coordinates. Subsequent computational analyses were performed in R. Detailed tracking procedures and trajectory analyses are described further in the supplementary materials and methods. Wilcoxon rank sum test was performed as described by Bauer (1972).

We thank V. Bindokas and the EMBO 3D Developmental Imaging Course for microscopy support, Adam Kuuspalu for animal care and E. L. Ferguson and V. E. Prince for helpful comments on the manuscript. We thank D. G. Ahn for generating and providing the Et(hand2:eGFP)ch2 enhancer trap line.