Role of Retinoic Acid Signaling, FGF Signaling and Meis Genes in Control of Limb Development

Abstract

The function of retinoic acid (RA) during limb development is still debated, as loss and gain of function studies led to opposite conclusions. With regard to limb initiation, genetic studies demonstrated that activation of FGF10 signaling is required for the emergence of limb buds from the trunk, with Tbx5 and RA signaling acting upstream in the forelimb field, whereas Tbx4 and Pitx1 act upstream in the hindlimb field. Early studies in chick embryos suggested that RA as well as Meis1 and Meis2 (Meis1/2) are required for subsequent proximodistal patterning of both forelimbs and hindlimbs, with RA diffusing from the trunk, functioning to activate Meis1/2 specifically in the proximal limb bud mesoderm. However, genetic loss of RA signaling does not result in loss of limb Meis1/2 expression and limb patterning is normal, although Meis1/2 expression is reduced in trunk somitic mesoderm. More recent studies demonstrated that global genetic loss of Meis1/2 results in a somite defect and failure of limb bud initiation. Other new studies reported that conditional genetic loss of Meis1/2 in the limb results in proximodistal patterning defects, and distal FGF8 signaling represses Meis1/2 to constrain its expression to the proximal limb. In this review, we hypothesize that RA and Meis1/2 both function in the trunk to initiate forelimb bud initiation, but that limb Meis1/2 expression is activated proximally by a factor other than RA and repressed distally by FGF8 to generate proximodistal patterning.

1. Introduction

Investigation of the mechanisms underlying limb development serves as a paradigm for understanding development in general. Many signaling and transcriptional pathways converge to generate growth and patterning of the limb [1]. Early studies found that the emergence of limb buds from the trunk is dependent upon fibroblast growth factor 10 (FGF10) signaling [2], whereas subsequent growth of the limb also requires FGF8 signaling [3,4]. During forelimb bud initiation, Tbx5 is required upstream of FGF10 [5,6,7] and retinoic acid (RA) is required upstream of Tbx5 [8]. During hindlimb bud initiation, Tbx4 and Pitx1 function upstream of FGF10 [9]. Subsequent proximodistal patterning of both forelimbs and hindlimbs was suggested from studies in chick embryos to require proximal-specific expression of Meis1 and Meis2 (Meis1/2) as well as RA signaling, which was proposed to activate Meis1/2 expression proximally [10,11,12]. However, genetic loss of RA signaling in mouse embryos using knockout mutations of RA-generating enzymes encoded by Aldh1a2 or Rdh10 resulted in no change in proximal limb Meis1/2 expression and normal limb patterning [8,13,14]. Here, we discuss the mechanisms of limb initiation and proximodistal patterning in the light of recent genetic studies on RA signaling and Meis1/2 function.

2. Requirement of Retinoic Acid for Forelimb Bud Initiation

2.1. Mechanism of Retinoic Acid (RA) Signaling

During early development when limb buds are forming, RA is generated by sequential metabolism of retinol (vitamin A) to retinaldehyde by RDH10 [15], followed by the metabolism of retinaldehyde to RA by ALDH1A2 [16]; as Rdh10 and Aldh1a2 are expressed in the trunk but not the proximal limb, RA enters the limb by diffusion from the trunk (Figure 1A). RA can be degraded by CYP26 enzymes encoded by Cyp26a1, Cyp26b1, or Cyp26c1 [17]; Cyp26b1 is expressed in distal limb after outgrowth has begun (Figure 1A). RA controls gene expression at the transcriptional level by serving as a ligand for nuclear RA receptors (RARs) that form a heterodimer complex with retinoid X receptor (RXR) when bound to an RA response element (RARE) [18]. Binding of RA induces a conformational shift in RAR which alters binding of coactivators that stimulate deposition of histone H3 K27 acetylation (H3K27ac) marks associated with gene activation, or binding of corepressors that stimulate deposition of histone H3 K27 trimethylation (H3K27me3) marks associated with gene repression [19,20] (Figure 1B,C).

2.2. Function of RA during Forelimb Budding

The mechanism through which RA signaling regulates limb development has been controversial as chick limbs treated with RA or RAR antagonists exhibit altered proximodistal patterning [10,11,12], whereas genetic loss of Aldh1a2 or Rdh10 in mice does not alter limb proximodistal patterning [8,13,14]. However, genetic loss of Aldh1a2 or Rdh10 does disrupt initiation of forelimb development in mice or zebrafish, although hindlimb development is not affected [8,13,15,21,22]. Mouse and zebrafish Aldh1a2 mutants lack expression of Tbx5 in trunk lateral plate mesoderm that forms the forelimb field. As Tbx5 is essential for forelimb initiation [5,6,7] via stimulation of epithelial-to-mesenchymal transition [23] and activation of Fgf10 [2], RA functions upstream of these important regulators of forelimb bud initiation. Other studies that support a function for RA in forelimb bud initiation include the treatment of chick embryos with the RA synthesis inhibitor disulfiram prior to forelimb bud (wing bud) establishment that prevents forelimb bud initiation [24], plus vitamin A-deficient rat embryos that exhibit forelimb hypoplasia [25] similar to Rdh10 knockout mice [15]. In addition, in chicken embryos treated with beads containing an RAR antagonist, a phenotype comparable to mouse Rdh10 mutant stunted forelimbs can be observed that leads to a shorter humerus [26].

Comparison of the forelimb fields of mouse Aldh1a2 and Rdh10 mutants revealed that RA activity, normally present throughout the trunk anteroposterior axis, is required to restrict Fgf8 expression to an anterior trunk domain in the heart and a posterior domain (caudal epiblast or tailbud) on either side of the forelimb field. Rdh10 mutants exhibit stunted forelimbs and lose early RA activity in the heart and forelimb domains needed to set the boundary of heart Fgf8 expression, but RA activity still exists caudally which functions to repress caudal Fgf8 and set its caudal expression boundary; this results in an ectopic Fgf8 expression domain that stretches posteriorly from the heart into the forelimb field, resulting in a forelimb field Tbx5 expression domain that is delayed and shortened along the anteroposterior axis [14]. Aldh1a2-/- embryos lack RA activity throughout the entire trunk, resulting in ectopic Fgf8 expression that enters the forelimb field from both the heart and the caudal regions, leading to complete loss of forelimb field Tbx5 expression and no appearance of forelimb buds [8]. These findings, plus the observation that cultured wild-type mouse embryos treated with FGF8 fail to activate Tbx5 in the forelimb field [14], demonstrate that the underlying cause of forelimb bud initiation defects in mutants lacking RA synthesis is excessive FGF8 activity in the trunk leading to a disruption of Tbx5 activation in the forelimb field (Figure 2). This model, in which RA functions permissively to allow forelimb Tbx5 activation by repressing Fgf8, is also supported by the observation that forelimb bud initiation can be rescued in aldh1a2 mutant zebrafish by introducing a heat-shock-inducible dominant-negative fibroblast growth factor (FGF) receptor [14].

Studies on chick embryos treated with RA combined with mouse enhancer reporter transgene studies suggested that RA signaling may also function instructively to directly activate forelimb Tbx5 via a potential forelimb enhancer located in intron 2 that contains several HOX-binding sites and a RARE [27,28]; however, this was a degenerate RARE with several base pair mismatches to the RARE consensus sequence [18]. When this potential enhancer (including the degenerate RARE) was deleted in mice using CRISPR/Cas9 gene editing, there was no effect on forelimb bud initiation or later forelimb development [29]. These deletion studies also demonstrated that enhancer reporter transgenes may not always identify endogenous enhancers, indicating that genetic deletion studies are also needed [30]. The challenge of finding limb enhancer regions near Tbx5 with a functional role in vivo is still ongoing and these studies could allow a better understanding of the underlying mechanism of Tbx5 activation. With regard to instructive mechanisms, as genetic studies in zebrafish demonstrate that loss of forelimb bud initiation in RA-deficient embryos can be fully rescued by elimination of excess FGF signaling [14], this finding provides evidence that there is no requirement for RA to function instructively to activate forelimb Tbx5 expression, only a permissive role. Future studies that identify an RA-independent enhancer required in vivo to activate Tbx5 during forelimb initiation would argue in favor of this hypothesis.

Overall, chick studies relying on treatment with signaling agents such as RA may not reveal normal functions of RA signaling. Similarly, treatment of chick interlimb trunk (between the forelimb and hindlimb buds) with FGF8 stimulates limb bud initiation [31,32], but it is clear to all in the field from mouse Fgf8 genetic loss-of-function studies that FGF8 is not required for limb bud initiation but instead for limb outgrowth and patterning [3,4]; Fgf8 is not expressed in trunk lateral plate mesoderm undergoing limb bud initiation and is only expressed in the limb after a substantial bud has already grown.

2.3. Mechanism through Which RA Represses Fgf8 in the Developing Trunk

Studies on the mechanism underlying RA repression of trunk Fgf8 have shown that Fgf8 is a direct target of RA signaling through an upstream RARE silencer that was shown to be required for repression of caudal Fgf8 expression in vivo either by transgene analysis or CRISPR/Cas9 deletion of the native RARE [33,34]. ChIP studies on Aldh1a2 knockout embryos suggest that RA-mediated Fgf8 repression caudally involves RA-dependent recruitment of nuclear receptor corepressors (NCOR1 and NCOR2), polycomb repressive complex-2 (PRC2) and histone deacetylase-1 (HDAC1) to the Fgf8 RARE plus nearby deposition of the repressive H3K27me3 mark [33,34]. In addition, RA signaling in trunk cells stimulates movement of the Fgf8 chromosomal region to the nuclear periphery, which is associated with gene repression [35].

Recently, ChIP-seq (H3K27ac and H3K27me3) analysis combined with RNA-seq analysis performed on wild-type vs. Aldh1a2-/- trunk tissue found that the previously known RARE upstream of Fgf8 is normally marked by H3K27me3, but this mark is lost in the RA-deficient mutant [36]. In this study, two additional RAREs near Fgf8 were also found to be marked by H3K27me3 when RA is present, highlighting the existence of additional potential Fgf8 RARE silencers that can be further studied to pursue the mechanism underlying RA repression of Fgf8 in heart tissues compared to caudal tissues already examined [33,34].

2.4. Loss of Trunk RA Signaling Does Not Affect Hindlimb Bud Initiation

Hindlimb bud initiation is not perturbed by a loss of trunk RA activity in Aldh1a2 and Rdh10 mutants [13,15]. Hindlimb buds do not express Tbx5, but instead use Tbx4 and Pitx1 to initiate expression of Fgf10 to stimulate outgrowth [9,37,38]. Expression of Tbx4 and Pitx1 is unaffected in the hindlimbs of mouse Aldh1a2 and Rdh10 mutants lacking RA activity [8,13].

3. Requirement of Meis Genes for Forelimb Bud Initiation

After the Drosophila Meis homolog was shown to be required for proximodistal patterning of fly limbs [39], subsequent studies on vertebrate Meis1 and Meis2 demonstrated that both are expressed throughout the trunk mesoderm (including lateral plate mesoderm that gives rise to limb mesoderm) and in the proximal region of both forelimb and hindlimb buds, suggesting a role in limb proximodistal patterning [10]. Studies on chick limbs showed that treatment with RA or RAR antagonists alters proximal-specific expression of Meis1/2, with RA expanding expression into the distal domain, and RAR antagonists eliminating proximal expression [10,11,12]. However, genetic loss of RA signaling in mouse does not alter proximal-specific expression of Meis1/2 despite loss of all RA signaling activity in the trunk and limb mesoderm [8,13,14]. In addition, no change in Meis2 expression in the interdigital tissues of the mouse was found in Rdh10 mutants lacking limb RA signaling [40]. Prior to forelimb bud initiation, RA is also not essential for Meis1/2 expression in trunk lateral plate mesoderm that gives rise to limb buds, but loss of RA does reduce Meis1/2 expression in somites [14,36].

In order to globally identify RA target genes in the trunk just prior to forelimb bud initiation, studies were performed that combined epigenetic ChIP-seq (H3K27ac and H3K27me3) with RNA-seq to compare E8.5 wild-type vs Aldh1a2-/- trunks lacking RA synthesis. Candidate targets were defined as genes with RA-regulated mRNA abundance that also have nearby RA-regulated H3K27ac (gene activation) or H3K27me3 (gene repression) marks associated with conserved RAREs. This approach identified many previously known RA target genes that control trunk formation, plus additional RA target genes were identified including Meis1 and Meis2 [36]. Meis1/2 were both found to have nearby RAREs that are highly conserved from mammals to reptiles or frogs [36]. Together with previous findings, these new findings suggest that Meis1/2 are normally activated by RA in the trunk somitic mesoderm but not in the lateral plate mesoderm or limbs, and that the effects of RA and RAR antagonist treatments on Meis1/2 expression in limb are due to off-target effects in which high concentrations of these reagents target Meis1/2 RAREs in limb tissue to usurp control of these genes and override their normal limb activators and repressors.

Genetic studies in mice have been performed to address Meis1/2 function during limb development. The Meis1 knockout is lethal at E11.5 due to hematopoietic defects, and the Meis2 knockout is lethal at E14.5 with defects in cardiac and cranial neural crest, but in both cases no defects in body axis or limb development were observed [41,42]. However, redundancy between Meis1 and Meis2 may have masked a defect. CRISPR/Cas9 gene editing was performed to generate Meis1/2 double mutants in mice, and dissection at E10.5 resulted in embryos that exhibit a body axis defect (small somites) with developmental arrest at E9.5, plus a lack of forelimb buds which are normally easily visible at E9.5; arrested development prevented analysis of hindlimb buds which develop after forelimb buds [36]. Thus, global knockout of Meis1/2 unexpectedly resulted in the loss of forelimb bud initiation, which indicates an important role in limb development, in addition to a role in somitogenesis (Figure 2).

4. Meis Genes and FGF Signaling Are Required for Limb Proximodistal Patterning but RA Signaling Is Dispensable

4.1. Chick Studies Support a Two-Signal Model for Limb Proximodistal Patterning

Early studies on chick limbs treated with RA suggested a role for RA in limb anteroposterior patterning [43]. However, subsequent studies in chicks provided evidence that RA is not required for limb anteroposterior patterning [44,45], but that sonic hedgehog (SHH) is the diffusible signaling factor required for limb anteroposterior patterning [46]; this requirement for SHH was confirmed in mouse knockout studies [47]. Other studies in chicks demonstrated that treatment of limb buds with RA, RAR antagonists, FGF8, or FGF receptor antagonists alters proximodistal patterning [10,11,12]. Mouse genetic loss-of-function studies verified a requirement for Fgf8 (and other Fgf genes) expressed distally in the apical ectodermal ridge (AER) to control limb proximodistal patterning, including restriction of Meis1/2 expression to the proximal limb [48]. However, loss of RA signaling by knockout of RA-generating enzymes encoded by Aldh1a2 or Rdh10 did not result in loss of limb proximodistal patterning or loss of proximal-specific expression of Meis1/2 [8,13,14].

Some researchers supporting the two-signal model raised criticisms, summarized in a review [49], about whether there is a total absence of RA, as measured using the RARE-lacZ RA-reporter transgene, in limb mesoderm of E9.5-E10.5 Aldh1a2 mutants (that need to be treated with a small dose of RA at E7.5 in order to survive to E10.5) or Rdh10 mutants (that survive to E10.5 without RA treatment). Aldh1a2 mutants treated with a very small dose of RA at E7.5 were shown to activate RARE-lacZ in the trunk mesoderm at E7.5-E8.5 and survive until E10.5, but at E9.5-E10.5 when limbs form, RARE-lacZ expression was not detected in the trunk lateral plate mesoderm or limb buds, indicating that the administered RA had been efficiently cleared [8]. In addition, RA titration studies on cultured Rdh10 mutants that normally have no expression of RARE-lacZ in limb or lateral plate mesoderm demonstrated that RARE-lacZ expression can be activated by as low as 0.25 nM RA, 100-fold lower than the normal level of limb RA, which is 25 nM, demonstrating that RARE-lacZ is a very sensitive RA-reporter [29]. Thus, as RA is reduced by at least 100-fold in limbs and lateral plate mesoderm of Aldh1a2 and Rdh10 mutants, it is reasonable to conclude that the concentration would be too low to provide completely normal patterning if RA is required as suggested by the two-signal model.

With regard to a potential interaction of RA and FGF8 during limb patterning, loss of limb RA in Rdh10 mutants does not alter the expression of Fgf8 in the AER [13]. This observation demonstrates that although RA does repress Fgf8 in the body axis to limit its expression to the heart and caudal regions [33,34], one should not assume that RA represses Fgf8 in other tissues.

Models based on the treatment of chick embryos with RA and RAR antagonists are weakened by the possibility of off-target effects. As RAR antagonists function as inverse-agonists that silence any gene near an RAR-bound RARE [50], they may dominantly repress genes that have a RARE nearby even though they normally use different regulatory elements for activation in a particular tissue. In the case of Meis1/2, the ability of RA and RAR antagonist treatments to effect expression in limb buds, even though loss of endogenous RA does not, can be rationalized by the recent discovery that these genes do require RA for full activation in trunk somites and both genes have functional RAREs [36]. Thus, retinoid treatment regimens may force effects on Meis1/2 expression in tissues that normally do not use RA to regulate Meis1/2.

Despite these conflicting results, a ‘two-signal model’ has been proposed in which RA generated in the trunk diffuses into the proximal region of the limb to promote proximal character by activation of Meis1/2 expression, with distal FGF8 signaling promoting distal character by repressing Meis1/2 expression [10,11,12]. This model is also supported by the observation that an RA-degrading enzyme encoded by Cyp26b1 is expressed in the distal limb under the control of FGF8, with CYP26B1 being required to prevent trunk RA from diffusing into the distal limb, which was reported to ectopically activate Meis1/2 expression distally [51,52]. In the Cyp26b1 knockout, Meis1/2 expression expands into the distal limb, similar to chick RA-treatment studies, but forelimb and hindlimb buds are truncated along the entire proximodistal axis, which is inconsistent with RA functioning to induce proximal identity [51]. In addition, in Cyp26b1 knockouts, the presence of endogenous RA in distal limbs where it should not normally exist results in a phenotype similar to exogenous RA teratogenesis with increased apoptosis and a block in chondrogenic differentiation, particularly to form the intricate skeletal structures of distal tissues such as hand/foot [53,54] or craniofacial structures [55]. Thus, the presence of endogenous RA in the proximal limb does not necessarily correlate with a designed function in proximodistal patterning of Meis1/2 expression, but instead may simply indicate diffusion overflow from the trunk where RA is required for body axis formation and forelimb initiation, with this RA being neither necessary nor harmful to proximal limb development. Expression of Cyp26b1 in the distal limb may not indicate a role in restricting RA signaling and sharpening an RA gradient to set the boundary of Meis1/2 expression, but may simply function to eliminate RA signaling distally where it is harmful to distal limb development. This would be similar to the function of the related RA-degrading enzyme CYP26A1 in the caudal body axis which functions to remove RA that interferes with body axis extension [56,57], thus restricting RA to the trunk/caudal border where RA is required for spinal cord development (mouse and zebrafish) and somitogenesis (mouse but not zebrafish) [16].

4.2. Mouse Genetic Studies Support a One-Signal Model for Limb Proximodistal Patterning

We previously suggested that the most parsimonious model for limb proximodistal patterning is a "one-signal model" that is driven by distal FGFs (including FGF8) that stimulate outgrowth, repress Meis1/2 to generate proximodistal patterning, and activate Cyp26b1 to remove distal RA from diffusing in from the trunk that would result in limb teratogenesis [18]. Recent studies reported that conditional genetic loss of Meis1/2 in the mouse limb results in proximodistal patterning defects [58]. Additionally, it was reported that restriction of Meis1/2 expression to the proximal limb is controlled by the repressive action of distal FGFs (including FGF8) expressed in the AER [58]. As the AER does not form until after the proximal limb is formed, thus delaying production of distal FGFs until after limb bud initiation [32], this results in early Meis1/2 expression throughout the limb. After the AER is established, Meis1/2 expression is repressed by distal FGFs, thus creating a boundary of Meis1/2 expression that leaves it in a proximal position as the distal limb continues to grow free of Meis1/2 expression (Figure 3).

5. Conclusions and Perspectives

New findings have clarified the mechanisms through which RA signaling, FGF signaling, and Meis1/2 genes control limb development. We know from recent mouse genetic loss-of-function studies that Meis1/2 are required for limb bud initiation (at least forelimb) as well as subsequent proximodistal patterning [36,58]. Thus, by combining all genetic studies it can be established that RA signaling functions permissively during forelimb initiation to allow expression of Tbx5 by repressing trunk Fgf8 that represses limb field Tbx5, but that RA is not required for limb patterning. Furthermore, genetic studies show that RA control of Meis1/2 occurs in the trunk for somitogenesis but not in limbs, whereas FGF signaling is required for both limb initiation (FGF10) and patterning (FGF8 and other distal FGFs), with distal FGFs restricting Meis1/2 expression to the proximal limb. With regard to limb patterning, these new genetic findings do not support a two-signal RA-FGF model for proximodistal patterning, but instead support a one-signal model in which a factor other than RA activates Meis1/2 expression during early limb bud outgrowth followed by action of distal FGFs to repress Meis1/2. As Meis1/2 is expressed in the trunk lateral plate mesoderm prior to limb bud outgrowth from this tissue, and as RA is not essential for Meis1/2 expression in trunk lateral plate mesoderm, it is possible that whatever factor activates Meis1/2 in the lateral plate mesoderm continues to allow expression as lateral plate mesodermal cells undergo epithelial-to-mesenchymal transition to form the limb bud. A final proof that RA is not required for Meis1/2 activation in the proximal limb would come from the identification of such a factor.

Determining the normal functions of RA during development is difficult as pharmacological studies and genetic loss-of-function studies often lead to conflicting results. Similarly, pharmacological studies using treatment with FGF8 led to erroneous conclusions about FGF8 function in limb bud initiation that were later reversed by mouse genetic studies. Thus, a greater reliance on genetic studies will provide clarity on the developmental pathways controlled by RA signaling, FGF signaling, and Meis1/2.