Evolution of repressive sequences within an enhancer contributed to morphological diversity in crucifer plants Authors: Alessandro Popoli, Remco A. Mentink, Lisa Brombach, Nora Papadima-Karanikou, Manuel Buendia-Monreal, Mingming Fang, Manas Joshi, Saiko Yoshida, Stefan Laurent, Peter Huijser, Miltos Tsiantis Published: 2025 ABSTRACT SignificanceEnhancers are regulatory DNA sequences that control gene expression and drive morphological evolution. Yet, how they diversify to create new expression domains remains unclear. We explore this by investigating an enhancer of the RCO plant homeobox gene, which arose through gene duplication and promotes leaf complexity. Compared to its ancestral counterpart, the RCO enhancer is less stringently required for RCO expression and is subject to more repression. This repression helped establish RCO’s expression pattern while avoiding potentially pleiotropic effects arising from its change in expression. Our work highlights regulatory features of a diversity-linked enhancer and underscores the importance of repressive sequences in morphological evolution. A key problem in biology is to identify the genetic basis for morphological differences between species. Diversification of cis-regulatory sequences such as enhancers has emerged as a major mechanism involved in morphological evolution (1–5). However, specific examples of how enhancers evolve to yield novel gene expression domains that underpin morphological diversity are lacking. The textbook paradigm of enhancers (6, 7) considers them to be discrete and highly modular, but recent work suggests that they can be strongly entangled with neighboring regulatory sequences (8) further complicating analyses of how enhancers diversify. CRISPR/Cas9 mutagenesis has opened up new opportunities to address these questions (9). However, although genomic areas that contribute to developmental gene regulation have been defined (10–13), CRISPR/Cas9 has not yet been systematically deployed to identify specific mechanisms through which enhancers functionally diversify to produce novel gene expression domains and morphological variation. Leaves provide an attractive system for understanding mechanisms of regulatory evolution because their shapes present considerable heritable variation, underpinned by changes in expression of developmental genes. Leaf form can be described as simple (if the leaf blade is undivided) or dissected (if the blade is divided into distinct leaflets) (14–16). Additionally, leaves can show intermediate complexity and elaborate outgrowths that are less pronounced than leaflets, referred to as lobes. It was previously shown that regulatory diversification after tandem gene duplication of the REDUCED COMPLEXITY (RCO) homeobox gene (Fig. 1A) underlies the emergence of complex (dissected and lobed) leaf shapes in the Brassicaceae (17, 18). This process involved the acquisition of an RCO gene expression domain comprising distinct expression foci at the leaf base, which contrasts with the distal expression domain of its ancestral paralog LATE MERISTEM IDENTITY 1 (LMI1) in the leaf margin and vestigial appendages called stipules (17). Mutations in the RCO enhancer result in weak RCO loss-of-function or RCO gain-of-function phenotypes. (A) Schematic representation of the LMI1/RCO locus in C. hirsuta. (B) Silhouettes of representative leaf 5 from wild-type (WT), lmi1−, and rco− reference alleles. The leaf-like organs resulting from the stipule outgrowths in lmi1− are indicated with red arrowheads. (Scale bar: 1 cm.) (C) Schematic representation of the CRISPR-Cas9-generated mutant alleles in the RCO enhancer. Blue arrows indicate the position and direction of the sgRNA target sequences used. The first five alleles indicate rcoenh mutants and the remaining three indicate rcoenh/lmi1enh double mutants. Yellow- and green-shaded boxes highlight sequence the deletion of which causes RCO gain-of-function (yellow, Repressor 64 or Rep64) or loss-of-function (green, Activator 69 or Act69), respectively. The function of these sequences is also tested in transgenic reporter gene assays in Fig. 4 D and E. (D) Normalized-Differential Margin Complexity (NDMC) of the terminal leaflet (TL) of leaf 5 from the mutants shown in (C). Increasing values indicate increasing complexity. The dashed line represents the WT average value. WT: n = 25, mut1: n = 13, mut2: n = 14, mut3: n = 13, mut4: n = 11, rcoenh−: n = 18, mut5: n = 13, mut6: n = 12, mut7: n = 7. (E) qPCR assay of the relative RCO expression levels of the alleles shown in (C), measured in 14-d-old seedling apices. The dashed line indicates average expression in the WT (=1). For (D and E), the effects of the mutations on expression and terminal leaflet margin complexity were tested for significance using a mixed-effects model (*P < 0.05, **P < 0.01, ***P < 0.001). WT: n = 7, mut1: n = 3, mut2: n = 7, mut3: n = 6, mut4: n = 6, rcoenh−: n = 7, mut5: n = 3, mut6: n = 3, mut7: n = 3, rco−: n = 3. (F) Silhouettes of representative leaf 5 of the mutants shown in (C). Black arrowheads indicate the dissection of the terminal leaflet in the RCO gain-of-function mutants, while red arrowheads indicate stipule outgrowths. (Scale bar: 1 cm.) (G) Photographs of representative lateral leaflets (first pair) of leaf 5 from WT, mut4, rcoenh-, and mut7. (Scale bar: 1 cm.) (H) Boxplots indicating petiolule narrowness (see Materials and Methods details “Leaf phenotype analysis”) of the alleles shown in (G). The effects of the mutations on petiolule narrowness were tested for significance using a linear model (*P < 0.05, **P < 0.01, ***P < 0.001). WT: n = 22, mut4: n = 22, rcoenh-: n = 36, mut7: n = 14. Boxplots (D, E, and H): boxes display the interquartile range (25th and 75th percentiles). Thicker line: median. The shift in gene expression from an LMI1 to an RCO pattern played a key role in crucifer leaf shape evolution. Transgenic experiments indicated that an approximately 500 bp enhancer element, RCOenh500, was sufficient to control this shift in gene expression domain, but causal sequence changes and mechanisms remain unknown (19). It is also unclear to what degree transgenic assays using reporter genes reflect the function of regulatory elements in their native genomic context—an issue that can potentially be resolved by studying CRISPR/Cas9 regulatory alleles. Leaves of rco loss-of-function alleles show reduced complexity since the RCO protein is a growth repressor that acts to separate distinct leaflets to form complex leaves (17, 20). Conversely, since LMI1 normally acts to repress stipule growth, loss of LMI1 function causes the conversion of vestigial stipules to ectopic leaves at the leaf base [Fig. 1B and Vuolo et al. (21)]. These phenotypes can therefore be used to evaluate the consequences of regulatory mutations of RCO and LMI1 and the mechanistic basis for how the distinct gene expression domain conferred by RCOenh500 evolved from its ancestral LMI1 counterpart. Here, we used CRISPR/Cas9 genome editing to mutagenize the RCO and LMI1 enhancers in Cardamine hirsuta. We assayed phenotypes and transcript levels of 17 regulatory alleles of the two enhancers, including chimeric LMI1/RCO enhancer alleles. We found that the more recently evolved RCO enhancer was subject to more negative regulation and, while the LMI1 enhancer was strictly required for LMI1 gene expression, RCO enhancer deletion caused only partial loss of function. We used reporter assays of these enhancer variants to demonstrate how a repressive sequence in the RCO enhancer delimits the RCO expression domain and how a second sequence contributes to activating RCO expression. The repressive sequence that shapes the distinct RCO expression domain at the leaf base arose in association with a local within-enhancer duplication. Our findings underscore the significance of gene repression in evolving a gene expression domain and show that repressive sequences underlying this effect can arise through sequence duplication. Our results also demonstrate how combining dense CRISPR/Cas9 mutagenesis with reporter gene assays is a powerful strategy to identify enhancer regulatory features that underlie morphological evolution. Results and Discussion Generation of Mutant Alleles in RCO and LMI1 Enhancers Using CRISPR/Cas9. To investigate the function of RCO and LMI1 enhancers in their native genomic context, we generated CRISPR/Cas9 alleles using eight single guide RNAs for each enhancer. In this way, we isolated five mutants in the RCOenh500 (Fig. 1C and SI Appendix, Fig. S1), seven mutants in the LMI1enh500 (Fig. 2A and SI Appendix, Fig. S3) and three carrying mutations in both enhancers (Figs. 1C and 2A and SI Appendix, Fig. S1). Additionally, we recovered two complex alleles (mut14 and mut15) where the LMI1 gene is deleted, resulting in the fusion of LMI1 upstream sequences to the RCOenh500, with one of them (mut14) creating a hybrid LMI1enh500/RCOenh500 (Fig. 3A and SI Appendix, Fig. S6). We subsequently investigated the leaf phenotypes of these alleles as well as their impact on LMI1 or RCO transcript levels using qRT-PCR (Figs. 1 D and E, 2 B and C, and 3 B and C), with the aim of understanding the functional properties of these enhancers and the regulatory logic underlying their diversification following gene duplication. Mutations in the LMI1 enhancer readily cause strong LMI1 loss-of-function phenotypes. (A) Schematic representation of the CRISPR-Cas9-generated mutant alleles in the LMI1 enhancer. Blue arrows show the position and direction of the sgRNA target sequences used. The first seven alleles indicate lmi1enh mutants, and the remaining three indicate rcoenh/lmi1enh double mutants. (B) Stipule Outgrowth Index (see Materials and Methods details “Leaf phenotype analysis”) of the mutants shown in (A). Increasing values indicate longer and more frequent stipule outgrowths. The dashed line indicates the absence of stipule outgrowth (=0). The effects of the mutations on stipule growth were tested for significance using a mixed-effects model (*P < 0.05, **P < 0.01, ***P < 0.001). WT: n = 3, mut8: n = 3, mut9: n = 3, mut10: n = 3, mut11: n = 6, mut12: n = 3, mut13: n = 3, lmi1enh-: n = 7, mut5: n = 3, mut6: n = 3, mut7: n = 6, lmi1−: n = 8. (C) Relative LMI1 expression levels of the alleles shown in (A) as determined by qPCR, measured in 14-d-old seedling apices. The dashed line indicates expression in the WT. The effects of the mutations on expression were tested for significance using a linear model (*P < 0.05, **P < 0.01, ***P < 0.001). N = 3 biological replicates for all the samples. Boxplots in (B and C): boxes display the interquartile range (25th and 75th percentiles). Thicker line: median. (D) Silhouettes of a representative leaf 5 of the mutants shown in (A). Red arrowheads indicate stipule outgrowths. (Scale bar: 1 cm.) Effect of CRISPR-Cas9-generated hybrid LMI1/RCO enhancers on leaf morphology and RCO expression. (A) Schematic representation of the two CRISPR-Cas9-generated mutant alleles, where targeting both the LMI1 and the RCO enhancer resulted in an LMI1 gene deletion. Yellow- and green-shaded boxes in RCOenh indicate the same regions as in Fig. 1C. (B) Stipule Outgrowth Index (see Materials and Methods details “Leaf phenotype analysis”) of the mutants shown in (A). Increasing values indicate longer and more frequent stipule outgrowths. The dashed line indicates absence of stipule outgrowth (=0). The effects of the mutations on stipule growth were tested for significance using a mixed-effects model (*P < 0.05, **P < 0.01, ***P < 0.001). N = 6 for all the samples. (C) Relative RCO expression levels of the alleles reported in (A) as shown by qPCR, measured in 14-d-old seedling apices. The dashed line indicates control expression in the WT. The effects of the mutations on expression were tested for significance using a linear model (*P < 0.05, **P < 0.01, ***P < 0.001). N = 3 biological replicates for all the samples. (D) Silhouettes of a representative leaf 5 of the mutants shown in (A). The black arrowhead indicates the dissection of the terminal leaflet in the RCO gain-of-function mutant, while red arrowheads indicate stipule outgrowths. (Scale bar: 1 cm.) (E) NDMC of the terminal leaflet of leaf 5 from WT and mut14. The dashed line indicates the WT. The effects of the mutation on terminal leaflet margin complexity were tested for significance using a mixed-effects model (**P < 0.01). WT: n = 25, mut14: n = 8. (F and G) Representative in situ hybridization micrographs of RCO (F) and LMI1 (G) transcripts on vegetative meristem sections of WT, mut5, and mut14 plants. Black asterisks indicate ectopic expression in the TLs, while red arrowheads show stipules. Samples observed by genotype and gene: LMI1 probe on WT = 5, LMI1 probe on mut5 = 1, LMI1 probe on mut14 = 2, RCO probe on WT = 4, RCO probe on mut5 = 13, RCO probe on mut14 = 3. (Scale bar: 100 μm.) (H) Adjusted leaf area values (see Materials and Methods details “Normalized leaf area for node effects”) of leaves from nodes 5, 6, 7, and 8 of WT and the alleles showing an RCO gain-of-function phenotype (mut2, mut5, mut6, and mut14) which overall show reduced leaf surface. The effects of the mutations on leaf area were tested for significance using a mixed-effects model (**P < 0.01, ***P < 0.001). WT: n = 92, mut2: n = 48, mut5: n = 41, mut6: n = 21, and mut14: n = 32. Boxplots in (B, C, E, and H): boxes display the interquartile range (25th and 75th percentiles). Thicker line: median. Characterization of Mutant Alleles in RCOenh500. We characterized eight rcoenh mutant alleles, three of which also carried mutations in LMI1enh500 (Fig. 1, SI Appendix, Figs. S1 and S2, and Dataset S1). Among these eight, one (rcoenh-) was a deletion of 755-bp that spans the entire enhancer sequence (Fig. 1C). This deletion allele resulted in a 50% reduction in transcript levels and a leaf phenotype where leaflets are incompletely separated from the rachis and instead form deep lobes (Fig. 1 D–H and SI Appendix, Fig. S1). This phenotype is much weaker than the rco reference allele, which is caused by a premature stop codon (17). Therefore, other regulatory sequences must act redundantly with RCOenh500 to promote RCO gene expression, as has been shown in other systems (22, 23). Two other alleles (mut4 and the double mutant mut7) yielded very weak loss-of-function phenotypes, where petiolules do not form properly, resulting in weaker leaflet separation defects than in the rcoenh- mutant (Fig. 1 G and H and SI Appendix, Figs. S1 and S2). No reduction in RCO transcript was detected in these two mutants by qRT-PCR (Fig. 1E). Three additional alleles (mut2, mut5, and mut6) had a conspicuous leaf phenotype showing increased dissection of the terminal leaflet (TL) and an increase in RCO transcript levels (Fig. 1 D–F, black arrowheads in F showing the TL dissection). This TL phenotype resembles transgenic lines where RCO is ectopically expressed in the terminal leaflet (20), suggesting that these are gain-of-function alleles, where elevated RCO expression causes increased terminal leaflet complexity. Notably, mut6 is a hybrid of sequences from the two enhancers with 124 bp from LMI1enh500 inserted into RCOenh500 (Fig. 1C), raising the possibility that this 124 bp sequence contributes to the elevated RCO expression in this mutant. Finally, one allele (mut3) showed increased RCO transcript levels and a tendency to increase TL complexity that was not statistically significant (Fig. 1 C–F). Overall, these results highlight two properties of the RCOenh500: First, it is partially, but not fully required for RCO gene expression, as an enhancer deletion allele yielded a mild phenotype and it reduced, but did not abolish gene expression. Second, it contains repressive sequences that limit RCO gene expression and negatively regulate leaf dissection. Mutations in such repressive sequences appear easier to recover than strong loss-of-function alleles in this experimental design. Characterization of Mutant Alleles in LMI1enh500. To understand the profile of LMI1enh500 response to mutation, we characterized ten LMI1enh500 alleles, three of which were double LMI1enh500/RCOenh500 mutants (Fig. 2, SI Appendix, Figs. S3–S5, and Dataset S2). The LMI1enh500 profile was very different to what we observed for RCOenh500. A 900-bp deletion allele encompassing the entire enhancer (lmi1enh−) reduced LMI1 transcript levels and gave a similarly strong phenotype to the lmi1− deletion allele generated for the purpose of this study (Fig. 2 A–D, SI Appendix, Figs. S3 and S4, and Dataset S3). In both alleles, vestigial stipules were fully converted to leaves (Fig. 2D). Four additional alleles resulted in reduced LMI1 transcript levels and strong lmi1 loss-of-function phenotypes (mut11, mut12, mut13, and mut7 which also carries a mutation in RCO, Fig. 2 A–D). These results indicate that the mutated regions contain sequences necessary for LMI1 expression and that the LMI1 enhancer sequence is more prone to resulting in loss of function upon mutation than its RCO counterpart. Additionally, our analysis of all seventeen rcoenh and lmi1enh mutant alleles (Figs. 1 and 2) indicates that the RCO enhancer is more prone to generate gain- rather than loss-of-function alleles upon mutagenesis, consistent with it having evolved more negative regulation than the ancestral LMI1 enhancer. Characterization of Complex Alleles with Concurrent LMI1enh500 and RCOenh500 Deletions. Next, we studied the two complex alleles arising from concurrent mutation of LMI1enh500 and RCOenh500 (Fig. 3 A–D, SI Appendix, Fig. S6, and Dataset S4). The first allele, mut14, arose through deletion of the LMI1 gene which resulted in a fusion of the LMI1 and RCO enhancers (400-bp of the 5′ of LMI1enh500, 101 of which are inverted, and 200-bp of the 3′ of RCOenh500, Fig. 3A). This mutant showed increased expression of RCO (Fig. 3C) accompanied by an increased TL dissection phenotype (Fig. 3 D and E), indicating that it is an RCO gain-of-function allele. The second allele, mut15, had a larger deletion of LMI1 (5.4-kb) compared to mut14, removing the entire LMI1enh500 and 350-bp upstream sequence (Fig. 3A). The leaf phenotype of mut15 showed both RCO (Fig. 3D) and LMI1 loss-of-function phenotypes (Fig. 3B and SI Appendix, Fig. S6), consistent with the large deletion eliminating the function of both genes. The fact that mut15, but not mut14, showed an RCO loss-of-function phenotype (Fig. 3D) indicates that the deleted RCO sequences contain information required for RCO expression but can be compensated by LMI1 upstream sequences in mut14 that are removed in mut15 (Fig. 3A) and, in combination with the RCO upstream sequence still present in mut14, can activate gene expression in the RCO domain. Strikingly, mut14 did not show an lmi1 loss-of-function phenotype despite deletion of the entire LMI1 gene (Fig. 3B and SI Appendix, Fig. S6). This observation indicates that the hybrid enhancer and regulatory sequence created in this deletion allele can drive RCO expression in a way that allows it to functionally substitute for LMI1. This is consistent with previous findings that LMI1 can complement the rco loss-of-function phenotype when expressed as a transgene in the RCO domain (17) and provides genetic evidence from editing in the native genomic context that differences in expression patterns of LMI1 and RCO genes underlie their different roles in development. Consistent with these ideas, in situ hybridization of RCO in mut14 and the previously described gain-of-function mut5 allele (Fig. 1D) showed ectopic RCO expression in stipules, which is not detectable in the wild type (Fig. 3F, red arrowheads on stipules and SI Appendix, Fig. S7). LMI1 transcripts could be detected in mut5, but not in mut14, consistent with the deletion of LMI1 in the mut14 allele (Fig. 3G). Additionally, RCO transcripts were detected in the terminal leaflets of mut14 and mut5 leaves, consistent with the terminal leaflet lobing phenotype in these alleles (Fig. 3F). Notably, all the alleles with increased RCO expression and higher terminal leaflet complexity (mut2, mut5, mut6, and mut14) had smaller leaves (Fig. 3H), indicating that elevated RCO expression causes pleiotropic effects on leaf development. Transgenic Validation of Regulatory Sequences within RCOenh500. To perform a fine dissection of the RCOenh500, we compared the eight rcoenh alleles to identify specific sequences that potentially underlie repression or activation of gene expression and evolution of the novel RCO gene expression domain. For example, mut5 harbors a deletion of 64-bp, which defines the smallest lesion that results in increased RCO transcript levels and a gain-of-function phenotype, and overlaps with the 232-bp deletion in mut2 that yields a similar increase in RCO transcript and leaf complexity (Fig. 1 C–F, region highlighted in yellow in C). Therefore, this 64 bp sequence likely contains negative regulatory elements that delimit the correct RCO expression domain. Comparing mut7 and mut4, which both yield loss-of-function phenotypes (Fig. 1 G and H), identified a shared 69-bp sequence (region between R5 and R6, highlighted in green in Fig. 1C), which likely contains positive regulatory elements for RCO. To test these predictions, we used reporter gene assays to evaluate whether these rcoenh sequence variants are in themselves sufficient to alter gene expression in the manner indicated by the CRISPR/Cas9 alleles. Specifically, we recreated the two candidate deletions causing RCO gain- and loss-of-function; first, the 64-bp deletion in mut5 that we named Rep64 for “Repressor 64” (the yellow bar in Fig. 1C) and second, the 69-bp deletion in mut4 that we named Act69 for “Activator 69” (the green bar in Fig. 1C). In these assays, we also recreated the hybrid RCO/LMI1 enhancer in mut14 to test whether this hybrid enhancer is sufficient to cause broadened or elevated RCO gene expression. We subsequently compared the gene expression driven by each mutant enhancer to the wild-type RCOenh500 (Fig. 4 A–F’). These are important experiments to further understand the functional significance of CRISPR/Cas9-defined sequences. For example, confirmation that the activation or repression conferred by these mutated RCOenh500sequences can act in isolation would indicate that they function as distinct modules operating, at least to some degree, independently of neighboring genomic sequences. Reporter gene validation of effects of regions Rep64 and Act69 identified through CRISPR indicates their modular function. (A and B) Expression of ChRCOp:NLS-3xGFP (A) and ChLMI1p:NLS-3xGFP (B) in C. hirsuta and Arabidopsis thaliana leaf primordia. (C) Expression conferred by the WT RCOenh sequence (631 bp) when placed upstream a cassette comprising the minimal 35 s promoter and NLS-3xGFP in C. hirsuta and A. thaliana leaf primordia. (C’) Normalized leaf primordium expression profile of the construct in (C) averaged from 10 independent transformants. (D) Expression conferred by a modified RCOenh sequence containing the 64bp g.o.f. deletion (Rep64, as in mut5; 567 bp) upstream the same min35s:NLS-3xGFP cassette in C. hirsuta and A. thaliana leaf primordia. (D’) Normalized leaf primordium expression profile of the construct in (D) averaged from 10 independent transformants. (E) Expression conferred by a modified RCOenh sequence containing the 69bp l.o.f. deletion (Act69, as in mut4; 562 bp) when placed upstream of the same min35s:NLS-3xGFP cassette in C. hirsuta and A. thaliana leaf primordia. (E’) Normalized leaf primordium expression profile of the construct in (E) averaged from 9 and 10 independent transformants for C. hirsuta and A. thaliana, respectively. (F) Expression conferred by the hybrid LMI1-RCO enhancer from mut14 (613 bp) when placed upstream of the same min35s:NLS-3xGFP cassette in C. hirsuta and A. thaliana leaf primordia. (F’) Normalized leaf primordium expression profile of the construct in (F) averaged from 8 and 10 independent transformants for C. hirsuta and A. thaliana, respectively. (A–F): Red arrowheads show the stipules. The white arrowheads in (D) indicate the fluorescence detected in the TL of C. hirsuta and in the upper sinus of A. thaliana. The black arrowhead in (D’) indicates ectopic expression in the C. hirsuta terminal leaflet. (Scale bar: 100 μm.) The number of independent transgenic lines showing the reported expression pattern is shown at the bottom right-hand side of the confocal micrographs. In these assays, the wild-type enhancer fragment drives gene expression in a domain faithful to a full reporter gene for RCO expression when positioned 5′ to a minimal 35S promoter (Fig. 4 A and C), as reported previously (19). In addition to imaging individual transgenic lines, we also estimated the average expression of each genotype to account for variation between independent transgenic lines (Fig. 4 C’–F’ and SI Appendix, Fig. S8). Overall, we observed expression patterns consistent with the corresponding CRISPR/Cas9 alleles (Fig. 4 C–F). Specifically, constructs carrying the mut5 deletion or the hybrid mut14 enhancer resulted in increased and broadened gene expression in distal parts of the leaf and in stipules where the wild-type RCOenh500 does not drive gene expression (Fig. 4 A, D, and F). Conversely, the mut4 deletion construct resulted in reduced expression relative to the wild-type RCOenh500 expression (Fig. 4 A and E). Thus, we conclude that Rep64 harbors negative regulatory elements, while Act69 contains elements for positive regulation (SI Appendix, Fig. S9 and Table S11). Each of these sequences acts as a distinct module to define the spatial pattern of RCO expression. Notably, the Rep64 deletion in mut5 removes half of a conserved 50-bp sequence duplication which distinguishes RCOenh500 orthologs from their LMI1enh500 counterparts (SI Appendix, Fig. S9 and Dataset S5). By replacing the duplicated sequence within RCOenh500 with its counterpart sequence from LMI1 enh500, which lacks the duplication, we observed ectopic gene expression in distal parts of the leaf and in stipules (SI Appendix, Fig. S10A). Therefore, this duplication within RCOenh500 contributes to delimiting the RCO expression domain via negative regulation. Finally, we created a construct that mirrors the CRISPR RCOenh500 enhancer deletion by deleting the RCOenh500 sequence in the context of the full RCO reporter gene (SI Appendix, Fig. S10B). We observed that this deletion almost completely abolishes GFP expression (SI Appendix, Fig. S10B). This does not match the reduced, rather than absent, RCO expression observed in the RCOenh500 enhancer deletion allele (Fig. 1E). This finding indicates that sequences acting redundantly with RCOenh500 are unlikely to reside in the 3.2 kb sequence used in the RCO reporter gene, or alternatively, if they reside there, they can only function in the native genomic context. In summary, our findings highlight the importance of combining CRISPR/Cas9 mutagenesis and reporter gene assays to obtain fine-grained information on enhancer function during development and evolution. Conclusions By comparing the genetic profile of mutations in an ancestral and derived enhancer, we provide insights into the genetic logic underlying enhancer sequence evolution, gene expression changes, and morphological diversity. The derived RCO enhancer has two main differences relative to its ancestral LMI1 counterpart. First, it is partially rather than strictly required for gene expression, resulting in RCO expression being better buffered to enh500 mutations. This feature may have helped to stabilize the novel RCO expression domain following gene duplication from LMI1, similar to previous findings (24). Second, and more importantly, RCOenh500 is subject to more negative regulation than LMI1enh500. When this negative regulation is lost, for example, in mut5 and mut14 deletion alleles, RCO is expressed in more distal parts of the leaf and in stipules, thus resembling the ancestral gene expression pattern of LMI1. We propose that this negative regulation within RCOenh500 was a pivotal mechanism to evolve the restricted RCO expression domain at the leaf base, which underpinned the evolution of complex leaf forms in crucifers (17, 20). Our approach allowed us to link genotypes generated by genome editing to morphological phenotypes by considering corresponding changes in gene expression—a connection often missing in the field. For some alleles, the correspondence between phenotypic, qRT-PCR, and reporter gene data was incomplete. Several reasons could account for this, including the inability of qRT-PCR to capture cell- and stage-specific gene expression information, position effects in transgenic reporter gene assays, and the influence of regulatory sequences not included in reporter genes (for details, see SI Appendix, Supplemental Text 1). The development of efficient transgenic knock-in strategies in plants for allele-specific expression analysis in the native genomic context using fluorescent tags will be an important next step toward addressing these issues. Sequence duplications at two levels likely contributed to the evolution of leaf complexity by RCO. First, duplication of the ancestral LMI1 gene including its regulatory sequences created the potential to evolve the new RCO gene expression domain. Second, a smaller duplication of a specific sequence within RCOenh500 accompanied and may have facilitated, the evolution of negative regulatory elements which helped delimit the novel RCO expression domain. Identifying specific transcription factors that bind sites (TFBS) in Rep64 and mediate these repressive effects, together with determining the evolutionary history of these TFBS sequences, will be an important follow-up to this study. In addition, the evolution of repressive sequences within the RCOenh500 may have limited the potential pleiotropic effects caused by novel expression of RCO, a potent growth repressor, since rcoenh alleles where these sequences are deleted (mut2, mut5, mut6, mut14) show reduced leaf size and altered leaf shape. Thus, our results demonstrate the importance of repressive sequences in shaping the gene expression output of enhancers and curtailing potentially pleiotropic effects of developmental genes during morphological evolution. Materials and Methods Full description of the materials used and protocol-level methods are shown in the SI Appendix, including: Experimental model and plant growth conditions; Cloning and generation of transgenic plants and mutants; Genotyping; Quantitative RT PCR; Sample preparation and imaging; Image registration and analysis; In situ hybridization assays; Leaf phenotype analysis; Bioinformatics, quantification; and statistical analyses. Supplementary Material Appendix 01 (PDF) Dataset S01 (TXT) Dataset S02 (TXT) Dataset S03 (TXT) Dataset S04 (TXT) Dataset S05 (TXT)