A screen for regeneration-associated silencer regulatory elements in zebrafish

SUMMARY

Regeneration involves gene expression changes explained in part by context-dependent recruitment of transcriptional activators to distal enhancers. Silencers that engage repressive transcriptional complexes are less studied than enhancers and more technically challenging to validate, but they potentially have profound biological importance for regeneration. Here, we identified candidate silencers from a screen examining abilities of DNA sequences to limit injury-induced gene expression in larval zebrafish after fin amputation. A short sequence (s1) on chromosome 5 near several genes that reduce expression during regeneration could suppress promoter activity in stable transgenic lines and diminish nearby gene expression in knock-in lines. High-resolution analysis of chromatin organization identified physical associations of s1 with gene promoters occurring preferentially during regeneration, and genomic deletion of s1 elevated expression of these genes after fin amputation. Our study provides methods to identify ‘tissue regeneration silencer elements’ (TRSEs) with potential to reduce unnecessary or deleterious gene expression during regeneration.

Graphical Abstract

eTOC:

Ando et al. report a screen in zebrafish to identify sequences that repress gene expression during tissue regeneration and find an element with silencing activity that physically contacts nearby genes and tempers their RNA levels during regeneration. The investigation of silencers can illuminate how regeneration programs are controlled in animals.

INTRODUCTION

Teleosts like zebrafish and urodele salamanders share the ability to regenerate major appendages like limbs and fins after amputation. A prevailing view is that regenerative capacity is not a result of these species retaining genes dedicated purely to appendage regeneration that mammals do not possess. Instead, what is likely to differ is how key programs such as inflammation, fibrosis, and morphogenesis are regulated in stump tissue upon injury.¹

Genome-wide profiling experiments in regeneration contexts have indicated that hundreds to thousands of genes change expression from the uninjured state (reviewed in ¹), with the majority of studies focusing on gene products that increase levels during regeneration. Investigations of amphibian limb regeneration have centered on induced factors in the wound epidermis or blastema that can stimulate cell division or pattern tissue. Similarly, genetic approaches to fin regeneration in zebrafish have largely uncovered functions for genes that increase expression after amputation. A number of recent studies indicate that these and other regenerative events are enabled in part by enhancer elements with properties that can direct gene expression in the presence of injury, maintain this expression for the duration of regeneration, and temper it as regeneration concludes, originally named tissue regeneration enhancer elements (TREEs) or damage-responsive elements.^2-4 TREEs have been experimentally characterized by transgenesis,^2-10 or inferred through epigenetic profiles,^11-15 in many species and regeneration contexts. Genetic deletion of TREEs can negligibly, partially, or completely abolish expression of a linked gene during regeneration, with undetectable to robust effects on regenerative responses.^2-5,9,16-18 Understanding the biology of TREEs can help elucidate upstream regulators of regeneration programs,¹⁵ suggest evolutionary strategies,^3,7,19 and inspire applications to modulate regeneration.²⁰Reductions in gene mRNA levels also take place during regeneration but have received comparatively little attention. Dampened gene expression could conceivably occur through a variety of mechanisms in uninjured and regenerating tissue, including the loss of a transcriptional activator complex at an enhancer, or new presence of repressor complexes recruited to silencer elements. Among the first silencers to be reported was a cis-acting sequence able to repress non-mating-type promoters in yeast,²¹ and subsequent studies have identified silencing elements in many other species, tissues, and settings.²² In some cases, repressive regulatory elements are bifunctional and can act as enhancers in other contexts.^23-28

Design

A limited number of studies of tissue regeneration have implicated sequences with silencing activity. These include sequences that suppress transcription of wingless in uninjured fly imaginal discs,¹⁰ and sequences within an enhancer that help restrict expression of transgenic reporters to endocardial cells after cardiac injury in zebrafish.²⁹ Whether silencer elements play broad roles in control of tissue regeneration is unexplored, a deficiency due at least in part to challenges in identifying and validating silencers as compared to enhancers.

Here, we designed a screening strategy to identify distal gene regulatory elements with the capacity to reduce regeneration-associated expression in amputated zebrafish fins. We identified candidate silencer sequences from published profiles of transcriptomes and accessible chromatin regions acquired in studies of zebrafish fin regeneration. We also took advantage of the impressive screening capacity of larval zebrafish to test the ability of these candidate sequences to silence injury-responsive gene expression. We found multiple sequences with silencing ability, one of which we named s1 and was located in an area with several nearby genes that reduce expression levels during fin regeneration. A panel of molecular genetic assays to manipulate and assess s1 indicated that it is both necessary and sufficient for silencing activity, and that it reduces expression of a linked set of genes on chromosome 5 during regeneration through regeneration-responsive DNA interactions. Our findings describe methodology for identifying silencer elements active in regenerating tissue.

RESULTS

Identification of candidate silencing elements associated with zebrafish fin regeneration

cis-regulatory sequences, including enhancers and silencers, are often associated with accessible chromatin structure. We predicted that silencing elements associated with regenerative events exist near gene targets that reduce RNA expression levels during regeneration (Figure 1A). To find potential candidate sequences, we assessed published profiles of gene expression (RNA-seq) and accessible chromatin (ATAC-seq) datasets using bulk tissue, purified fibroblasts, or purified osteoblasts from uninjured and regenerating zebrafish caudal fins (Figure 1B).^5,8 We looked for genes that reduce expression and have a distal sequence in the vicinity that increases in chromatin accessibility during regeneration. This strategy is based on the assumption that a class of silencing elements would acquire or increase presence of a repressive transcriptional complex during regeneration, further increasing chromatin accessibility from that in the uninjured tissue (although other possible mechanisms exist, see Discussion). We identified 653 regions with increased accessibility (log₂ fold change > 0.5) at 4 days post amputation (dpa) in whole fin tissue and linked to genes that reduce RNA levels 3-fold or more, compared to uninjured fins (0 dpa) (Figures 1C and 1D; Table S1A).⁵ Most sequences displaying increased accessibility during regeneration were in intergenic regions (59.4%), 3 kb upstream of transcriptional start sites (13.2%), or within introns (17.8%). The remaining regions were located in UTRs, exons, or immediate downstream regions of genes (12.6%) (Figure 1E). Examples of these regions include distal sequences of distinctly increasing accessibility during regeneration (marked by asterisks in Figures 1F and 1G) linked to nos2a and cyp24a1, genes that sharply reduce RNA levels during regeneration. A total of 1,159 differential ATAC-seq peaks were detected at 1 dpa (Figures S1A and S1D; Table S1B).⁵ Comparisons of chromatin profiles from purified fibroblasts indicated 878 regions at 4 dpa compared to 0 dpa, and those from purified osteoblasts indicated 950 regions (Figures S1B, S1C, S1E, and S1F).^5,8

Figure 1. Identification of candidate gene silencing elements.

(A) Cartoon showing hypothesized silencer activity during tissue regeneration.

(B) Transcriptomics and accessible chromatin profiles for identifying regions with increasing chromatin accessibility near downregulated genes (candidate silencers). After fin amputation, a wound epidermis forms at ~1 dpa, and a blastema forms shortly thereafter. At 4 dpa, blastemal cells containing fibroblasts and osteoblasts continue to proliferate.

(C) Dot plot of differential ATAC-seq peaks near differential transcripts from published whole fin 0 vs 4 dpa RNA-seq and ATAC-seq datasets. ⁵ Only the most dynamic ATAC-seq peak per gene that is located 3 kb or more away from the transcription start site (TSS) was plotted. 653 candidate silencers are indicated in the upper left quadrant labeled with orange color.

(D) Average tracks (top) of read counts per million reads (CPM) of ATAC-seq signals of the identified candidate silencers from the upper left quadrant in (C) and their heatmaps indicating regions of increased accessibility within 2 kb (bottom). Average signals increased ~3-fold in 4 dpa whole fin tissues compared to 0 dpa.

(E) Genomic profile of candidate silencers for whole fin 0 vs 4 dpa data set. Over half are located 3 kb away from TSS of nearby genes.

(F and G) Browser tracks indicating distal regions of increasing accessibility during regeneration (marked by asterisks) linked to nos2a (F) and cyp24a1 (G), genes that sharply reduce RNA levels in 4 dpa fins compared to 0 dpa.

(H and I) ChromVAR analysis predicts transcription factor binding sites by assessing all (H) or intergenic (I) candidate silencers in whole fin 0 vs 4 dpa ATAC-seq and RNA-seq datasets. INSM1, FOXO6, TCF7l1 and SIX1 (red) were identified as enriched motifs preferential to candidate silencers. See Methods for details on ChromVAR analysis.

ChromVAR analysis indicated enriched transcription factor (TF) motifs within open chromatin regions near downregulated genes (candidate silencers) from 4 dpa ATAC-seq datasets of whole fin tissue (Figure 1H). Among the top hits of regions with increased accessibility are predicted binding motifs of the AP-1 complex subunits, such as Batf (Basic leucine zipper transcription factor, ATF-like), Jun, and Fos, known to promote cell proliferation and survival. Epas1 (Endothelial PAS domain protein 1b), also known as Hif-2 (Hypoxia-inducible factor-2), was also among the top hits. Binding motifs for these factors are also enriched in the analysis of all increasing ATAC-seq peaks including candidate enhancers,⁵ suggesting they are not specific for regeneration-related gene silencing. Relatively enriched transcription factor motifs include those for Insulinoma-associated 1 (Insm1), a conserved zinc-finger transcription factor that was reported to suppress a genetic program for progenitor cell proliferation during zebrafish retinal regeneration,³⁰ and Forkhead box subclass O 6 (FoxO6), a conserved transcription factor that regulates various cellular processes like memory consolidation in the hippocampus, gluconeogenesis in the liver, and skull growth via Hippo signaling.^31-34 To exclude common motifs in all regulatory elements and focus on motifs enriched in distal silencers, we repeated chromVAR analysis on the candidate silencers that are located in intergenic regions or 3 kb up/downstream of downregulated genes (Figures 1E and S1H-S1J). This filtering revealed enriched predicted binding motifs including those of TCF7l1, a member of the Tcf/Lef family of transcription factors that are known to be involved in fin regeneration,^35-38 in candidate silencers from 4 dpa bulk tissue; SIX homeobox 1 (SIX1), necessary for fast-twitch muscle specification,³⁹ in candidate silencers from 4 dpa bulk tissue; Zic family member 1 (ZIC1), necessary to promote neural crest fate,⁴⁰ in candidate silencers from 1 dpa bulk tissue; and Activating transcription factor 3 (ATF3), which is induced in cardiac fibroblasts in response to hypertensive stimuli, in candidate silencers from 4 dpa fibroblasts (highlighted in red, Figures 1I, S1K and S1L; Tables S1C-S1E).⁴¹ Enriched motifs in candidate silencers from 4 dpa osteoblasts include a predicted motif of LIM homeobox transcription factor 1, beta (LMX1B) which is an anti-osteogenic factor and is required for bone formation during mouse digit tip regeneration (highlighted in red, Figure S1M; Table S1F).⁴²

Enriched Gene Ontology terms associated with putative target genes near candidate regeneration-associated silencers in these datasets include circadian rhythm, regulation of cell differentiation and response to hormone from 1 and 4 dpa bulk tissue datasets (Figures S2A-S2C); hormone-mediated signaling pathway, circadian rhythm, and brain development in comparison of 4 dpa bulk tissue and fibroblast data (Figures S2D-S2F); and response to hormone and cellular response to endogenous stimulus from 4 dpa bulk tissue and osteoblast datasets (Figures S2G and S2H).

Larval screen for silencing of injury-responsive gene expression

We predicted that bona fide contextual silencing elements identified from these analyses would possess the capacity on their own to suppress gene expression during regeneration. Therefore, we designed a screen to test if candidate sequences can suppress a promoter that itself can direct injury-activated expression. We prioritized sequences to use in this screen based on the extent of reduction in RNA levels of the putative target gene, the dynamism and size of the associated ATAC-signal, and the nature of the gene. These prioritized regions are each located within a 200 kb region surrounding the potential target gene.

Several years ago, we reported that the 2 kb (P2) upstream region of lepb acts as a minimal promoter when paired with an enhancer, but also can direct reporter protein fluorescence upon amputation of the larval fin fold, either in stable lines or F0 larvae (Figures 2A and 2B). P2 on its own does not direct expression of a reporter gene in regenerating adult zebrafish fins.³ To identify sequences that silence regeneration-associated expression from P2 in larvae, we subcloned 133 of the candidate elements upstream of a P2:EGFP construct and injected each construct separately into 50-100 one-cell embryos (Figures 2C and 2D). Genome integration of DNA constructs often occurs after the first cleavage in zebrafish embryos, resulting in mosaicism. Then, after fin fold amputation at 3 days post fertilization (dpf), we examined induced EGFP fluorescence in 4 dpf F₀ mosaic larvae as an output of the ability of a candidate sequence to silence injury-induced expression from the P2 promoter. We found that, typically, >50% of animals injected at the one-cell stage with a control P2:EGFP construct displayed readily detectable presence of injury-induced fluorescence (Figures 2E, S3A and S3B; Table S2A).

Figure 2. Larval screen for the ability to silence an injury-responsive promoter during fin fold regeneration.

(A, B) The 2 kb lepb minimal promoter (P2) directs expression of an EGFP cassette in amputated larval fin folds but not adult fins. EGFP expression is readily assessed in regenerating fin folds of injected F₀ larvae. Two examples shown in lower panels of (B) were considered EGFP⁺ responses.

(C) Scheme for silencer screen. EGFP fluorescence after fin fold amputation at 3 dpf is typically observed in ~50% or more of 4 dpf F₀ larvae injected with P2:EGFP DNA. Effective candidate silencer sequences subcloned upstream of this construct (right) are expected to measurably reduce this percentage.

(D) Flow chart for silencer screen. 3640 candidate silencers (top distal candidate silencer per gene) were identified from all four datasets of whole fin tissue 0 vs 4 dpa, whole fin tissue 0 vs 1 dpa, fin fibroblast 0 vs 4 dpa, and fin osteoblast 0 vs 4 dpa. 133 candidate silencers were manually selected based on the extent of reduction in RNA levels, the dynamism and size of the nearby ATAC-signal, and the nature of the gene. Four injection trials revealed 3 silencers with ratios of EGFP⁺ larvae consistently at 25% or lower.

(E, F) Ratios of EGFP⁺ individuals in control (E, n = 12) and all tested candidate silencer-P2:EGFP-injected embryos (F) after the 1^st screening are indicated with blue bars. EGFP⁺ cells were observed in 25% or fewer larvae in cases of 16 silencer candidates indicated by the gene closest to the candidate sequences (blue bars) among 133 examined sequences. Ratios of EGFP⁺ individuals corresponding to 6 candidate silencers were consistently <25% after a second screening (green bars). s1 and s2 exist within a 100 kb region, and the more dynamic element s1 is indicated with a red asterisk.

(G) Ratios of EGFP⁺ individuals in the 6 silencer construct-injected embryos after the 3^rd and 4^th screening are indicated with orange and purple bars, respectively. In cases of s1-s3, EGFP⁺ cells were observed in 25% or fewer larvae consistently.

(H, I) The runx1 enhancer (REN) and cfos minimal promoter direct expression in regenerating larval hearts after partial cardiac ablation by nitroreductase (NTR) under the control of cmlc2 promoter and prodrug (metronidazole, MTZ) which is converted by NTR into a cytotoxin. Cardiac EGFP expression is assessed in injected F₀ larvae at 5 dpf after MTZ treatment from 3 to 4 dpf and is typically observed in ~50% of 5 dpf F₀ larvae injected with REN-cfos:EGFP control DNA (red arrow in lower image).

(J) Scheme for cardiac silencer candidates. EGFP fluorescence after MTZ treatment and cardiomyocyte ablation is typically observed in ~50% or more of 5 dpf F₀ larvae injected with REN-cfos:EGFP DNA. Effective candidate silencer sequences (right) are expected to reduce this percentage.

(K, L) Ratios of control (K, n = 5) and all tested candidate silencer-REN-cfos:EGFP constructs (L). The 1^st screening is indicated with blue bars. EGFP⁺ cells were observed in 25% or fewer larvae in cases of 13 silencer candidates, as labeled by the gene closest to the candidate sequence among 35 examined sequences. Ratios of EGFP⁺ individuals corresponding to two candidate silencers (s4 and s5) were consistently lower than 25% after a second screening (green bars, indicated with a red asterisk).

(M) Flow chart for silencer screen. 609 candidate silencers were identified from two heart regeneration datasets assessing epicardium (uninjured vs 3 dpa and uninjured vs 7 dpa), or cardiomyocytes (uninjured vs 7 dpi and uninjured vs 14 dpi). 35 candidate silencers were manually selected based on the extent of reduction in RNA levels, the dynamism and size of the nearby ATAC-signal, and the nature of the gene. 1^st and 2^nd larval screens identified two silencers (s4 and s5) for which ratios of EGFP⁺ larvae were constantly 25% or lower.

From the larval screen, 16 of 133 constructs containing a different candidate silencer resulted in 25% or fewer larvae with detectable regeneration-associated fluorescence (Figure 2F; Table S2A). Interestingly, injection of some candidates caused higher injury-induced EGFP intensity or greater ratios of EGFP-positive larvae than the control P2:EGFP construct, suggesting that they may possess some level of enhancer activity (blue bars in Figure 2F). One can speculate that these sequences might act as enhancers while in their endogenous genome location with respect to a nearby target gene, but that this activity could also have an effect of inhibiting expression of a different proximal gene. We repeated the larval screen using the 16 filtered candidate sequences to assess reproducibility, finding that 6 of 16 candidates limited regeneration-associated EGFP expression (green bars in Figure 2F) (Figure 2D). Among these with strong silencing activity are a region located 91 kb downstream of the gene smarca1 and 190 kb upstream of the transcription start site (TSS) of tenm1 (smarca1+91k) (Figure S3C), and a region located 3 kb upstream of the TSS of smarca1 (smarca1-3k) (Figure S3D). Others are linked to the genes pdcd4b (pdcd4b-93k), bnip3 (bnip3-2k), ucmaa (ucmaa+9k), and olfml2ba (olfml2ba-1k) (Figure S3E). Importantly, two further repeats of injections with constructs containing the 6 elements revealed that, of these, only smarca1+91k, smarca1-3k, and pdcd4b-93k, now referred to as silencer 1 - silencer 3 (s1-s3), consistently behaved as silencing elements (Figures 2D and 2G). This result punctuates experimental variation inherent in this type of mosaic screen.

To test if silencing candidates can be similarly identified from other tissues, we attempted to establish a similar pipeline for heart regeneration. We employed a dataset describing chromatin regulation in cardiomyocytes during zebrafish heart regeneration, ⁴ from which we could find hundreds of candidate regeneration-associated silencer elements (Tables S1G and S1H). By asking these elements to silence expression from a runx1-linked enhancer and promoter construct that is responsive to a transgenic cardiac injury system in larvae (Figures 2H-2K), we found two candidate sequences with silencing activity, referred to as silencer 4 and silencer 5 (s4 and s5) (Figures 2L, 2M, S3F, and S3G; Table S2B). Many tissues in larval zebrafish show strong regenerative responses,^43,44 and thus we expect the high-throughput and visual strengths of zebrafish can enable screens for gene regulatory silencers in most regenerating tissues.

We were intrigued that two candidates, s1 and s2, only 94 kb away from each other, displayed the strongest silencing ability in larval screens. Moreover, we were able to detect 8 genes in an ~827 kb region containing s1 and s2 that displayed lower RNA levels during regeneration than in uninjured fins (Figure S3H): arhgap25, bmp10, si:dkey-13n15.11, sh2d1aa, tenm1, smarca1, GAB3, and arhgef9a. Therefore, we focused our further experiments on this region, and on the more dynamic element of the two, s1.

The s1 candidate silencer represses gene expression in stable transgenes and at endogenous loci in knock-in mutants

To test whether s1 is sufficient to silence gene expression from promoters in stable lines, we established transgenic lines with or without an 850 bp region containing s1 subcloned upstream of a constitutive ubiquitin promoter (ubi:Brainbow),⁴⁵ a tryptophan hydroxylase 1b (tph1b) promoter that directs expression in regenerating fin fibroblasts (tph1b:mCherry),⁴⁶ or the P2:EGFP containing the LEN TREE linked to the lepb gene (LENP2:EGFP), which directs reporter expression in regenerating fin or heart at larval and adult stages.³ We generated at least 2 stable lines for each construct, and in all cases we observed visually lower or undetectable expression in fin regenerates compared to control transgenic lines (Figures 3A and 3B). Because expression from the ubiquitin promoter was silenced by s1 in larval and uninjured adult contexts in addition to during fin regeneration, we infer that s1 can have broad silencing ability when placed adjacent to a promoter inserted randomly into the genome and with which it might not normally interact.

Figure 3. s1 represses expression from transgenes or endogenous genes.

(A and B) Transgene constructs to test influence of s1 on a ubiquitous promoter (ubiquitin), injury-induced promoter (tph1b), or TREE plus minimal promoter (LENP2:EGFP). In each case, and in multiple stable lines, constructs containing the s1 silencer exhibit reduced expression of transgenic reporter genes (lower panels in B).

(C) s1 was inserted by non-homologous end joining into a location 22 kb upstream of kita, a gene required for normal stripe pigmentation.

(D) qPCR analysis of kita RNA levels in wild-type and s1 knock-in mutant kita^s1/s1 fins. Three fins each were pooled, and three biological replicates were performed. Paired two-tailed Student t-tests were performed to indicate significance. Paired two-tailed Student t-tests, n = 3. Data are mean ± S.D.

(E) Homozygous s1 insertions near the kita locus (middle) reduced the number of melanocytes to near those levels in homozygous kita null mutants (right).

(F) Quantification of experiments in (E). Pigment cells in a 1 mm² trunk area were counted in three WT, kita^s1/s1, and kita^b5/b5 fish each. Paired two-tailed Student t-tests, n = 3. Data are mean ± S.D.

(G) Knock-in of s1 sequence 1.5 kb upstream of fgf20a, a gene required for blastema formation.

(H) fgf20a expression measured by qPCR is reduced to varying extents in fgf20a s1 knock-in fins compared to wild-types at 1 and 7 dpa. Three fins each were pooled, and three biological replicates were performed. Paired two-tailed Student t-tests, n = 3. Data are mean ± S.D.

(I) Knock-in of s1 sequence 83 kb downstream of the transcriptional start site of lamb1a, a gene required for normal formation of the regeneration epidermis.

(J) lamb1a expression measured by qPCR is reduced to varying extents in lamb1a s1 knock-in fins compared to wild-types at 1 and 7 dpa. Three fins each were pooled, and three biological replicates were performed. Paired two-tailed Student t-tests, n = 3. Data are mean ± S.D.

(K and L) Fin regeneration at 33°C (K) or at 28.5°C (L) was assessed longitudinally for each animal, and clutchmate controls were used. A ratio of the lengths of the original structures that were removed was calculated for each animal at each timepoint (see Methods). Normalized fin lengths indicated slight differences in knock-ins compared to wild-types (see Figures S4A and S4B). Wald’s tests were used for statistical comparisons of functional indicators among groups from 2 to 21 or 28 dpa, preserving individual animal performance over time and yielding a single p value. (n = 10). Data are mean ± S.D.

Transgene constructs are artificial in several ways, and they do not conclusively test the capacity to silence an endogenous gene. An arguably more rigorous test for silencer activity inserts the sequence near the promoter of a gene at its endogenous location in the genome. We first used CRISPR-Cas9 techniques to insert s1 22 kb upstream of the TSS of kita, a gene required for normal melanocyte development and animal pigmentation (Figure 3C).⁴⁷ This insertion site and those for genes below were chosen to avoid disturbing promoters, including those of non-target genes, or other predicted regulatory elements, and in considering predicted gRNA efficiencies. Quantitative PCR (qPCR) using adult fins of homozygous s1 knock-in (KI) mutant fish (kita-s1KI^pd384, referred to as kita^s1/s1) revealed that endogenous kita RNA levels were <50% those of wild-types (Figure 3D). Moreover, s1 visibly altered pigmentation in body stripes of adult kita^s1/s1 zebrafish, giving an appearance of the stripes that was similar to those of homozygous kita^b5 (null) mutants,⁴⁷ and densities of pigment cells were decreased in kita^s1/s1 mutants to levels of null mutants (Figures 3E and 3F). To test effects of s1 knock-in near genes known to be induced during fin regeneration, we used gene editing to insert s1 upstream of the genes fgf20a or downstream of lamb1a. Each of these genes has been identified in forward genetic screens for mutations that cause heritable defects in fin regeneration.^48-50 Insertion of s1 1.5 kb upstream of the fgf20a TSS or 83 kb downstream of the lamb1a TSS demonstrably reduced the extent of gene induction in regenerating fins of homozygous s1 insertion animals (fgf20a-siKI^pd385 and lamb1a-s1KI^pd0386, referred to as fgf20a^s1/s1 and lamb1a^s1/s1) to varying extents based on the time after amputation (Figures 3G-3J). We also carefully measured fin regeneration in s1 insertion mutant animals, following individual animals longitudinally from 2 to 21 or 28 dpa and measuring regeneration as a ratio of the amputated portion. We found slight, measurable differences in the lengths of fgf20a^s1/s1 regenerates or lamb1a^s1/s1 rays when carried out at 28.5°C or 33°C, but these differences were not to the extent we considered biologically significant (Figures 3K, 3L, S4A and S4B). Silencer knock-in experiments are likely to be heavily influenced by location of the inserted silencer. Our experiments indicate that s1 knock-ins can influence gene expression and the functions of linked genes, to varying extents.

We assessed conservation of s1 in vertebrates at primary sequence and functional levels. We found sequence conservation of s1 within the carp genome, but not within fugu or coelacanth genomes, nor within amphibian and mammalian genomes we examined (Figure S4C), suggesting rapid change during species evolution. To test if zebrafish s1 sequences are functionally recognized by the transcriptional machinery of mammals, we transfected HEK293T cells with P2:EGFP, LENP2:EGFP, or REN-cfos:EGFP plasmids. We found that inclusion of s1 in these constructs was sufficient to reduce the number of cells expressing EGFP (Figures S4D-S4K). These results indicate that s1 sequences of zebrafish origin recruit a similar or overlapping set of mammalian proteins as in their typical species context. They also suggest it is possible that silencing sequences identified from regenerating zebrafish tissues may have similar applications to control gene expression in tissue regeneration strategies as recently shown for TREEs.^20,56

We assessed published chromatin signature profiles from studies of fin regeneration, finding that the s1 region is less methylated and does not acquire measurable histone modifications in multiple organs and uninjured and regenerating fins (Figures S4L-S4N),^8,51,52 suggesting that s1 is continuously occupied to some extent. To identify s1 subregions necessary to repress expression, we first assessed chromVAR analysis of those sequences with increasing accessibility in the vicinity of genes that reduce levels during fin regeneration (4 dpa bulk whole fin tissues). This analysis indicated multiple enriched predicted motifs (Figure 1I), several of which are contained in s1 and associated with repressor binding from other studies, including motifs for Bhlhe40, DLX4, ESRRG, Klf4, Nr5a2, ZIC4 and CTCFL (Figures 4A and 4B). To identify which motif-containing regions of s1 are most essential, we then employed a stable line of s1-ubi:Brainbow zebrafish to perform tiling deletion assays using a panel of gRNA pairs. We expected that the broad expression normally directed by ubi would make it easiest to detect derepression in mosaic F₀ animals, which we assessed as adults (Figure 4C). From injections of 7 pairs, we identified a 91-bp region that is essential for s1 silencer activity and contains many predicted transcription factor binding motifs (Figures 4D and 4E). Inspection of ATAC-seq signals in these 91 nucleotides indicated a subregion of signal avoidance in samples from regenerating versus uninjured fins, suggestive of increased factor binding (ATAC-seq tracks in Figure 4F). To infer which transcription factors bind in context, we performed Transcription factor Occupancy prediction by Investigation of ATAC-seq Signal (TOBIAS) to provide a footprinting analysis for transcription factor binding to our candidate silencers (lollipops in Figure 4F, Binding score >= 10) during regeneration. We found that the factors REST and Nr5a2 were the only factors with predicted sites in the 91 bp essential region of s1 that showed significantly increased binding during regeneration (Figures 4G-4I; Table S3).⁵³ In total, our transgenic and knock-in assays indicate that the s1 sequence, which includes a short 91-bp critical region, can silence gene expression in multiple genomic locations, both during regeneration and also in uninjured tissues.

Figure 4. Identification of a minimal region within s1 required for silencing.

(A, B) Predicted transcription factor binding motifs in an 850 bp region of s1 that may bind repressive complexes, based on ChromVAR analysis of candidate silencers. See Methods for details on ChromVAR analysis.

(C) Eight gRNAs were designed to target sequences in s1, and 7 different pairs were injected into one-cell s1-ubi:Brainbow embryos, with the goal to identify essential silencing sequences in adult animals.

(D) Images of examples of adult fish from experiments in (C). Those animals injected with guides 5 and 6 at the one-cell stage, deleted a 91 bp region, consistently displayed de-repressed fluorescent protein expression. We counted fish which has several body parts with higher expression (white arrow heads) as a de-repressed fish indicated with (+). The other fish indicated with (neg) were not distinguishable from uninjected control s1-ubi:Brainbow animals.

(E) Quantification of ratios of animals indicating de-repressed gene expression from experiments in (C, D).

(F) (Top) Transcription factors with a binding score >10 predicted by ChromVAR located within the critical 91 bp region of s1 that possesses silencing ability. (Bottom) Tracks of read counts per million reads (CPM) in all ATAC-seq signals from whole fin tissues show some avoidance at the 91-bp region during regeneration (1 dpa and 4 dpa) compared to uninjured fins (0 dpa), suggesting binding of transcription factor(s). See Methods for details on ChromVAR analysis.

(G) TOBIAS footprinting volcano plot for the global changes of merged ATAC-seq signals of whole fin 0 vs 4 dpa data and HiCAR 0 vs 4 dpa data. Among the enriched binding factors (F), REST and Nr5a2 were also identified as significantly differential binding factors during regeneration by footprinting.

(H, I) Differential footprinting of REST and Nr5a2 transcription factors. In both cases, their observed distributions and means (4 dpa) shifted to higher binding score (log2 fold change) compared to the background distributions and means (0 dpa), suggesting that REST and Nr5a2 are significantly differential binding factors at 4 dpa.

s1 makes context-dependent gene contacts during regeneration

While experiments to this point indicated that s1 is sufficient to repress gene expression when inserted ectopically in the genome, they do not address how s1 behaves in its endogenous location via its typical neighboring regulatory sequences and gene target(s). It has been established that cis-regulatory elements control distal gene expression through dynamic changes in chromatin looping. To assess contacts that s1 may have with other regions of its chromosome during regeneration, we performed HiCAR, a genome-wide analysis that detects chromatin contacts with accessible DNA regions like regulatory elements from small amounts of tissue.⁵⁴ To our knowledge, chromosome confirmation capture (3C)-based approaches had not been applied in models of complex tissue regeneration. To identify interactions with cis-regulatory elements, we focused only on intra-chromosomal looping, finding 23,668, 19,141 and 22,865 putative chromatin loops from chromatin samples of 0, 1, and 4 dpa regenerating fin tissues, respectively (Figures 5A-5C). Comparisons of the different timepoints indicated 16,600 interactions detected at 0, 1, and 4 dpa, and 5,022 regeneration-responsive interactions only detected in 1 and/or 4 dpa regenerating fins (Figure 5D), suggesting that ~20% of all cis regulatory element-associated chromatin interactions in fins occur in response to injury. In addition, identified HiCAR loops indicated a relationship between differential ATAC-seq peaks and differentially expressed distal genes (not nearest genes), defining 718 and 1465 candidate regeneration-associated silencers, and 850 and 1352 candidate TREEs, from whole fin 0 vs 4 and 0 vs 1 dpa datasets, respectively (Figures 5E and 5F; Tables S4A and S4B). Figure S5A-S5L show examples of a candidate silencer and candidate enhancer, linked to the genes nfil3-6 and vcana, respectively, with increasing ATAC-seq peaks and more frequent interactions with putative target genes at 1 and/or 4 dpa compared to 0 dpa. The candidate silencers tested in the F₀ larval fin screen have a range in HiCAR loop numbers and tend to maintain chromatin interactions from uninjured through regeneration, except in several cases including s1 (Figure S5M-S5O and Table S2A).

Figure 5. HiCAR indicates increased physical interactions of s1 with nearby gene loci during regeneration.

(A-D) HiCAR sequencing identified 23,668, 19,141 and 22,865 putative chromatin loops from chromatin samples of 0 (A), 1 (B), and 4 dpa (C) fin tissues. Comparison of the different timepoints indicating 16,600 interactions detected at 0, 1, and 4 dpa, and 5,022 interactions detected only in regenerating fins (D).

(E and F) Candidate regeneration-associated silencers and TREEs with increasing accessible chromatin regions near interacting domains of HiCAR differential loops, with the other end interacting with TSS of differentially expressed transcripts. Assessed from RNA-seq, ATAC-seq and HiCAR datasets of whole fin tissue 0 vs 4 dpa (E) and 0 vs 1 dpa (F). 718 and 1465 candidate silencers (orange boxes) and 850 and 1352 candidate enhancers (green boxes) were identified from the whole fin 0 vs 4 dataset and the 0 vs 1 dpa dataset, respectively.

(G and H) Chromatin interactions in the ~1 mb region including s1 from HiCAR of whole fin tissue at 0, 1, and 4 dpa. The signals in the 2D matrix indicate raw signals normalized by zero-truncated poisson distribution, regression coefficients, read counts, GC content, mappability score, genomic distance and the number of cuts (MAPS), showing that there are more long-range interactions (black-outlined squares) at 4 dpa compared to 0 and 1 dpa (G). They correspond to the 1D loops

(H). s1 is in one of the topologically associating domains (TADs, black dashed lines) and has more frequent chromatin loops at 4 dpa compared to 0 and 1 dpa, many that interact with the promoter region of tenm1 gene located in the same TAD (G).

(I and J) Chromatin accessibility in the ~1 mb region including s1 from ATAC-seq of whole fin 0 vs 1 dpa and 0 vs 4 dpa. s1 has increasing ATAC-seq peaks at 1 dpa (I) and 4 dpa (J) compared to each corresponding 0 dpa control.

(K and L) Transcription of genes in the ~1 mb region including s1 based on RNA-seq of whole fin 0 vs 1 dpa and 0 vs 4 dpa. tenm1 has reduced levels at 1 (K) and 4 dpa (L) compared to each corresponding 0 dpa control.

s1 silencer is indicated with green dotted line in G and hazel solid line in H-L. tenm1, a putative target gene of s1, is indicated with a gray boxed area.

Sequencing depth was sufficient for resolution of chromatin looping at 5 to 10 kb, and topologically associating domains (TADs, black dashed lines in Figure 5G) were readily detected in chromatin interaction matrices generated by these datasets. We focused on a 1 mb region surrounding s1, identifying many interactions that could be visualized in chromatin matrix heatmaps, including chromatin loops that emerged at 4 dpa and were not detectable from 0 or 1 dpa samples (Figures 5G, 5H and S5P). Importantly, the HiCAR loop caller MAPS normalized the chromatin contact matrix against accessibility; thus, regeneration-associated loops indicate more frequent chromatin interactions during regeneration (Figure 5H). The majority of 4 dpa loops in the s1 region indicated contact between s1 and the gene tenm1, a gene encoding a highly conserved type II transmembrane glycoprotein known to be widely expressed throughout the nervous system. Notably, tenm1 RNA levels drop during regeneration, measured by RNA-seq as reduced by 80% at 4 dpa (Figures 5K and 5L), a molecular change we inferred might be the result of contacts with s1, before gradually increasing at later timepoints (see Figure 6J). We considered based on observations to this point that s1 could act as an insulator, an element that mediates long-range chromatin looping and prevents interactions between regulatory elements and genes located in a different TAD. Indeed, published ChIP-seq analysis of CTCF binding sites revealed potential binding at the s1 locus in 24 hpf embryos.⁵⁵ However, our HiCAR analysis indicated that s1 is located within the same TAD as the genes it contacts and/or reduce levels during regeneration, indicating that it is not working as an insulator during fin regeneration (Figure 5G). Taken together, our data implicate the emergence of 3D chromosome conformational changes during regeneration that position s1 near one or more downregulated genes within its TAD more frequently than in uninjured fins.

Figure 6. Deletion of s1 changes expression of nearby genes.

(A) A ~5 kb region including the s1 silencer was deleted using genome editing.

(B, C) Regenerating fins from wild-type (WT), s1^+/Δ, and s1^Δ/Δ clutchmates were assessed longitudinally for each animal at 2-28 dpa. A ratio of the lengths of the original structures that were removed was calculated for each animal at each timepoint (see Methods). Normalized fin lengths indicated no differences in knockouts compared to WT. Wald’s tests compared groups from 2 to 28 dpi as in Figures 3K and 3L. (n = 10). Data are mean ± S.D.

(D-F) Volcano plots of RNA-seq data from WT and s1^Δ/Δ sibling fins comparing each timepoint of 0, 1, and 4 dpa. tenm1 (red) has the most significantly higher levels in s1^Δ/Δ siblings versus wild-type clutchmates, by p value. Other differentially expressed genes (fold change > 2, p-value < 0.00005) located within 3 mb from s1 on chromosome 5 (green; 14 genes at 0 dpa, 7 genes at 1 dpa, 5 at 4 dpa) anywhere on chromosome 5 (blue; 11 genes at 0 dpa, 2 genes at 1 dpa, 2 genes at 4 dpa), or on all other chromosomes are indicated (black; 32 genes at 0 dpa, 7 genes at 1 dpa, 6 genes at 4 dpa), suggesting that s1 presence preferentially impacts genes in its vicinity.

(G) Heatmap of relative expression of several genes in a 1.7 mb region including s1, from bulk RNA-seq of WT and s1^Δ/Δ sibling fins at 0, 1, and 4 dpa. Red-outlined values represent genes with a 10% or higher expression levels in mutant fins.

(H) Tracks of tenm1 in RNA-seq profiles of WT (blue) and s1^Δ/Δ sibling (red) fin regenerates at 0, 1, and 4 dpa. tenm1 expression is higher in mutants by ~2.5-fold at 0 dpa, ~4.3-fold at 1 dpa, and ~2.8-fold at 4 dpa.

(I) Heatmap of qPCR analysis of arhgap25, tenm1, smarca1 and asb12a transcripts in WT and s1^Δ/Δ fins at 0, 1, and 4 dpa. Three fins each were pooled, and three biological replicates were performed. Paired two-tailed Student t-tests, n = 3. Data are mean ± S.D. Red-outlined values represent genes with a 0.05 or lower p-value and that have higher levels in mutant fins.

(J) tenm1 expression measured by qPCR is reduced at 4 dpa in WT fins and returns gradually during regeneration. At 28 dpa, its levels near those of 0 dpa. In s1^Δ/Δ fins, the expression is increased at all timepoints. Three fins each were pooled, and three biological replicates were performed. Paired two-tailed Student t-tests, n = 3. Data are mean ± S.D.

Deletion of s1 de-represses local regeneration-associated gene expression

To determine whether s1 is required for regulation of neighboring genes during regeneration, we used gene editing to remove it from the genome (Figure 6A). Homozygous mutant animals appeared grossly normal through adulthood. We amputated fins and assessed several indicators of regeneration as compared to wild-type sibling controls. We found that fin regeneration was measurably normal in mutants at 28.5°C or at 33°C (Figures 6B, 6C, and S6A-S6D). Thus, s1 itself is not required for grossly normal fin regeneration.

To determine whether s1 is essential for normal regulation of nearby genes during regeneration, several of which reduce levels during regeneration, we assessed gene expression by bulk RNA sequencing and qPCR analysis. Comparisons between 0, 1, and 4 dpa regenerates of s1^Δ/Δ and wild-type siblings indicated similar transcriptomes. Yet, remarkably, several genes near the s1 locus or on the same chromosome (chr5) as s1 displayed higher levels in 0 dpa or regenerating fins in the s1 deletion background (Figures 6D-6G and S6E-S6G; Tables S5A-S5C). In fact, tenm1 showed a robust increase in s1^Δ/Δ regenerates with the most statistically significant changes from wild-type RNA levels in the entire transcriptome (~4-fold at 1 dpa and ~3-fold at 4 dpa, compared to wild-type siblings) (Figure 6H). qPCR identified increases in RNA levels of 4 s1-linked genes in s1^Δ/Δ regenerates, including increases in tenm1 levels at several regeneration timepoints (Figures 6I and 6J). We conclude that s1 is a silencer gene regulatory element that increases contacts with nearby genes including tenm1 during regeneration, and that it is required to suppress levels of tenm1 and other genes in the region during fin regeneration.

DISCUSSION

Tissue regeneration is a symphony of biological signal control. Hundreds to thousands of genes change levels in each of potentially dozens of cell types in response to injury. Similarly impressive additional changes occur during the regeneration period itself. The extent of these changes is influenced by proximity to the injury site and other factors, and the cumulative signature of changes might be unique for every individual cell involved in regeneration. These signatures ostensibly define how and to what extent a cell will participate in regeneration. For several years it has been accepted that regeneration-associated enhancers, or TREEs, provide instructions for context-preferred or -specific induction of gene expression during regeneration. Here, we show evidence that silencers with regeneration-associated, gene-repressing activity, which we now refer to as Tissue Regeneration Silencer Elements (TRSEs), can be identified by way of a screening strategy in larval zebrafish followed by a panel of molecular genetic experiments.

Our experiments indicate that, when located out of its normal context, the s1 sequence itself acts as a general silencer of adjacent promoters regardless of cell type, developmental stage, or injury state. Furthermore, s1 has some silencing effects on genes in its vicinity in the absence of injury, as RNA levels of genes like tenm1 increase in s1 mutants even at 0 dpa. However, we find chromatin looping structures in the s1 region that are more frequently detected during regeneration than without injury. We interpret this to mean that 3D structure, in addition to other local non-coding sequences, has critical impact on the extent to which s1 can regulate genes in context. Importantly, several genes in the s1 region exist that show silencing during regeneration and also have increased RNA levels during regeneration in s1 mutant animals. Not all of these genes are identified as physically interacting with s1 in our HiCAR assays, which might be explained by detection limitations of this method when applied in regenerating zebrafish fins. Nevertheless, our combined evidence indicates a silencing module during regeneration involving several genes regulated by s1. Indeed, two sequences with silencing activity (s1 and s2) were identified from this region, making it possible that several regulatory elements can contribute to silencing activity in this region and more broadly. The identity of such a complex of regulatory sequences, target genes, and binding proteins could emerge from higher resolution chromosome interaction assays, proximity labeling assays and proteomics, and/or a series of individual and combinatorial deletions in the region. Our data implicate potential binding within such a complex by the transcription factor Nr5a2, which has been described to regulate totipotency and pluripotency networks during development among other functions.⁵⁷

We speculate that a role of the silencing we describe could be to strategically downregulate genes whose products may facilitate tissue maintenance but have an inhibitory effect on regeneration. We also acknowledge that its function may be more mundane, such as to reduce expression from a cluster that is not relevant to regenerative events. Another possibility is that activation of genes in other areas of the chromosome or genome indirectly leads to s1-mediated silencing at this locus, possibly through positioning at the nuclear periphery.⁵⁸ Of note, s1 deletion had no biologically significant effects on fin regeneration. This could indicate that other silencers in the region normally cooperate to achieve full silencing of biologically impactful gene expression, or exist simply to provide redundancy. This may be yet another form of redundancy during regeneration that has been recognized in various other studies in which coding sequences for genes expected to have roles have been deleted without effects on regeneration.^3,46 Evidence of redundancy during regeneration contributes to an idea that there is appreciable robustness to regeneration and its ability to occur without obvious abnormality during genetic perturbations. Such robustness is an interesting topic in itself, and its study could lead to a new understanding of signal control during regeneration. s1 is the sole candidate TRSE that we tested extensively by molecular genetics from this screen, and thus we can only expect that other TRSEs exist with shared and unique features, some of which we expect will have more discernable requirements during regeneration.

Limitations of the Study

We acknowledge limitations of this study. First, we selected as primary candidates for TRSEs those sequences displaying increased accessibility at a sequence near genes that sharply reduce expression during regeneration. We interpreted these as events involving a transcriptional repressor complex binding selectively to a silencer during regeneration. However, we recognize there may be dynamic binding of a repressor without changing chromatin accessibility in a region near a gene that reduces expression, and we incorporated some of these candidate regions into our larval screen. We also recognize that regions near genes that experience a reduction in gene expression from the uninjured to regenerating context might instead occur by loss or displacement of an activating factor at an enhancer element, theoretically recognizable by a loss of ATAC signal (lower left quadrant in Figure 1C). This is a potentially important set of elements that can also be tracked. Second, in this era of single-cell genomics, we used bulk omics datasets to identify candidate silencers. This likely contributed noise to our screen, and we expect that single-cell multi-omics analyses would improve the quality of screening and more sensitively identify cell-type-specific regulatory elements and their target interactions in regenerating tissues.⁵⁹ Finally, we used a larval strategy to test the silencing activity of sequences identified from regenerating adult tissues. We acknowledge that a larval screen for silencing elements might mispresent adult tissue in some way, and that false negatives are likely. Yet, we expect some or most sequences to be active at both stages, plus there are massive benefits to early-stage screening in time and costs, and that scanning of the whole animal for all expression domains (including interesting, unexpected domains) is most feasible in larvae. Refinement of the screen toward a higher TRSE yield might include use of a docking site for transgene insertion, optimization of the injury-responsive promoter, and/or use of a genetic injury model.

STAR★METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Kenneth Poss (ken.poss@duke.edu).

Materials Availability

Reagents generated in this study will be shared upon request.

Data and Code Availability

The HiCAR of zebrafish fins and bulk RNA-seq of s1^Δ/Δ and WT sibling zebrafish fins have been deposited in the GEO database under ID code GEO: GSE231771 and GSE231956, respectively.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Zebrafish

Wild-type (WT) or transgenic zebrafish of the outbred Ekkwill (EK) strain ranging from 3 to 12 months were used for all experiments. Approximately equal sex ratios were used for experiments. Fish were housed at approximately 4 fish per liter in Pentair aquarium racks and fed three times daily, and the water temperature was maintained at 27.5°C. Fish were kept on a 14/10 light/dark cycle. Fish were anesthetized in 0.75% v/v 2-phenoxyethanol (Sigma-Aldrich) in fish water for regeneration experiments. Fins were amputated at 50% of their original length using a scalpel (Sterile Scalpel Blades #22, Feather). The amputation plane was determined to be the midpoint between the shortest ray of the caudal fin and where it attaches to the body. The fin was removed by rolling the surgical blade from its tip towards its handle. The fish were then returned to their tank, or to beakers in the case of genotyping. The fin clippings were discarded or collected for genotyping. All experiments with zebrafish were approved by the Institutional Animal Care and Use Committee at Duke University.

METHOD DETAILS

Motif Analysis

ATAC-Seq reads were trimmed using Trim Galore (0.4.1, with -q 15) then mapped with bowtie2 (2.2.5, with parameters --very-sensitive)⁷⁰ to zebrafish genome (UCSC danRer10 or 11 refSeq annotation). The mapped reads were filtered by a MAPQ greater than 30 using samtools (v 1.5)⁶⁹ and duplicated reads were removed by picard (v 1.91). The peaks were called by MACS2 (v2.1.0, with --nomodel --shift 37 --extsize 73 -q 0.05 -g 1.5e9).⁶⁸ The bioconductor package DiffBind (v 2.14.0)⁶⁷ was employed for differential open region analysis. A fold change greater than two and a P-value of less than 0.05 was used to filter the significant differential peaks. The enriched motifs were analyzed using chromVAR (v 1.8.0).⁶⁶ The motifs (Figure 1H) were a collection of jasper2018, jolma2013 and cisbp_1.02 from package motifDB (v 1.28.0) and were merged by a distance smaller than 1e-9 calculated by MotIV::motifDistances function (v 1.42.0). The best curated human motifs (downloaded from github.com/Dowell-Lab/TFEA) were used for the distal motif enrichment (>3k from TSS, Figure 1I, S1K-S1M and 4A). JASPAR motifs (non-redundant, vertebrate, 2022) were used to find the enriched motifs in the 91 bp region of s1 (Figure 4F).⁷⁹

Screen for Effects of Candidate Fin Regeneration Silencer Elements

Candidate silencing elements were mainly identified as a 140 bp – 4 kb sequence covering more than 1.5-fold increasing ATAC-seq peaks near genes reduced by 33% or more during fin regeneration. ATAC-seq and RNA-seq datasets of whole fin tissue 0 vs 4 dpa, whole fin tissue 0 vs 1 dpa, fin fibroblast 0 vs 4 dpa, and fin osteoblast 0 vs 4 dpa were referenced.^5,8 We prioritized candidate silencer sequences based on the extent of reduction in RNA levels, the dynamism and size of the nearby ATAC-signal, and the nature of the gene. Candidate silencer sequences were amplified from 3 dpf zebrafish genomic DNA using the primers listed in Table S2A, followed by subcloning via Gateway Cloning using Gateway LR Clonase II (Thermo Fisher Scientific) upstream of a P2:EGFP flanked by I-SceI sites.³ The constructs with or without candidate silencers in an injection solution (500 pL of 26.7 ng/μL), 333 U/mL I-SceI (NEB), 0.5x CutSmart buffer and 0.05% Phenol Red (Sigma-Aldrich) were injected into 50-100 one-cell zebrafish embryos. Mosaic EGFP expression in F₀ larvae at 1 dpa after a larval tail amputation at 3 dpf was assessed, and ratios of larvae with EGFP induction in response to injury were calculated (Figures 2D, 2E, 2F and S3B). If sequences have repressive effects, we estimated that only 0 to 25 of 100 injected embryos would fluoresce after larval tail amputation at 3 dpf. We repeated the screen for the sequences that showed a 25% ratio or lower, or if they displayed other potentially interesting expression features.

Screen for Effects of Candidate Cardiac Regeneration Silencer Elements

Most candidate silencing elements were identified as 500 bp – 3 kb sequences, with features of a more than 1.5-fold increasing chromatin signals during regeneration, near genes reducing levels by more than 33% during heart regeneration, from a dataset of cardiomyocyte-specific histone H3.3 profiling capturing sites of nucleosome turnover in regenerating cardiomyocytes,⁴ or more than 1.5-fold increasing ATAC-seq peaks near genes reducing levels by more than 33% during heart regeneration from epicardial ATAC-seq and RNA-seq datasets at 0, 3, and 7 dpa.⁶⁰ Candidates were subcloned via Gateway assembly into a construct containing a runx1-linked enhancer and mouse cfos promoter (REN-cfos:EGFP) that directs injury-responsive expression in larvae after cardiac injury (Figure 2H and 2I).⁴ Primers were listed in Table S2B. The constructs with or without candidate silencers (500 pL of 26.7 ng/μL), 333 U/mL I-SceI (NEB), 0.5x CutSmart buffer and 0.05% Phenol Red (Sigma-Aldrich) solution were injected into 50-100 one-cell cmlc2:mCherry-N-2A-Fluc^pd71 embryos in which cardiomyocytes are ablated by nitroreductase (NTR) and prodrug (metronidazole, MTZ) which are converted by NTR into cytotoxin. We assessed mosaic EGFP expression in F₀ larvae at 5 dpf after 10 mM MTZ treatment for 24 hours from 3 – 4 dpf, and ratios of larvae with EGFP induction in response to injury were calculated (Figure 2K and 2L). If sequences have repressive effects, we estimate that only 0 to 25 of 100 injected embryos would fluoresce at 5 dpf after cardiomyocyte ablation at 3 - 4 dpf. We repeated the screen for the sequences which showed a 25% or lower ratio.

Generation of Transgenic Reporter Lines

ubi:Brainbow^pd380, s1-ubr.Brainbow^pd381, s1-LENP2:EGFP^pd382 and s1-tph1b:mCherry^pd383 were generated using I-SceI transgenesis. The 850-bp s1 fragment was subcloned via Gateway Cloning using Gateway LR Clonase II (ThermoFisher Scientific) upstream of a 3.5 kb ubiquitin promoter with Brainbow,^45,80,81 upstream of LEN and lepb 2 kb promoter (P2) minimal promoter with EGFP³, or upstream of 7 kb tph1b upstream regulatory sequences and mCherry.⁴⁶ Two flanking I-SceI sites were included in each vector. Primers are provided in Table S2A. 500 pL of 26.7 ng/μL reporter construct, 333 U/mL I-SceI (NEB), 0.5x CutSmart buffer, and 0.05% Phenol Red (Sigma-Aldrich) solution was injected into each one-cell zebrafish embryo. F₀ founders were identified by PCR genotyping of male spermand outcrossed F₁ larvae and adult tissues.

Generation of Silencer Knock-in Zebrafish Lines

To insert an s1 cassette near kita (kita-s1KI^pd384, referred to as kita^s1/s1), fgf20a (fgf20a-s1KI^pd385, referred to as fgf20a^s1/s1), or lamb1a (lamb1a-s1KI^pd386, referred to as lamb1a^s1/s1) locus, potential gRNA target sites were identified using the web program CHOPCHOP (http://chopchop.cbu.uib.no/index.php). Genomic DNA sequences retrieved from Ensembl GRCz10 or z11 (https://useast.ensembl.org/Danio_rerio/Info/lndex) were used for the target site searches. Target sequences were selected that had no predicted off-target sites in the reference genome and no mismatches at loci to be targeted based on the whole-genome sequencing data of the laboratory EK strain. Target-specific Alt-R crRNA and common Alt-R tracrRNA were synthesized by IDT, and each RNA was dissolved in duplex buffer (IDT) as 1 μg/uL stock solution. Stock solutions were stored at −80°C. To prepare the crRNA:tracrRNA duplex, equal volumes of 1 μg/uL Alt-R crRNA and 1 μg/uL Alt-R tracrRNA stock solutions were mixed together and annealed by heating in a PCR machine: 95°C, 5 min; followed by gradual cooling by sitting the tube on the bench for a few minutes. The 1 μg/μL crRNA:tracrRNA duplex stock solution was mixed with equal volume of 1 μg/μL Cas9 protein (PNA BIO), incubated at 37°C for 5 min and kept on ice prior to injection. Based on validating the efficiency by heteroduplex mobility assay, gRNAs targeting 22 kb upstream, 1.5 kb upstream, or 83 kb downstream of the transcription start site (TSS) of kita, fgf20a, or lamb1a, respectively, were used for the following injections with a donor DNA plasmid. The gRNA target sequences are listed in Table S6A. s1 sequence was cloned via golden gate assembly into a vector with two universal gRNA target sites (see Table S6A) flanking the s1 insert for excision from the vector and non-homologous end joining at the specific locus. Then, 200 ng/μL gene specific crRNA:tracrRNA, 200 ng/μL universal crRNA:tracrRNA, 400 ng/μL Cas9 protein, and 40 ng/μL s1 donor vector were co-injected into one-cell stage embryos. To screen F₀ founders, genomic DNA was extracted from male sperm and genotyped with KAPA2G Genotyping Mix (Kapa Biosystems) using the primers, a gene specific forward primer and a reverse primer in s1 sequence (see Table S6B). F₀ founders were identified by PCR genotyping of male sperm and outcrossed F₁ larvae and adult tissues.

Length Measurements of Fin Regenerates

Anesthetized zebrafish were imaged using a Zeiss AxioZoom microscope at the time points indicated. All raw images were processed using either ZEN (Zeiss) or FIJI⁶² software. Lengths of fin regenerates were measured using ZEN Blue software (Zeiss), averaging the length of four regenerating fin rays each (2^nd and 3^rd most lateral rays of each lobe). PRISM software was used to perform paired two-tailed Student t tests on fin measurement data and generate graphs (GraphPad PRISM 9.5.1). For a longitudinal analysis of fin regeneration, image size must be consistent throughout the time course. 10 mm x 10 mm brightfield images were acquired using a Zeiss AxioZoom V16 dissecting microscope (12.5x zoom) with a Zeiss Axiocam 506 color camera using ZEN 2012 Blue Edition software (ver. 1.1.2.0). Fish were anesthetized with 0.075% 2-Phenoxyethanol (Sigma Aldrich) and placed on a 1% agarose gel-filled dish (Alkali Scientific) to be imaged. Image files were exported as tiff files, without compression or conversion to 8 bit. Adobe illustrator files were generated for each genotype were made that tracked individual fish throughout the time course (uninjured through 28 dpa). Fish were identified by the unique pigmentation pattern on their caudal fins. In order to measure lengths of the uninjured fins, 2 dpa fin images with demarcators indicating the amputation plane were overlaid onto images of the uninjured fin. The overlaid images were then manipulated to match fin ray segments proximal to the amputation plane.

After generating the Adobe files, files were exported as tiff files without compression. FIJI software (ver. 1.53k Java 1.8.0_172) was used to perform measurements. When each file was opened, the scale was set by using the line tool to draw a straight and flat line from one side of an individual fin image to the other side. The known distance was 10 mm from edge to edge of the image. In order to measure fin length, 4 separate fin rays are measured: the second and third rays from the ventral edge, and the second and third rays from the dorsal edge. A line was drawn using the line tool from the amputation plane to the tip of the fin ray. If the fin ray branches, the line was drawn to the middle/average of the branches. By averaging fin ray lengths per fish and normalizing this data to the uninjured lengths, we generated a longitudinal analysis of caudal fin regeneration based on the measurements.

RNA Isolation and Quantitative PCR

Caudal fin tissues from three fish were pooled for each biological replicate and collected in Tri-Reagent (Sigma). RNA was isolated by ethanol precipitation followed by purification with an RNA Clean & Concentrator Kit-5 (Zymo). cDNA was synthesized from 1 ug of total RNA with a First Strand cDNA Synthesis Kit (Roche). qPCR was performed on a Roche LightCycler 480 using LightCycler 480 SYBR Green I Master Mix (Roche) with primers listed in Table S6C. qPCR analyses of kita^s1/s1, fgf20a^s1/s1, lamb1a^s1/s1 and s1^Δ/Δ tissues were completed in technical triplicates for each timepoint. All experiments normalized transcript expression levels to actb2 as a housekeeping gene and then normalized the levels to 0 dpa as a control. PRISM software was used to perform statistical analysis with paired t tests and generate graphs (GraphPad PRISM 9.5.1).

Conservation Tests of Zebrafish s1 Sequence Across Species

DNA sequences for the s1 silencer and smarca1 gene loci of different species were retrieved from Ensembl database (https://www.ensembl.org/index.html). A conservation test was performed using the DNA sequence alignment tool mVISTA (http://genome.lbl.gov/vista/mvista/submit.shtml)⁸⁴ for multiple sequence alignments to investigate whether s1 sequence is evolutionarily conserved. The calculation window and conservation identity were set to 100 bp and 70%, respectively.

Assay for Repressive Ability of Zebrafish Silencer in Human Cells

To test the repressive ability of zebrafish s1 silencer in mammals, P2:EGFP, s1-P2:EGFP, LENP2:EGFP, s1-LENP2:EGFP, REN-cfos:EGFP, and s1-REN-cfos:EGFP plasmids were transfected with PEI MAX transfection reagent (Polysciences, cat#24765) in HEK293T cells. pcDNA3-Venus-AKAluc (Riken BRC DNABank, RDB15781) was used as control of transfection rate. Each group of plasmids were transfected for two replicates. In each replicate, two wells were transfected at the same time to exclude the possibility that transfection rates are different. For transfection, HEK293T cells were plated 16-24 hours into 12-well plates. After transfection, cells were cultured in 37°C incubator for another 16-24h before fixation and imaging. Cells were fixed and stained with DAPI. Imaging was performed on Zeiss 700 confocal and were analyzed by using FIJI. All the images were taken under the same setting and the proportion of EGFP⁺ cells were counted. For statistical analysis, unpaired Student’s t tests were used for comparisons between transfected constructs with or without s1 silencer.

Footprinting Analysis

To predict transcription factor binding in addition to the motif enrichment analysis, we conducted footprinting analysis using TOBIAS (v.0.12.0)⁵³ on the merged signals of ATAC-seq of whole fin 0 vs 4 dpa data and HiCAR of whole fin 0 vs 4 dpa data.⁶⁸ Tn5 bias, prediction of TF binding sites, and visualization of differentially bound TFs was performed with TOBIAS. JASPAR motifs (non-redundant, vertebrate, 2022) were used to scan for binding sites.⁷⁹

HiCAR Library Preparation and Sequencing

Biological triplicate pools of caudal fin clips were collected at 0 dpa (freshly amputated), 1 dpa, or 4 dpa time points using a scalpel, from 10 fish per pool. Fin tissues were frozen in liquid nitrogen and ground using a mortar and pestle on dry ice for 5 min. The tissues were crosslinked with 1% formaldehyde (Fisher Chemical) for 10 min at room temperature. ~100,000 crosslinked cells per replicate were used in the following steps (crosslink, Tn5 tagmentation, CviQI digestion, in situ ligation, reverse crosslink, DNA purification, NlaIII digestion, circularization, and DNA library amplification by PCR) as previously described.⁵⁴ Sequencing of the HiCAR libraries was performed on a DNBSEQ-MGI2000, 100 bp PE, by BGI Americas.

HiCAR datasets were processed by nf-core/hicar pipeline (v 1.0.0, with parameter -- cutadapt=true --macs_gsize='1.3e9' --ucscname='danRer11', danRer11 genome and annotation files, https://github.com/nf-core/hicar). Briefly, reads were aligned to danRer11 reference genome using bwa mem⁷² with flags -SP. Alignments were parsed, and paired end tags (PET) were generated using the pairtools (v0.3.0) (https://github.com/mirnylab/pairtools). PET with low mapping quality (MAPQ < 10) was filtered out. PET with the same coordinate on the genome or mapped to the same digestion fragment were removed. Uniquely mapped PETs were flipped as side 1 with the lower genomic coordinate and aggregated into contact matrices in the cooler format using the cooler tools (v0.8.11)⁷¹ at delimited resolution (5 kb, 10 kb, 50 kb, 100 kb, 250 kb, 500 kb, 1 mb, 25 mb, 50 mb, 100 mb). The dense matrix data were extracted from cooler files and visualized using trackViewer (v. 1.37.12).⁷⁵ The R1 and R2 reads signals around TSS or peaks were calculated with UCSC utilities (v377) before PET flipping. TADs were defined by Homer (v4.11).⁷³ Chromatin loops were called by MAPS (v1.1.0).⁷⁴ The raw signals are normalized by zero-truncated poisson distribution, regression coefficients, read counts, GC content, mappability score, genomic distance and the number of cuts.

Generation of Silencer Deletion Zebrafish Lines

s1^A/A was generated by injecting, 500 pL of 250 ng/μL crRNA:tracrRNA pair flanking a ~4.9 kb region including s1 silencer (see Table S6A), 500 ng/μL Cas9 Protein (PNA BIO), 0.05% Phenol Red (Sigma-Aldrich) solution was injected into one-cell zebrafish embryos. The gRNA target sites were identified and validated following the same method described above. Embryos were genotyped at 2-4 dpf using primers in Table S6B to determine gRNA efficiency. To screen F₀ founders, genomic DNA was extracted from male sperm and genotyped with KAPA2G Genotyping Mix (Kapa Biosystems) using primers flanking the deleted region including s1 sequence (see Table S6B). F₀ founders were identified by PCR genotyping of male sperm and outcrossed F₁ larvae and adult tissues.

RNA-seq Library Preparation and Sequencing

Fins were homogenized by a stainless-steel bead (5 mm, Cat. No. 69989) in Tri-reagent and the RNA was extracted by chloroform followed by isopropanol precipitation. The remaining DNA was digested by DNase. RNA was purified using a Zymo RNA clean/concentrator kit. cDNA was prepared by Maxima H-minus RT reverse transcription kit. PCR was performed using Kapa HiFi HotStart Master Mix and adapter primers. The amplified cDNA was purified using 0.6x SPRI beads. cDNAs were tagged using Tn5 transposase and amplified by Q5 polymerase. Size selection was performed with gel-cutting and extraction, and 300 - 800 bp DNA fragments were collected for sequencing. Libraries were sequenced using a DNBSEQ-MGI2000, 100 bp PE, by BGI Americas. RNA-seq reads were trimmed by Trim Galore (v 0.6.7, with cutadapt v 3.5) and mapped with STAR (v 2.7.10, with parameters --twopassMode Basic and supplying the Ensembl danRer11 annotation)⁸² to zebrafish genome (danRer11). The mapped reads were counted using R/Bioconductor GenomicAlignments package (v 1.30.0).⁸³ Bioconductor package DESeq2 (v 1.34.0)⁶⁵ was employed to analyze differential expressions (DE) with genotype information. The enrichment analysis were performed by Bioconductor clusterProfiler package (v4.2.2).⁷⁸

QUANTIFICATION AND STATISTICAL ANALYSIS

Clutchmates were randomized into different treatment groups for each experiment except experiments using s1^Δ/Δ and WT siblings. No animal or sample was excluded from the analysis unless the animal died during the procedure or was deformed/destroyed due to fixation or dissection. All experiments were performed with at least two biological replicates, and at least three samples were used for each experiment unless otherwise indicated. Sample sizes, statistical tests, and p values are indicated in the figures or the legends. All statistical values are displayed as Mean ± S.D. (Standard Deviation). Statistics tests were calculated using two-tailed Student’s t-tests when normality test was passed or Mann-Whitney Rank Sum test otherwise using Prism software, or Walds test for longitudinal analysis of fin length measurements.

Supplementary Material

1

Supplementary Table S1. Candidate silencers from RNA-seq and ATAC-seq of fin and heart tissues and their motif enrichment analysis (related to Figures 1 and 2)

(A) Candidate silencers from RNA-seq and ATAC-seq of whole fin tissue 0 vs 4 dpa

(B) Candidate silencers from RNA-seq and ATAC-seq of whole fin tissue 0 vs 1 dpa

(C) ChromVAR enriched motifs in candidate silencers from RNA-seq and ATAC-seq of whole fin tissue 0 vs 4 dpa

(D) ChromVAR enriched motifs in candidate silencers from RNA-seq and ATAC-seq of whole fin tissue 0 vs 1 dpa

(E) ChromVAR enriched motifs in candidate silencers from RNA-seq and ATAC-seq of fin fibroblast 0 vs 4 dpa

(F) ChromVAR enriched motifs in candidate silencers from RNA-seq and ATAC-seq of fin osteoblast 0 vs 4 dpa

(G) Epicardium uninjured vs 3 dpa and uninjured vs 7 dpa

(H) Cardiomyocyte uninjured vs 7 dpi and uninjured vs 14 dpi

2

Supplementary Table S2. Control and candidate silencers tested in F₀ larval fin and cardiac regeneration screen (related to Figures 1 and 2)

(A) Fin regeneration

(B) Heart regeneration

3

Supplementary Table S3. Differential binding motifs in footprinting analysis (related to Figure 4)

4

Supplementary Table S4. Candidate silencers and enhancers defined by HiCAR loops from RNA-seq, ATAC-seq, and HiCAR of whole fin tissues (related to Figure 5)

(A) 0 vs 4 dpa

(B) 0 vs 1 dpa

5

Supplementary Table S5. Differential expressed transcripts from RNA-seq of whole fin wild-type vs s1 KO (related to Figure 6)

(A) 0 dpa

(B) 1 dpa

(C) 4 dpa

6

Supplementary Table S6. Lists of sequences for gRNAs and primers (related to STAR Methods)

(A) List of gRNA for generating s1 knock-in/knockout lines or deleting short sequences in s1

(B) List of primers for genotyping s1 silencer knock-in or knockout lines

(C) List of primers for qPCR

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
ATAC-seq/RNA-seq of zebrafish fins and fin fibroblasts	Thompson et al.5	GEO: GSE146960
ATAC-seq of zebrafish fin osteoblasts	Lee et al.8	GEO: GSE126700
RNA-seq of zebrafish fin osteoblasts	Lee et al.8	GEO: GSE126701
Whole genome bisulfite sequencing of zebrafish fins	Lee et al.8	GEO: GSE126702
RNA-seq of zebrafish cardiomyocytes	Goldman et al.4	GEO: GSE81865
H3.3 ChIP-seq of zebrafish cardiomyocytes	Goldman et al.4	GEO: GSE81862
ATAC-seq/RNA-seq of zebrafish epicardium	Cao et al.60	GEO: GSE89444
HiCAR of zebrafish fins	This paper	GEO: GSE231771
RNA-seq of s1Δ/Δ and WT sibling zebrafish fins	This paper	GEO: GSE231956
Experimental models: Cell lines
Human: HEK293T cell line	ATCC	CRL-3216
Experimental models: Organisms/strains
Zebrafish: Tg(cmlc2:mCherry-N-2A-Fluc)pd71	Chen et al.61	pd71
Zebrafish: Tg(ubi:Brainbow)pd380	This paper	pd380
Zebrafish: Tg(s1-ubi:Brainbow)pd381	This paper	pd381
Zebrafish: Tg(LENP2:EGFP)pd129	Kang et al.3	pd129
Zebrafish: Tg(s1-LENP2:EGFP)pd382	This paper	pd382
Zebrafish: Tg(tph1b:mCherry-NTR)pd275	Tornini et al.46	pd275
Zebrafish: Tg(s1-tph1b:mCherry-2a-NTR)pd383	This paper	pd383
Zebrafish: kita-s1KIpd384	This paper	pd384
Zebrafish: kitab5	Parichy et al.47	b5
Zebrafish: fgf20a-s1KIpd385	This paper	pd385
Zebrafish: lamb1a-s1KIpd386	This paper	pd386
Zebrafish: s1pd387	This paper	pd387
Recombinant DNA
Plasmid: P2:EGFP	Kang et al.3	N/A
Plasmid: REN-cfos:EGFP	Goldman et al.4	N/A
Plasmid: pcDNA3-Venus-AKAluc	RIKEN BRC repository	N/A
Plasmid: s1-P2:EGFP	This paper	N/A
Plasmid: LENP2:EGFP	Kang et al.3	N/A
Plasmid: s1-LENP2:EGFP	This paper	N/A
Plasmid: s1-REN-cfos:EGFP	This paper	N/A
Software and algorithms
FIJI	Schindelin et al.62	https://imagej.nih.gov/ij/
Prism 9	GraphPad	https://www.graphpad.com/features
Zen	Zeiss	https://www.zeiss.com/microscopy/en/products/software/zeiss-zen-lite.html
Integrative Genomics Viewer (IGV)	Robinson et al.63	https://software.broadinstitute.org/software/igv/
Juicebox	Robinson et al.64	https://aidenlab.org/juicebox/
DEseq2	Love et al.65	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
ChromVAR	Schep et al.66	https://github.com/GreenleafLab/chromVAR
DiffBind	Ross-Innes et al.67	https://bioconductor.org/packages/release/bioc/html/DiffBind.html
MACS2	Zhang et al.68	https://github.com/macs3-project/MACS
Samtools	Li et al.69	https://samtools.sourceforge.net/
bowtie2	Langmead and Salzberg 70	https://bowtie-bio.sourceforge.net/bowtie2/index.shtml
Cooler	Abdennur and Mirny 71	https://github.com/open2c/cooler
bwa mem	Li and Durbin72	https://bio-bwa.sourceforge.net/
Homer	Heinz et al.73	http://homer.ucsd.edu/homer/
MAPS	Juric et al.74	https://github.com/ijuric/MAPS
trackViewer	Ou and Zhu75	https://bioconductor.org/packages/trackViewer
motifStack	Ou et al.76	https://bioconductor.org/packages/release/bioc/html/motifStack.html
ChIPpeakAnno	Lihua et al.77	https://bioconductor.org/packages/ChIPpeakAnno
clusterProfiler	Yu et al.78	https://bioconductor.org/packages/clusterProfiler
TOBIAS	Bentsen et al.53	https://github.com/loosolab/TOBIAS

Highlights:

• A larval screen identifies candidate silencers during zebrafish fin regeneration

• s1 contains sequences essential to repress expression from nearby genes

• s1 makes physical contacts with genes that reduce expression during fin regeneration

• s1 is essential for regeneration-associated reductions in expression of nearby genes