Three-dimensional chromatin mapping of sensory neurons reveals that promoter-enhancer looping is required for axonal regeneration

Abstract

The in vivo three-dimensional genomic architecture of adult mature neurons at homeostasis and after medically relevant perturbations such as axonal injury remains elusive. Here, we address this knowledge gap by mapping the three-dimensional chromatin architecture and gene expression program at homeostasis and after sciatic nerve injury in wild-type and cohesin-deficient mouse sensory dorsal root ganglia neurons via combinatorial Hi-C, promoter-capture Hi-C, CUT&Tag for H3K27ac and RNA-seq. We find that genes involved in axonal regeneration form long-range, complex chromatin loops, and that cohesin is required for the full induction of the regenerative transcriptional program. Importantly, loss of cohesin results in disruption of chromatin architecture and severely impaired nerve regeneration. Complex enhancer-promoter loops are also enriched in the human fetal cortical plate, where the axonal growth potential is highest, and are lost in mature adult neurons. Together, these data provide an original three-dimensional chromatin map of adult sensory neurons in vivo and demonstrate a role for cohesin-dependent long-range promoter interactions in nerve regeneration.

Keywords: cohesin; nerve regeneration; promoter loops; regeneration program; three-dimensional chromatin architecture.

Fig. 1.

Loss of cohesin impairs nerve regeneration. (A and B) Representative micrographs and quantification of SCG10 intensity at the indicated distances from the lesion site in sciatic nerves from AAV-GFP or AAV-Cre-GFP injected Scc1flox/flox mice at 7 d following SNC. The asterisk marks the lesion site. (Scale bar, 1 mm.) (mean ± SEM of n = 5 nerves from four AAV-GFP mice and n = 6 nerves from four AAV-Cre-GFP mice; **P < 0.05, ****P < 0.0001, two-way Anova, Sidak’s multiple comparisons test). (C) Bar charts of the regeneration index in AAV-GFP or AAV-Cre-GFP injected Scc1flox/flox mice at 7 d after SNC (mean ± SEM of n = 5 nerves from four AAV-GFP mice and n = 6 nerves from four AAV-Cre-GFP mice; two-sided unpaired Student’s t test). (D) Micrographs showing GFP (green) and CTB (magenta) signal in DRG 4 d after injection of CTB in the tibialis anterioris and gastrocnemius muscle 14 d after SNC. Arrowheads mark GFP-positive neurons. (Scale bar, 100 μm.) (E) Bar graphs of the CTB positive neurons as percentage of GFP expressing neurons (mean ± SEM of n = 4 mice; two-sided unpaired Student’s t test). (F) Representative micrographs of PGP9.5 immunostaining with DAPI counterstaining in the interdigital hind-paw skin from AAV-GFP or AAV-Cre-GFP injected Scc1flox/flox mice 18 d after SNC. The dashed lines indicate the boundary between the epidermis and dermis. (Scale bar, 100 µm.) (G and H) Quantification of the number of intraepidermal fibers (IEDF) per millimeter of interdigital skin and the percentage of IEDF versus dermal fibers (DF) (mean ± SEM of n = 5 skins from four AAV-GFP mice; n = 7 skins from four AAV-Cre-GFP mice; two-sided unpaired Student’s t test).

Fig. 2.

Cohesin-dependent and independent genes. (A) Dot plot of the semantically clustered GO biological process categories of the up-regulated (red) and down-regulated (blue) genes in the indicated conditions. Color code reflects the P-value (modified Fisher’s exact P ≤ 0.005) and the size of the dot the gene count in each category (gene count > 6). (B) Pie charts of the expression of constitutive (Top) and injury-activated (Bottom) genes in Rad21 KO neurons in naïve and injured conditions (n = 3 independent samples; FDR < 0.05). (C) Line plots of the expression in the indicated conditions of the cohesin-independent and dependent (inducible and noninducible) genes (TPM = transcripts per million). (D) Heatmaps of the semantically clustered GO biological process categories of the cohesin-dependent and independent genes. Color code reflects the P-value (modified Fisher’s exact P ≤ 0.001).

Fig. 3.

Loss of cohesin leads to a decrease in chromatin interactions. (A) Schematic of the chromatin mapping in DRG neurons. (B) Genome-wide plot of the averaged insulation scores around strong boundaries (the size of flanks around each strong boundary is 400 Kb) in the indicated conditions. (C) P(s) curve (contact probability versus genomic distance) within and between genomic domains of length 300 to 500 Kb for the indicated conditions. (D) Average Hi-C contact matrices at 10-Kb resolution of the indicated number of 3D chromatin domains of length 300 to 400 Kb in the indicated conditions. (E) Bar charts of the number of 3D chromatin domains showing an average increased, decreased, and unchanged frequency of contacts (n = 3 independent samples; FDR < 0.05). (F) Odds ratios analysis of the association between up- and down-regulated genes and differentially expressed genes present in increased, decreased, or unchanged domains; the numbers in red represent the P-value given by two-sided Fisher’s exact test.

Fig. 4.

Regenerative genes reside within cohesin-dependent chromatin domains. (A) Heatmap of the semantically clustered GO biological process categories of the down-regulated genes in Rad21 KO injured neurons residing within domains that were either lost or showed a decreased contact frequency and unchanged domains. Color code reflects the P-value (modified Fisher’s exact P ≤ 0.001). (B) Heatmap showing the residence of RAGs within genomic domains that were either lost or showed a decreased interaction frequency in Rad21 KO neurons. (C) Network visualization of the injury-activated, cohesin-dependent genes (red) (cohesin is depicted in orange). Cohesin-dependent genes are preferentially associated with lost domains (blue = 314) and domains with decreased frequency (green = 188) with respect to unchanged domains (gray = 84). Edges connecting the genes to their genomic domains define the fold change of Hi-C interactions in SNC_KO versus SNC_WT. For a better visualization, all the domains on each chromosome are visualized as a unique circle. (D) Example of Hi-C maps of the contact frequency in the indicated conditions within 3D genomic domains. The contacts between the three successive bins (the one containing TSS and its two neighbors) and bins within a genomic distance of 500 Kb are extracted from 5-Kb Hi-C contact matrices. (E) Cohesin facilitates the formation of 3D genomic domains where regenerative genes are coregulated. Loss of cohesin disrupts the architecture of 3D genomic domains impairing the activation of the regenerative program.

Fig. 5.

Cohesin-dependent genes engage in longer chromatin loops in DRG neurons in response to nerve injury. (A and B) Box plot of the number of significant chromatin interactions (A) and the genomic distance between them (B) for inactive (TPM = 0 and first quartile) and active (from second to forth quartile) genes in the indicated conditions. (C) Box plot of the loop length for up-regulated (FDR < 0.05, FC > 1.5) genes after injury in WT and Rad21 KO neurons in the indicated conditions. P-values are computed from two-sided unpaired Student’s t test; n.s. = not significant. (D and E) Box plot of the promoter loop length for cohesin-dependent and independent (D) and inducible and noninducible (E) genes in the indicated conditions. P-values are computed from two-sided unpaired Student’s t test; n.s. = not significant. (F) Box plot of the number of promoter loop length per gene for cohesin-dependent and independent genes in the indicated conditions. P-values are computed from two-sided unpaired Student’s t test. (G) Examples of promoter loops at functionally relevant genes. (H) Cohesin facilitates the formation of E-P chromatin loops that allow the activation of the regenerative program after nerve injury.

Fig. 6.

Injury-responsive genes display a more immature chromatin looping signature after injury. (A) Odds ratios analysis of the association between up-, down-regulated genes, cohesin-dependent, and independent genes, constitutive genes, versus genes associated to fetal cortical plate and adult neuron E-P loops; the numbers in red represent the P-value given by two-sided Fisher’s exact test. (B) Heatmap of the semantically clustered GO biological process categories of the up-regulated genes associated with fetal cortical plate and down-regulated genes associated with adult neuron E-P loops. Color code reflects the P-value (modified Fisher’s exact P ≤ 0.001). (C) Box plot of the promoter loop length for injury-activated, cohesin-dependent and independent genes, and constitutive genes, in fetal cortical plate and adult neurons. FC = fold change. P-values are computed from two-sided unpaired Student’s t test. (D) Examples of promoter loops at functionally relevant genes.