Integrating Hi-C and FISH data for modeling of the 3D organization of chromosomes

Abstract

The new advances in various experimental techniques that provide complementary information about the spatial conformations of chromosomes have inspired researchers to develop computational methods to fully exploit the merits of individual data sources and combine them to improve the modeling of chromosome structure. Here we propose GEM-FISH, a method for reconstructing the 3D models of chromosomes through systematically integrating both Hi-C and FISH data with the prior biophysical knowledge of a polymer model. Comprehensive tests on a set of chromosomes, for which both Hi-C and FISH data are available, demonstrate that GEM-FISH can outperform previous chromosome structure modeling methods and accurately capture the higher order spatial features of chromosome conformations. Moreover, our reconstructed 3D models of chromosomes revealed interesting patterns of spatial distributions of super-enhancers which can provide useful insights into understanding the functional roles of these super-enhancers in gene regulation.

Fig. 1

The schematic overview of GEM-FISH, which applies a divide-and-conquer strategy to reconstruct the 3D organization of a chromosome by systematically integrating both Hi-C and FISH data. a The 3D chromosome model at TAD-level resolution is calculated by integrating Hi-C and FISH data as well as prior biophysical knowledge of a 3D polymer model. b The 3D conformations of individual TADs are determined using the intra-TAD geometric restraints derived from the input Hi-C map and prior biophysical knowledge of polymer models. c The final complete 3D structure of the chromosome is obtained by assembling the modeling results from the previous two steps, i.e., placing the previously determined intra-TAD conformations into the TAD-level resolution model through translation, rotation, and reflection operations. More details can be found in the main text

Fig. 2

The modeling results of human Chromosome 21 (Chr21). a The TAD-level resolution 3D structure of Chr21 calculated by GEM-FISH, where each dot represents the center of a TAD. b The final 3D structure of Chr21 reconstructed by GEM-FISH. The visualization in (a) and (b) was performed using UCSF Chimera. c, d The relative error matrices of the TAD-level resolution models computed by GEM-FISH using both Hi-C and FISH data, and by GEM using only Hi-C data, respectively. e, f The relative error matrices of the final models computed by GEM-FISH using both Hi-C and FISH data, and GEM using only Hi-C data, respectively

Fig. 3

Assignment of TADs to A/B compartments for Chrs 20, 21, and 22. a–c Assignment of TADs for Chrs 20 (a), 21 (b), and 22 (c) using the experimental FISH data (obtained from ref. ). d–f Assignment of TADs of the 3D chromosome models calculated by GEM-FISH using both Hi-C and FISH data for Chrs 20 (d), 21 (e), and 22 (f). g–i Assignment of TADs of the 3D chromosome models calculated by GEM using only Hi-C data for Chrs 20 (g), 21 (h), and 22 (i)

Fig. 4

The modeling results of GEM-FISH for the human X-Chromosome (including both active state ChrXa and inactive state ChrXi). a, b Visualization of the final 3D models of ChrXa (a) and ChrXi (b) in UCSF Chimera. (c) Comparison of the compactness of TADs between the 3D models of active (ChrXa) and inactive (ChrXi) states of human X-Chromosome. N = 40 TADs for both Chrs Xa and Xi. *p-value < 10⁻¹³, one-tailed Wilcoxon rank-sum test. d, e The assignment of TADs of ChrXa (d) and ChrXi (e) to the A/B compartments. f, g Projection of the 3D convex hull plots of the A and B compartments of ChrXa (f) and ChrXi (g) to the XZ plane. For the boxplots, the top and bottom lines of each box represent the 75th and 25th percentiles of the samples, respectively. The line inside each box represents the median of the samples. The upper and lower lines above and below the boxes are the whiskers

Fig. 5

Regions belonging to the same subcompartment tend to colocalize in the 3D space. a, c, e Visualization of the regions belonging to subcompartments B1 (cyan) and B2 (magenta) in the 5 Kbps-resolution 3D models reconstructed by GEM-FISH for Chr20, Chr21, and Chr22, respectively. Only regions that belong to either B1 or B2 are shown. The visualization was performed using UCSF Chimera. b, d, f Boxplots on the densities of regions belonging to subcompartments B1 and B2 for Chr20, Chr21, and Chr22, respectively. N_B1 = 39, 20, 44 genomic segments belonging to subcompartment B1 for Chrs 20, 21, and 22, respectively. N_B2 = 23, 25, 9 genomic segments belonging to subcompartment B2 for Chrs 20, 21, and 22, respectively. *p-value < 0.007, **p-value < 10⁻⁴, ***p-value < 10⁻⁴. All tests were performed using the one-tailed Wilcoxon rank-sum test. For the boxplots, the top and bottom lines of each box represent the 75th and 25th percentiles of the samples, respectively. The line inside each box represents the median of the samples. The upper and lower lines above and below the boxes are the whiskers. Red points marked by ‘+’ represent outliers, which represent the observations beyond 1.5 times interquartile range away from the top or the bottom of the box

Fig. 6

An experimental evidence that genomic loci from the same subcompartment tend to colocalize. a The ChIP-Seq profiles of different histone marks for the three loci L1, L2, and L3 examined in the FISH experiment. Locus L3 is relatively enriched with the mark H3K27me3 and depleted from the mark H3K36me3, and hence considered belonging to subcompartment B1. On the other hand, the two loci L1 and L2 are depleted from the marks H3K27me3, H3K27ac, H3K36me3, H3K4me1, and H3K79me2, and hence considered belonging to subcompartment B2. b One example of the experimental FISH images for the two loci L1 and L2 (top), and for the two loci L1 and L3 (bottom). c The cumulative distribution function (CDF) curves of the distances between L1 and L2 and between L1 and L3, indicating the tendency of the two loci L1 and L2 (belonging to the same subcompartment B2) to lie spatially closer to each other than the other two loci L1 and L3 (belonging to subcompartments B2 and B1, respectively), in spite of the smaller genomic distance between L1 and L3 than between L1 and L2. Data from 35 individual imaged copies of Chr21 were used to generate (c)

Fig. 7

Super-enhancers tend to lie on the surface of the 3D models of chromosomes. a–c Super-enhancers (cyan) lie on the surfaces of the 3D models reconstructed by GEM-FISH for Chrs 20, 21, and 22, respectively. For Chr21 (b), four of the five super-enhancers were found in the gene-rich G-band q22.3 region, which is shown in red. The visualization was performed using UCSF Chimera