Genome-wide absolute quantification of chromatin looping

Abstract

3D genomics methods such as Hi-C and Micro-C have uncovered chromatin loops across the genome and linked these loops to gene regulation. However, these methods only measure 3D interaction probabilities on a relative scale. Here, we overcome this limitation by using live imaging data to calibrate Micro-C in mouse embryonic stem cells, thus obtaining absolute looping probabilities for 36,804 chromatin loops across the genome. We find that the looped state is generally rare, with a mean probability of 2.3% and a maximum of 26% across the quantified loops. On average, CTCF-CTCF loops are stronger than loops between cis-regulatory elements (3.2% vs. 1.1%). Our findings can be extended to human stem cells and differentiated cells under certain assumptions. Overall, we establish an approach for genome-wide absolute loop quantification and report that loops generally occur with low probabilities, generalizing recent live imaging results to the whole genome.

Figure 1.. Overall scheme for genome-wide absolute quantification of chromatin loops from Micro-C.

a, Flowchart describing genome-wide absolute loop quantification by calibrating Micro-C dot strength against loops measured in live imaging. b, Absolute looping probability is defined as the probability of interactions excluding the expected background. The background is primarily a function of genomic separation. c, AbLE quantifies Micro-C dot strength by summing the Micro-C signal originating from a chromatin loop excluding the local background, which is estimated in a separate step. The size of the local region, which can vary from 25–200 kb, is chosen to be proportional to $\sqrt{s}$ (see Methods). d, Diagram of simulated region with features (bottom) and simulated Micro-C map (top) from 3D polymer simulation. Micro-C bin size = 1 monomer, corresponding to 1 kb. Color scale is linear. e, Plot of ground-truth absolute looping probability from 3D polymer simulations vs. AbLE scores from the simulated Micro-C map. Gray dotted line indicates best-fit line of proportionality. Examples of simulated Micro-C dots corresponding to loops with various looping probabilities are shown on right. Micro-C color scale is linear. f, Micro-C dots corresponding to loops previously measured with live imaging; diagrams of engineered loci for live imaging shown below Micro-C maps. Micro-C bin size = 1 kb. Color scale is linear. Coordinates are given in terms of the modified genomes for the engineered cell lines (see Fig. S9a). g, Looping probability vs. AbLE score for Fbn2 loop, Npr3 loop, and control condition (Fbn2 with depletion of RAD21). Gray line indicates best-fit line of proportionality; light gray shaded area indicates 95% confidence interval. Circles indicate AbLE scores of individual Micro-C replicates; error bars indicate bootstrapped 95% confidence intervals in BILD estimates.

Figure 2.. Global analysis of chromatin looping probabilities.

a, Flowchart of methods used to perform global analysis of chromatin looping probabilities. Five mESC Micro-C datasets were combined to form a merged dataset with superior signal-to-noise ratio compared to the deepest dataset previously available. Then, loops were called using Mustache and filtered for quantifiability. Filtered loops were quantified and classified for further analysis. b, Distribution of looping probabilities and relationship between probability vs. size among 36,804 filtered loops. In the box plot, whiskers extend to 0.5^th and 99.5^th percentiles; all outliers beyond the whiskers are plotted as individual points. c, Distribution of loop classes as determined by presence of CTCF/cohesin, enhancers, or promoters at loop anchors. “Other” refers to anchors that did not overlap any of these features. d, Distributions of looping probabilities and relationship between probability vs. size among pure CTCF-CTCF loops and pure cis-regulatory loops. e, Mean looping probability for all possible combinations of loop anchor classes. Under the “inclusive” definition, a loop anchor is said to be a CTCF- and cohesin-bound site, enhancer, or promoter if that feature is present, regardless of whether other features are present. Under the “exclusive” definition, a loop anchor is said to have a feature if only that feature is present; loops with two or more features at either anchor are omitted from the analysis. C = CTCF/cohesin, E = enhancer, P = promoter, and O = other. Among all CTCF-CTCF loops (defined with the inclusive definition), the effect of CTCF motif orientation is also analyzed.

Figure 3.. Chromatin looping probabilities are correlated with epigenomic features at loop anchors.

a, Epigenomic features at each anchor of each loop are quantified by summing the values from the associated assay (ChIP-seq, ATAC-seq, or GRO-seq) in a 5-kb window centered on the anchor. Micro-C and select tracks are shown for an example loop. Micro-C bin size = 1 kb. Color scale is linear. b, Looping probability vs. feature strengths at anchors for six ChIP-seq features. x- and y- values represent ChIP signals at left (L) and right (R) anchors and are binned into deciles; color scale represents the mean absolute looping probability within each bin. Bins in which the standard error of the mean (σ/n) exceeds 0.2 are colored white due to their high uncertainty. c, Schematic of linear models for predicting looping probability from epigenomic features. Top: OLS linear regression model with 87 regressors. Bottom: dimensionality-reduced model with 2 latent variables. Models are trained on a subset containing 80% of loops; R² values are calculated the remaining 20% of data. d, Scatter plot of all filtered loops in latent variable space, colored by absolute looping probability. e, Scatter plots of loops in latent variable space, with loops in specific classes highlighted in black and other loops colored gray. f, Weight vectors defining the transformation to latent variable space. Due to symmetry between the left and right anchors, the weights for each feature represents both the weight for the left and right anchors (e.g., “CTCF” represents the weights for regressors CTCF_L and CTCF_R). g, Goodness-of-fit R² of OLS models for predicting absolute looping probability from all 87 regressors, within loop classes individually. Models are trained on a subset containing 80% of loops; R² values are calculated the remaining 20% of data.

Figure 4.. Human embryonic stem cells (hESCs) and human foreskin fibroblast cells (HFFs) exhibit similar chromatin looping trends as mESCs.

a, Flowchart of method to estimate absolute looping probabilities in hESC and HFF cells. In each cell type, loops are called from Micro-C maps using Mustache, filtered for quantifiability, and quantified using AbLE. Quantifications are converted to absolute units using the mESC calibration. b, Looping probability distributions and statistics for mESC loops (Fig. 2b,d) hESC loops, HFF loops, and shared loops (loops common to both hESCs and HFFs) in hESCs and HFFs for all loops and for CTCF-CTCF loops. Box plot whiskers extend to 0.5^th and 99.5^th percentiles; outliers beyond the whiskers are plotted as individual points. Due to their high uncertainty, some loops with extremely low probabilities (below 10⁻³-10⁻⁴) are not shown, but these account for less than 3% of loops in each category. c, Scatter plot of looping probabilities of shared loops in HFFs vs. hESCs. Dashed line indicates the line x = y.