Does multi-way, long-range chromatin contact data advance 3D genome reconstruction?

Abstract

Background: Methods for inferring the three-dimensional (3D) configuration of chromatin from conformation capture assays that provide strictly pairwise interactions, notably Hi-C, utilize the attendant contact matrix as input. More recent assays, in particular split-pool recognition of interactions by tag extension (SPRITE), capture multi-way interactions instead of solely pairwise contacts. These assays yield contacts that straddle appreciably greater genomic distances than Hi-C, in addition to instances of exceptionally high-order chromatin interaction. Such attributes are anticipated to be consequential with respect to 3D genome reconstruction, a task yet to be undertaken with multi-way contact data. However, performing such 3D reconstruction using distance-based reconstruction techniques requires framing multi-way contacts as (pairwise) distances. Comparing approaches for so doing, and assessing the resultant impact of long-range and multi-way contacts, are the objectives of this study.

Results: We obtained 3D reconstructions via multi-dimensional scaling under a variety of weighting schemes for mapping SPRITE multi-way contacts to pairwise distances. Resultant configurations were compared following Procrustes alignment and relationships were assessed between associated Procrustes root mean square errors and key features such as the extent of multi-way and/or long-range contacts. We found that these features had surprisingly limited influence on 3D reconstruction, a finding we attribute to their influence being diminished by the preponderance of pairwise contacts.

Conclusion: Distance-based 3D genome reconstruction using SPRITE multi-way contact data is not appreciably affected by the weighting scheme used to convert multi-way interactions to pairwise distances.

Keywords: Conformation capture; Multi-dimensional scaling; Pairwise distance; Procrustes alignment; SPRITE.

Fig. 1

Contact frequency by maximum distance between reads for standard Hi-C [3] and SPRITE stratified by k-mer size

Fig. 2

Barplots of a k-mer sizes (green > 3, red 3, black 2) and b distances (quintiles with first quintile black and fifth quintile cyan) of k-mers by chromosome. The first quintile ranged from 18 to 27%, the second from 17 to 26%, the third from 18 to 24%, the fourth from 17 to 24%, and the fifth from 5 to 27%

Fig. 3

Procrustes RMSEs as a function of length, percentage many-contacts and percentage long-distance contacts. Least squares fits relate these features to RMSEs obtained using the N-W 3D reconstruction as target, with black and green designating aligned U-W and O-W 3D reconstructions, respectively

Fig. 4

Procrustes RMSEs for a standard Hi-C reconstruction target compared to SPRITE reconstructions using U-W (black U), N-W (red N), and O-W (green O)

Fig. 5

Heatmaps depicting contact matrices (of square root transformed counts) for the Hist1 region of Chromosome 6. Scales (not shown) differ according to assay type, weighting scheme and threshold (saturation) value. Darker pixels are for higher contact values, except white is for saturation. Regions of interest are defined by pairs of dark green, lime green and turquoise lines

Fig. 6

Same as Fig. 5. but for the contact matrices for the three consecutive CTCF anchored loops on Chromosome 8