Normalization of a chromosomal contact map

Abstract

Background: Chromatin organization has been increasingly studied in relation with its important influence on DNA-related metabolic processes such as replication or regulation of gene expression. Since its original design ten years ago, capture of chromosome conformation (3C) has become an essential tool to investigate the overall conformation of chromosomes. It relies on the capture of long-range trans and cis interactions of chromosomal segments whose relative proportions in the final bank reflect their frequencies of interactions, hence their spatial proximity in a population of cells. The recent coupling of 3C with deep sequencing approaches now allows the generation of high resolution genome-wide chromosomal contact maps. Different protocols have been used to generate such maps in various organisms. This includes mammals, drosophila and yeast. The massive amount of raw data generated by the genomic 3C has to be carefully processed to alleviate the various biases and byproducts generated by the experiments. Our study aims at proposing a simple normalization procedure to minimize the influence of these unwanted but inevitable events on the final results.

Results: Careful analysis of the raw data generated previously for budding yeast S. cerevisiae led to the identification of three main biases affecting the final datasets, including a previously unknown bias resulting from the circularization of DNA molecules. We then developed a simple normalization procedure to process the data and allow the generation of a normalized, highly contrasted, chromosomal contact map for S. cerevisiae. The same method was then extended to the first human genome contact map. Using the normalized data, we revisited the preferential interactions originally described between subsets of discrete chromosomal features. Notably, the detection of preferential interactions between tRNA in yeast and CTCF, PolII binding sites in human can vary with the normalization procedure used.

Conclusions: We quantitatively reanalyzed the genomic 3C data obtained for S. cerevisiae, identified some of the biases inherent to the technique and proposed a simple normalization procedure to analyse them. Such an approach can be easily generalized for genomic 3C experiments in other organisms. More experiments and analysis will be necessary to reach optimal resolution and accuracies of the maps generated through these approaches. Working with cell population presenting highest levels of homogeneity will prove useful in this regards.

Figure 1

The different steps of the original genomic 3C experiment in yeast and their associated biases [13]. A) Experimental steps. 1: Yeast cells are fixed with formaldehyde. 2: the genome is digested using a 6 cutter restriction enzyme (RE1; red double-headed arrows). 3: extraction of protein/DNA complexes and ligation in diluted conditions that favor DNA-end interactions and religation within the same complex. During this process, some RF will simply circularize (i), while others will religate in their original orientation (ii). Religation products are also expected between non-collinear restriction fragments (iii), whereas collinear RF separated by one, or more, RF will also interact together (iv). 4: de-crosslinking and DNA purification. 5: digestion of DNA products using a frequent 4 cutter restriction enzyme (RE2; black double-headed arrows). 6: DNA is ligated in diluted conditions, favoring intra-molecular circularization of single DNA molecules. Remaining linear fragments are degraded. 7: DNA circles containing a RE1 site are re-opened using RE1. 8: short DNA sequences, containing EcoP15I recognition site and a biotinylated nucleotide are added at both ends of the linear fragments. 9: circularization of linear fragments. 10: EcoP15I digestion of the DNA segments 25 bp apart from the enzyme recognition site. 11: pull-down of the DNA fragments containing biotinylated nucleotides. 12: amplification of the DNA fragment isolated and sequencing. B) Pie-chart representation of the different types of events obtained at step 3: religations, long range intra, long range inter, loops (from 50 millions pair-end sequences analyzed from the HindIII-MspI condition A and B experiments). C) Quantification of the fragment length bias. D) Quantification of the GC bias. E) Quantification of the circularization length bias.

Figure 2

Normalized intra-chromosomal contact map of S. cerevisiae. The color scale represents the normalized interaction frequencies between fragments which is calculated with the Sequential Component Normalization. A) Matrices of the sixteen chromosomes from S. cerevisiae. The strongest interactions are at the diagonale i.e. for close fragments along the chromosome. B) The normalized interaction score is calculated with the SCN method and taking into account the effect of the genomic distance. C) Zoom on chromosomes X, XI and XII. Chromosome XII is spatially segregated in two compartments by the rDNA locus.

Figure 3

Normalized inter-chromosomal contact map of S. cerevisiae. The color scale represents the normalized interaction frequencies between fragments which is calculated with the Sequential Component Normalization. A) Matrix of the sixteen chromosomes from S. cerevisiae. B) Zoom on chromosomes VII and XVI. C) Zoom on chromosomes IV and XIII.

Figure 4

Normalized inter-chromosomal contact map of S. cerevisiae.A) Inter-chromosomal contact map of chromosomal arms ranked according to their size, from the shortest (left) to the longest (right). The white empty squares correspond to specific emphasis on the five shortest arms (B), and on chromosome XII (C).

Figure 5

Receiver operating curves to assess 3D colocalization of genomic elements for the yeast contact map. Receiver operating curves (ROC) were used to assess 3D colocalization of different genomic elements. Data from Duan et al. [13] (left column) and normalized data (right column) were used. A) Centromeres, Telomeres, early origins of replication give positive signal with both types of data. B) The group of tRNA was assessed for 3D colocalization. Two clusters proposed by [13] were assessed with both data: cluster 1 of tRNA genes proposed to colocalize near rDNA and cluster 2 of tRNA genes proposed to colocalize near centromeres. The data from [13] give a positive signal contrary to the data normalized with SCN.

Figure 6

Receiver operating curves to assess 3D colocalization of genomic elements for the human contact map. Receiver operating curves (ROC) were used to assess 3D colocalization of different genomic elements for the human contacts map of Lieberman et al[8]. Non normalized data (left column) and normalized data (right column) were used. Only Telomeres give positive signal when using the non normalized data (curves for Centromeres, PolII are superimposed with the CTCF curve). When using the data normalized with SCN, all genomic elements tested give positive signal to the ROC test (curve for PolII is superimposed with CTCF curve).