Single-cell absolute contact probability detection reveals that chromosomes are organized by multiple, low-frequency yet specific interactions

At the kilo- to mega-base pair scales, eukaryotic genomes are partitioned into self-interacting modules or topologically associated domains (TADs) that associate to form nuclear compartments. Here, we combined high-content super-resolution microscopies with state-of-the-art DNA labeling methods to reveal the variability in the multiscale organization of the Drosophila genome. We found that association frequencies within TADs and between TAD borders are below ~10%, independently of TAD size, epigenetic state, or cell type. Critically, despite this large heterogeneity, we were able to visualize nanometer-sized epigenetic domains at the single-cell level. In addition, absolute contact frequencies within and between TADs were to a large extent defined by genomic distance, higher-order chromosome architecture, and epigenetic identity. We propose that TADs and compartments are organized by multiple, small frequency, yet specific interactions that are regulated by epigenetics and transcriptional state.


Introduction
Results 75 Multiple, low-frequency interactions mediate TAD assembly and insulation 76 A major mechanism for TAD formation in mammals involves the stable looping 77 of TAD borders 8 . Stable looping between TAD borders was also recently proposed to  Supplementary Fig. 1f-h). Remarkably, the width of these distributions 95 was comparable to the mean distance between TAD borders, revealing a high 96 degree of structural variability, independently of TAD size or epigenetic state (Fig. 1c 97 and Supplementary Fig. 1i). Further, the linear relation between dispersion and 98 physical distance ( Supplementary Fig. 1i-j) suggests that this variability is regulated 99 6 by the polymer properties of the chromatin fiber. 100 Next, we quantified the absolute contact probability between consecutive 101 borders by integrating the probability distance distributions below 120 nm (99% 102 confidence interval obtained from single library two-color control experiments, Fig. 1c   103 and Supplementary Fig. 1e). Notably, the contact probability between consecutive 104 TAD borders was below 10%, independently of cell type or of the epigenetic state of 105 the TAD being flanked (Fig. 1d). Consistently, Hi-C contact frequencies between 106 consecutive TAD borders vs. random genomic loci were indistinguishable (Fig. 1e). 107 These results, combined with the lack of enrichment of CTCF and cohesin at TAD 108 borders in Drosophila 20 , suggest that TAD assembly does not involve stable loops in 109 flies, but rather can be explained by an 'insulation-attraction' mechanism 21 . This 110 model may provide an alternative explanation for the formation and maintenance of 111 more than 50% of metazoan TADs whose boundaries are not formed by looping 112 interactions as defined by Hi-C experiments 8 . 113 In agreement with this model, absolute contact probabilities within TADs and 114 between their borders were similar (Fig. 1f and Supplementary Fig. 1k), with 115 inactive/repressed TADs displaying higher contact probabilities than active TADs (7 ± 116 1% vs. 2.7 ± 1%). Contact probabilities within TADs were in all cases considerably 117 higher than with neighboring TADs (Fig. 1f), indicating that stochasticity is locally 118 modulated at the TAD level. Of note, contacts across TAD borders were not 119 uncommon (~3%, Fig. 1f), implying frequent violations of boundary insulation at TAD 120 borders. These results indicate that confinement of chromatin into TADs may require 121 only small differences in absolute contact probabilities (~2 fold). Thus, condensation 122 of chromatin into TADs may arise from a multitude of low-frequency, yet specific, 123 intra-TAD contacts. 7 Infrequent, long-range interactions modulate chromatin folding 125 Recent Hi-C studies suggested that stable clustering between neighboring 126 active TAD borders regulates transcriptional programs that persist during 127 development 7 . We directly tested this hypothesis by measuring the contact 128 probabilities between non-consecutive TAD borders (Fig. 2a). Hi-C contact 129 frequencies among TAD borders increased nonlinearly with absolute contact  Supplementary Fig. 2d), in accordance with our microscopy results. Furthermore, the 147 frequency of contacts between non-consecutive TAD borders genome-wide was 148 similar to that of random genomic loci for both cell types (Fig. 2d)

243
To quantify this phenomenon, we calculated the percentage of compartments In this work, we showed that genome organization in Drosophila is not driven 261 by stable or long-lived interactions but rather relies on the formation of transient, low-262 frequency contacts whose frequencies are modulated at different levels. Stochasticity 263 is modulated locally at the TAD level by specific intra-TAD interactions, and globally 264 at the nuclear level by interactions of TADs of the same epigenetic type.

265
Furthermore, stochasticity is also regulated between cell-types. These modulated 266 stochasticities reveal a novel mechanism for the spatial organization of genomes.    in nm for a specific chromatin type and cell type (Fig. 4f-g).

511
Clustering of domains of different epigenetic marks was defined as the ratio In-situ Hi-C data processing and normalization 517 Hi-C data was processed using an in-house pipeline based on TADbit 61 . First, quality 518 of the reads was checked using the quality_plot() function in TADbit, which is similar 519 to the tests performed by the FastQC program with adaptations for Hi-C datasets.

520
Next, the reads are mapped following a fragment-based strategy as implemented in 521 TADbit where each side of the sequenced read was mapped in full length to the 522 reference genome (dm3). After this step, if a read was not uniquely mapped, we 523 assumed the read was chimeric due to ligation of several DNA fragments. We next 524 searched for ligation sites, discarding those reads in which no ligation site was found.

525
Remaining reads were split as often as ligation sites were found. Individual split read 526 fragments were then mapped independently. Next, we used the TADbit filtering  Table 6).