Emerging Evidence of Chromosome Folding by Loop Extrusion

1Mammalian interphase chromosomes exhibit both cell type- and locus-specific organizations that manifest characteristic patterns on Hi-C maps. These include square areas of enriched contact frequency along the diagonal, termed topologically associating domains (TADs), often elaborated with peaks at their corners, grids of peaks within and between TADs, and enriched lines or tracks of contact frequency emanating from a boundary (Fig. 1C; for reviews, see; ). We distinguish TADs from compartmental segments of the genome, which also appear as squares along the diagonal of Hi-C maps but differ in that they associate to form a checkered pattern in cis and in trans. Indeed, TADs, peaks, and tracks have an independent mechanistic origin from the patterns associated with the compartmental segregation of active and inactive chromatin, and we discuss the interplay of these two mechanisms elsewhere. TAD boundaries are frequently demarcated by binding sites of the transcription factor CTCF, and are enriched for the structural maintenance of chromosomes (SMC) complex cohesin. Functionally, TADs are believed to demarcate coherent cis neighborhoods of gene-regulatory activity and hence are crucial for development. To explain how such neighborhoods can be formed, we put forward a mechanism based on a still-hypothetical process of loop extrusion.

2Here we present emerging evidence that interphase chromosomes are organized by loop extrusion, an active ATP-dependent process that allows nanometer-size molecular machines to organize chromosomes at much larger scales. We review how loop extrusion by cohesins can explain the formation of TADs, peaks, and tracks visible in interphase Hi-C maps. We then detail specific predictions made by the polymer model of loop extrusion, and discuss recent experimental perturbations to CTCF and cohesin that test these predictions and provide strong support for the loop extrusion mechanism. Although we focus on comparisons to mammalian interphase Hi-C experiments, loop extrusion likely plays important roles in other organisms and parts of the cell cycle. We also discuss imaging experiments, single-molecule experiments, and a possible molecular mechanism of loop extrusion.

3We frame our discussion around how we originally implemented the mechanism of loop extrusion limited by directional barriers as a polymer model. In the process of loop extrusion, loop extruding factors (LEFs) translocate along the chromosomes, holding together progressively more genomically distant loci along a chromosome, thus producing dynamically expanding chromatin loops (see Supplemental Movie 1).

4LEF translocation is either halted by encounters with other LEFs or probabilistically halted at specific genomic loci that contain extrusion barriers. We assume that if halted only on one side, a LEF may continue to extrude chromatin from its other side. LEFs continue to extrude until they dissociate from the chromatin fiber, releasing the extruded loop, as they dynamically exchange with the nucleoplasm.

5The minimal system of LEFs limited by extrusion barriers that we implement is defined by four parameters (Fig. 1A):

6lifetime on chromatin (sec)

7velocity along the chromatin fiber (kb/sec)

8separation between LEFs (kb)

9permeability of the extrusion barriers (probability)

10For comparison to ensemble-averaged Hi-C experiments, that capture a snapshot of contacts occurring at a particular point in time, it is also useful to define the product of lifetime and velocity, processivity (kb), which indicates the average size of a loop that a LEF would extrude if left unobstructed. Motivated by observations of CTCF motif orientations at TAD boundaries and at peaks, we implement barriers as being directional, (i.e., halting LEFs approaching it from only one side). Barriers can be modeled as either halting LEFs as long as the blocking factor is present, or stalling them until LEF dissociation (see Supplemental Movie 1). In our models, the permeability can be thought to represent the probability that a barrier locus is occupied by a blocking factor.

11To compare predictions from our simulations with experiments, we generate a simulated ensemble of chromatin conformations for a given set of parameters (Fig. 1B). To accurately capture features of chromatin folding at high resolutions we typically use monomers representing several nucleosomes to simulate 10–50 Mb of chromatin. From these conformations we can extract experimentally relevant observables. These include maps of contact frequency that can be compared to Hi-C contact maps, as well as distributions of spatial distances between pairs of loci, that can be compared with FISH experiments. From the simulated contact maps, we can then quantify features such as TADs, peaks, and contact frequency decay, as done for experimental Hi-C maps. By comparing simulated and experimental features, we can then define a set of wild-type parameters, from which perturbations, and hence predictions, can be made.

12The mechanism of loop extrusion limited by directional barriers recapitulates many features of interphase chromosome folding visible in Hi-C maps (Fig. 1C), including:

13TADs: regions of enriched contact frequency between neighboring barriers

14Tracks: lines emerging from one side of a barrier

15Peaks and grids of peaks, occurring between proximal barriers in cis but not between chromosomes

16Presence of inward-oriented CTCF motifs at TAD boundaries and at peak bases

17Further support comes from site-specific disruptions of TAD boundaries and peak bases, which respectively result in merging of adjacent TADs and orientation-dependent losses of peaks. To our knowledge, no alternative mechanism of interphase chromosome organization currently agrees with all the above.

18Although we focus here on interphase loop extrusion, we note that loop extrusion by SMCs appears to have important consequences in mitosis, where the term was coined and first mathematically modeled. The closely related concepts of reeling, facilitated tracking, loop expansion and progressive loop enlargement have a rich history. Loop extrusion also appears relevant in bacteria. There are also related proposals for interphase loop extrusion, which we discuss briefly below.

19We note that although the terms “contact,” “loop,” and “interaction” are often used interchangeably in the chromosome organization literature, they are often used to describe very different features of Hi-C contact maps. In the context of loop extrusion, we reserve the term “loop” in the very narrow sense of two regions of a continuous chromatin fiber brought together by a LEF at a given point in time. Moreover, simulations and data analyses show that peaks of contact frequency in interphase Hi-C maps are not consistent with stable chromatin loops. Therefore, we refrain from using “loop” to describe any feature of Hi-C contact maps.

20The stochastic nature of loop extrusion poses an experimental challenge for testing predictions from the model. Extruded loops are not directly visible via population-average Hi-C approaches because they are located at different genomic positions in different cells at any given time. Even with single-cell Hi-C methods an individual pair of loci linked by an extruding loop would not appear particularly different from any other captured contact. Visualization of extruded loops by microscopy is similarly challenging due to their continually changing locations both along the genome and in 3D space. Direct confirmation that a particular chromatin loop has been extruded in vivo will require methods that can simultaneously track multiple DNA loci as well as the loop extruders themselves. Nevertheless, much of the strongest evidence to date supporting the role of loop extrusion in interphase comes from changes in Hi-C maps upon perturbations that affect specific components of the loop extrusion machinery.

21To make experimental predictions, we must first identify components of the interphase loop extrusion machinery with their biological candidates. Several lines of evidence make us hypothesize that cohesin complexes play the role of LEFs, and CTCF plays the role of an extrusion barrier. Cohesin is enriched at TAD boundaries in interphase and is highly homologous to condensins, the main complexes responsible for compacting mitotic chromosomes. CTCF is enriched at TAD boundaries at preferentially oriented motifs, and, compared with other transcriptional regulators, binds relatively stably to its cognate sites (for review, see ). With these identities, we discuss how our model of loop extrusion predicts different outcomes for three perturbations: depletion of CTCF, depletion of cohesins, and increased processivity of cohesins (Fig. 2).

22For the depletion of the LEF, cohesin, our simulations display two phenomena (Fig. 2B) (i) the loss of TADs and associated Hi-C peaks; and (ii) decompaction of chromatin at the scales of individual extruded loops (<200 kb). Changes in local compaction, in turn, can be studied by observing changes in the contact probability, P(s), as a function of genomic separation, s. Local compaction is seen as a region of P(s) with a shallow slope (~100–500 kb), which we refer to as the shoulder (Fig. 2A); decompaction leads to reduction or loss of the shoulder region. We note that our models predict that a sharp decrease in LEF processivity would similarly lead to a loss of TADs, peaks, and compaction.

23For the depletion of site-specific extrusion barriers, as imposed by CTCF, our simulations also predict the loss of TADs and associated Hi-C peaks (Fig. 2C). However, our simulations predict that other consequences of this perturbation should be very different from depletion of LEFs. This is because in our model, extrusion barriers only impose an instructive function (i.e., their major effect is on the localization of extruded loops rather than on their sizes or abundance). We therefore predict little effect on overall compaction, and hence little change in the P(s) curve. This differentiates our predictions for CTCF depletion from those for cohesin depletion.

24For the depletion of a cohesin unloading factor, like Wapl, our model predicts that the consequent increased processivity and number of LEFs would lead to several phenotypes (Fig. 2D): (i) peaks at corners of TADs become stronger and appear between more distal barrier loci, creating extended grids of peaks; (ii) the orientational preference of barrier loci will become weaker, as LEFs halted at a directional barrier for long durations can stop traffic from the opposing direction as well. Finally, (iii), our model predicts that sufficiently increased coverage by extruded loops will overcompact chromosomes. In Hi-C this would be detected as an extension of the shoulder in P(s), as opposed to how it recedes in the case of cohesin depletion. Macroscopically, sufficient compaction would cause chromosomes to condense into a prophase-like state with a cohesin-rich central scaffold.

25Crucially, our model predicts that the loss of cohesin loop extruders and the loss of CTCF extrusion barriers should both lead to the loss of TADs and Hi-C peaks, yet in completely distinct fashions. Furthermore, increased processivity of cohesin extruders is predicted to manifest distinct phenotypes on Hi-C maps and macroscopic chromosome organization.

26Whereas perturbing CTCF and cohesin dynamics is crucial for testing predictions of loop extrusion, depletion of such essential complexes poses many experimental challenges. For CTCF, cells begin dying after ~4 days of stringent depletion. For cohesin, there are additional challenges related to its role in sister chromatid cohesion and chromosome segregation during mitosis, and its multiple dynamically exchanging subunits and regulators that can be present in different abundances and likely have unique impacts on loop extrusion dynamics. Despite these challenges, recent studies have achieved modulation of cohesin and CTCF that result in dramatic changes, consistent with predictions from polymer models of loop extrusion (Table 1).

27Consistent with our predictions for decreasing the number of active LEFs, depletion of the cohesin loader Nipbl (Scc2) and acute degradation of the cohesin kleisin Rad21 (Scc1) during interphase led to both: (i) complete erasure of TADs and Hi-C peaks (ii) and decompaction, as evidenced by loss of the P(s) shoulder (Fig. 3A). Decompaction is further supported by imaging, showing loss of H2B clustering by PALM following both RNAi knockdown of NIPBL and AID-mediated degradation of Rad21. We note that earlier Hi-C studies saw limited impact following the depletion of Rad21, potentially due to incomplete depletion.

28A corollary of the Nipbl depletion result is that cohesin must be constantly loaded on chromatin to maintain TADs and associated corner peaks. Consistently, TADs and Hi-C peaks are both rapidly lost upon AID-mediated degradation of Rad21 (<3 h) and reestablished after auxin wash-off (40–60 min). These consequences follow directly from our loop extrusion models, and the turnover time of cohesin in G1 (~5–30 min).

29Future studies will be useful to dissect the dynamics of the processes and the potential role of Nipbl beyond that of a loader. In particular, although Nipbl depletion appears to have a dramatic effect on extrusion, knockout of its cofactor Mau2 (Scc4) appears to have a much weaker effect on loading yet a fairly strong effect on processivity. Moreover, we note that different components of the interphase extrusion machinery could be limiting at different concentrations and in different contexts. We hypothesize that, via its consequences on loop extrusion, modulation of the levels of various cohesin subunits and interactors can serve to fine-tune overall gene regulation across cell-types and tissues.

30Consistent with our predictions for the loss of site-specific barriers to extrusion, acute auxin-induced degradation of CTCF in mESCs and HeLa cells led to a dramatic loss of TADs and Hi-C peaks (Fig. 3B). However, the P(s) curve did not change, implying that although demarcation of contact-insulating boundaries in Hi-C maps was lost, the same degree of chromatin compaction was maintained. In support of the dynamic exchange of LEFs in our model, the effect of CTCF depletion was fully reversible following a 2-day wash-off period. We note that stringent dosage depletion was necessary to observe dramatic insulation defects: even a 15% preservation of CTCF showed a relatively mild phenotype. Similar loss of TADs and peaks were reported in vivo for an inducible CTCF knockout in cardiomyocytes. Weaker effects have also been reported recently and earlier, but this may have been due to relatively inefficient depletion or lower starting levels of CTCF.

31The predicted lack of decondensation following CTCF depletion is further supported by imaging. PALM shows little difference in H2B clustering. Imaging of FISH probes at selected loci upon CTCF degradation show that inter-TAD distances increased, whereas intra-TAD distances remained the same. Together these results are consistent with global compaction levels being unchanged but with diminished insulation across CTCF sites. Importantly, the lack of chromatin decompaction in CTCF depletion rules out models in which CTCF is strictly required for the loading of chromatin-bound cohesin and any ensuing cohesin-mediated loops. Instead, the differences in imaging and Hi-C maps upon CTCF versus cohesin depletion are consistent with the loop extrusion model we describe, in which CTCF barriers serve an instructive function and cohesin is loaded onto chromatin and can compact chromosomes through extrusion even in the absence of CTCF.

32Consistent with our predictions for increasing the processivity and density of active LEFs, depletion of the cohesin unloader Wapl led to multiple phenotypes observed in Hi-C maps and by imaging. For Hi-C (Fig. 3C) this includes (i) strengthened peaks at TAD corners, (ii) emergence of new peaks between boundaries at greater separations, creating extended grids of corner peaks; (iii) a weakened correspondence between these features and CTCF motif orientation. Increased local compaction upon Wapl depletion is reflected by (iv) extension of the shoulder in the P(s) curve and, (v) the emergence of prophase-like vermicelli chromatids via imaging. This remarkable observation provides further evidence for a universal molecular mechanism—loop extrusion—underlying both metaphase and interphase chromosome organization.

33Depletion of another component of the cohesin unloading machinery, Pds5A and Pds5B (Pds5A/B), led to many of the same phenotypes. However, there were also intriguing differences that may prove instructive for determining exactly how CTCF halts the progression of cohesin along the chromosome—for example, Pds5 may instruct directional cohesin stalling, and competition between the two HAWK family proteins, Nipbl and Pds5, may regulate cohesin translocation velocity. The observation that Wapl depletion appears to largely rescue the Hi-C phenotype of Mau2 depletion provides further support to the proposal that the Nipbl/Mau2 “loading complex” also has roles in promoting cohesin processivity for loop extrusion. Finally, consistent with loop extrusion simulations with increased processivity, the joint depletion of Wapl and Pds5A/B showed even stronger effects in terms of shifting the shoulder in P(s) and in the emergence of vermicelli.

34Collectively, the congruence of both Hi-C and imaging experiments following the perturbation of CTCF and cohesin dynamics strongly supports the role of loop extrusion in interphase. Future simulations and experiments will be valuable for probing the consequences of multiple simultaneous perturbations.

35Although providing strong support for chromosome folding by loop extension in vivo, the studies discussed above do not directly probe the molecular details of loop extrusion. Molecularly realizing the process of loop extrusion presents a considerable challenge, namely, that the protein complexes performing loop extrusion need to track consistently in cis along chromatin, over large distances (up to tens-of-thousands of nucleosomes) without falling off. Moreover, the substrate, chromatin, is highly disordered due to nucleosomes and other DNA-bound proteins, likely posing a greater challenge than tracking along microtubules performed by cytoplasmic motors. Here we discuss how recent single molecule experiments argue that loop extrusion likely occurs via an active process, driven by molecular motors. Although many of these observations were made with condensin and bacterial SMCs, they illustrate that loop-extrusion is a plausible mechanism of action for the whole family of SMC proteins, including cohesin.

36Recently, showed that a single yeast condensin has motor activity and is able to translocate processively along naked DNA in vitro. Using a DNA curtain assay, they found individual condensin complexes travel unidirectionally, rapidly (~4 kb/min) and processively (~10 kb) in an ATP-dependent manner with 10 nm steps (30 bp on naked DNA). As previous single-molecule studies only reported sliding dynamics of SMCs (;;;; for review, see ), the directional translocation observed by is incredibly important.

37The high structural homology of cohesin to condensin makes it likely that the same physical mechanism would govern its processive motion, in addition to its established role of mediating sister chromatin cohesion. Indeed, the ability of these SMCs to compact chromosomes appears to be remarkably coherent over evolutionary timescales and cellular contexts. Due to its dual roles, and more elaborate set of subunits, however, reconstituting this activity for cohesin may be more difficult in vitro. Nevertheless, we believe that the in vitro observations of ATP-driven processive condensin translocation argue against the likelihood of motor-free mechanisms of SMC processivity in general, including for cohesin.

38Although strongly supporting the loop extrusion mechanism, the single-molecule experiments leave open several questions of how loop extrusion can work in vivo:

39How can SMCs translocate on chromatinized rather than naked DNA?

40How can translocation result in loop extrusion?

41Is the measured speed of translocation sufficient to generate TADs and peaks?

42Do cells have sufficient ATP budgets to support extrusion during interphase?

43In particular, it remains to be understood how SMC complexes can translocate on chromatin fibers rather than naked DNA. Translocations while maintaining constant contact with DNA may not always be possible due to the complexity of chromatin fiber and abundance of DNA-bound proteins. Although the size of an SMC complex (~50 nm) exceeds that of a single nucleosome (~10 nm), nucleosomes would constitute challenging obstacles for SMC translocation if maintaining constant contact with DNA is required for translocation.

44A possible solution comes from the structural similarity of SMC domain organization to that of kinesin and myosin motors that walk on microtubules and actin, which suggests that SMCs can similarly walk on chromatinized DNA. Importantly, a walking mechanism would allow translocation where obstacles such as nucleosomes and other DNA-bound proteins can be passed over, avoiding disruptions of the underlying nucleosomal array. During each step of the walking process, one SMC head can remain DNA-bound, whereas the other hops forward and rebinds nearby DNA (Fig. 4A). SMC walking is consistent with the rapid and flexible dynamics of their arms, and the 10 nm step size would allow passing over nucleosomes (e.g., by hopping from linker to linker) and other DNA-bound complexes, avoiding the need for unwinding nucleosomal DNA or nucleosome eviction (Fig. 4B).

45A walking mechanism would be greatly aided by the known ability of SMCs to topologically entrap DNA, which can ensure that the walker tracks in cis, along the same chromatin fiber (Fig. 4C). Pseudo-topological entrapment can similarly help maintain extrusive cohesins on the same DNA molecule over long genomic distances. In other words, SMC complexes may translocate along the chromatin fiber and accomplish loop extrusion as shackled walkers (Fig. 4A).

46An important open question is how CTCF, and possibly other chromatin-bound proteins, can halt cohesin translocation whereas nucleosomes do not, when they are fairly similar in size. Although they probed diffusive sliding rather than processive tracking dynamics, report that cohesin can rapidly slide over some DNA-bound proteins and nucleosomes, but becomes obstructed by DNA-bound CTCF and transcriptional machinery; a similar, yet more restrictive, dependence of sliding on the size of DNA-bound factors has been reported in other single-molecule studies probing sliding dynamics. This suggests that CTCF blocks translocation of cohesin by a specific mechanism rather than by steric exclusion—for example, by inhibiting the ATPase action of the cohesin machinery directly or via other cohesin interactors (e.g., via Pds5) and potentially in concert with cofactors. Alternatively, CTCF may recruit additional cofactors to increase its physical size or pose a greater challenge for walking due to its DNA binding geometry.

47Multiple possibilities exist as to how the translocation of motor complexes along a chromosome can realize the process of loop extrusion. These include (1) a single translocating motor attached to a chromatin anchor; (2) two connected motors translocating in opposite directions; (3) a single motor that switches between two chromatin fiber substrates (Fig. 4D). These architectures for the action of SMC motors can lead to different consequences for the processive dynamics of extrusion. Unidirectional extrusion could result from a single motor-and-anchor architecture. Bidirectional extrusion would emerge from the latter two possibilities. We note there are multiple possibilities for how many SMC complexes are required to realize motor activity (Fig. 4E), either as monomers or, potentially, oligomers. One advantage of two-motor extrusion is that it naturally allows one motor to continue extruding if the other becomes blocked. Because models discussed here and elsewhere (in;;,) assumed independent bidirectional extrusion, it remains unclear if one-sided loop extrusion is sufficient to form TADs, peaks, and tracks, as well as to compact mitotic chromatids.

48The measured rates of stepping and step sizes for condensin agree well with the expectations of the loop extrusion theory in interphase for cohesin. Using ~2 steps/sec and ~10-nm step size measured in vitro, this gives ~18 kb/min if cohesin moves one nucleosome per step (~150 bp). This is further doubled if cohesin extrusion occurs via a two-motor mechanism, yielding ~36 kb/min. These values are compatible with the ~10–30 kb/min predicted by polymer models as sufficient to generate TADs and corner peaks in vivo. There are several ways to arrive at this estimate. The first involves dividing the size of the largest TADs (~1 Mb) by time to reestablish TADs following exit from mitosis (~0.5–2 h, ~30 min) or following auxin wash-off (~30 min). Alternatively, one can use the processivity of cohesin estimated from fitting Hi-C data with loop extrusion models (~200–400 kb), and divide this by the cohesin turnover time (~5–30 min).

49We note that pushing by RNA Pol II alone, at its reported velocities, would be too slow (~1.5–3 kb/min). The observation of cohesin-dependent features in both active and inactive chromatin, as well as the transcriptionally inactive maternal zygotic pronucleus, further argues against Pol II providing the primary motive force for loop extrusion.

50A simple estimate shows that the energy burden of ATP consumption by loop-extruding cohesins in interphase is negligible as compared to ATP production in a mammalian cell. Again using 2 ATP per sec per SMC complex, and the total number of actively extruding cohesin molecules, either measured (~100,000 per cell [Hansen AS, pers commun]) or estimated from fitting simulations to Hi-C data (~1 loop-extruder/200 kb, i.e., ~60,000/diploid G2 cell), one obtains a very low rate of ATP consumption (<2 × 105 ATP/sec). This constitutes <0.02% of the 109 ATP/sec production rate by a fibroblast. Thus the energy burden of chromosome organization by cohesin is marginal.

51While in proofs, a paper appeared that reported a direct observation of loop extrusion in vitro by single purified yeast condensin complexes on DNA. In their experiments, condensins extruded loops of up to tens of kilobases at a speed of up to 1.5 kb/sec in an ATP-dependent fashion. Surprisingly, the extrusion observed was strictly one-sided, which prompts further investigation. Overall, this exciting new study provides the first direct evidence of active loop extrusion by SMC complexes.

52Although the key role of molecular motors in the cytoplasm is broadly appreciated, there is now a growing appreciation for loop extrusion by SMC complexes as an active processes organizing and compacting chromatin in the nucleus. Analogous to the myriad uses for the contractile dynamics of active actin and tubulin networks, we hypothesize that interphase loop extrusion has been repurposed for a variety of biological ends, including targeting VDJ recombination, and regulation of enhancer–promoter interactions.

53Hi-C maps and videos are available at http://mirnylab.mit.edu/projects/emerging-evidence-for-loop-extrusion.