(12/12) This has been a wonderful and close collab with @bloodgenes.bsky.social & lab. All credit to Varshini who led all computational work and put it all together and to Chun-Jie who led many experiments with important contributions from all the authors. We'd love feedback and discussion!

(11/n) Other "3D CREs" have been described incl. beautiful work on Facilitators (Kassouf/Higgs) & Range Extenders (Kvon). A key distinction, Matchmakers promote looping indirectly: Matchmakers mostly don't themselves form loops. Instead they promote crosser loops through targeted cohesin loading.

(10/n) Due to key role of cohesin in E-P regulation www.biorxiv.org/content/10.6... , it may be very important to allocate extrusion near key genes in erythropoiesis, where most of the genome is compacted and silenced and an increasingly specialized gene expression program is adopted.

(9/n) Working in primary human donor-derived cells prevents degron tagging of cohesin. But we perturbed STAG1/2&NIPBL in primary cells and STAG2 in myeloid cells and re-analyzed Blobel lab NIPBL&SMC3 erythroblast data. All perturbations consistent with Matchmaker cohesin-dependence.

(8/n) Matchmakers strengthen during erythropoiesis and correlate with stronger expression of key erythroid genes. KO of key matchmaker-binding TF NFE2 leads to modest reduction in matchmaking and nearby gene expression.

(7/n) Inspired by fountain work, we tested if moderate targeted cohesin loading can explain Matchmakers: stronger crosser loops without visible fountains. Varshini confirmed this w polymer sims. Interestingly, erythropoiesis lead to chromatin compaction, and compaction hides fountains in 3D maps.

(6/n) In contrast to fountains, Matchmakers do not show "fountain pattern" at single loci. Instead, Matchmakers are composed of “crosser loops”. Matchmakers are "altruistic": they help nearby elements form loops largely without themselves engaging as a loop anchor.

(5/n) Prev work identified “orthogonal extrusion stripes” in 3D maps named “plumes” (de Wit), “jets” (Merkenschlager), “fountains” (Mirny, Meister, Ercan et al) and similar aggregate patterns (Vahedi, Xue, et al) Key distinction: Fountains are extrusion stripes at single loci, but not Matchmakers.

(4/n) As expected, pile-up analysis on CTCF sites and promoters show clear insulation. Varshini clustered eryTFs into 4 clusters and saw strong “anti-insulation” at clusters 3+4. We call cluster 3+4 eryTF "MatchMakers" because they match make loops across themselves, without forming loops directly

(3/n) Recently @bloodgenes.bsky.social lab ID’d key ‘eryTF’ CREs www.science.org/doi/full/10.... When Varshini examined 3D structure around these CREs she discovered “anti-insulation” pattern: Loops that cross these CREs are stronger than loops that don't --> This is opposite of insulation.

(2/n) Collab with @bloodgenes.bsky.social began with genetic variation: We mapped ultra-high-res 3D genome structure across erythropoiesis using primary human donor derived cells. We can see the E-P loop that FDA-approved cure for Sickle cell disease targets (Fig 1D), but we also see new loops!

We know little about proteins required for long-range compartment 3D interactions, but this @kyleeagen.bsky.social lab preprint shows that NSD3 can make Mb-scale long-range compartment-like interactions: www.biorxiv.org/content/10.6...

Masahiro and I were fortunate to contribute some RCMC analyses to this beautiful paper from Koska and Wysocka that comprehensively dissects the determinants of promoter competition: www.nature.com/articles/s41...

@mileshuseyin.bsky.social and the lab have put together a comprehensive protocol for genome-wide Micro-C and for Region-Capture Micro-C in @natprot.nature.com : www.nature.com/articles/s41... See also the GitHub for a user-friendly end-to-end computational pipeline: github.com/ahansenlab/M...

And thanks so much to Adrian Henggeler and @fenaochs.bsky.social for writing such a thoughtful News&Views: www.nature.com/articles/s41...

The overall conclusion remains the same: Looping probabilities are globally very rare: mean is 1.2% in mESCs and 2.2-2.8% in 4 human cell lines using Micro-C data from www.biorxiv.org/content/10.1... Consistent with E-P loops forming through transient contact: www.biorxiv.org/content/10.6...

We have generated a "mESC mega merge" Micro-C dataset with 54 Billion (!) unique ligations. This map is freely available on GEO. We have also annotated 65,929 consensus loops in mESCs (CTCF, E, P, other) available as a table. The 54B Micro-C map approaches RCMC-resolution genome-wide!

We use BILD-quantified live-cell imaging data to calibrate Micro-C to get absolute quantification (e.g. this loop is looped 5% of the time). We now have 5 calibration points instead of 3, increasing robustness. The synTAD datapoint is the 339CECP cell line from www.biorxiv.org/content/10.6...

Excited to see James' Genome-wide Absolute Quantification of Looping paper out in @natsmb.nature.com : www.nature.com/articles/s41... This has been in collaboration with @lucagiorgetti.bsky.social @leonidmirny.bsky.social @zechnerlab.bsky.social labs. Brief thread below on some key updates

Very interesting new preprint from @jengreitz.bsky.social lab arguing that rather than there being significant enhancer-promoter compatibility, promoters simply differ in their enhancer responsiveness. If promoters are responsive, they respond to all enhancers. www.biorxiv.org/content/10.6...

(15/n) This has been a very close tri-lab collab with @leonidmirny.bsky.social @zechnerlab.bsky.social All credit to - Harvey led experiments and many analyses and LSTM ML - Henrik led inference and developed VEPI - Jack developed Fyrtarn, lattice-processing pipeline - And rest of team!

(14/n) LIMITATIONS - We studied n=1 E-P pair --> generality is TBD - CRE-rich many E and many P regions may behave differently - Each of 5 estimates has limitations: ~25-42 nm contacts that last ~10-20 sec is our best estimate, but there is uncertainty - Same for 0.3-1 sec Time Gate - Others too.

(13/n) Our data points to time gating - E-P contacts too brief (<0.3-1sec) appear to be txn unproductive --> this is likely important for E-P selectivity See also cited work from @lucagiorgetti.bsky.social @elphegenoralab.bsky.social @leonidmirny.bsky.social Dan Larson and others in this area!

(12/n) t(E-P) ~ 10-20 sec Discussion - E-P contacts are transient (~10-20 sec) but still stabilized above and beyond random diffusive contacts - Consistently, we see E-P dots in RCMC even without CTCFs. - t(E-P) matches residence time of many TFs and general transcriptional machinery

(11/n) R(E-P) ~ 25-42 nm Discussion - Isotropic action-at-a-distance is difficult to reconcile with E-P selectivity due to volume density of Es+Ps in nucleus. - Direct E-P bridging by txn complexes consistent with biochemistry - But we studied <2kb E, long super-enhancers may behave very differently

(10/n) Integrating all 5 estimates: - R(E-P) ~ 25-42 nm - t(E-P) ~ 10-20 sec - Time Gate ~ 0.3-1 sec - All point to transient contact mechanism for E-P interactions explaining why they are easy to miss. - W. @voslab.org did structural modeling --> 25-42 nm consistent w. e.g. Mediator-dimerization

(9/n) ESTIMATE 5 - Perturbations - Cohesin depletion --> near-complete loss of txn - Add insulating CTCF sites --> 90-97% drop in transcription - Polymer sims can only explain if R(E-P)~36 nm and by adding Time Gate: E-P events <0.3-1sec get filtered out and are transcriptionally unproductive.

(8/n) ESTIMATE 4 - Can a minimal mechanistic model explain the data? - VEPI: Variational E-P inference fully parameterizes mechanistic model - Captures cross-correlation between E-P contact and MS2 bursts - Estimates t(E-P) < 20 sec consistent with transient contact.

(7/n) ESTIMATE 3 - Does E-P proximity precede transcriptional bursts? - Train LSTM ML model on live-cell E-P+MS2 trajectories - E-P proximity predicts bursts - E-P proximity events last ~15 sec t(E-P) - E-P events match co-localization control --> R(E-P) ~ contact to 30 nm

(6/n) ESTIMATE 2 - Adding CTCF sites to E and P of 339kb E-P pair increases expression 20x. - But it only mildly decreases E-P 3D distance from median 246 nm to 167 nm. - Can only explain this if R(E-P) is very small, we estimate ~29 nm.