Oncofusion

start / graph / hyp-ctcf-occupancy-splits-break-formation-from-rearrangement-fate

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q105-ctcf-adjudication

CTCF/cohesin occupancy at the KMT2A exon-11 anchor is both the cause of the break and the chaperone of its correct repair - so losing it gives FEWER breaks and MORE rearrangements

also called the formation-versus-fate hypothesis · occupancy as ligation chaperone · more rearrangements from fewer breaks

The same protein clamp that invites the cutting enzyme to the KMT2A gene may also be what holds the cut ends together so they can be sewn back correctly. If so, taking the clamp away produces fewer cuts, but each cut is far more likely to end up as a cancer-causing rearrangement.

The same thing, for a scientist

Two published models make CTCF occupancy at one 1 kb element look contradictory - a TOP2 break hotspot on one account [E2][E3], a rearrangement brake on the other [E1]. They conflict only through an untested auxiliary premise, that rearrangement rate follows break rate at this element. **This hypothesis says that premise is false here, and names the reason: the anchor is a ligation chaperone for the break it causes.** CTCF/cohesin occupancy recruits and traps TOP2, so occupancy raises the break rate [E2]; and the same complex holds the cut ends in register, so occupancy lowers the fraction of breaks that resolve by illegitimate joining. The net effect of removing it is therefore fewer breaks with a higher conversion ratio, and the published sign of the net - more inherited copy gain and probe separation after siCTCF [E1] - is an empirical fact that needs no contradiction of the fragility model. **This is not a bare possibility.** The step it posits is already separated mechanistically at this class of element - trapping of the TOP2 cleavage complex and its conversion to an irreversible break are distinct processes with distinct requirements, conversion depending on proteasomal degradation and TDP2 phosphodiesterase activity [E2] - and the direction it posits is already demonstrated at this locus for another factor: TDP2 loss raises TOP2-induced translocations at *MLL* by changing how breaks are rejoined rather than how many form, and TDP2-dependent joining even creates a specific therapy-related junction class [E4]. **Prior probability is set at 0.35** - the mechanism is plausible and precedented and the specific claim about CTCF at this element rests on nothing directly measured.

Go deeper
The statement, and the two numbers it predicts

At GRCh38 chr11:118,488,159-118,489,206 - a CTCF+RAD21+SMC3 co-occupied anchor spanning Ensembl canonical KMT2A exon 11 5:

  • dB/dO > 0: break rate B increases with occupancy O, because occupancy recruits and traps TOP2 23.
  • dR/dB|O < 0 in O: the conversion of a break to an inherited rearrangement R falls as occupancy rises, because the complex tethers the ends.
  • Net: dR/dO = (dR/dB)(dB/dO) + B·d(dR/dB)/dO can be negative even though dB/dO is positive. Removing occupancy predicts B down, R up, R/B up.

The quantity that has never been reported, in any system, for any factor, is R/B at this element. Every paper in this dispute reports B or R and none reports both 123.

Why a tethering role is the natural candidate rather than an invented one. Cohesin is a tethering complex; the anchor is where two chromatin segments are held together; and TOP2 cleavage at an anchor happens inside a structure whose job is proximity. A break made inside a holder is a break whose ends do not diffuse. Nothing in this argument requires a new activity - it requires only that the structure already described as the cause of the break also constrains its ends, which is the default expectation for any protein-DNA clamp and is the opposite of what the fragility literature has had reason to test.

What would kill it

Stated as results, not hedges. All at the same element, in the same cells.

  1. A CTCF degron that lowers breaks AND lowers rearrangements. Then occupancy is not doing two jobs; it is doing one, and Gray et al.'s siCTCF result is an artefact of knocking down a near-pan-essential protein. This is the cleanest kill and it is the primary endpoint of exp-ctcf-degron-breaks-and-rearrangements-same-cells-kmt2a.
  2. A CTCF degron that raises breaks. Then proposition (i) fails for CTCF at this element, the reconciliation is unnecessary, and the fragility class statement simply does not generalise here. The hypothesis is not needed and should be withdrawn rather than rescued.
  3. R/B unchanged under any occupancy perturbation. The hypothesis is specifically a claim about the ratio. If the ratio is invariant while B and R move together, formation and fate are not separable at this element and the premise this hypothesis attacks was true after all.
  4. The anchor is not occupied in any haematopoietic cell. Then the whole dispute is epithelium-specific, and a hypothesis about how the anchor behaves in leukaemogenesis has no substrate. Every occupancy statement available is from ENCODE cross-cell-type clusters or from RPE epithelium 5.
  5. The rearrangement events are not rearrangements. Gray et al.'s readouts are FISH probe separation and copy number, and the authors state the second hit is unidentified. If junction sequencing shows the events are extrachromosomal or non-reciprocal products with no KMT2A fusion junction, then R in this hypothesis is not the quantity the disease is made of, and the hypothesis is about a different phenomenon than the one it claims to explain 1.
  6. A junction-class shift with no rate change. If CTCF loss changes the junction homology signature without changing event counts, then occupancy governs the kind of repair and not its fidelity, and the hypothesis should be restated in those terms - which is exactly the pattern TDP2 shows at this locus 4.
  7. Cohesin removal moves R in the opposite direction to CTCF removal. Then the chaperone is CTCF specifically, not the anchor, and the structural argument that motivates the hypothesis is wrong even if its prediction survives.
What it would explain if true, and what it would cost

It would collapse two entries in this graph into one mechanism. mechanism-loop-anchor-dsb-hotspots and mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement would stop being rivals and become the formation and fate halves of a single account of the same element.

It would make the Broeker skew mechanistic rather than descriptive. De novo breakpoints concentrate in the centromeric half of the BCR and therapy-related breakpoints in the telomeric half; the anchor sits at 70.4% along it, in the telomeric half 5. A mechanism in which chemotherapy lowers occupancy at one telomeric anchor, reducing local breakage but licensing illegitimate joining, is a candidate for why the therapy-related distribution differs from the de novo one - which this graph has held as an inference with no named actor.

It would sharpen the biomarker claim in the wrong direction, and that is a real cost. Atkin et al. propose mapping DSBs in patients' normal cells to identify individuals susceptible to KMT2A rearrangement and to therapy-related malignancy 3. If R/B is the variable that matters and it moves opposite to B, then a high break count could mark a well-guarded locus and the biomarker's sign is in question. That is an uncomfortable consequence and it is a reason to test rather than to assume.

It predicts something checkable without a new experiment. If the anchor rather than CTCF alone is the chaperone, cohesin-mutant leukaemias should show an altered KMT2A rearrangement rate or junction structure. This graph already holds gene-stag2, gene-rad21 and gene-smc3 entries and the recombinome carries 3,401 sequenced junctions 6 - a cohort analysis, not a bench experiment.

Facts and figures
statementAt the CTCF+RAD21+SMC3 anchor spanning Ensembl canonical KMT2A exon 11 (GRCh38 chr11:118,488,159-118,489,206), occupancy raises the rate of TOP2-mediated double-strand breaks and lowers the fraction of those breaks that resolve into an inherited rearrangement, because the same complex that recruits and traps TOP2 also holds the cut ends in register for faithful rejoining. Removing occupancy therefore produces FEWER breaks and MORE rearrangements - a rise in rearrangements per break.
originliterature-gap
prior probability0.35
what it would reorganisemechanism-loop-anchor-dsb-hotspots and mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement would stop being rivals and become the formation and fate halves of one account of one element; and breakpoint-kmt2a-bcr-architecture's de novo versus therapy-related breakpoint skew would acquire a named actor for its therapy-related half.
uncomfortable consequenceIf rearrangements per break is the variable that matters and it moves opposite to break count, then a high DSB count at a well-guarded anchor could mark LOW rather than high rearrangement risk, and the sign of the patient break-mapping biomarker proposed by Atkin et al. is in question. That consequence is a reason to test the hypothesis, not a reason to prefer it.
lineage warningNo CTCF occupancy or breakage measurement at this element exists in a B-lineage cell, in an infant, or in a KMT2A::AFF1 context. The protective evidence is epithelial, the directional fragility evidence is mouse fibroblast, mouse B cell and human colorectal, and the only patient measurement is adult myeloid. This hypothesis is about a therapy-induced mechanism and this graph's index disease is de novo infant t(4;11) with an in-utero origin and no prior anthracycline.
predictions
[
  "A CTCF degron at this element lowers END-seq or DSBCapture break signal and raises FISH break-apart and locus copy gain in the same cells, so that rearrangements per break rises.",
  "A RAD21 degron lowers break signal at the element at least as much as the CTCF degron, because cohesin is the more predictive occupancy variable for TOP2 breakage.",
  "Junction sequences recovered after occupancy loss are enriched for illegitimate, low-homology joins relative to those recovered from unperturbed cells under the same etoposide exposure.",
  "Doxorubicin or etoposide, which lower CTCF at this locus, produce the same signature - a fall in local break signal with a rise in inherited events - rather than the intuitive rise in both.",
  "Cohesin-mutant leukaemias (STAG2, RAD21, SMC3) show an altered KMT2A rearrangement rate or junction-homology distribution relative to cohesin-wild-type cases.",
  "The effect is anchor-specific: a matched CTCF+cohesin anchor with no rearrangement history shows the break decrease without the rearrangement increase."
]
falsifiers
[
  "A CTCF degron at this element that lowers breaks AND lowers rearrangements - occupancy is doing one job, and the published siCTCF rise is an artefact of depleting a near-pan-essential protein.",
  "A CTCF degron that RAISES breaks at this element - proposition (i) fails for CTCF here, no reconciliation is needed, and this hypothesis should be withdrawn rather than rescued.",
  "Rearrangements per break unchanged under any occupancy perturbation - formation and fate are not separable at this element and the premise this hypothesis attacks was true.",
  "The anchor is unoccupied by CTCF in every haematopoietic cell tested - the dispute is epithelium-specific and this hypothesis has no substrate in leukaemogenesis.",
  "Junction sequencing shows the FISH events are extrachromosomal or non-reciprocal products with no KMT2A fusion junction - then the R in this hypothesis is not the object the disease is made of.",
  "Cohesin removal moves rearrangement rate in the OPPOSITE direction to CTCF removal - the structural tethering argument that motivates the hypothesis is wrong even if its prediction survives.",
  "Gothe et al. turn out to contain a CTCF or cohesin depletion arm at the MLL locus with a result incompatible with the ratio prediction - that paper's full text was not read by this beat and is the largest unchecked source."
]
experiments
[
  "exp-ctcf-degron-breaks-and-rearrangements-same-cells-kmt2a"
]
analyses
[
  "ana-kmt2a-exon-11-ctcf-anchor-is-one-cohesin-bound-element"
]
How this could bite us back
  • This hypothesis's most actionable-sounding reading is the most dangerous one: that lowering CTCF at the locus is what chemotherapy does wrong. It is a hypothesis, the direct measurement does not exist, and the underlying data are cell-culture and mouse cytogenetics. Nothing here bears on whether a child should receive doxorubicin or etoposide, and therapy-related AML after topoisomerase II poisons is an already-quantified risk weighed in protocol design.
  • If the hypothesis is true, 'increase occupancy to protect the locus' buys fewer rearrangements at the price of more breaks, at an element built from proteins near pan-essentiality (CTCF -1.827 across all lines in this project's DepMap scan). An intervention aimed at the anchor is therefore worse than it sounds, and the only tractable node in the chain remains a genome-wide methyltransferase.
  • G9a/GLP inhibition is the intervention this mechanism family points at, and UNC0642 is an in vivo chemical probe rather than a clinical agent, supported here by a mouse arm with a cytogenetic endpoint at n = 3-4 per group and no leukaemia-incidence readout. Compensation is an obvious risk: the probe inhibits G9a and GLP together, and the founding study's own result is that EHMT1 depletion does not phenocopy EHMT2 depletion.
  • The hypothesis, if adopted, would put the sign of a proposed patient susceptibility biomarker in doubt. Anyone using it that way should note that it is speculative, that the biomarker rests on n = 3 versus 3, and that the correct response to both is the measurement, not a clinical inference from either.
Still open
  1. Is there any published measurement of rearrangements per break at any translocation hotspot, for any factor? If the quantity has been reported somewhere it would calibrate what an interesting change looks like.
  2. Does cohesin physically constrain TOP2-generated ends, or is the tethering argument only structural plausibility? A biochemical or single-molecule result on end mobility at a cohesin-bound break would move the prior substantially in either direction.
  3. Can Canela et al. 2019's deposited HCT-116 RAD21-degron END-seq be reanalysed at chr11:118,488,159-118,489,206 to give the break half of the prediction for free, in a human cell, today?
  4. Do the 3,401 recombinome junctions carry enough homology-signature information to test the junction-class prediction retrospectively against cohesin-mutation status, if the two datasets could be joined?
  5. If the prediction holds, does it imply that a TOP2 poison given with an agent that STABILISES the anchor would be safer than one given alone - and is that separable from simply giving less TOP2 poison?
Every source behind this entry (6)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Gray et al. report that H3K9me1/2 balance at the MLL/KMT2A locus, set by KDM3B-G9a crosstalk, regulates amplification and rearrangement; that KDM3B depletion raises H3K9me1/2 and reduces CTCF occupancy at the locus, promoting amplification and rearrangements; that depleting CTCF alone is sufficient; that doxorubicin suppresses KDM3B and CTCF protein and that KDM3B or CTCF overexpression rescues Dox-induced alterations; with readouts of DNA FISH copy gain and probe break-apart, locus qPCR and ChIP in human RPE cells extended to KG1a, primary HSPCs and mice, the copy gain transient at 12 h and inherited at 72 h - and with no double-strand-break assay and no mention of cohesin, RAD21 or SMC3 anywhere in the paper.

    Epigenetic balance ensures mechanistic control of MLL amplification and rearrangement. · Gray ZH, Chakraborty D, Duttweiler RR, Alekbaeva GD, Murphy SE, Chetal K, Ji F, Ferman BI, Honer MA, Wang Z, Myers C, Sun R, Kaniskan HU, Toma MM, Bondarenko EA, Santoro JN, Miranda C, Dillingham ME, Tang R, Gozani O, Jin J, Skorski T, Duy C, Lee H, Sadreyev RI, Whetstine JR · Cell 2023
    PMID 37788669DOI
    What this evidence cannot show

    Full text read from PMC10591855 on 2026-09-22; term counts over the deposited body and figure legends, not the supplement. The events are copy gain and probe separation, not a sequenced KMT2A fusion junction, and the authors state the second hit is unidentified - so 'rearrangement' in this evidence item is not the same object as a t(4;11). 'Depleting CTCF is also sufficient' is one siRNA arm against a protein at -1.827 pan-essentiality in this project's DepMap scan. Every chromatin measurement is epithelial.

    Evidence that argues against this (1)
    • The same paper's own citation for its doxorubicin arm reports that CTCF genomic binding is largely unchanged under stress even as the protein pool is degraded, with the loss confined to an SC-35 nuclear-speckle species - so the protein-level arm of the chain does not by itself deliver an occupancy change, and the locus ChIP has to carry it alone. Lehman BJ et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Canela et al. 2019 separate two steps in TOP2-induced breakage: trapping of the TOP2 cleavage complex, which is transcription-independent and requires pre-existing binding of cohesin to DNA, and the conversion of trapped complexes into irreversible TOP2-linked DSBs, which is accelerated about two-fold at transcribed relative to non-transcribed loci and depends on proteasomal degradation and TDP2 phosphodiesterase activity; and they show that acute cohesin loss lowers breakage - SMC3 depletion by CRE in mouse embryonic fibroblasts and RAD21 depletion by auxin-inducible degron in human HCT-116 both cause etoposide-induced END-seq signal to drop precipitously genome-wide, proportional to the cohesin lost (27% RAD21 reduction, 26% END-seq reduction) - concluding by quantitative modelling that cohesin binding and transcriptional activity alone predict the kinetics of TOP2-induced DSBs.

    Topoisomerase II-Induced Chromosome Breakage and Translocation Is Determined by Chromosome Architecture and Transcriptional Activity. · Canela A, Maman Y, Huang SN, Wutz G, Tang W, Zagnoli-Vieira G, Callen E, Wong N, Day A, Peters JM, Caldecott KW, Pommier Y, Nussenzweig A · Molecular Cell 2019
    PMID 31202577DOI
    What this evidence cannot show

    Read in full text from PMC8170508 on 2026-09-22. The perturbation is COHESIN, not CTCF; the paper contains no MLL or Kmt2a analysis at all (zero occurrences of either string); the HCT-116 arm is G1-arrested with lovastatin; and the depletion is partial. The two-step separation is the part this hypothesis rests on, and it is a genome-wide kinetic result, not a measurement at the KMT2A anchor. Nothing in the paper proposes or tests a tethering or repair-chaperone role for the anchor - that is this hypothesis's addition.

    Evidence that argues against this (1)
    • The paper's own conclusion is that cohesin binding and transcription suffice to predict TOP2 DSB kinetics, which is a formation model with no fate term - so if the same two variables also predicted rearrangement outcome, the ratio this hypothesis turns on would be constant and the hypothesis unnecessary. Canela A et al. 2019
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Atkin et al. measured genome-wide DNA breakage by low-input DSBCapture in normal non-blast cells from AML patients and found breakage at the KMT2A intron 10 / exon 11 region (hg38 chr11:118,488,500-800) 2- to 10-fold higher in each KMT2A-rearranged remission patient than in each non-fusion patient (significant 3.2-fold group average), identified 304 TOP2-sensitive, 404 CTCF, 517 RAD21 and 256 SMC3 sites preferentially break-prone in fusion-driven patients, found low-dose etoposide ex vivo raised breakage at these regions in the fusion patients only with no difference in genome-wide median breakage, and proposed that measuring DSBs in patients enables monitoring of individuals susceptible to chromosomal abnormalities and to therapy-related malignancy caused by TOP2 poisons.

    PMID 34405474DOI
    What this evidence cannot show

    n = 3 per arm, adult myeloid, cross-sectional, and cannot distinguish a constitutional set point from a consequence of leukaemia and its treatment. No CTCF perturbation and no rearrangement readout. This item is cited here as the clearest statement of the premise this hypothesis attacks - that break-proneness predicts rearrangement susceptibility - and that premise is the paper's proposal, not its measurement.

    Evidence that argues against this (1)
    • If this hypothesis is right the sign of this biomarker is in question, because a high break count at a well-guarded anchor would mark low rather than high rearrangement risk - which is a consequence uncomfortable enough that it should be tested before either the biomarker or the hypothesis is relied on. Gomez-Herreros F et al. 2017
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Gomez-Herreros et al. show that TOP2-induced DSBs are rejoined in part by TDP2-dependent non-homologous end-joining, that TDP2 rejoins DSBs induced during transcription-dependent TOP2 activity in breast cancer cells and at the translocation hotspot MLL, that TDP2 suppresses chromosome rearrangements induced by TOP2 and reduces TOP2-induced chromosome translocations arising during gene transcription, and that TDP2-dependent non-homologous end-joining is nonetheless implicated in the formation of a rare subclass of translocations previously associated with therapy-related leukaemia and characterised by junction sequences with 4 bp of perfect homology.

    TDP2 suppresses chromosomal translocations induced by DNA topoisomerase II during gene transcription. · Gomez-Herreros F, Zagnoli-Vieira G, Ntai I, Martinez-Macias MI, Anderson RM, Herrero-Ruiz A, Caldecott KW · Nature Communications 2017
    PMID 28794467DOI
    What this evidence cannot show

    Abstract read in full on 2026-09-22; body and figures not read, so the magnitude of the translocation increase, whether breakage was quantified in the same cells, and which MLL region was assayed are not established here. System is breast cancer cells. This establishes that rearrangement fate can move independently of break formation AT THIS LOCUS for a repair factor - it does not establish that CTCF is such a factor, which is the whole content of this hypothesis and is unmeasured.

    Evidence that argues against this (1)
    • The same paper shows a repair factor can be protective in aggregate and causal for one junction class, so 'occupancy guards repair' could be true on average and false for the junction type that actually produces KMT2A fusions - which would make the hypothesis correct about the ratio and wrong about the disease. Gomez-Herreros F et al. 2017
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The element in dispute is a single CTCF+cohesin co-occupied anchor: ENCODE clustered ChIP-seq places CTCF at GRCh38 chr11:118,488,159-118,489,206 (score 1000) with RAD21 at 118,488,428-118,488,933 (1000) and SMC3 at 118,488,454-118,488,896 (1000), reproduced in GRCh37 as CTCF 118,359,222-118,359,582 (1000), RAD21 118,359,253-118,359,515 (1000) and SMC3 118,359,255-118,359,540 (953); Ensembl canonical exon 11 of ENST00000534358 lies inside it at 118,488,614-118,488,760; the hg38 chr11:118,488,500-118,488,800 patient break-prone window lies wholly inside it; and the anchor sits at 70.4% along the canonical exon-8-to-exon-14 span of the breakpoint cluster region, in the telomeric half that Broeker et al. found carries 75% of therapy-related and 26% of de novo breakpoints.

    What this evidence cannot show

    ENCODE clusters summarise many cell types and the score is capped at 1000; a score of 1000 means reproducibly strongly bound across contributing experiments, NOT that CTCF is bound in a haematopoietic progenitor, and no leukaemia or CD34+ experiment is identified. TOP2B is absent from the cluster set at this element, so the 62-fold CTCF+RAD21+TOP2B fragility class cannot be asserted here. The centromeric/telomeric assignment applies Broeker's 4.57/3.87 kb split proportionally to an exon span because the BamHI fragment's coordinates are not published. Full derivation in ana-kmt2a-exon-11-ctcf-anchor-is-one-cohesin-bound-element.

    Evidence that argues against this (1)
    • The identification of Gray et al.'s site with this cluster is an inference from ENCODE occupancy strength, not from a coordinate Gray published; if their 'exon 11' is the DCAL exon 11 - canonical exon 10 - the site is instead a much weaker score-514 CTCF cluster 3.5 kb away and is not the patient break-prone window at all, which would make the two sides describe two elements and remove the premise of this hypothesis. NCBI Gene Expression Omnibus 2024
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Meyer et al. report genomic breakpoint data from 3,401 acute leukaemia patients analysed between 2003 and 2022, with the major breakpoint cluster region BCR1 mapped between KMT2A intron 7 and exon 13 carrying 3,336 of the breakpoints (98%), subdivided into subregions A (exon 9 - intron 9, 1,761 bp, 37.2% of breaks), B (exon 10 - intron 10, 679 bp, 19.8%) and C (exon 11 - intron 11 - exon 12 - intron 12 and exon 13, 5,026 bp, 39.5%) in their own exon numbering, with the functional cut between A-B and C; and that breakpoints within KMT2A intron 11 are associated with poorer outcome than upstream breakpoints, the cut point being exactly at the borderline between intron 10 and exon 11.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, Burmeister T, Groger D, Sutton R, Venn NC, Cazzaniga G, Corral Abascal L, Tsaur G, et al., Marschalek R · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    Read in full text from PMC10169636 on 2026-09-22, subregion definitions extracted verbatim. This exon numbering is Ensembl canonical PLUS ONE, established arithmetically in ana-kmt2a-exon-11-ctcf-anchor-is-one-cohesin-bound-element; so Meyer's 'intron 10 / exon 11' cut point is canonical intron-9/exon-10 and is 3,752 bp from the CTCF anchor, NOT at it. Cited here because it is the cohort resource that makes the cohesin-mutant junction prediction answerable without new experiments; the 3,401 junctions are from a diagnostic pipeline and their cohesin-mutation status is not part of the published dataset.

    Evidence that argues against this (1)
    • The recombinome's subregion break densities may reflect repeat content rather than any chromatin element - the same paper records four Alu elements in its intron 9, three transcriptionally active - and 39.5% of breaks spread over 5,026 bp of subregion C is a lower density than 19.8% over 679 bp of subregion B, so the anchor sitting in subregion C is not by itself evidence of elevated local break density. Meyer C et al. 2023
Connected to