The relapse-initiating persister state — present at diagnosis, identifiable before a single dose is given, and defined by basal glucocorticoid-response activation plus small size plus quiescence
also called relapse-initiating cells · persister state · therapy-resistant subpopulation at diagnosis · prednisone-resistant subpopulation · MRD-like state
In infant KMT2A-rearranged leukaemia, the cells that will cause relapse are already present at diagnosis and can be told apart from the rest by reading single cells: they are smaller, they sit still rather than dividing, and their steroid-response genes are already switched on before any steroid is given. Counting them at diagnosis predicted relapse better than the test currently used.
The same thing, for a scientist
This is the entry the beat was dispatched to find: a state of the cells that survive induction in infant KMT2A-rearranged ALL that has been **characterised**, not merely counted. Candelli et al. applied two independent single-cell RNA sequencing technologies to **diagnostic** bone marrow and peripheral blood from infants with MLL-rearranged ALL, scoring each leukaemic cell against a prednisone-dependent signature derived from an independent study. That classified individual cells as resistant or sensitive, and **the proportion of the two predicted future relapse in individual patients better than the current best indicator** — which in this disease is the in vivo response to prednisone monotherapy [E1]. The resistant cells are described concretely: **basal activation of the glucocorticoid response, smaller size, and a quiescent gene expression programme with stemness properties** [E1]. Two things make this important beyond its own result. First, the state exists **before treatment**, so it is selected rather than induced, which changes what a combination is supposed to do. Second, it is **measurable at diagnosis** from a sample that is already taken. The independent paired-timepoint study in paediatric B-ALL agrees on the direction — residual cells are complex and relapse cells shift to poorly differentiated states, with hypoxia-pathway activation in residual cells, resistant clones and MLL-rearranged disease [E2] — and a 96-patient pan-leukaemia single-cell multiomic study finds a chemotherapy-resistant **HSPC-like** subpopulation associated with poor outcome across subtypes, controlled by a HOXA/AP1/CEBPA regulatory network [E3]. Three independent studies, three cohorts, one recurring shape: a quiescent, stem-like, pre-existing minority.
What was measured, and why the design matters more than usual
The scoring signature was prednisone-dependent and derived from an independent study, not from the relapse outcomes it was then used to predict 1. That is the difference between a predictive signature and a circular one, and it is why this result is worth more than the many expression signatures that correlate with outcome in the cohort that generated them.
Two separate single-cell technologies were used, which controls for platform-specific dropout — a real concern when the claim is about a minority population.
The samples are diagnostic bone marrow and peripheral blood. Nothing was sampled at MRD, so the claim is not 'these are the MRD cells'; it is 'these cells at diagnosis have the transcriptional properties that MRD cells are known to have, and their abundance predicts relapse'. The distinction is worth keeping, and the independent paired study that did sample residual disease found the hypoxia axis rather than the glucocorticoid axis as its headline 2 — agreement on the existence of a distinct residual state, not on its definition.
Basal glucocorticoid-response activation is the strange part
Quiescence and stemness in a therapy-resistant compartment are expected; this graph already holds cell-state-quiescence and cell-state-stem-like-self-renewal for them. Basal activation of the glucocorticoid response, before exposure, is not expected, and it points somewhere specific.
In this disease glucocorticoid resistance already has a proposed intrinsic mechanism: an MLL-AF4 to NG2 to FLT3 to AP-1 to NR3C1 axis, recorded in cell-state-pro-b-arrest. A cell whose GC-response genes are already on could be a cell that has already run and exhausted that programme, or one in which the receptor is engaged without productive downstream death. Either reading predicts that the response to the first steroid dose is partly determined before it is given, which is exactly what the in vivo prednisone-response test measures at the whole-patient level and what this measurement improves on at the single-cell level.
It also creates an uncomfortable possibility for combination design. If the surviving cells are those with the glucocorticoid programme already engaged, adding more glucocorticoid is the least likely thing to help, and the BCL-2-plus-MCL-1-plus-dexamethasone triple combination that this graph records as the most effective preclinical regimen in infant KMT2A-rearranged ALL has a mechanism that must be acting on a different compartment, or must be re-sensitising this one.
Smaller size — a property with an assay attached
Cell size is not a transcriptional inference: it is a physical property, it is measured on every flow cytometer as forward scatter, and it is the property centrifugal elutriation separates on.
That matters for this beat specifically. cell-state-unsynchronisable-cell-cycle records that KMT2A-rearranged lines cannot be chemically synchronised and that centrifugal elutriation worked. If the relapse-initiating cells are smaller, then elutriation is not only a workaround for the synchronisation problem — it is a candidate enrichment method for the persister compartment itself, usable on primary material without antibodies or dyes, and nobody appears to have tried it.
The caveat is that small size in a quiescent cell is partly a consequence of not having grown for a cycle, so size and quiescence are not independent measurements of two things.
What is still missing, stated because the brief asked whether any state has been characterised rather than counted
Characterised: the diagnostic relapse-initiating compartment in infant KMT2A-rearranged ALL — transcriptome, size, cycle status, and a predictive use 1.
Counted only: the actual MRD cells in infant KMT2A-rearranged ALL. The paired diagnosis/residual/relapse single-cell study that does exist is paediatric B-ALL broadly, with MLL-rearranged cases as a subgroup of unstated size 2. This project's epi-kmt2a-b-all-partner-and-mrd-clearance-across-ages holds the clinical MRD numbers. Nobody has published single-cell profiling of MRD cells sampled from infants with KMT2A-rearranged ALL at end of induction.
Not attempted: any measurement of this state under menin inhibition. Revumenib and ziftomenib are now given to children with this disease, and whether the persister compartment is depleted, spared or enriched by them is unmeasured. The same sample that supplies an MRD result could supply the answer.
| state definition | A minority subpopulation present in the diagnostic sample whose transcriptional programme predicts survival of induction and subsequent relapse; in infant KMT2A-rearranged ALL defined by basal glucocorticoid-response activation, smaller cell size and a quiescent programme with stemness features |
|---|---|
| measured where | diagnostic bone marrow and peripheral blood, infants with MLL-rearranged ALL; two independent single-cell RNA sequencing platforms |
| predictive claim | quantification of resistant versus sensitive cells at diagnosis predicted relapse better than the existing best indicator, which is the in vivo response to prednisone monotherapy |
| signature provenance | the prednisone-dependent scoring signature was derived from an independent study, not from the relapse outcomes it predicts |
| gap | no published single-cell profiling of MRD cells sampled from infants with KMT2A-rearranged ALL at end of induction; and no measurement of this compartment under menin inhibition |
| untried enrichment route | centrifugal elutriation separates on size and sedimentation velocity, works in KMT2A-rearranged lines where chemical synchronisation fails, and the persister cells are reported as smaller — an antibody-free enrichment nobody has attempted |
defining features
[ "basal activation of the glucocorticoid response, before exposure", "smaller cell size (a physical property, measurable as forward scatter and separable by elutriation)", "quiescent gene expression programme", "stemness properties" ]
related but distinct states
{
"hypoxia-defined residual state": "PMID 35145224, paediatric B-ALL broadly, sampled at true MRD",
"HSPC-like resistant state": "PMID 40089994, pan-leukaemia, 96 patients, HOXA/AP1/CEBPA network",
"reversible dormant PDX state": "PMID 27916615, recorded in cell-state-quiescence"
}atlas refs — the numbers behind the chart
[
{
"label": "Relapse-initiating cells identifiable at diagnosis in MLL-rearranged infant ALL",
"pmid": "34304246",
"title": "Identification and characterization of relapse-initiating cells in MLL-rearranged infant ALL by single-cell transcriptomics.",
"authors": "Candelli T, Schneider P, Garrido Castro P, et al.",
"venue": "Leukemia",
"year": 2022,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "Paired diagnosis/residual/relapse single-cell atlas of paediatric B-ALL",
"pmid": "35145224",
"title": "Elucidating minimal residual disease of paediatric B-cell acute lymphoblastic leukaemia by single-cell analysis.",
"authors": "Zhang Y, Wang S, Zhang J, et al.",
"venue": "Nat Cell Biol",
"year": 2022,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "Pan-leukaemia HSPC-like chemoresistant subpopulation across 96 paediatric patients",
"pmid": "40089994",
"title": "Single-cell panleukemia signatures of HSPC-like blasts predict drug response and clinical outcome.",
"authors": "Chen C, Xu J, Sussman JH, et al.",
"venue": "Blood",
"year": 2025,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
]- If the resistant compartment already has its glucocorticoid programme switched on at diagnosis, then intensifying steroid — the historical response to poor prednisone response in this disease — is targeting the cells least able to respond to it.
- The state is present before treatment, so it is selected rather than induced, and a combination designed to prevent the emergence of resistance is solving a problem that has already happened by the time the patient arrives.
- A diagnostic single-cell assay that predicts relapse better than prednisone response would, if adopted, move infants between risk arms on the basis of an unvalidated small-cohort result; the cost of a false low-risk assignment in this disease is high.
- Targeting quiescent, stem-like, small cells means targeting the cells most similar to normal haematopoietic stem cells in an infant, which is the population least able to tolerate collateral damage.
- Are the relapse-initiating cells of Candelli, the hypoxia-activated residual cells of Zhang and the HSPC-like blasts of Chen the same cells under three descriptions? All three predict poor outcome, none has been scored on another's data, and the three signatures are jointly testable on any one of the three public datasets.
- Does menin inhibition deplete, spare or enrich this compartment? Children are receiving revumenib now; a diagnostic and an end-of-induction single-cell sample from the same child on a menin inhibitor would answer it, and would be the first measurement of a menin inhibitor's effect on the compartment that causes relapse.
- Can the persister compartment be enriched by size alone? The cells are reported as smaller and centrifugal elutriation is demonstrated to work in KMT2A-rearranged cells; an antibody-free physical enrichment from primary marrow would make every downstream assay possible on a population currently too rare to study.
- Why is the glucocorticoid response already active in the cells that go on to resist glucocorticoids? Either the programme is engaged and unproductive, or it has already been run and exhausted, and the two predict opposite answers to whether more steroid helps.
- Does the predictive value survive prospective validation? A per-patient relapse prediction from a small single-cell cohort is the shape of result that most often fails to replicate, and no prospective test has been reported.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Two different single-cell RNA sequencing technologies applied to diagnostic bone marrow and peripheral blood biopsies from infants with MLL-rearranged ALL, scoring cells against an independently derived prednisone-dependent signature, classified individual leukaemic cells as treatment-resistant or sensitive; quantification of the two groups predicted future relapse in individual patients better than existing risk stratification, and leukaemic cells associated with high relapse risk were characterised by basal activation of the glucocorticoid response, smaller size, and a quiescent gene expression programme with cell stemness properties.
Identification and characterization of relapse-initiating cells in MLL-rearranged infant ALL by single-cell transcriptomics · Candelli T, Schneider P, Garrido Castro P, Jones LA, Bodewes E, Rockx-Brouwer D, et al., Stam RW · Leukemia 2022What this evidence cannot show
The number of infants profiled is not stated in the abstract, and single-cell studies of a rare disease are characteristically small; a per-patient relapse prediction from a small cohort is a hypothesis-generating result until validated prospectively, which has not been reported. The classification depends entirely on one externally derived prednisone signature, so the resistant population is defined by a single drug's transcriptional correlate rather than by function — no cell was shown to survive treatment or to initiate relapse experimentally, despite the entry's name. 'Smaller size' in a quiescent cell is partly a consequence of not having completed a growth cycle, so size and quiescence are not independent observations. Fusion partner is not resolved. Read: abstract only.
Evidence that argues against this (1)
- The independent study that actually sampled residual disease reaches a different definition of the surviving state. Profiling 161,986 single cells at diagnosis, residual disease and relapse in paediatric B-ALL, the changes in residual cells were explicitly described as more complicated than the relapse shift, and the axis that emerged was HYPOXIA - activated in residual cells, in resistant clones and in MLL-rearranged B-ALL - with hypoxia-pathway inhibition sensitising cells to chemotherapy. Glucocorticoid-response activation is not the reported headline there. Two studies of the surviving compartment in overlapping diseases have named two different programmes, and they have not been reconciled; a signature built from one would not necessarily identify the other. Read: abstract only. Zhang Y et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In 161,986 single-cell transcriptomes with paired B-cell-receptor sequencing from paediatric B-ALL sampled at diagnosis, residual disease and relapse, leukaemic cells at relapse shifted toward poorly differentiated states while changes in residual cells were more complicated, and differential analysis highlighted hypoxia-pathway activation in residual cells, in resistant clones and in B-ALL with MLL rearrangement; both in vitro and in vivo models showed that inhibition of the hypoxia pathway sensitised leukaemic cells to chemotherapy.
Elucidating minimal residual disease of paediatric B-cell acute lymphoblastic leukaemia by single-cell analysis · Zhang Y, Wang S, Zhang J, Liu C, Li X, Guo W, et al., Cheng T · Nat Cell Biol 2022What this evidence cannot show
Paediatric B-ALL as a whole; the MLL-rearranged subgroup's size, age range and fusion partners are not stated, and infants are not identified separately, so this is not an infant KMT2A-rearranged study. Residual cells are rare and their isolation imposes selection and dissociation biases that no single-cell protocol removes. A hypoxia signature is a transcriptional correlate, not an oxygen measurement. Read: abstract only.
Evidence that argues against this (1)
- The hypoxia arm of this result is contradicted by genetics in the myeloid form of the KMT2A-rearranged lesion: conditional Hif-1a deletion in mouse MLL-AF9 AML combined with standard chemotherapy produced faster disease progression, and the authors warn explicitly against combining HIF inhibition with chemotherapy in AML. Since the residual-cell finding here is precisely an argument for adding hypoxia-pathway inhibition to chemotherapy, the two point in opposite directions, and the difference is myeloid genetics versus lymphoid pharmacology with no experiment bridging them. Read: abstract only. Velasco-Hernandez T et al. 2019
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Single-cell multiomic analysis of diagnostic specimens from 96 paediatric patients with acute lymphoblastic, myeloid and lineage-ambiguous leukaemias, integrated with bulk RNA sequencing and clinical data, identified a prevalent chemotherapy-resistant subpopulation resembling haematopoietic stem and progenitor cells that was associated with poor clinical outcomes across all subtypes investigated; the HSPC-like blasts carry a core transcriptional regulatory network combinatorially controlled by HOXA, AP-1 and CEBPA whose signature predicts chemotherapy response and long-term outcome, and shared candidate targets against them include FLT3, BCL2 and the PI3K pathway.
Single-cell panleukemia signatures of HSPC-like blasts predict drug response and clinical outcome · Chen C, Xu J, Sussman JH, Vincent T, Tumulty JS, Yoshimura S, et al., Tan K · Blood 2025What this evidence cannot show
Pan-leukaemia by design, so the KMT2A-rearranged fraction of the 96 patients is not stated and no fusion-specific claim can be drawn; a signature that generalises across subtypes is by construction not subtype-specific. Diagnostic specimens only, so 'chemotherapy-resistant' is inferred from outcome association rather than from profiling cells that survived chemotherapy. HOXA is a defining feature of KMT2A-rearranged disease generally, which raises the possibility that a HOXA/AP1/CEBPA network scores KMT2A-rearranged blasts as HSPC-like for reasons of lineage programme rather than of stemness. Read: abstract only.
Evidence that argues against this (1)
- A pan-leukaemia HSPC-like signature and the KMT2A-rearranged-specific developmental state described elsewhere are not obviously the same population and may be in tension. Single-cell transcriptomics against a developing bone marrow reference found KMT2A-rearranged infant B-ALL to be uniquely dominated by an EARLY LYMPHOCYTE PRECURSOR state, a defined position in lymphoid development rather than an undifferentiated HSPC one, and the same study established that the rearrangement arises before haematopoietic specification so that cell of origin cannot be inferred from transcriptional state at all. If the blasts are ELP-like as a class, a subpopulation scored HSPC-like may be reporting distance along one reference axis rather than a distinct resistant compartment. Read: abstract only. Khabirova E et al. 2022
identified in diagnostic marrow and blood of infants with MLL-rearranged ALL by two single-cell platforms; abundance predicted relapse better than in vivo prednisone response
the relapse-initiating cells carry a quiescent programme with stemness features, which is the state cell-state-quiescence defines
two named definitions of the surviving compartment that have never been scored against each other: glucocorticoid-primed and quiescent at diagnosis, versus hypoxia-activated at residual disease
the pro-B arrest entry records the MLL-AF4/NG2/FLT3/AP-1/NR3C1 glucocorticoid-resistance axis; this state shows that axis already engaged at diagnosis in the cells that go on to relapse
the persister cells are smaller, and centrifugal elutriation separates on size and is the one method demonstrated to work in KMT2A-rearranged cells — an untried antibody-free enrichment route
- supports →When infant leukaemia comes back, it comes back early and while treatment is still running
a cellular account of why two-thirds of infants relapse, typically during treatment: the compartment is present before the first dose
clonal genetics and single-cell transcriptomics converge with no shared method: the relapse compartment is present and quiescent at diagnosis
arm B tests whether the small quiescent relapse-associated compartment can be enriched from primary marrow by size alone
the compartment a turnover-limited drug reaches last, qualified
DEK phosphorylation was identified by comparing normal HSPCs with LSC-enriched populations
one of the 13+3 entries whose single-cell citations the index could not see
two independently derived descriptions of a pre-existing quiescent stem-like minority that predicts relapse; the pan-leukaemia HSPC-like paper is already cited there
CREBBP alterations in paediatric ALL are relapse-specific and largely therapy-induced, peaking at 65 percent in early relapse and acquired by a clone that survived induction
MEF2C S222 phosphorylation is a marker of PRIMARY chemoresistance in patient AML and MARK/SIK inhibition restores chemosensitivity only in phospho-positive specimens (PMID 29431698)
the absence of TOP2A from G0 is the mechanistic form of the persister problem for anthracycline and etoposide exposure in this disease
CtIP's B-cell requirement is proliferation-conditional, so a quiescent persister is the state in which it is least needed
- ← related toCTNNB1 (beta-catenin) — the Wnt link inside DotCom, and the one piece of KMT2A biology that is not chromatin
BET-inhibitor resistance in MLL-AF9 leukaemia emerges from LSCs and is in part increased Wnt/beta-catenin signalling; pathway suppression restores I-BET sensitivity in vivo (PMID 26367796) (Q89 2026-09-21)
AZD1775 in the KMT2A::AFF1 line RS4;11 gave the highest inhibition of recovery of proliferation among ALL subtypes tested, disrupting a core KMT2A-RUNX1-MYC network by single-cell RNA-seq and ATAC-seq - and the escape is a reversible NON-GENETIC drug-tolerant state defined by pre-B-cell-fate factors, lipid metabolism and pre-BCR signalling (PMID 38822412). No compartment was measured in either state (Q95 2026-09-22)