Risk: CLEC12A-directed therapy ablates normal granulocytes and disables an inhibitory brake on NET formation - the toxicity is mature-myeloid, not stem-cell
also called CLL-1 CAR agranulocytosis · CD371 on-target off-tumour toxicity · MICL blockade NET risk
CLEC12A therapies are sold on the promise that they spare the blood-forming stem cells, and they mostly do - but CLEC12A is also on ordinary neutrophils, the white cells that fight infection. In the first trial of a CLEC12A CAR, every patient lost their blood counts and two died of infection.
The same thing, for a scientist
**The stem-cell-sparing argument is not the safety argument.** CLEC12A's therapeutic pitch rests on absence from the CD34+CD38- HSC gate [see `gene-clec12a`], and that claim broadly holds. It is irrelevant to the toxicity that has actually been observed, which comes from the **mature myeloid compartment** where CLEC12A is abundant. **Observed clinical toxicity.** First-in-human CLL-1 CAR-T, 10 adults with R/R AML: all 10 developed CRS (6 high grade), **severe pancytopenia in all 10**, and **two deaths from severe infection due to chronic agranulocytosis**. The authors state plainly that CLL-1 is highly expressed on normal granulocytes and propose bridging transplant as the rescue [E1]. KITE-222, 12 patients, had a milder profile - one DLT of prolonged grade 4 neutropenia/thrombocytopenia, no grade >=3 CRS - but also no efficacy, so the two studies do not bracket the risk of an *effective* CLEC12A agent [E2]. **An unmeasured second risk: the receptor's day job.** MICL/CLEC12A is an inhibitory ITIM receptor that recognises DNA in neutrophil extracellular traps; **loss or inhibition of MICL function causes uncontrolled NET formation** via ROS-PAD4 and an auto-inflammatory feedback loop, and humans carrying inhibitory anti-MICL autoantibodies have exacerbated NET-driven pathology in rheumatoid arthritis, lupus and severe COVID-19 [E3]. A therapeutic binder that engages CLEC12A without killing the cell is functionally an anti-MICL antibody on that cell. **No CLEC12A clinical programme read for this entry reports having measured NET formation, and the risk is therefore unquantified rather than excluded.** **A third, smaller one: the target is not AML-restricted within myeloid disease.** CLL-1 is on 10-60% of CD34+ cells in CML and MDS [E4].
Mitigations, and what each one costs
- Bridge to transplant. Proposed by the authors of the first-in-human study 1. It works, and it converts a targeted therapy into a transplant indication - which for an infant with KMT2A-rearranged disease is a very different risk-benefit conversation.
- Gated co-targeting. Use CLEC12A only as a costimulatory input behind a primary antigen, as in the ADGRE2/CLEC12A design, which reached off-target HSPC toxicity comparable to a CD19-CAR and prevented antigen escape in xenografts 5. Cost: complexity, and the toxicity floor it reached was a floor, not zero.
- Cytokine attenuation. The same work mitigated HSPC toxicity by reducing CAR-T-derived interferon-gamma 5, and an asymmetric low-affinity CD3 arm is the bispecific equivalent. Cost: potency.
- Nothing on the NET risk. There is no proposed mitigation because there is no measurement. The cheapest first step is to measure NET markers (citrullinated H3, cell-free DNA) in banked samples from an existing CLEC12A trial.
Why this risk is worse in the index disease than in adult AML
Every clinical observation above is in adult relapsed/refractory AML, where patients arrive with damaged marrow and where prolonged agranulocytosis is an accepted cost. Infant KMT2A-rearranged leukaemia is a different bargain: the patients are under one year old, infection is already the dominant non-relapse cause of death on infant protocols, and there is no CLEC12A expression measurement in infant KMT2A-rearranged B-ALL at all. Importing an adult-AML safety profile into that population is not supported by anything in this entry.
| kind | toxicity |
|---|---|
| severity | high in the observed adult R/R AML setting (2 of 10 deaths from infection in chronic agranulocytosis in one study); unquantified in children and unmeasured in infants |
| compartment | mature myeloid (granulocytes, monocytes) - NOT the haematopoietic stem cell compartment |
| second mechanism | functional blockade or engagement of an inhibitory ITIM receptor that restrains NET formation; unmeasured in any clinical CLEC12A programme read here |
| mitigation | bridge to allogeneic transplant; gated co-targeting with CLEC12A as a costimulatory rather than primary antigen; reduction of CAR-derived interferon-gamma; low-affinity CD3 arms in bispecifics. None addresses the NET risk. |
affects
[ "gene-clec12a", "mechanism-menin-inhibition-induces-clec12a" ]
unmeasured
[ "NET formation in treated patients", "any paediatric CLEC12A safety data", "any infant KMT2A-rearranged data" ]
- **The marketing claim and the observed toxicity are about different cells.** 'Spares haematopoietic stem cells' is true and does not predict the agranulocytosis that has actually killed patients.
- **A safety profile from a failed trial is not reassurance.** KITE-222's mild toxicity came with no efficacy; on-target toxicity scales with on-target activity, so the favourable half of that result cannot be kept while discarding the unfavourable half.
- **The menin combination could make this worse, not better.** If menin inhibition induces CLEC12A [see `mechanism-menin-inhibition-induces-clec12a`], it may induce it on normal and recovering myeloid cells as well as on blasts, widening the on-target/off-tumour overlap precisely when the CAR is most active.
- **An unmeasured risk is not a small risk.** The NET mechanism has a *Nature* paper behind it and zero measurements in any CLEC12A trial; recording it as 'no evidence of harm' would be a misuse of an absence.
- Do CLEC12A-engaging binders agonise, antagonise or ignore MICL's inhibitory signalling on the neutrophils they do not kill? No CLEC12A programme cited here states which, and the answer determines whether the NET risk is real.
- Is the two-death signal from the first-in-human study reproducible? KITE-222 was much milder but also ineffective, so the field has no safety observation from an EFFECTIVE CLEC12A agent.
- What would a CLEC12A agent do to an infant's already-fragile granulocyte compartment on an infant ALL backbone, where infection is the leading non-relapse cause of death? There is no data at all.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In the first-in-human phase I study of CLL-1 CAR-T cells in 10 adults with relapsed/refractory AML, all patients developed cytokine release syndrome (6 of 10 high grade), severe pancytopenia occurred in all patients, and two patients died of severe infection due to chronic agranulocytosis; the authors note that CLL-1 is also highly expressed on normal granulocytes and propose bridging haematopoietic stem cell transplantation to rescue long-term agranulocytosis due to off-target toxicity.
First-in-human phase I study of CLL-1 CAR-T cells in adults with relapsed/refractory acute myeloid leukemia. · Jin X, Zhang M, Sun R, Lyu H, Xiao X, Zhang X, et al. · J Hematol Oncol 2022What this evidence cannot show
Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. n=10, single arm, adults, median follow-up 173 days. Heavily pre-treated R/R AML patients have prior marrow damage and prior infections, so attributing the two deaths solely to on-target CLEC12A toxicity is not established by the abstract - the authors' own attribution is to off-target toxicity on granulocytes, and that is an interpretation, not a controlled comparison.
Evidence that argues against this (1)
- KITE-222, a larger and more recent CLL-1 CAR study (n=12), reported no grade >=3 CRS and only one cytopenia DLT - a substantially milder profile. The difference may be construct, dose or population, but it means the two-death signal is not reproducible across CLEC12A CAR programmes and should not be treated as the expected outcome of the modality. Daver N et al. 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In the KITE-222 phase 1 study of a CLL-1-directed CAR T in 12 patients with relapsed/refractory AML, one patient experienced a dose-limiting toxicity of prolonged grade 4 neutropenia and thrombocytopenia, all patients experienced grade >=3 adverse events, none had grade >=3 cytokine release syndrome, one had grade >=3 ICANS, and clinically meaningful responses were lacking across dose levels.
Phase 1 Study of KITE-222, an Autologous CLL-1-Directed CAR T-cell Therapy in Patients with Relapsed/Refractory Acute Myeloid Leukemia. · Daver N, Blachly JS, Ghobadi A, Advani A, Muffly L, Garciaz S, et al. · Clin Cancer Res 2026What this evidence cannot show
Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. **A safety profile measured in a trial with no efficacy is not the safety profile of an effective agent** - the cells largely did not do their job, and on-target toxicity scales with on-target activity. This is the single most important caveat in this entry and it cuts against reading KITE-222 as reassuring.
Evidence that argues against this (1)
- The one DLT occurred in the only patient to receive two courses of lymphodepleting chemotherapy, which confounds attribution of the prolonged cytopenias to the CAR rather than to the conditioning. Xiu Y et al. 2018
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
MICL (CLEC12A) directly recognises DNA in neutrophil extracellular traps and this recognition is vital to regulate neutrophil activation; loss or inhibition of MICL functionality leads to uncontrolled NET formation through the ROS-PAD4 pathway and an auto-inflammatory feedback loop, and inhibitory anti-MICL autoantibodies are detected in patients with rheumatoid arthritis, lupus and severe COVID-19.
Recognition and control of neutrophil extracellular trap formation by MICL. · Malamud M, Whitehead L, McIntosh A, Colella F, Roelofs AJ, Kusakabe T, et al. · Nature 2024What this evidence cannot show
Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. Mouse models plus human autoantibody association; the human data are associative. **The step from 'function-blocking autoantibody' to 'therapeutic CAR or bispecific binder' is an inference made by this entry, not by the paper** - a binder could be agonistic, antagonistic or neutral at the receptor, and none of the CLEC12A therapeutic programmes cited here reports which.
Evidence that argues against this (1)
- In the same paper, dysregulated NET release was PROTECTIVE during systemic Aspergillus fumigatus infection. In a profoundly neutropenic leukaemia patient at high fungal risk, the sign of a MICL-blockade effect on outcome is genuinely uncertain and could be favourable. Xiu Y et al. 2018
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
CLL-1 showed variable expression in 10-60% of CD34+ cells in chronic myelogenous leukaemia and myelodysplastic syndrome, so the antigen is not restricted to AML within myeloid neoplasia.
C-type lectin-like molecule-1: a novel myeloid cell surface marker associated with acute myeloid leukemia. · Bakker AB, van den Oudenrijn S, Bakker AQ, Feller N, van Meijer M, Bia JA, et al. · Cancer Res 2004What this evidence cannot show
Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. The CML and MDS denominators are not given in the abstract, and a 10-60% range across cases is wide enough that it describes heterogeneity rather than a central estimate.
Evidence that argues against this (1)
- For a patient being treated for AML, CLL-1 expression on a co-existing MDS clone is arguably on-target rather than off-target, since MDS clones in that setting are frequently the reservoir the AML arose from. Whether this is a risk or a benefit depends on the clinical scenario. Xiu Y et al. 2018
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A gated CAR design combining an attenuated ADGRE2-CAR with a CLEC12A chimeric costimulatory receptor preferentially engaged ADGRE2-positive/CLEC12A-positive leukaemic stem cells over ADGRE2-low/CLEC12A-negative normal HSPCs, prevented antigen escape in AML xenografts, and produced off-target HSPC toxicity similar to that of a CD19-CAR which could be mitigated by reducing CAR T cell-derived interferon-gamma.
Cooperative CAR targeting to selectively eliminate AML and minimize escape. · Haubner S, Mansilla-Soto J, Nataraj S, Kogel F, Chang Q, de Stanchina E, et al. · Cancer Cell 2023What this evidence cannot show
Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. Preclinical: xenograft and humanised mice, no clinical data. The design uses CLEC12A as a costimulatory input rather than as the killing antigen, so its favourable toxicity does not transfer to a primary CLEC12A CAR. The comparison 'similar to a CD19-CAR' is not zero.
Evidence that argues against this (1)
- This design's whole premise is that CLEC12A alone was not selective enough to drive killing safely - which is itself an argument that single-antigen CLEC12A targeting carries a toxicity burden the field has already decided to engineer around. Xiu Y et al. 2018
the CLEC12A target sits in the NK receptor family and on normal myeloid cells
on-target off-tumour toxicity of every CLEC12A-directed modality
inducing the antigen may widen the on-target/off-tumour overlap onto normal myeloid cells
observed in CLL-1-directed CAR T programmes
no CLEC12A safety or expression data exist in infants; infection is already the leading non-relapse cause of death on infant protocols
- related to →Lineage switch as antigen escape from CD19-directed therapy in KMT2A-rearranged leukemia
the two antigen-directed failure modes in this disease: lineage switch away from CD19, and antigen-negative escape from CLEC12A