Phase-2 dossier: CBX8 - the candidate this review would kill, and the reasons are all numeric
also called CBX8 dossier · CBX8 no-go · chromobox 8 phase-2 verdict
CBX8 looked promising on paper and even has a crystal structure with an inhibitor in it. But when we looked at the actual data, cancer cells do not need it, the two best papers about it disagree with each other, and the chemistry that exists is a hundred times too weak. We recommend not starting this one.
The same thing, for a scientist
CBX8 fails on three independent axes at once, which is why it is the recommended no-go. (1) Dependency: DepMap Chronos mean is positive (+0.101) - knocking CBX8 out slightly helps the average cancer line - and its co-dependency with MEN1 is 0.017 pooled, -0.002 within-tissue, 0.053 in leukaemia lines, i.e. indistinguishable from zero on every cut. (2) Chemistry: 15 ChEMBL rows at pChEMBL 6 or above with a best reported Kd of 120 nM, against a chromodomain pocket whose best public probes are CBX7-selective. (3) Literature: the graph's own gene-cbx8 entry records that two good papers flatly disagree about whether CBX8 is required, and the DepMap data sides with the paper that says no. The one favourable fact - very low normal-tissue expression (0.85 TPM whole blood) - is a therapeutic-index argument for a target that does not appear to be a target.
What structure exists
More than the dependency data deserves. Verified against RCSB this cycle: 3I91, human CBX8 with an H3K9 peptide, X-ray 1.55 A; 5EQ0, the CBX8 chromodomain in complex with the inhibitor UNC3866, X-ray 1.18 A 1. A 1.18 A inhibitor co-crystal is better structural support than most targets in this beat enjoy.
This is the trap. Structural availability is not a reason to run a programme. It is the cheapest of the five things a programme needs, and the only one CBX8 has.
The dependency data, which is the reason to stop
In the executed co-dependency scan, CBX8's mean DepMap gene effect across cell lines is +0.101 - positive, meaning CRISPR knockout is very slightly favourable to the average line - with leukaemia selectivity of +0.005 2. In the lineage-conditioned follow-up, its correlation with MEN1 is 0.017 pooled, -0.002 within-tissue, and 0.053 in leukaemia lines 3.
There is no cut of this dataset in which CBX8 behaves like a member of the menin module. The negative controls in the same analysis (ACTB pooled r 0.007, GAPDH 0.044) score comparably.
The literature disagreement, and why DepMap breaks the tie
The graph's biology entry gene-cbx8 records this as the one place in the KMT2A::AFF1 interactome where two good papers flatly disagree about whether CBX8 is required, with published work also placing CBX8 in tension with BCOR. When a field disagrees and a large unbiased screen sides clearly with one position, the screen should be believed unless someone can name the reason the screen would be blind to the effect. Nobody has named one here.
The named possibility worth recording: DepMap is a proliferation screen in immortalised lines over two to three weeks. A CBX8 requirement that is specific to leukaemia stem cells, or to the fetal-origin cell state of infant ALL, would be invisible in it. That is a legitimate objection, and it is the single condition under which this no-go should be revisited.
Normal tissue: the one good number
GTEx v8: CBX8 is the least expressed candidate in this beat by a wide margin - 0.85 TPM whole blood, 0.44 TPM minimum in skeletal muscle, 11.9 TPM maximum in testis 4. gnomAD pLI 0.99991, LOEUF 0.363 5. If CBX8 were a dependency, this profile would make it a good target. It is not, so the number is decorative.
The verdict, written so it can be overturned
NO-GO. Do not start a CBX8 programme this phase.
The pre-registered condition for reopening: a CRISPR or degrader experiment in primary infant KMT2A::AFF1 blasts or a patient-derived xenograft - not a cell line - showing at least 50% reduction in leukaemia-initiating cell frequency by limiting-dilution transplant. That is the only assay class DepMap cannot see. If that experiment is run and is negative, CBX8 should be marked closed rather than dormant.
| target gene | gene-cbx8 |
|---|---|
| uniprot | Q9HC52 |
| chromodomain aromatic cage that reads methylated histone H3 lysine | |
| phase2 verdict | KILL. This is the candidate this review recommends stopping on the evidence. |
| compute backend | None. No further computational work is warranted before the reopening experiment. |
family selectivity liability
[ "CBX2", "CBX4", "CBX6", "CBX7" ]
structures verified this cycle
[
{
"pdb": "3I91",
"title": "Human chromobox homolog 8 (CBX8) with H3K9 peptide",
"method": "X-ray",
"resolution_A": 1.55
},
{
"pdb": "5EQ0",
"title": "Chromodomain of CBX8 in complex with inhibitor UNC3866",
"method": "X-ray",
"resolution_A": 1.18
}
]essentiality
{
"chronos_mean": 0.101,
"leukemia_selectivity": 0.005,
"pooled_r_vs_MEN1": 0.0172,
"within_tissue_r_vs_MEN1": -0.0016,
"leukemia_only_r_vs_MEN1": 0.0529,
"negative_control_comparison": {
"ACTB_pooled_r": 0.0071,
"GAPDH_pooled_r": 0.0443
}
}normal tissue gtex v8 tpm
{
"whole_blood": 0.852,
"min": {
"tissue": "Muscle_Skeletal",
"tpm": 0.44
},
"max": {
"tissue": "Testis",
"tpm": 11.943
},
"n_tissues": 54
}gnomad
{
"pLI": 0.99991,
"loeuf": 0.363
}chemical matter
{
"chembl_target": "CHEMBL3232684",
"rows_pchembl_ge_6": 15,
"best": {
"molecule": "CHEMBL4635096",
"type": "Kd",
"value_nM": 120,
"pchembl": 6.92
}
}go no go
{
"verdict": "NO-GO this phase",
"reasons": [
"DepMap mean gene effect positive (+0.101)",
"co-dependency with MEN1 at negative-control level on every cut",
"15 ChEMBL rows, best 120 nM, against a family-promiscuous fold",
"primary literature is in direct disagreement and the unbiased screen sides against the requirement"
],
"reopening_condition": "A limiting-dilution transplant experiment in primary infant KMT2A::AFF1 blasts or a PDX showing at least 50% reduction in leukaemia-initiating cell frequency on CBX8 loss. This is the only assay class DepMap is blind to. A negative result there closes the target permanently.",
"cost_of_reopening_experiment": "high - primary material plus in vivo limiting dilution; approximately one year"
}- CBX8 sits opposite BCOR in a published conflict; inhibiting CBX8 could shift the balance in the direction that favours disease rather than against it.
- CBX chromodomains are a redundant paralogue family, so a selective CBX8 ligand may be silenced by CBX2/4/6/7 compensation while a pan-CBX ligand carries Polycomb-wide toxicity.
- Spending a cycle on CBX8 has an opportunity cost measured in the LEDGF programme not being run. Declining a target is itself an intervention with consequences.
- Is there any published CBX8 requirement that was measured in leukaemia-initiating cells by limiting dilution rather than in bulk proliferation, which would be the one form of evidence DepMap cannot contradict?
- Does the CBX8-BCOR tension recorded in gene-cbx8 mean that CBX8 inhibition would push cells in the direction of disease rather than away from it?
- Would a CBX-family-wide degrader behave differently from a CBX8-selective ligand, given the paralogue redundancy that could mask a single-gene knockout?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
Two CBX8 chromodomain structures were verified by direct query of the RCSB data API this cycle: 3I91 (CBX8 with H3K9 peptide, X-ray 1.55 A) and 5EQ0 (CBX8 chromodomain with inhibitor UNC3866, X-ray 1.18 A).
RCSB PDB data API entry records for 3I91 and 5EQ0 · RCSB PDB 2026What this evidence cannot show
UNC3866 is a chemical probe developed primarily against CBX7; its presence in a CBX8 co-crystal demonstrates ligandability of the fold, not a CBX8-selective starting point.
Evidence that argues against this (1)
- Two chromodomain structures establish the reader module is crystallisable. Chromodomains are small, well-behaved and bind short methylated peptides, which makes them easy to solve and hard to drug selectively — the same fold recurs across the CBX paralogues and selectivity, not structure availability, is the limiting problem. RCSB PDB 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In this project's executed DepMap co-dependency scan, CBX8 has a positive mean gene effect (+0.101) and a leukaemia selectivity of +0.005, so CRISPR knockout is not deleterious to cancer lines and is not preferentially deleterious to leukaemia lines.
Oncofusion KG analysis: DepMap co-dependency scan against MEN1 and DOT1L (analyses/depmap-codependency/results.md) · Iqbal Lab Oncofusion Knowledge Graph 2026What this evidence cannot show
DepMap is a bulk proliferation screen in immortalised lines over two to three weeks; it cannot detect a requirement confined to leukaemia-initiating cells or to a fetal-origin cell state.
Evidence that argues against this (1)
- This caveat is the whole counter-case for CBX8 and is why the no-go carries an explicit reopening condition rather than being final. source 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
CBX8's co-dependency with MEN1 is indistinguishable from zero on every cut of the lineage-conditioned analysis: 0.017 pooled, -0.002 within-tissue, 0.053 in leukaemia lines - comparable to the ACTB and GAPDH negative controls.
Oncofusion KG analysis: does the MEN1 co-dependency signal survive lineage adjustment? (analyses/depmap-lineage-conditioned/results.md) · Iqbal Lab Oncofusion Knowledge Graph 2026What this evidence cannot show
Co-dependency correlation measures shared requirement, not physical interaction; a genuine physical partner that is functionally redundant would also score zero here.
Evidence that argues against this (1)
- A co-dependency indistinguishable from zero on every cut is the strongest single statement in this dossier and it argues against the target rather than for it. The entry should carry that as the headline; a null co-dependency plus low normal-tissue expression is a target nobody has a reason to pursue, not a clean one. DepMap / Open Targets 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
CBX8 is the least expressed candidate in this beat in normal tissue - 0.85 TPM whole blood, 0.44 TPM minimum in skeletal muscle, 11.9 TPM maximum in testis, across all 54 GTEx v8 tissues.
GTEx Portal API v2, medianGeneExpression for CBX8, datasetId gtex_v8 · GTEx Consortium 2026What this evidence cannot show
Values near 1 TPM are close to the noise floor of bulk RNA-seq and should not be over-interpreted as evidence of absence.
Evidence that argues against this (1)
- GTEx TPM is BULK tissue RNA. It gives neither protein abundance nor cell-type resolution within a tissue, so a low whole-blood figure is compatible with high expression in the progenitor compartment that matters here, and a high figure in a bulk tissue does not localise the risk. Low bulk expression is also double-edged for a chromatin reader: a protein present at low copy number can still be rate-limiting, and low TPM does not imply a wide therapeutic window. GTEx Portal 2026
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
CBX8 is constrained against loss of function in humans (pLI 0.99991, LOEUF 0.363).
gnomAD GraphQL API, gnomad_constraint for CBX8 · Broad Institute 2026What this evidence cannot show
Constraint on a gene that DepMap says is dispensable in cancer lines is a reminder that population constraint reflects developmental and organismal roles, not cell-autonomous proliferation.
Evidence that argues against this (1)
- A gnomAD constraint metric is not a safety signal for a drug, and this beat uses it as one across every dossier. pLI and LOEUF measure intolerance to HETEROZYGOUS loss-of-function across a whole lifetime from conception, in a population sample. A drug gives partial, reversible occupancy in an adult for weeks. High constraint says the gene matters in development; it does not predict on-target toxicity of transient pharmacological inhibition, and this project has already had to make the same correction for germline MEN1 penetrance in the endocrine beat. CBX8 at pLI 0.99991 is being read as a caution while its co-dependency is zero, so the two metrics point in opposite directions and neither is decisive. gnomAD, Broad Institute 2026
- E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
ChEMBL holds only 15 activity rows against CBX8 (CHEMBL3232684) at pChEMBL 6 or above, with a best reported affinity of Kd 120 nM.
ChEMBL REST API, activities for target CHEMBL3232684 with pchembl_value >= 6 · EMBL-EBI ChEMBL 2026What this evidence cannot show
120 nM against a chromodomain is respectable for a probe and inadequate for a lead; and CBX-family cross-reactivity is the norm rather than the exception.
Evidence that argues against this (1)
- A ChEMBL activity-row count measures how much medicinal chemistry has been DONE, not how tractable the target is. The best reported affinity comes from the most optimised compound in whichever series was pursued, so it is a property of the investment rather than of the site, and comparing row counts across targets ranks attention rather than druggability. Fifteen rows is not evidence of intractability; it is evidence that almost nobody has tried. ChEMBL, EMBL-EBI 2026
CBX8 sits at negative-control level in both analyses (-0.002 within-tissue)
SW2_110A probe data and the Tan-versus-Maethner contradiction bear on the dossier
registers the no-go verdict and its single reopening condition
the executed DepMap data sides against a CBX8 requirement; no-go this phase
positive mean gene effect and near-zero selectivity come from this executed scan
- related to →BCOR — the other side of the CBX8 conflict, and a target you would be inhibiting in the direction of disease
the published CBX8/BCOR tension that makes the direction of any CBX8 intervention uncertain
the no-go verdict rests entirely on this data pull and inherits its blindness to stem-cell requirements
supplied the sub-1-TPM whole-blood profile
not funded this phase; the no-go is a decision, not an omission