A tighter CXXC domain fails and a weaker one works: affinity rank does not predict function in the MLL-AF9 swap series
also called Risner swap · CGBP beats DNMT1 on affinity and loses on function
Four different CpG-gripping modules were put into the fusion in place of its own. The one that gripped DNA more tightly could not make leukaemia and the one that gripped more weakly could. How tightly the module holds DNA is not what decides whether the fusion works.
The same thing, for a scientist
Risner 2013 measured DNA binding of four isolated CpG-recognition domains by fluorescence polarisation against one Hoxa9-derived 12-mer, then substituted each into MLL-AF9 and tested colony formation and leukaemogenesis. MLL CXXC is tightest at Kd 0.64 uM; CGBP/CXXC1 is second; DNMT1 CXXC is roughly 34-fold weaker than MLL; MBD1 CXXC and the MBD1 MBD show no binding. Functionally, MLL(DNMT1 CXXC)-AF9 gives robust colony formation and in vivo leukaemogenesis comparable to MLL-AF9, while the tighter CGBP substitution and both MBD1 substitutions are dead. The authors state the conclusion: DNA binding to the relevant target sequence is required but high affinity alone is not sufficient. Separately, and decisively for any argument that reasons from affinity to localisation, ALL substituted domains - including the non-functional ones - still targeted MLL-AF9 to the Hoxa9 locus by ChIP. This is the strongest available replacement for the affinity-based version of the too-weak-to-hold argument, because it is an experiment in the fusion, in vivo, in which affinity and function are directly dissociated.
The dissociation, in one table
All affinities: GST-tagged isolated domain, fluorescence polarisation against fluorescein-GGGTCGCGGGAG (a two-CpG Hoxa9 site), 50 mM Tris / 150 mM KCl / 1 mM DTT / 10 uM ZnCl2 1.
| domain swapped into MLL-AF9 | DNA binding | relative affinity | colony formation | leukaemia in vivo |
|---|---|---|---|---|
| MLL CXXC (1147-1203) | Kd 0.64 uM | 1 (tightest) | yes | yes |
| CGBP/CXXC1 (161-217) | binds | ~7-fold weaker than MLL | no | no |
| DNMT1 CXXC (645-700) | binds | ~34-fold weaker than MLL | yes, robust | yes |
| MBD1 CXXC-1 (168-224) | none detectable | - | no | no |
| MBD1 MBD (1-75) | none detectable | - | no | no |
The rank order of affinity and the rank order of function disagree in the middle of the table. CGBP binds DNA roughly five-fold more tightly than DNMT1 and cannot make leukaemia; DNMT1 binds more weakly and can. The authors' own words 1: "for a C XX C domain to function in the context of an MLL-AF9 fusion protein, DNA binding activity to relevant target DNA sequence is required, but a high DNA binding affinity alone is not sufficient."
What did predict function was which specific CpG residues at Hoxa9 stayed unmethylated - a property shared by MLL and DNMT1 CXXC and not by CGBP CXXC.
Why this replaces the affinity argument rather than merely qualifying it
This project's standing claim has been: the CXXC contact is too weak to hold the fusion on chromatin by itself, so menin, LEDGF and the AT-hooks must supply the grip. The premise was a number, 44 uM, and ana-cxxc-dna-affinity-ladder shows that number is the extreme of a nine-point series.
The conclusion does not need the premise. Three results in this one paper carry it better:
- Localisation is not affinity-determined. All substituted domains, including the two that bind no DNA at all, still targeted MLL-AF9 to the Hoxa9 locus by ChIP 1. If a fusion carrying a CpG-binding-dead MBD domain still arrives at Hoxa9, the arriving is not being done by the CpG-binding module. That is a direct, in-fusion measurement of exactly what the graph wanted to argue, and it does not mention a Kd.
- Function is not affinity-determined. The CGBP/DNMT1 inversion above.
- The requirement is a function, not a surface. What the fusion needs is protection of particular CpGs from methylation, which a paralogous domain performs as well as MLL's own.
So the correct form of the project's claim is: the CXXC domain is required for the fusion to work and is not what determines where the fusion goes. That is stronger than the affinity claim, it is directly measured in the index mechanism, and it is immune to every objection the affinity ladder raises.
| script | none - a reading of one primary paper's figures and text |
|---|---|
| status | run |
| result summary | Affinity rank and functional rank are inverted in the middle of the MLL-AF9 CXXC swap series: CGBP CXXC binds DNA more tightly than DNMT1 CXXC and cannot support leukaemogenesis, while DNMT1 CXXC (34-fold weaker than MLL CXXC) can. All swaps, including two that bind no DNA, still targeted the fusion to Hoxa9 by ChIP. The operative variable is protection of specific CpGs from methylation, not affinity. |
| fp conditions | 50 mM Tris-HCl, 150 mM KCl, 1 mM DTT, 10 uM ZnCl2; fluorescein-labelled GGGTCGCGGGAG; three independent experiments; Kd from anisotropy versus log[protein] fitted to a one-site sigmoid |
| system | murine MLL-AF9 retroviral transduction of c-Kit+ bone marrow progenitors; B6/SJL CD45.1+ recipients |
| shown for fusion | MLL-AF9 in mouse; nothing in KMT2A::AFF1 or B-lineage |
| replaces argument | The affinity-magnitude version of 'the CXXC contact is too weak to hold the fusion'. The surviving form is: the CXXC domain is required for the fusion to work and is not what determines where the fusion goes. |
inputs
[ "PMID 23990460 full text (PMC3795288)", "PMID 20010842 full text (PMC2908503)", "PMID 18483194 full text (PMC2396713)" ]
swap series
[
{
"domain": "MLL CXXC",
"residues": "MLL 1147-1203",
"kd_uM": 0.64,
"relative_kd": 1,
"colony_formation": "yes",
"leukaemia_in_vivo": "yes",
"targets_hoxa9_by_chip": "yes"
},
{
"domain": "CGBP/CXXC1 CXXC",
"residues": "CGBP 161-217 (163-215 in the fusion)",
"relative_kd": "about 5x weaker than MLL (second tightest)",
"colony_formation": "no",
"leukaemia_in_vivo": "no",
"targets_hoxa9_by_chip": "yes"
},
{
"domain": "DNMT1 CXXC",
"residues": "DNMT1 645-700 (649-697 in the fusion)",
"relative_kd": "34.4x weaker than MLL",
"colony_formation": "yes, robust",
"leukaemia_in_vivo": "yes",
"targets_hoxa9_by_chip": "yes"
},
{
"domain": "MBD1 CXXC-1",
"residues": "MBD1 168-224 (172-221 in the fusion)",
"relative_kd": "no binding detectable",
"colony_formation": "no",
"leukaemia_in_vivo": "no",
"targets_hoxa9_by_chip": "yes"
},
{
"domain": "MBD1 MBD",
"residues": "MBD1 1-75",
"relative_kd": "no binding detectable",
"colony_formation": "no",
"leukaemia_in_vivo": "no",
"targets_hoxa9_by_chip": "yes"
}
]- A ligand that occupies the MLL CXXC fold without abolishing CpG protection would be inactive, and a ligand that abolishes it invites functional substitution by DNMT1's own CXXC-containing machinery - a resistance route the paralogue-toxicity column does not capture.
- If localisation survives a CpG-binding-dead domain, a CXXC-directed agent may show perfect target engagement and no change in fusion occupancy, which would read as a failed hypothesis when it is a mis-specified endpoint.
- The whole series is murine myeloid. A B-lineage infant KMT2A::AFF1 leukaemia may weight the CXXC contribution differently, and nothing in this table would detect that.
- Would the swap series behave the same way in human KMT2A::AFF1 in a B-lineage context? Every functional result here is murine MLL-AF9 myeloid, and this graph's commonest documented synthesis failure is generalising exactly that way.
- Genome-wide, how far do the dead swaps actually get? The Hoxa9 ChIP is binary; CUT and RUN on the five swap constructs would show whether occupancy tracks affinity quantitatively even though function does not.
- What does the CGBP domain do differently that kills the fusion, given that it binds DNA better? Risner nominates residues shared by MLL and DNMT1 but not the others; nobody has tested them.
- If protection from methylation rather than binding is the requirement, does a DNMT-inhibitor-treated cell rescue a CXXC-dead fusion? That is a direct test of the mechanism and costs a decitabine arm on the existing constructs.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In MLL-AF9, substituting the MLL CXXC domain with the DNMT1 CXXC domain (roughly 34-fold weaker DNA binding) preserves colony formation and in vivo leukaemogenesis, while substituting the CGBP/CXXC1 domain (which binds DNA more tightly than DNMT1) abolishes both; all substituted domains including the non-binding MBD1 ones still targeted the fusion to the Hoxa9 locus by ChIP; and the discriminating property was protection of specific Hoxa9 CpG residues from methylation, not DNA-binding affinity.
Functional specificity of CpG DNA-binding CXXC domains in mixed lineage leukemia. · Risner LE, Kuntimaddi A, Lokken AA, et al. · J Biol Chem 2013What this evidence cannot show
Murine MLL-AF9 retroviral model in c-Kit+ bone marrow progenitors, not human KMT2A::AFF1 and not B-lineage - the standard system-mismatch for this graph (rule 73). The affinities are of GST-tagged isolated domains against one 12-mer, so the affinity axis of the comparison inherits every limitation in ana-cxxc-dna-affinity-ladder. The swapped constructs replace MLL amino acids 1150-1201, and each donor domain was designed with the same number of residues flanking its CXXC cysteines as MLL's, which controls length but not context. Full text read via the PMC HTML route (PMC3795288).
Evidence that argues against this (1)
- The ChIP result showing all domains still target Hoxa9 is the strongest datum here and also the least controlled: a ChIP enrichment at one locus in a transduced population does not measure occupancy quantitatively, and a fusion arriving at Hoxa9 by residual mechanisms while doing nothing there is consistent with the same data. A quantitative genome-wide comparison of the swap series - which nobody has run - could show that the dead swaps are targeted far less efficiently and that affinity does matter after all, just not enough to invert a binary ChIP. Kerry J et al. 2017
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Cierpicki et al. conclude from their own data that the CXXC domain does not recognise DNA bases beyond the CpG dinucleotide and is therefore not solely responsible for specific localisation to the Hoxa9 locus, with sites carrying multiple CpG motifs bound only 2- to 4-fold more tightly.
Structure of the MLL CXXC domain-DNA complex and its functional role in MLL-AF9 leukemia. · Cierpicki T, Risner LE, Grembecka J, et al. · Nat Struct Mol Biol 2010What this evidence cannot show
The authors' conclusion, read in the paper's Results; the supporting ITC panel is in a Supplementary Table that is behind the Europe PMC open-access gate and was not read. This is a specificity argument and is independent of the affinity magnitude, which is why it survives the affinity ladder unchanged.
Evidence that argues against this (1)
- Erfurth 2008 measured a 9-fold affinity spread across five CpG-containing oligonucleotides at fixed construct, salt and method, with the MLL-protected Hoxa9 sites tighter. So the domain's affinity landscape across CpG-containing DNA is not flat even if no base-specific readout outside the CpG has been identified structurally, and the mechanism of that spread is unresolved. Erfurth FE et al. 2008
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Structure-guided point mutants that abolish CXXC-DNA binding (C1188D) in MLL-AF9 increase CpG and H3K9 methylation at Hoxa9, reduce Hoxa9 transcripts, abolish immortalisation and abolish leukaemia in mice, establishing that the DNA-binding function of the domain is required even though its affinity is not what ranks the paralogues.
Structure of the MLL CXXC domain-DNA complex and its functional role in MLL-AF9 leukemia. · Cierpicki T, Risner LE, Grembecka J, et al. · Nat Struct Mol Biol 2010What this evidence cannot show
Murine MLL-AF9, not the index fusion. C1188D is designed to introduce electrostatic repulsion and was confirmed folded by HSQC, which is the right control, but a single designed charge reversal adjacent to the zinc-coordinating cluster is a strong perturbation and the mutants were not tested for menin, LEDGF or polycomb binding.
Evidence that argues against this (1)
- The C1188D result is a loss-of-function, and Risner's swap series is the matched gain-of-function: put a different DNA-binding domain in and leukaemia comes back with a domain 34-fold weaker, or fails to come back with a domain that binds tighter. Read together they show requirement without affinity-dependence, and neither experiment alone would have shown that. Risner LE et al. 2013
The direct in-fusion dissociation of affinity from function is the strongest support for the replacement argument.
The CGBP/CXXC1 CXXC domain binds unmethylated CpG DNA more tightly than DNMT1's and cannot support MLL-AF9 leukaemogenesis.
The DNMT1 CXXC domain is about 34-fold weaker than MLL's and supports MLL-AF9 leukaemogenesis.
Sharpens the dossier's existing functional-substitution warning with the affinity numbers that make it counterintuitive.
- related to →The human CXXC/CpG-reading family is exactly twelve proteins, and TET2 is not one of them
Places four of the twelve human CXXC domains on one affinity scale measured in one assay.
If affinity does not rank function, a competition model among CXXC readers cannot be settled by affinity either.
Row 8 of the ladder is the MLL arm of this swap series' affinity panel.
If read, it would either corroborate or contradict the rank order this replacement argument rests on.