The CXXC-DNA affinity is a ten-point ladder from 44 uM to 33 nM, not two numbers 1,300-fold apart, and every step but the last is under 5.5-fold

Read the table, not the endpoints. Sorted by affinity, every adjacent step from row 1 to row 9 is 5.5-fold or smaller - most are under 2-fold - and the only large step is the final 19-fold to Birke's SPR, which is precisely the step this beat could not resolve. The 1,300-fold "discrepancy" is the length of a well-populated ladder plus one unexplained rung, and this graph had been quoting the bottom rung as if it were the value.

Three pairs isolate variables almost cleanly:

• Rows 1 and 6 differ in salt and method, and 7.7-fold. Both tagless, both a 12-bp palindrome with a single CpG, both a solution equilibrium measurement of the isolated domain. What changes is 300 mM NaCl against 150 mM NaCl, and NMR titration against EMSA. This is the closest thing the literature has to a direct measurement of the salt term, which nothing else in this beat could price - and it says the ionic-strength choice alone is worth roughly an order of magnitude on this contact.
• Rows 8 and 9 differ in three things and 5.5-fold. Same laboratory, same protein span 1147-1203, the same 12-mer oligonucleotide GGGTCGCGGGAG. What changes is the GST tag (absent / present), the salt (300 mM NaCl / 150 mM KCl) and the readout (ITC / FP). Since rows 1 and 6 put roughly 8-fold on the salt drop alone, the GST tag is worth very little here - possibly nothing. The explanation the field and this graph both reached for is bounded almost to zero by this pair.
• Rows 2 and 8 differ only in the DNA sequence, and 9-fold. Same laboratory, same construct, same salt, same method, same temperature. The two Hoxa9 sites MLL actually protects from methylation bind at 3.5 and 7.4 uM; the two matched sites it does not protect bind at 32 and 15 uM.
• Rows 9 and 10 differ in four things and 19-fold: 41 extra C-terminal residues, six CpGs instead of two, a 52-bp duplex instead of a 12-mer, and an immobilised ligand read kinetically instead of a solution equilibrium. Nothing published separates those four, which is the one genuine hole this beat could not close.

The 44 uM measurement is the least physiological point in the series, on every axis at once. It uses the shortest construct (57 aa, the minimal fold, with none of the flanking basic sequence), the weakest DNA ligand (a designed palindrome with one CpG, not a CpG island), and 300 mM NaCl - roughly twice the monovalent ionic strength of a cell nucleus. A protein-DNA contact carried by an electropositive surface is the most salt-sensitive class of interaction there is; the three residues whose alanine substitution costs most in that very paper are Arg1154, Lys1185 and Lys1193 1.

And the salt term is now measured, not merely predicted. Liang 2021 titrated a tagless, chemically synthesised CXXC domain against a 12-bp single-CpG palindrome at 150 mM NaCl and obtained 5.7 +/- 0.2 uM 7. Against Cierpicki's 44 uM on the same class of ligand with the same tag state, that is 7.7-fold for halving the salt (confounded with NMR against EMSA, and with temperature). The single most physiologically apt tagless measurement of this contact is 5.7 uM, and this graph has never cited it.

That is not a criticism of Cierpicki et al. 300 mM NaCl is a defensible choice for an NMR titration of a small basic domain against DNA - it suppresses non-specific association and keeps the sample behaved. The error is downstream: quoting a number measured under conditions chosen for tractability as though it were the affinity of the contact in a nucleus.

And the discrepancy was resolved in print in 2008, by the same laboratory, in a paper this graph does not cite. Erfurth et al. wrote 2, verbatim: "Previous measurements of binding of the MT domain to CG elements, using surface plasmon resonance (SPR), yielded a K d of 3 x 10-8 M, approximately 2 orders of magnitude tighter than the ITC measurements reported here. This set of experiments used a GST fusion with MT and a DNA site with six CG residues each separated by 6-8 nt. The dimerization of GST has been shown to influence fused DNA binding domains. The potential in these experiments for a bivalent interaction of the GST-MT with two sites on the DNA could result in the increase in affinity seen." The same paragraph gives the MT domain's boundaries as MLL amino acids 1147-1244.

So the field noticed in 2008. What the field did not have, and what rows 8 and 9 of this table now supply, is a bound on how much of the gap GST can account for. Those two differ by 5.5-fold across tag plus a salt halving, and rows 1 and 6 attribute roughly 8-fold to the salt halving on its own - so the GST tag's own contribution is small and may be nil. The 2008 explanation is the one everybody repeated, including this graph's own caveats last cycle, and it is very close to being wrong.

What dies. The sentence this graph repeats - "44 uM is a weak interaction ... three orders of magnitude weaker than the menin contact" - does not survive. It compares the weakest of nine measurements of the CXXC-DNA contact against the reference measurement of the menin contact, and the two were made at 300 mM and 50 mM NaCl respectively (see ana-the-menin-affinity-comparison-was-made-at-one-sixth-the-salt). Taking the nearest-to-physiological tagless measurement of each, the ratio is 5.7 uM against 53 nM: 108-fold, not a thousand-fold - and the menin figure still enjoys a third of physiological salt. With Risner's GST-tagged 0.64 uM it is twelve-fold. On Birke's geometry the CXXC contact is tighter than the menin contact. No entry should state a fold-difference between these two contacts without naming the ionic strength of both.

What survives, and is stronger than what it replaces. The conclusion that the CXXC domain cannot be the fusion's address is untouched, because it never depended on the affinity being weak:

1. Affinity rank does not predict function in this exact system. The CGBP/CXXC1 CXXC domain binds unmethylated CpG DNA more tightly than the DNMT1 CXXC domain, and it is the DNMT1 domain that substitutes for MLL CXXC in MLL-AF9 and causes leukaemia while the CGBP domain does not. The authors state it: "DNA binding activity to relevant target DNA sequence is required, but a high DNA binding affinity alone is not sufficient" 4. See ana-affinity-rank-does-not-predict-cxxc-function.
2. In the same experiment, every swapped domain - including the non-functional ones - still targeted MLL-AF9 to the Hoxa9 locus by ChIP 4. Localisation did not track DNA-binding affinity even within a controlled swap series.
3. Cierpicki's own conclusion is about specificity, not strength: the CXXC domain "does not recognize DNA bases beyond the CpG dinucleotide and therefore is not solely responsible for specific localization to the Hoxa9 locus", and sites with multiple CpG motifs gain only 2-4 fold 1. A domain that reads a two-base motif present tens of millions of times in the genome cannot specify an address at any affinity - and a tighter one is worse, because it is more completely titrated by bulk CpG.
4. Genome-wide, the fusion under-occupies the address it supposedly reads. In SEM cells MLL-AF4 sits on 20% of unmethylated CpG sites while CFP1 sits on 50% and KDM2B on 89%, and fusion occupancy correlates with menin at r2 = 0.96 6.

So the argument must be re-founded on specificity and on measured occupancy, and stop being an argument about a number. The entries that made the affinity argument have been rewritten accordingly.

The contribution of KMT2A 1204-1244 to DNA binding has never been measured. Every published measurement either stops at 1203 or 1214, or is Birke's GST-MT, which changes four things at once. The residual Risner-to-Birke step is 19-fold and confounds: (i) the 41 extra C-terminal residues, a strongly basic, UniProt-disordered patch (AYLQKQAKAVKKKEKKSKTSEKKDSKESSVVKNVVDSSQKP - 13 lysines, 5 acidic residues, net charge about +8); (ii) six CpGs against two; (iii) a 52-bp duplex against a 12-mer; (iv) ligand immobilised on a chip and read kinetically, against a solution equilibrium.

This is resolvable with one afternoon of ITC or MST - see exp-tagless-side-by-side-cxxc-versus-mt-domain. Until it is run, no entry should attribute the gap to GST dimerisation as though that were established: rows 8 and 9 bound the GST-plus-salt contribution at 5.5-fold, and rows 1 and 6 put most of that on the salt, leaving the tag with very little.

A fifth independent measurement exists and could not be read. PMID 29276034 (Xu et al., Structure 2018) systematically measured the DNA-binding selectivities of all human CXXC domains and deposited 4NW3, the MLL CXXC-DNA crystal structure. All five retrieval routes failed for it this cycle - see concept-pmid-29276034-holds-a-fifth-cxxc-measurement-and-is-unreadable.

A note on how this beat found its sixth paper, because it bears on the reliability of the rest. Liang 2021 was not found by any of this beat's PubMed queries. It was already in this graph, as a counter-evidence note inside exp-cxxc-reader-depletion-versus-kmt2a-fusion-occupancy - one of the six entries the beat was dispatched to correct. The queries missed it because MLL[tiab] does not match the token MLL1: MLL[tiab] AND CXXC[tiab] returns 21 records and excludes it, while MLL1[tiab] AND CXXC[tiab] returns 8 and includes it. See concept-mt-domain-and-cxxc-domain-are-two-names-for-one-region-and-the-literatures-do-not-intersect. The ladder should be read as ten measurements found, not ten measurements that exist.