Our Family's Research Initiative — Investigating EEF1A2 D91N

Status, in one paragraph

A molecular dynamics investigation comparing wild-type and D91N eEF1A2 is being carried out with an independent research collaborator on high-performance computing resources. It is ongoing. It has produced no validated findings, and nothing on this page should be read as a result.

No functional defect has been demonstrated

We state this prominently because it is the single easiest claim for a family-led project to overstate. A simulation that has not been analysed cannot support a mechanistic conclusion, and a simulation that has been analysed still cannot demonstrate a functional defect on its own — see the limitations section below. When validated findings exist, they will appear on the research updates page with their analysis date, and the pre-registered plan will be published alongside them so the two can be compared.

1 · The molecular dynamics investigation

What this method can address

Molecular dynamics integrates the equations of motion for every atom in a solvated protein model, producing a trajectory — a film of how the structure moves. For a question like this one it can, in principle, speak to: Established

Conformational dynamics — which parts of the protein move, how far and how often.
Local structural stability — whether a region holds its fold or frays.
Hydrogen bonding — which contacts are present and what fraction of the time they are occupied.
Solvent accessibility and water penetration — whether water reaches regions that should be buried, which is often how a structural defect first shows itself.
Switch-region flexibility — directly relevant here, since D91N sits adjacent to the switch II helix.
Magnesium coordination geometry — whether the ion's coordination shell is maintained.
Wild-type versus mutant comparison — the only comparison that means anything, since absolute values from a simulation are not interpretable on their own.

What it cannot address — stated before any results exist

A simulation is a hypothesis generator, not evidence of disease mechanism. Specifically:

It samples nanoseconds to microseconds. The events that matter biologically — folding, degradation, a ribosome stalling for seconds — are far slower and are simply not in the trajectory.
It depends entirely on the starting structure. A model built from the wrong nucleotide state cannot answer a question about nucleotide-dependent behaviour.
It depends on the force field, which is an approximation, and magnesium–protein interactions are among the harder things for standard force fields to reproduce accurately.
Differences between two short trajectories are frequently noise. Without replicates treated as the statistical unit, almost any pair of simulations will appear to differ somewhere.
It contains no ribosome, no tRNA, no chaperone, no proteasome and no cell. It cannot tell you what a neuron does.

The honest ceiling: a well-designed simulation can tell us which class of mechanism remains open, and therefore which experiment is worth someone's time and money. That is genuinely useful, and it is much less than a result.

Why the analysis plan was fixed in advance

A trajectory contains thousands of measurable quantities. If you look first and choose your measure afterwards, you will find a difference — and you will have learnt nothing, because you would have found one either way. The protective step is to commit to the analysis before seeing the data.

The plan therefore specifies, in advance: the systems to be compared (wild type and D91N, across nucleotide states), replicates as the statistical unit rather than individual frames, a defined set of observables with thresholds fixed beforehand, a positive control gate — does the wild-type model distinguish its nucleotide states at all? If it does not, the simulation cannot answer the question and no comparison of the mutant is interpretable — and a decision tree whose branches include, explicitly, "no robust difference".

That last branch is the point. A pre-registered plan that cannot return a null result is not a plan.

A second methodological commitment: the primary endpoint is a difference of differences rather than a difference. Comparing mutant and wild type within each nucleotide state, and then comparing those two comparisons, means that systematic errors inherited from the modelled starting structure largely cancel — and the result is a single number rather than a gallery of plots to choose from.

2 · The D91N structural hypothesis

Replacing aspartate with asparagine at position 91 removes a negative charge while leaving the residue's size and shape close to unchanged. Aspartate carries a carboxylate; asparagine carries a neutral amide. In a structural element whose job is to read the identity of a bound nucleotide, removing a charge is a plausible way to break a signal while leaving the architecture intact.

The consequences worth investigating — and each of these is a question, not a claim: Hypothesis

Loss or rearrangement of electrostatic interactions that the carboxylate previously made.
Altered hydrogen bonding, in number, geometry or occupancy.
Local structural rearrangement around the switch II helix.
Changed thermodynamic stability of the fold.
Altered conformational dynamics — the protein still folding correctly but moving differently.
Altered coupling between nucleotide state and protein conformation: the possibility that the mutant is, functionally speaking, no longer reading which nucleotide it holds.

Three things we are explicitly not assuming

That D91 coordinates magnesium. One study places this variant directly adjacent to the switch II helix, not within the catalytic centre, another lists the position as overlapping the eEF1B binding site, and none states that it contacts magnesium or the nucleotide. UniProt annotates residues 91–94 as the G3 region and lists no GTP or magnesium binding at residue 91. No magnesium-coordination claim appears on this site.
That D91 participates in catalytic GTP hydrolysis. Same reason. Adjacency in a model is not a catalytic role.
That findings from other variants transfer. G70S lies in switch I and behaves as essentially non-functional in mice; D252H disrupts binding to the exchange factor eEF1B and argues for a toxic gain of function; E122K was stable in neurons and unstable in muscle in the same animal. These are three different stories. D91N is a fourth, and it has not been told.

Open verification items

These are unresolved gaps in our own groundwork, listed because a reader deserves to know where the floor is thin. They are tracked and will be closed and dated on the updates page.

Open items as of the last review of this page.
Item	Why it matters
Read the supplementary structural figure and results of the cohort paper that places D91N adjacent to switch II.	The main text gives the placement but not the distances or contacts behind it.
Identify which experimental structures exist for eEF1A2, and in which nucleotide state.	The structural models we have identified carry GDP. If no experimentally determined GTP-state structure exists, a GTP-state model is built rather than observed — which changes how much weight its magnesium geometry can carry.
Record the simulation's own run parameters: nucleotide state, force field, water model, ion parameters, length, number of replicates.	Without these the trajectory is not interpretable by anyone else, and not reproducible by us.

3 · Future updates, figures and trajectories

This section is reserved and currently empty. When validated results exist it will hold: the pre-registered analysis plan as it was fixed; the run parameters; the observables with their thresholds; figures with their underlying data; and the trajectory files or an archive link, so that the analysis can be repeated by someone who does not trust ours.

If the result is a null — no robust difference between wild type and mutant — that will be published here with the same prominence as any other outcome.

Mechanistic hypotheses Experimental roadmap Collaborate with us

References

Rendered from the site bibliography. Each entry states how far the source was actually read — a database record, an abstract, or the full text — because that distinction changes how much weight a citation deserves.