Molecular Architecture and Function Mechanism of Tri-heteromeric GluN1-N2-N3A NMDA Receptors

Reviewer #1 (Public review):

Summary:

The previous evidence for NMDARs containing N1, N2, and N3 subunits (t-NMDARs) was weak. All previous results could be explained by mixtures of di-heteromeric receptors. The authors here set out to identify t-NMDARs both in vitro and in the brain.

Strengths:

The single-channel recording is quite convincing because the authors could reproduce previous results in their system, but could also then add new observations. It is quite hard (if not impossible) to obtain the N1-N2A-N3A result at 100 µM Glu/Gly from a mixture, because the N1-N2A diheteromer has such a high open probability. Therefore, any idea that this might be, in fact, two receptors (GluN1-N2A and GluN1-N3A) is trivially falsified. The authors might prefer to make this argument based on the reduction of open probability, which cannot be achieved from a mixture masquerading as a single channel.

With regard to crosslinker usage in brain tissue, these are very impressive attempts, which I applaud. The fluorescence images of the brain sections look convincing. But the bands corresponding to N2-N3 crosslinked subunits from neurons or the brain are faint. I would want more information to be convinced that these faint bands come from GluN2-N3 dimers.

Weaknesses:

In the first part of the paper, where the CryoEM structure is determined, it's not really clear to me the extent to which Fab binding might bias the position of the ATDs (and even then the arrangement of each subunit within the whole complex). Then, much later at the end of the results, there is a structural analysis that claims to be integrative (Figure 7) but does not obviously rely on any other data than the structures, but does mention this point about the Fabs. The results could be rearranged to make these points clearer.

I have my biggest doubts about the crosslinking of native receptors. For the biochemistry from neurons or brain tissue, this is a very ambitious idea that has been hard to execute over the past 15-20 years. The authors use AzF for the obvious reason that this was done before in NMDARs. The constructs that have been assembled are neat. But AzF is a really bad crosslinker. The authors attribute the weak bands to subunit mobility, but the minor abundance is more likely due to the strong constraints on AzF crosslinking and its unsuitable photochemistry in general (very easily activated with room light, for example).

There is no information at all given about the wavelength, intensity, duration of UV exposure, and how, for example, the right exposure was determined. How were the samples protected in between?

Reviewer #2 (Public review):

Summary:

The authors purified and solved by cryo-EM a structure of tri-heteromeric GluN1/GluN2A/GluN3A NMDA receptors, whose existence has long been contentious. Using patch-clamp electrophysiology on GluN1/GluN2/GluN3A NMDARs reconstituted into liposomes, they characterized the function of this NMDAR subtype. Finally, thanks to site-targeted crosslinking using unnatural amino acid incorporation, they show that the GluN2A subunit can crosslink with the GluN3A subunit in a cellular context, both in recombinant systems (HEK cells) and neuronal cultures and in vivo.

Strengths:

The NMDAR GluN3 subunit is a glycine-binding subunit that was long thought to assemble into GluN1/GluN2/GluN3 tri-heteromeric receptors during development, acting as a brake for synaptic development. However, several studies based on single subunit counting (Ulbrich et al., PNAS 2008) and ex vivo/in vivo electrophysiology have challenged the existence of these tri-heteromers (see Bossi, Pizzamiglio et al., Trends Neurosci. 2023). A large part of the controversy stems from the difficulty in isolating the tri-heteromeric population from their di-heteromeric counterparts, which led to a lack of knowledge on the biophysical and pharmacological properties of putative GluN1/GluN2/GluN3 receptors. To counteract this problem, the authors used a two-step purification method - first with a strep-tag attached to the GluN3 subunit, then with a His tag attached to the GluN2 subunit - to isolate GluN1/GluN2/GluN3 tri-heteromers from GluN1/GluN2A and GluN1/GluN3 di-heteromers, and they did observe these entities in Western blot and FSEC. They solved a cryo-EM structure of this NMDAR subtype using specific FAbs to identify the GluN1 and GluN2A subunits, showing an asymmetrical, splayed architecture. Then, they reconstituted the purified receptors in lipid vesicles to perform single-channel electrophysiological recordings. Finally, in order to validate the tri-heteromeric arrangement in a cellular system, they performed photocrosslinking experiments between the GluN2A and GluN3 subunits. For this purpose, a photoactivatable unnatural amino acid (AzF) was incorporated at the bottom of GluN2A NTD, a region embedded within the receptor complex that is predicted to be in close proximity to the GluN3 subunit. This is an elegant approach to validate the existence of GluN1/GluN2/GluN3 tri-hets, since at the chosen AzF incorporation position, crosslinking between GluN2A and GluN3 is more likely to reflect interaction of subunits within the same receptor complex than between two receptors. They show crosslinking between GluN2A and GluN3 in the presence of AzF and UV light, but not if UV light or AzF were not provided, suggesting that GluN2A and GluN3 can indeed be incorporated in the same complex. In a further attempt to demonstrate the physiological relevance of these tri-heteromers, they performed the same crosslinking experiments in cultured neurons and even native brain samples. While unnatural amino acid incorporation is now a well-established technique in vitro, such an approach is very difficult to implement in vivo. The technical effort put into the validation of the presence of these tri-heteromers in vivo should thus be commended.

Overall, all the strategies used by this paper to prove the existence of GluN1/GluN2/GluN3 tri-heteromers, and investigate their structure and function, are well-thought-out and very elegant. But the current data do not fully support the conclusions of the paper.

Weaknesses:

All the experiments aiming at proving the existence of GluN1/GluN2/GluN3 tri-heteromers rely on the purification of these receptors from whole cell extracts. There is therefore no proof that these receptors are expressed at the membrane and are functional. This is a limitation that has been overlooked and should be discussed in the manuscript. In addition, in the current manuscript state, each demonstration suffers from caveats that do not allow for a firm conclusion about the existence and the properties of this receptor subtype.

(1) In Cryo-EM images of GluN1/GluN2A/GluN3A receptors, the GluN3 subunit is identified as the subunit having no Fab bound to it. How can the authors be sure that this is indeed the GluN3A subunit and not a GluN2A subunit that has not bound the Fab? Does the GluN3A subunit carry features that would allow distinguishing it independently of Fab binding? In addition, it is surprising that the authors did not incubate the tri-heteromers with a Fab against GluN3A, since Extended Figure 3 shows that such a Fab is available.

(2) Whether the single-channel recordings reflect the activity of GluN1/GluN2/GluN3 tri-heteromers is not convincing. Indeed, currents from liposomes containing these tri-heteromers have two conductance levels that correspond to the conductances of the corresponding di-heteromers. There is therefore a need for additional proof that the measured currents do not reflect a mixture of currents from N1/2A di-heteromers on one side, and N1/3A di-heteromers on the other side. What is the purity of the N1/3A sample? Indeed, given the high open probability and high conductance of N1/2A tri-heteromers, even a small fraction of them could significantly contribute to the single-channel currents. Additionally, although the authors show no current induced by 3uM glycine alone on proteoliposomes with the N1/2A/3A prep (no stats provided, though), given the sharp dependence of N1/3A currents on glycine concentration, this control alone cannot rule out the presence of contaminant N1/3A dihets in the preparation.

Finally, pharmacological characterization of these tri-heteromers is lacking. In vivo, the presence of tri-heteromeric GluN1/GluN2/GluN3 tri-heteromers was inferred from recordings of NMDARs activated by glutamate but with low magnesium sensitivity. What is the effect of magnesium on N1/2A/3A currents? Does APV, the classical NMDAR antagonist acting at the glutamate site, inhibit the tri-heteromers? What is the effect of CGP-78608, which inhibits GluN1/GluN2 NMDARs but potentiates GluN1/GluN3 NMDARs? Such pharmacological characterization is critical to validate that the measured currents are indeed carried by a tri-heteromeric population, and would also be very important to identify such tri-heteromers in native tissues.

(3) Validation of GluN1/GluN2/GluN3 tri-heteromer expression by photocrosslinking: The mixture of constructions used (full-length or CTD-truncated constructs, with or without tags) is confusing, and it is difficult to track the correct molecular weight of the different constructs. In Figure 6, the band corresponding to a putative GluN3/GluN2A dimer is very weak. In addition, given the differences in molecular weights between the GluN2 subunits and GluN3, we would expect the band corresponding to a GluN2A/GluN2B to migrate differently from the GluN2A/GluN3 dimer, but all high molecular weight bands seem to be a the same level in the blot. Finally, in the source data, the blots display additional bands that were not dismissed by the authors without justification. In short, better clarification of the constructs and more careful interpretation of the blots are necessary to support the conclusions claimed by the authors.