Altered surface mGluR5 dynamics provoke synaptic NMDAR dysfunction and cognitive defects in Fmr1 knockout mice

Metabotropic glutamate receptor subtype 5 (mGluR5) is crucially implicated in the pathophysiology of Fragile X Syndrome (FXS); however, its dysfunction at the sub-cellular level, and related synaptic and cognitive phenotypes are unexplored. Here, we probed the consequences of mGluR5/Homer scaffold disruption for mGluR5 cell-surface mobility, synaptic N-methyl-D-aspartate receptor (NMDAR) function, and behavioral phenotypes in the second-generation Fmr1 knockout (KO) mouse. Using single-molecule tracking, we found that mGluR5 was significantly more mobile at synapses in hippocampal Fmr1 KO neurons, causing an increased synaptic surface co-clustering of mGluR5 and NMDAR. This correlated with a reduced amplitude of synaptic NMDAR currents, a lack of their mGluR5-activated long-term depression, and NMDAR/hippocampus dependent cognitive deficits. These synaptic and behavioral phenomena were reversed by knocking down Homer1a in Fmr1 KO mice. Our study provides a mechanistic link between changes of mGluR5 dynamics and pathological phenotypes of FXS, unveiling novel targets for mGluR5-based therapeutics.

Similarly, hyperactivity could artifactually reduce freezing time in the contextual fear conditioning test.
The potential confound of higher locomotor activity can be easily evaluated. (a) Number of seconds spent exploring each object should be shown, in addition to the discrimination index, for the novel object recognition phase. (b) Number of seconds spent exploring each of the two identical objects during the familiarization phase should be shown. (c) An independent open field test for exploratory locomotion must be conducted, and presented with standard activity parameters.
There is no explanation about how the behavioral data shown in Figure 10 derives from a different experiment than the behavioral data shown in Figure  Full statistical results must be added (e.g. F value, degrees of freedom, stringent posthoc analyses). Stating only the p value for ANOVA and t-tests is insufficient.
The authors ignore the very large body of literature showing normal cognitive scores in Fmr1 mice on many learning and memory tasks conducted by many independent labs, including the novel object recognition and fear conditioning assays which were conducted in the present studies. The authors are encouraged to read the behavioral literature more fully.
Absence of the full cognitive datasets, and of the necessary open field control experiment, represent serious flaws in this manuscript.

Reviewer #3 (Remarks to the Author):
This manuscript details interesting experiments investigating the interaction of mGluR5 and NMDA receptors in a mutant mouse model of fragile X (the Fmr1 KO). Specifically, the authors explore the dynamics of receptor diffusion in cultured neurons (using Q-dots), finding that the mobility of mGluR5 and the NMDAR are increased in the Fmr1 KO neurons. No change is seen in the mobility of GluA2, suggesting that the effect is somewhat specific to the mGluR-NMDAR complex. Based on the previous results of the Huber lab, they then show that disrupting mGluR5-Homer binding mimics and occludes the increased receptor trafficking in the Fmr1 KO. Additionally, the co-localization of mGluR5, NMDAR and Homer1 is increased in in the synapses of the Fmr1 KO neurons.
In addition to the in vitro work, the authors examine the whether the NMDAR component of LTD is altered in young (P11-15) hippocampal slices from Fmr1 KO mice. Specifically, they measure NMDAR EPSCs during stimulation of group 1 mGluRs using the agonist DHPG. They find that the NMDAR component of DHPG-LTD is deficient in the Fmr1 KO, and this is mimicked by disruption of the mGluR-Homer interaction. To show that enhancing the mGluR-Homer interaction could rescue the deficit in NMDAR function in the Fmr1 KO, they injected shRNA to the dominant negative Homer 1a (or scrambled shRNA), and measured the NMDAR component of DHPG-LTD in the Fmr1 KO. They also performed novel object recognition and contextual fear conditioning on injected animals, and show a remarkable rescue of both behvaioral deficits.
Overall, this study is an interesting addition to the literature regarding the regulation of mGluR-NMDAR interaction, and the role of this in the fragile X mouse model. The in vitro Qdot experiments are particularly compelling. There are, however, a number of drawbacks that should be addressed, particularly with respect to the electrophysiology and behaviour experiments.
1.) The in vitro experiments using Q-dots show a convincing increase in the mobility of mGluR5 and NMDARs in the Fmr1 KO neurons. However, these experiments use Mitotracker labelling to define "synaptic sites" without providing verification using a more conventional synaptic marker (i.e., synaptophysin, PSD-95). This is important because all of the work shown in Figures 5-7 is dependent upon the "synaptic" association. These data would be much more robust if there was validation using another method, and if the total colocalization of mGluR5 and GluN1 was shown (not just the "synaptic" fraction).
2.) The LTD experiments are interesting in light of the specific focus on NMDAR EPSCs. However, there are a few issues. First, the developmental time-point at which these experiments are performed is P11-15, at which point mGluR-LTD is presynaptic (Nosyreva and Huber, 2005). The authors, however, interpret their effects in terms of a postsynaptic mechanism. This should be clarified by including measurements of PPR. These experiments should also be accompanied by standard LTD measurements to ensure that they are indeed inducing DHPG-LTD with their protocol. 3.) In Figure 7, both the mGluR5ct and control peptides seem to have an effect on the Fmr1 KO. This should be discussed. Accordingly, the electrophysiology shown in Figure 8 should include a peptide-treated Fmr1 KO group. 4.) The in vivo injection experiments require a number of controls, including an estimation of the efficiency of the Homer1a knockdown, and WT groups for the electrophysiology experiments.

5.)
The behaviour experiments seem to be the weaker parts of this study. The descriptions of the work are not very thorough, and the data appear hastily put together. For example, the methods state that behaviour experiments were only performed in the Fmr1 KO, yet there is data shown for WT. Were these experiments performed blind to genotype? Were the groups age-matched and counterbalanced for treatment? These details are essential for assessing the results, especially since the effects shown are so uncharacteristically large.
Moreover, many of these findings have been previously published. The deficits in contextual fear conditioning have been described previously (Eadie et al. 2012, Auerbach et al. 2011. While it is true that reduction of Homer1a in the Fmr1 KO has not been found to rescue these specific phenotypes, it has been thoroughly studied with respect to other phenotypes (Huber). Additionally, the deficit in contextual fear conditioning has been attributed to a hypofunction of NMDARs in the dentate gyrus (Eadie et al. 2012). How does this fit with the authors' other findings, which are focused on CA1?
The reviewer is right to mention that the diffusion coefficients measured in this study are different from the ones previously reported, although those differences remain moderate (as an example, median diffusion coefficient of 0.084 µm 2 /s for AMPAR and 0.037 µm 2 /s for NMDAR in Groc et al., 2004 versus 0.095 µm 2 /s for AMPAR and 0.11 µm 2 /s for NMDAR in the present study). Several factors can explain these differences: i) the former data (Sergé et al., 2002;Groc et al., 2004) were obtained in Banker cultures (neurons only) while mixed cultures (neurons and glia together) were used here, ii) the developmental stages were different between studies (3-8 days in vitro for Sergé et al., 2002, 9-11 days in vitro for Groc et al., 2004 days in vitro in the present study), iii) the species were different (rats in Sergé et al., 2002 andGroc et al., 2004 versus mice here), iv) the frequencies of image acquisitions were also different (25 Hz in Sergé et al., 2002 and15 Hz in Groc et al., 2004 versus 20 Hz in the present study), as well as v) the culture and recording medium. Each of these factors may account for these differences between studies. However, one should keep in mind that our data match recent publications performed in a similar configuration (see Ladépêche et al., 2013 andDupuis et al., 2014), that absolute diffusion values should be handled with caution since they depend heavily on the methods used to perform singleparticle tracking measurements, and that the same recording parameters were carefully kept along this study to allow unbiased comparison between the different experimental conditions.

Comment:
Are their problems with the specificity of antibodies? The antibodies used here are similar to the one used in previous studies?

Response:
The antibodies used to track AMPAR (mouse monoclonal anti-GluA2 antibody, #MAB387, Millipore) and NMDAR (rabbit polyclonal anti-GluN1 antibody, #AGC-001, Alomone labs) have been already used in many other studies by the groups of Drs. L. Groc and D. Choquet and demonstrated to be specific using various approaches (e.g. Opazo et al., 2010;Ladépêche et al., 2013;Zhang et al., 2013;Dupuis et al., 2014). The specificity of mGluR5 antibody (mouse monoclonal anti-NH 2 -mGluR5 antibody) has been previously demonstrated by Mion et al., 2001, using both Western blotting and cell surface immunolabelling under non permeabilized condition. We have now added this information in the Methods section (page 11, Tracking and Surface Diffusion Calculation" (Line 385)).

Comment:
The surface labeling in supplementary figure 6 look very dense and rather unspecific, could this be a reason?

Response:
We do not believe the signal to be particularly dense or unspecific for the following reasons. The antibodies used have been shown to be specific using various approaches (e.g. specificity was tested against competing peptides, immunolabeling on HEK cells expressing recombinant NMDAR etc.) (please see response to comment above). Furthermore, the values for the synaptic fraction of GluN1 are similar to those reported in the literature (Ladepeche et al., 2013, PNAS), and therefore seem within a normal range. In addition, the imaging parameters (e.g. threshold of the signal) were based on previous experience with this method and the respective antibodies, and were set to eliminate noise. Comment: For the statistics of the mobility data it would be nice to show distributions averaged between different experiments to report the variability between experiments. One way could be to include the experimental variation in the cumulative representation?
We are not certain about the specific way this could be addressed. Below we attached a plot representing the mean squared displacement (MSD, i.e Figure 3b,c. As shown on the histograms and cumulative fractions, extra-synaptic and synaptic GluN1-NMDAR diffusion coefficients were still found to be significantly enhanced in the Fmr1 KO condition compared to WT, which confirms our initial observation.

Comment:
The cutoff for immobile receptors should be mentioned.

Fmr1 KO
This information is already given in the Results section (page 4, paragraph "Exaggerated lateral diffusion of mGluR5 at synapses in Fmr1 KO neurons"(line 114)).

Comment:
For many experiments the number of experiments (cells) and their origin of how many different cultures is not given.

Response:
We have now included this information in each figure legend.

E. Conclusions: robustness, validity, reliability
Despite the mentioned problems above, the multitude of experiments, particular the use of interaction side specific peptides, seem to confirm the main conclusion that the interaction between mGluR5 and Homer1A is critical to the functional consequences reported here and by others.

Response:
We thank the reviewer for this evaluation of our study.

F. Suggested improvements: experiments, data for possible revision
As stated above, I would appreciate a revision of the data given on the dynamics of the involved receptors, particular in respect to previous work. In general the use of surface dynamics of key molecules as transmitter receptors can teach us a lot about molecular mechanisms behind and this work could be another nice example. However, large differences in the dynamics between experiments let wonder whether the method is robust enough to support the conclusions made in the paper.

Response:
As mentioned above, the diffusion coefficients measured in this study are indeed slightly different from previously reported values (see Sergé et al., 2002;Groc et al., 2004) which may be explained by several factors varying from one study to another, including i) the type of neuronal cultures used, ii) the developmental stages studied, iii) the animal species, iv) the acquisition frequencies, and v) the recording medium. Each of these factors may account for these differences between studies. Our data however match recent publications performed in a similar configuration (see Opazo et al., 2010;Dupuis et al., 2014), and all single-particle tracking experiments were conducted in the same recording configuration to allow unbiased comparison between the different experimental conditions. Most of all, when considering the apparent discrepancy in diffusion coefficients between our dataset and previous reports, one should keep in mind that absolute diffusion values should be handled with caution since they depend heavily on the methods used to perform single-particle tracking measurements, and thus vary from one study to another for a same subtype of receptor depending on the experimental configuration, although they fall in a similar range.

G. References: appropriate credit to previous work?
Previous work is appropriate cited

H. Clarity and context: lucidity of abstract/summary, appropriateness of abstract, introduction and conclusions
The paper is clear written and results discussed in the context.

Reviewer #2 (Remarks to the Author):
Summary statement This manuscript addresses further contributions of the mGluR5 receptor in hippocampal synaptic and behavioral phenotypes of Fragile X mutant mice, through a Homer 1a mechanism.

Comment:
At the end of the Results section, page 7 line 229, mention is made of a "second generation" Fmr1 mouse model. Differences in the mutation and breeding are briefly described later under Animals, page 10 lines 345-347. Apparently the mice used in the present experiments are different than the widely used Fmr1 line of mice. A full description of the new line of mutant mice must appear at the beginning of the manuscript and in the abstract. Rationale for the different type of mutation, how the mutation was generated, and the quality and quantity of genetic differences between the conventional Fmr1 mutation, must be carefully explained. Comparisons of the full set of phenotypic differences between the present mouse and the conventional Fmr1 line must be spelled out. The single reference 61 to a 2006 publication is insufficient.

Response:
We thank the reviewer for suggestions for improving the clarity of the manuscript. We have now added the following information: "This model was generated by deletion of both the promoter and exon 1 of the Fmr1 gene, as described previously (Mientjes et al., 2006). These mice are distinct from the original Fmr1 KO (Fmr1 tm1Cgr ) mouse line (Bakker et al., 1994), because they are deficient for both Fmr1 mRNA and FMRP protein.

Comment:
The novel object recognition task was conducted using non-standard methods. In fact, a spatial component has been added, by placing the second object in a different arm of an Lshaped maze, instead of presenting the first and second objects in adjacent locations within an open field.

Response:
As mentioned below by reviewer 2, there is significant body of evidence showing no, or very limited cognitive phenotypes in the mouse model of FXS. This issue has been discussed in a recent review (Gross, Hoffmann, Bassell and Berry-Kravis, 2015 Neurotherapeutics 2(3):584-608) which highlighted the importance of establishing robust phenotypic tests which are reproducible between labs. To this extent, we believe that the use of this non-classical configuration can be an advantage, because we have been able to reproduce a phenotype established previously in the Fmr1 tm1Cgr mutant in an FVB background in an independent lab (Busquets-Garcia et al., 2013, Nature Medicine 19: 603-607) This test has been already described for evaluating the THC-induced memory impairment in the object recognition task (Puighermanal et al., 2009Nat. Neurosci. 12, 1152-1158and Puighermanal et al., 2013Neuropsychopharmacology 38, 1334-1343 and in mouse models of Alzheimers disease (Ill-Raga et al., 2015 andAso E. 2015, J Alzheimers Dis. 43(3):977-91). The non-classical configuration used here has a number of advantages over object exploration in a classical open-field: the maze is non-anxiogenic (the dark-coloured, nonreflective walls of the maze and narrow corridor, limit anxiety, which may be a confounding phenotype in certain open field configurations). In addition, the dim light conditions present in the L-maze may encourage self-directed exploration (e.g. as described in Mun et al., 2015 Scientific rep 5:17697). Moreover, since x,y exploration is limited by the limited arena size, the mice are encouraged to spend more time in exploring new features each time the arena is presented (namely objects) and object exploration time is improved (Aso et al., 2015). In addition, this configuration reduces the habituation sessions required for the classical open field configuration. The inventing laboratory have obtained a utility model to permit the commercialization of this kind of maze for the study of object recognition (#U200802592 Title: Device for testing cognition in animals. Inventors: A. Busquets-Garcia, R. Maldonado,

A. Ozaita)
To limit any confounds related to a spatial component (as raised above) we balanced the presentation of the novel object by randomly assigning it to the top or bottom arm of the maze. By scoring the exploration of objects on day 1 of the task (the acquisition phase of the task in which two identical objects were presented) we also verified that there was no bias with respect to one arm or another.

Comment:
This difference is important because Fmr1 mice have been frequently reported to show hyperactivity. If the mutants move throughout the L-shaped arena at higher rates, they are likely to explore both arms more than their WT controls, and concomitantly are likely to spend less total time with the objects.

Response:
To address this question, we performed posthoc automated tracking using Ethovision software (Noldus) and analyzed the total distance moved, velocity and percentage of time active and inactive during exploration in the L-maze during habituation and on the training and test of object exploration. We found no differences between the genotypes for any of these parameters. These data are presented in Figures 13 -Supplementary Figure. We performed the same analysis for the AAV injected cohort and once again found no difference between the different groups. These data are presented in Figure 18 -Supplementary Figure. Based on this analysis and also the data related to object exploration on the first day (see below), we see no evidence supporting the potentially confounding effect of a hyperactivity phenotype.
Once again, we would also like to point out that the 'hyperactivity' phenotype is also somewhat inconsistently reported within the literature and may largely depend on light conditions, housing conditions, testing method, duration and genetic background. For example, Yan et al., 2004 performed analysis (open field in the dark phase and under red light illumination for 5 minutes per day over a 5 days period; Fmr1 tm1Cgr model on FVB/C57Bl/6J hybrid background) and reported no increased activity. On the other hand, Bhattacharya et al., (2012) reported differences in the exploration of peripheral sectors of an open field (under 800 lux illumination, exploration during 15 mins during the light phase and using the Fmr1 tm1Cgr model on the C57Bl/6J background). It is likely that the differences in locomotor activity in the latter case may be confounded by the strong light conditions, which may create an aversive or anxiogenic environment. Using actimetry, Pietropaolo et al., 2011 reported an increase in activity over 1 hr in the Fmr1 tm1Cgr model on a FVB background only (no genotype effect was observed in the C57Bl/6 background). Spencer et al., 2005 reported increased distance travelled over a 30 minute exploratory period, using the Fmr1 tm1Cgr model on a C57Bl/6J background, but precise lighting conditions were not reported, so it is difficult to evaluate the possible confounds of anxiety (which is also a phenotype of this model).

Comment:
Similarly, hyperactivity could artificially reduce freezing time in the contextual fear conditioning test.

Response:
We are aware of these possible confounds and for this reason analysed freezing during the first session (acquisition). We found no difference in freezing scores between wildtype and Fmr1 KO animals, suggesting that both groups learned from the experience. For the sake of brevity and because, as Reviewer 3 points out, the contextual fear memory phenotype is not novel, these data were not reported in the original manuscript (but are now reported here - Figure 15 -Supplementary Figure).
Nonetheless to respond more fully to this criticism (in the context of the previous comments), we also performed a more extensive analysis of activity during the first 2 minutes of exploration. To do this we initially tried to automatically track movement of the animal using Ethovision analysis software. However, due to the poor contrast between the animal and the conditioning chamber, in particular due to the rods of the floor, this approach was not successful. Therefore we manually scored exploration. Since the conditioning chamber is a richer environment than the open field arena due to the textured floor, non-uniform junction between walls and floor and also due to the olfactory and visual cues, exploratory behaviour is more complex and composed of walking, supported and non-supported rearing, nosepokes, head-turns without full body movement and periods of grooming. We manually scored a subset of these behaviours: walking and rearing. We found no difference in these activitybased measures between wild-type and Fmr1 -/y animals ( Figure 15 -Supplementary Figure).

Comment:
The potential confound of higher locomotor activity can be easily evaluated. (a) Number of seconds spent exploring each object should be shown, in addition to the discrimination index, for the novel object recognition phase. (b) Number of seconds spent exploring each of the two identical objects during the familiarization phase should be shown. (c) An independent open field test for exploratory locomotion must be conducted, and presented with standard activity parameters. (Figures 11,12, We analysed open field exploration during a 30 minute period, under 40 lux illumination (measured in the centre of the arena) and analysed total distance moved and velocity ( Figure  14 -Supplementary Figure). We found no genotype effect for these measures over the period analysed. To confirm these findings, we repeated the experiment using an independent group of age-matched animals (male, Fmr1 -/y and wildtype littermates of equivalent age) and this time recorded exploration over 60 mins. Once again, we found no significant difference in total distance moved and velocity, as well as periods of activity and inactivity (not shown).

Comment:
There is no explanation about how the behavioral data shown in Figure 10

Response:
We apologise for the confusion with respect to the source of the data presented in these two figures. Since the object recognition memory and contextual fear memory phenotypes had not been previously demonstrated in the second generation Fmr1 KO model, we first performed pilot experiments using naïve (non-injected) mice of equivalent age to those used for the shRNA experiments. The data presented in the old Figure 10 (now Figures 11 and 15 -Supplementary Figures) pertained to data from these pilot experiments. Having confirmed this phenotype, we repeated the experiments using a separate cohort of animals, stereotaxically injected at P21 and tested at 12 weeks of age. This data is now presented in Figure 10 (previously Figure 9). We have now amended the figure legends and in addition added the following to the main text to further clarify this point: "Since these phenotypes have not previously been investigated in the second generation Fmr1KO mouse model, we initially performed a pilot experiment to determine whether we could recapitulate these phenotypes." Lines 231-233, page 7

Comment:
Full statistical results must be added (e.g. F value, degrees of freedom, stringent posthoc analyses). Stating only the p value for ANOVA and t-tests is insufficient.

Response:
We have now included further information regarding the statistic in each figure legend.

Comment:
The authors ignore the very large body of literature showing normal cognitive scores in Fmr1 mice on many learning and memory tasks conducted by many independent labs, including the novel object recognition and fear conditioning assays which were conducted in the present studies. The authors are encouraged to read the behavioral literature more fully.

Response:
We thank the reviewer for their suggestions for improving our manuscript. We are aware of the rich body of literature related to the behavioural characterization of the Fmr1KO mouse. Unfortunately, due to the limited number of references permitted by the journal (70) and the diverse approaches presented in the manuscript, we had to reduce our discussion of the relevant literature to a minimum.
With respect to the issues raised about the normal cognitive scores in Fmr1KO mice, we are aware of these conflicting results, as well as the suggestion that genetic background may play a role in the manifestation of certain phenotypes (e.g. Pietropaolo et al., 2011;Spencer et al. 2011). As pointed out above, one of the current and future challenges for research in this domain is to identify phenotypes, or phenotyping approaches, that are reproducible between laboratories. To this extent, we find it to be a strength of our findings that we were able reproduce phenotypes using methods established by independent laboratories using the Fmr1 tm1Cgr mouse in an FVB background (Busquets-Garcia et al., 2013;Gomis-González et al., 2016;Oddi et al., 2015).
With respect to the contextual fear memory data published previously, we would like to point out that the methods used here differ from those of Paradee et al., (1999), Dobkin et al., (2000, van Dam et al., (2000), Peier et al., (2005), in the respect that no auditory cues were presented. In addition, the conditioning described in our manuscript occurred over a single session, as described previously (Frankland, Bontempi et al., 2004) and not over successive days as described by Eadie et al., 2012. In addition, although  With respect to novel object recognition, we would like to point out that this paradigm has recently been highlighted as one of the more promising behavioural tools for investigating cognitive defects in the Fmr1KO mouse (Gross et al., 2015) e.g Ventura et al., 2004;Seese et al, 2014;Bhattacharya et al., 2012. In particular, the presentation of only two objects and the testing of long-term memory (24h) present major refinements over previous work. As mentioned above, the reproducibility of the approach employed in our manuscript and its utility for examining pharmacological correction is also an advantage.
Finally, we would like to make the point that a full behavioural characterization of the second-generation Fmr1 KO model was not the major focus of this manuscript. Given the constraints of the journal (limit 70 references) we hope that the Reviewer will accept our apologies for not being able to discuss these points in greater detail, with reference to the appropriate literature, in the current manuscript.

Comment
Absence of the full cognitive datasets, and of the necessary open field control experiment, represent serious flaws in this manuscript.

Response:
We appreciate the suggestions from the Reviewer to improve the presentation and interpretation of our work. We have now provided analysis for all the questions presented above. We hope that the Reviewer will understand that a more complete behavioural characterization is beyond the scope of the current work and will be addressed in a later article.

Summary:
This manuscript details interesting experiments investigating the interaction of mGluR5 and NMDA receptors in a mutant mouse model of fragile X (the Fmr1 KO). Specifically, the authors explore the dynamics of receptor diffusion in cultured neurons (using Q-dots), finding that the mobility of mGluR5 and the NMDAR are increased in the Fmr1 KO neurons. No change is seen in the mobility of GluA2, suggesting that the effect is somewhat specific to the mGluR-NMDAR complex. Based on the previous results of the Huber lab, they then show that disrupting mGluR5-Homer binding mimics and occludes the increased receptor trafficking in the Fmr1 KO. Additionally, the co-localization of mGluR5, NMDAR and Homer1 is increased in the synapses of the Fmr1 KO neurons.
In addition to the in vitro work, the authors examine whether the NMDAR component of LTD is altered in young (P11-15) hippocampal slices from Fmr1 KO mice. Specifically, they measure NMDAR EPSCs during stimulation of group 1 mGluRs using the agonist DHPG. They find that the NMDAR component of DHPG-LTD is deficient in the Fmr1 KO, and this is mimicked by disruption of the mGluR-Homer interaction. To show that enhancing the mGluR-Homer interaction could rescue the deficit in NMDAR function in the Fmr1 KO, they injected shRNA to the dominant negative Homer 1a (or scrambled shRNA), and measured the NMDAR component of DHPG-LTD in the Fmr1 KO. They also performed novel object recognition and contextual fear conditioning on injected animals, and show a remarkable rescue of both behavioral deficits.
Overall, this study is an interesting addition to the literature regarding the regulation of mGluR-NMDAR interaction, and the role of this in the fragile X mouse model. The in vitro Q-dot experiments are particularly compelling. There are, however, a number of drawbacks that should be addressed, particularly with respect to the electrophysiology and behaviour experiments.

Response:
We thank the reviewer for this overall positive evaluation of our work and will address the specific comments in detail.

Comment:
1.) The in vitro experiments using Q-dots show a convincing increase in the mobility of mGluR5 and NMDARs in the Fmr1 KO neurons. However, these experiments use Mitotracker labelling to define "synaptic sites" without providing verification using a more conventional synaptic marker (i.e., synaptophysin, PSD-95). This is important because all of the work shown in Figures 5-7 is dependent upon the "synaptic" association.

Response:
The use of mitotracker as a synaptic marker has been validated previously Groc et al., 2004 by examining co-localisation with the synaptic marker synaptotagmin. This approach has been extensively used for localizing synaptic sites during single particle imaging by the teams of Laurent Groc (Groc et al., 2007) and Daniel Choquet (Opazo et al., 2010). In the context of fragile X syndrome, it is preferable to use this approach instead of transfection, because transfection with a synaptic marker such as homer or PSD-95 may precociously alter synapse development. Since we wanted to examine native receptors this was an important factor to avoid. In addition, a number of mRNAs encoding synaptic markers are targets of FMRP (Darnell et al., 2011) and their expression may thus be altered in Fmr1KO cultures further element of complexity to an already difficult mechanism to be disentangled. Finally mitotracker labeling has the advantage of being rapid (1 min) and this is important in the context of live imaging where neuronal health is crucial.

Comment:
These data would be much more robust if there was validation using another method, and if the total co-localization of mGluR5 and GluN1 was shown (not just the "synaptic" fraction).

Response:
We used QD and triple immunocytochemistry for demonstrating an increased permanence of mGluR5 and NR1 at synapses and an increased co-localization of mGluR5 and NR1, respectively. The analysis of data from both experiments leads to the same conclusion. The small yet statistically significant increased level of mGluR5/NMDAR1 co-clustering at synaptic sites, makes very unlikely the detection of this with other methods such as immunoprecipitation. The increased functional interaction between, GluR5 and NR1 could be studied by BRET. This method is not used in any of the laboratories involved in the present study, and would require transfection, perturbing the system and making even more difficult to compare WT and KO. In the present study we decided to study naïve receptors only.

Response:
The shRNA and control shRNA constructs were validated previously in the CNS (in vitro and in vivo) Tappe et al., 2006. Nature Medicine 12, 677 -681. In addition to this validation, we have now provided data showing an estimation of knockdown efficiency in hippocampal neurons. In cultured hippocampal neurons, homer1a mRNA levels are reduced by 65-68% (depending on the choice of normalizing gene) with respect to expression levels present in AAV-scr infected (control) neurons (Figure 21 -Supplementary Figure b). Following stereotaxic delivery, an estimated 97% of neurons within the stratum pyramidale of CA1 were transduced with our AAV vectors expressing shRNA and control constructs (estimated by co-expression of the reporter gene, eGFP) (Supplementary Figure 21c).

Comment:
5.) The behaviour experiments seem to be the weaker parts of this study. The descriptions of the work are not very thorough, and the data appear hastily put together. For example, the methods state that behaviour experiments were only performed in the Fmr1 KO, yet there is data shown for WT. Were these experiments performed blind to genotype? Were the groups age-matched and counterbalanced for treatment? These details are essential for assessing the results, especially since the effects shown are so uncharacteristically large.

Response:
We apologise for these omissions in the presentation of this data. We have now included a fuller description of the methods used as well as a substantial number of controls as requested by reviewer 2. All behavioral experiments were performed using age-matched cohorts of WT and Fmr1-/y littermates. Treatments were balanced within litters and with respect to genotype and semi-randomized with respect to the assignment of treatments to individual animals. Contextual fear conditioning and object memory experiments were performed and analyzed blind to genotype and treatment. We have now included changes to the methods section to reflect this information. Please see: "Methods", "Behavioral Testing" section,"Novel object recognition task" and "Contextual fear conditioning task" subsections, page 17.
In addition, we would like to note that the mice were injected at post natal age 21 days and the behavioral training started 9 weeks after virus infusion. To this extent it is important to note that Homer1a levels are significantly upregulated in the third week of postnatal life and peak between 3-5 weeks of age. Moreover, it is important to note that homer1a is upregulated in response to activity and that circuit level activity is perturbed in fragile X syndrome, likely leading to an overall exacerbation of homer1a signaling. Since the refinement of neuronal circuits is critically regulated by circuit level activity, we believe that early and prolonged intervention during a critical period could explain why these effects are so pronounced.

Comment:
Moreover, many of these findings have been previously published. The deficits in contextual fear conditioning have been described previously (Eadie et al. 2012, Auerbach et al. 2011).
While it is true that reduction of Homer1a in the Fmr1 KO has not been found to rescue these specific phenotypes, it has been thoroughly studied with respect to other phenotypes (Huber). Additionally, the deficit in contextual fear conditioning has been attributed to a hypofunction of NMDARs in the dentate gyrus (Eadie et al. 2012). How does this fit with the authors' other findings, which are focused on CA1? New finding?

Response:
We did not intend to claim that the behavioral phenotypes presented here are novel. The description of these phenotypes in naïve (non-injected) animals was included to demonstrate that we could recapitulate these phenotypes in the second-generation mouse model (as we have done previously in Zhang et al., 2014), in order to validate the phenotype prior to demonstrating its correction. Additionally, although the very elegant work of Kimberly Huber's team has indeed demonstrated rescue of anxiety and seizure susceptibility, we find it important to note that the primary feature of Fragile X syndrome is intellectual disability. Instead, as noted in a recent review (Gross, Hoffmann, Bassell and Berry-Kravis, 2015, Neurotherapeutics)

the majority of behavioral characterizations of the Fmr1 KO mouse have focused on autism-associated features, including anxiety. However, as FXS is an intellectual disability disorder, the development of therapeutic strategies for the improvement of cognitive defects in FXS is also a priority. In this sense, we feel that our findings showing the correction of two hippocampal-dependent cognitive phenotypes is extremely pertinent to for the development of novel therapeutic strategies.
However, most importantly, our work describes two extremely novel findings: the higher surface mobility of mGluR5 and the altered plasticity of NMDA receptors, as a consequence of mGluR5/Homer disruption. The functional interaction between mGlu5 and NMDA and their reciprocal modulation is well established, but never investigated in the context of FXS. We also provide a strong correlation between these findings and the correction of the aforementioned behavioural phenotypes.
Whilst it is indeed true that hypofunction of NMDAR in the dentate gyrus has been associated with a defect in contextual fear conditioning, these findings are not mutually exclusive with our own work. The NMDA/AMPA ratio has not been studied before in the CA1 region using our protocols. Indeed, hypo-functionality of NMDA receptors have been detected in other regions of FXS such as anterior pyriform cortex (Gocel andLarson, 2012), prefrontal cortex (Martin et al., 2016) and nucleus accumbens of striatum (Neuhofer et al., 2015). Indeed, taken together, they lend support to the idea that dysfunction of NMDAR and mGluR5 can be associated in FXS and both receptors and their interaction might be considered as a new avenue for FXS therapy.

Reviewer #1 (Remarks to the Author):
The authors tried to answer my methodological questions mostly in bringing arguments that experiments have been done in the past in slightly different systems and experimental configurations and recent publications of the group show comparable receptor mobility in neurons. I have to say I disagree with both arguments. First, for the reference given by the authors the mobility of NMDAR in Dupuis et al. 2014 is reported to be under control conditions around 0.03-0.04 µm²/s (Fig.1), whereas in the current manuscript even in synapses NMDAR under control conditions are much more mobile (D > 0.1 µm²/s). Second, the argument that within mixed cultures, receptors are even faster than in glia-free Banker type cultures sound counterintuitive. There I would expect to see the opposite, since particular synapses are probably enwrapped by glia endfeets and might rather impose constrain of the QD-labeled receptors that like to enter or leave the synaptic membrane. Third, the authors claim that diffusion might be influenced by the age of the culture. Surprisingly, the culture of previous studies were between 3-11 DIV, whereas here the cultured cells where 12-15 DIV. If at all, I would expect to see lower diffusion coefficients since the surface dynamics of proteins decrease with the maturation of neuronal networks (Borgdorff, Choquet 2003, Groc et al. 2006. Forth, the authors seem to be aware that absolute values of diffusion coefficients are highly vulnerable but do claim that the effects they report are specific for the effects they describe. I would take this point never as an argument to defend my data, rather I would expect a more careful analysis and design of the experiments to circumvent such high variability.
Here i would like to see an improvement on the data rather than bringing arguments that do not hold.

Reviewer #2 (Remarks to the Author):
The behavioral data presentation is now more complete. The authors have successfully addressed the potential hyperactivity confound with additional experiments. Relevant new data appear in Supplementary Figures 13-15.
Statistical presentation remains insufficient. What is needed are actual ANOVA F values, degrees of freedom, and p values for each factor, and for numerical values obtained from each posthoc comparison. Instead, what appears in the figure legends are means + standard errors, which are unnecessary since already shown in the data, and the names of the statistical tests used, which should have already been summarized in the Statistics paragraph of the Methods section, lines 648-657.
In the Response to Reviewer 2, the authors clearly explain the Fmr1 mutant model which they used in the present study. However, this information still does not appear early within the manuscript. It is essential that the text states and describes this line of mice, to differentiate it from the more standard Fmr1 knockout that readers will immediately assume were used. Explicit naming and/or description of this specific Fmr1 mutation must appear briefly in the Abstract and Introduction, with a reminder to the readers in the Discussion section, rather than only in the Methods section which readers may not read carefully.
Several of the references to the original generation and phenotyping of this Fmr1 mutant model, which appear in the Response list, should appear in the manuscript reference list, and be cited within sentences describing this Fmr1 line. Either the authors could request that the journal editor allow them to exceed the 70 reference limit, or the necessary additional references can be placed into Supplementary Materials.
In the Response to Reviewer 2, the authors explain that their modifications of the novel object recognition task were deliberate, to enhance detection of a cognitive phenotype, and based on previous publications using these methods. However, again, the reader does not receive sufficient alerts that this is not the standard novel object recognition assay. Clear statements must be added to the Abstract, Introduction, Methods, Figure legends, Results and Discussion sections, emphasizing that these were modifications of the standard novel object recognition task, in which complexity was introduced by adding a spatial compartmentalization feature. Further, the authors are strongly encouraged to use a different name for their novelty test, which better describes their method, and which clearly distinguishes their method from standard novel object recognition.
Supplementary Figure 22 is stated as proving that no object bias existed for exploring the lego versus the bottle cap. However, the p values obtained in this simple t-test is p=.06, which is very close to statistical significance. Ns are only 4 per object. The authors are encouraged to expand this essential control experiment with larger Ns. If an innate preference for one object emerges, the authors are encouraged to choose different objects.
Differences between how the authors conducted fear conditioning, and how fear conditioning was conducted by the many other labs that did not detect fear conditioning deficits in Fmr1 knockout mice, must be clearly stated within the Discussion section.
Because the literature on lack of cognitive deficits in Fmr1 mice is a major source of controversy in the field, greatly increased emphasis on how the present behavioral assays differ from most of the standard literature on Fmr1 mice is essential.
At present there is almost no description of the behavioral results in the Discussion section. The single sentence in lines 329-330 is entirely insufficient. If the authors want to state conclusions about cognitive deficits and rescues, relevant to the primary diagnostic symptom of Fragile X syndrome, considerably more text must be devoted to interpreting their behavioral findings appropriately.
If the authors are not as interested in the behavioral component, as indicated by a statement in the Response that behavioral characterization was not a major focus of the manuscript, then the behavioral figures should be removed entirely.

Reviewer #3 (Remarks to the Author):
The authors have carefully considered the reviewers' comments, and the manuscript has been strengthened significantly. This is particularly true of the new behavior data and more thorough explanation. They should be commended for a well revised manuscript that will make an interesting addition to the literature.

Reviewer #4 (Remarks to the Author):
Their reported values for the diffusion coefficient of GluN1 are indeed a little higher than reported in comparable studies, but the authors provide plausible arguments for the difference in the measured diffusion coefficient. The precision and reproducibility of these experiments depends greatly on the experimental conditions (type, age, species of culture); detection (antibody-labeled endogenous receptors, or exogenously tagged); methods used to track receptor (QDs or beads); acquisition parameters (frame rate etc.) and can be highly variable. But in fact, as they show in the appendix in the second rebuttal, the numbers are not terribly different. The skewed, non-Gaussian distributions in diffusion coefficients make that the median and mean values report very different aspects of the population, and I believe this might have caused some of the confusion. What is most important, is that the experimental conditions within experiments are the same. Based on the presented information and data, the authors convincingly demonstrate that the lateral diffusion of mGluR5 and NR1 is altered in Fmr1 KO cultures, and that in particular mGluR5 diffusion is increased in "synaptic areas".
I do want to comment on the use of MitoTracker as a marker for synapses (as was noted by reviewer 2). Although MitoTracker has "extensively been used", at best it is a very rough estimate of synaptic sites; 60-70% overlap with synaptotagmin according to Groc et al., 2004, meaning that in 30-40% of the cases, synaptic tracks are not correctly assigned. This most likely reflects the fact that the majority of spines does not contain mitochondria (see e.g. Li et al., Cell 2004). Moreover, the morphology of mitochondria is highly variable and does not relate at all to synapse shape. And, important for this study, mobility, location, shape and number of mitochondria are all influenced by neuronal activity. So, even if MitoTracker was the best possible approach for estimating synaptic location in these experiments, the authors should at least demonstrate that the ability of MitoTracker to label synaptic sites is similar between wild-type and Fmr1 KO cultures, in which the activity of synapses is severely altered.
We thank the Reviewers for their consideration of our manuscript and for their suggestions for improvement. Below, we present a point-by-point response to their comments. Modifications to the manuscript are indicated by highlighting in yellow in the main text, and where appropriate, below.

Reviewer #1 (Remarks to the Author):
Comment: The authors tried to answer my methodological questions mostly in bringing arguments that experiments have been done in the past in slightly different systems and experimental configurations and recent publications of the group show comparable receptor mobility in neurons. I have to say I disagree with both arguments. First, for the reference given by the authors the mobility of NMDAR in Dupuis et al. 2014 is reported to be under control conditions around 0.03-0.04 µm²/s (Fig.1), whereas in the current manuscript even in synapses NMDAR under control conditions are much more mobile (D > 0.1 µm²/s).

Response:
We apologize for not replying more fully to this question in the first revision. To address this concern, we now provide a supplementary  Bard et al., 2010, for which higher values (0.11 and 0.32 µm2/s) were obtained). This comparison should convincingly demonstrate that our value is very similar to those obtained by these other studies despite any differences in the type-(Banker/mixed) or age   Finally, following his or her initial evaluation of our manuscript, the Reviewer asked us to discuss the discrepancy between our results and those reported previously in the literature. Given the aforementioned discussion, we hope that we have now established that this difference is negligible. Nonetheless, we have added the following sentence to the Discussion (page 8, in which we also cite Renner et al., which was not cited previously : "Noteworthy, although these mobility values may be influenced by a number of experimental variables, our measures for WT neurons are consistent with those reported under similar conditions (e.g. 24)" In summary, we hope to have provided convincing arguments, well supported by the existing literature, that our mobility values are in the observed range of reported values, therefore corroborating the validity of our conclusions.

Comment:
Second, the argument that within mixed cultures, receptors are even faster than in glia-free Banker type cultures sound counterintuitive. There I would expect to see the opposite, since particular synapses are probably enwrapped by glia endfeets and might rather impose constrain of the QD-labeled receptors that like to enter or leave the synaptic membrane.

Response:
We agree that these assumptions would appear to be intuitive, however, as an analysis of the literature (please see Supplementary Table)  Comment: Third, the authors claim that diffusion might be influenced by the age of the culture. Surprisingly, the culture of previous studies were between 3-11 DIV, whereas here the cultured cells where 12-15 DIV. If at all, I would expect to see lower diffusion coefficients since the surface dynamics of proteins decrease with the maturation of neuronal networks (Borgdorff, Choquet 2003, Groc et al. 2006.

Response:
We believe that our responses may have been misconstrued and we apologize to not being clearer in our explanation. Here, we simply pointed out that there is a range of parameters that might affect the diffusion values obtained. In particular, we cited this as one explanation for why our values were not the same as Sergé et al., J Neuroscience, 2002, in which the cultures used were very different in age (3-8 DIV) compared to our study.
However, in addition to the age, it is important to note that the experimental conditions and analysis criteria employed in these two studies are completely different. First, in Sergé et al., 2002 a transfected myc-tagged or HA-tagged receptor construct was used and the tagged receptor was expressed under the control of a hybrid (non-native) promoter. In our case native (endogenously expressed) receptors were assayed. It is possible that the tagged protein does not have the same dynamics within the membrane. It is also possible that over-expression of the receptor from a plasmid-based construct, or indeed the transfection itself alters the physiology of the neurons, their state of maturity or leads to a greater accumulation of receptor protein in the membrane, so it is not entirely equivalent to the normal developmental trajectory of a neuron. However, as mentioned above, the most critical difference between this early work and our own is that 0.5 µm beads conjugated to the myc antibody were used instead of QD-based detection these beads are ~20 times larger than QDs! In addition, they did not distinguish between synaptic and extra-synaptic mobility (see also our point above). All of these factors may lead to the absolute diffusion values being different between their study and our study and for this reason we felt that a direct comparison of these values is not valid. As mentioned above, there is a more recent study (Renner et al., 2010) reporting diffusion values for mGluR5 that are comparable to ours and for which the experimental conditions are similar.
Comment: Forth, the authors seem to be aware that absolute values of diffusion coefficients are highly vulnerable but do claim that the effects they report are specific for the effects they describe. I would take this point never as an argument to defend my data, rather I would expect a more careful analysis and design of the experiments to circumvent such high variability.

Response:
We believe that this is a misunderstanding. We think that there might be a confusion between the variability of values reported in the literature, which we clearly acknowledge despite very similar values obtained in the various studies (see above), and our inter-experiment variability, which was carefully tested and controlled for. With respect to vulnerability, we would like to insist on the fact that this is not the case. Indeed, we were meticulous in ensuring that our experimental conditions were the same throughout (including maintaining a strict breeding strategy so that cultures were produced from WT and KO embryos at the same time and that the dissection was always performed at the same time of day by the same person). The antibody incubation and imaging parameters were also strictly controlled. Therefore the variability between experiments in our condition was minimal.

Comment:
The behavioral data presentation is now more complete. The authors have successfully addressed the potential hyperactivity confound with additional experiments. Relevant new data appear in Supplementary Figures 13-15.

Response:
We thank the reviewer for this statement.

Comment:
Statistical presentation remains insufficient. What is needed are actual ANOVA F values, degrees of freedom, and p values for each factor, and for numerical values obtained from each posthoc comparison. Instead, what appears in the figure legends are means + standard errors, which are unnecessary since already shown in the data, and the names of the statistical tests used, which should have already been summarized in the Statistics paragraph of the Methods section, lines 648-657.

Response:
We decided to give p-values and the types of statistical tests for each data in the figure legends (rather than the statistics section), since they vary depending on the type of experiment performed. We now provide these additional statistical values (F-values etc.)

in the corresponding figure legends
Comments: In the Response to Reviewer 2, the authors clearly explain the Fmr1 mutant model which they used in the present study. However, this information still does not appear early within the manuscript. It is essential that the text states and describes this line of mice, to differentiate it from the more standard Fmr1 knockout that readers will immediately assume were used. Explicit naming and/or description of this specific Fmr1 mutation must appear briefly in the Abstract and Introduction, with a reminder to the readers in the Discussion section, rather than only in the Methods section which readers may not read carefully.

Response:
We have now integrated this information throughout the manuscript as suggested by the Reviewer. Specifically, the following mentions are made:

Introduction (page 3, lines 92-95):
"The majority of these experiments were performed using the second-generation Fmr1 KO mouse line, which lacks both Fmr1 mRNA and FMRP 33 Certain electrophysiological experiments were performed both in the second-and first-generation 34 mutants, demonstrating good comparability between these models."

Results (page 4, lines 110-112):
"Here we used a single nanoparticle (quantum dot; QD) imaging approach to track surface mGluR5 in live hippocampal neurons derived from second-generation Fmr1 KO and WT mouse embryos (12-15 days in vitro)".
Results (page 7, lines 238-242): "Since these phenotypes have not previously been investigated in the second generation Fmr1KO mouse model, we initially performed a pilot experiment with a separate batch of experimentally naïve animals to determine whether we could recapitulate these phenotypes. Consistent with the aforementioned studies employing the first-generation Fmr1 KO mouse line, the second-generation model, used here, exhibited similar decreases in the discrimination index (DI) in the NOR task… and in the percentage of freezing compared with WT mice following retrieval of CFC memory …"

Discussion (page 10, lines 347-349):
Our study using the second generation model replicates findings reported for the first generation Fmr1 tm1Cgr model, suggesting that these tests are robust and reproducible between independent laboratories.
Methods (page 11, lines 399-404) Second generation Fmr1 KO mice 33 were mostly used in our study. This model was generated by deletion of both the promoter and exon 1 of the Fmr1 gene, as described previously 33 . These mice are distinct from the original Fmr1 KO (Fmr1 tm1Cgr ) mouse line 34 , because they are deficient for both Fmr1 mRNA and FMRP protein. This mouse model has largely been used for physiological, molecular, and anatomical studies (e.g. 60, 76-81), however limited behavioral characterization has also been performed (e.g. 60, 82-85).

Methods (page 16, ljnes 615-616)
"For the electrophysiological recordings of NMDAR currents from Fmr1 KO mice (second generation) …" Comments: Several of the references to the original generation and phenotyping of this Fmr1 mutant model, which appear in the Response list, should appear in the manuscript reference list, and be cited within sentences describing this Fmr1 line. Either the authors could request that the journal editor allow them to exceed the 70 reference limit, or the necessary additional references can be placed into Supplementary Materials.

Response:
Again, as stated above, we now added several references of studies that used this mouse model, and also additional information regarding this mouse model (see results and methods e.g. Methods, page 11 lines 403-404).

Comments:
In the Response to Reviewer 2, the authors explain that their modifications of the novel object recognition task were deliberate, to enhance detection of a cognitive phenotype, and based on previous publications using these methods. However, again, the reader does not receive sufficient alerts that this is not the standard novel object recognition assay. Clear statements must be added to the Abstract, Introduction, Methods, Figure legends, Results and Discussion sections, emphasizing that these were modifications of the standard novel object recognition task, in which complexity was introduced by adding a spatial compartmentalization feature. Further, the authors are strongly encouraged to use a different name for their novelty test, which better describes their method, and which clearly distinguishes their method from standard novel object recognition.