AMPA-receptor specific biogenesis complexes control synaptic transmission and intellectual ability

AMPA-type glutamate receptors (AMPARs), key elements in excitatory neurotransmission in the brain, are macromolecular complexes whose properties and cellular functions are determined by the co-assembled constituents of their proteome. Here we identify AMPAR complexes that transiently form in the endoplasmic reticulum (ER) and lack the core-subunits typical for AMPARs in the plasma membrane. Central components of these ER AMPARs are the proteome constituents FRRS1l (C9orf4) and CPT1c that specifically and cooperatively bind to the pore-forming GluA1-4 proteins of AMPARs. Bi-allelic mutations in the human FRRS1L gene are shown to cause severe intellectual disability with cognitive impairment, speech delay and epileptic activity. Virus-directed deletion or overexpression of FRRS1l strongly impact synaptic transmission in adult rat brain by decreasing or increasing the number of AMPARs in synapses and extra-synaptic sites. Our results provide insight into the early biogenesis of AMPARs and demonstrate its pronounced impact on synaptic transmission and brain function.

an enzymatic function or is it a scaffold function? And what do es the Ferric C helate Reductase like protein actually do?. The 'functional' experiments don't really tell us much other than FRRS1l and C PT1c are required to produce the AMPA receptor at the membrane. So overall, I find that I am left with a rather superficial insight into the function of FRRS1l and C PT1c and their interaction with AMPA receptors.
Issues to be addressed: 1. Statistics. There are no statistics for most data and figures, except for physiology. This needs to be corrected throughout and significant differences reported.
2. Data deposition. The proteomic data should be provided in supplementary tables and deposited in public domains. See next comment.
3. Figure 1a. Shows a heat map of protein abundance from multiple APs. It is not clear i n the figure legend or methods how the numbers were generated to produce this heat map. The units (mol. Abundance [log a.u.]) require clear explanation.
4. Figure 1a,b. Why were not the same assays performed with antibodies to C PT1c? This appears incomplete and the data would be valuable for their hypothesis about the mutual localization and interactions.
5. Figure 2. Refers to 4 different anti-FRRS1l proteins and two different anti-C PT1c antibodies and presents mean +-SEM of 5 and three Aps. This is confusing: what do the error mean? How did the different antibodies perform? The key with red/blue is also unclear.
6. The abstract says "FRRS1l and C PT1c …specifically and cooperatively bind to the GluA1 -4 proteins of AMPA receptor". C an they clarify if this protein is really specific to AMPA receptors or is it that the AMPA receptor is the major binding partner in the brain. Expression databases indicate that FRRS1l and C PT1c are expressed in many non-neuronal tissues where there are not thought to be functional AMPA receptors. C ould it be that the FRRS1l and C PT1c are more general rather than AMPAR -specific? Further to this point see the next two comments. 7. They human phenotypes. While it is possible that the phenotypes are due to altered expressio n of AMPA receptors, I would advise caution in how this is explained (until mouse genetic experiments show some related phenotypes with electrophysiology and rescue). The FRRS1l protein can have other functions that could lead to the intellectual disability phenotype and the authors have not separated these possibilities.
8. Figure 3a. Please label the genotype of all family members in each pedigree. Are all affected individuals homozygous for the FRRS1l mutation? Are any unaffected individuals homozygous for any of the FRRS1l mutations? 9. Figure 4a. Sac1 immunoblot of anti-GluA AP samples (second panel from left). This immunoblot looks under-exposed compared to the other anti-GluA AP panels. The Sac1 band in the GluA1/2-C PT1/FRRS1l lane is faint. There are even fainter bands in the GluA1/2-Sac1 lane and the GluA1/2-Sac1-FRRS1l lane.
10. Page 5. Sentence beginning "Interestingly, these AMPAR subunits displayed…" is confusing. Please re-write. 11. Page 7. First two lines. "And suggest the importance of FRRS1l-containing AMPARs for normal brain development and function". Brechet shows that FRRS1l is found in at least two distinct complexes: a complex that lacks AMPARs and complex that contains AMPARs. It is not known if the mutations in FRRS1l affect the AMPAR containing or the AMPAR-lacking complexes? I recommend removing this statement from the results section.
12. Page 7. Paragraph beginning "Further analysis by confocal…". The authors suggest the confocal data in figure 7b shows C PT1c expression causes a redistribution of FRRS1l from the surface to an intracellular membrane compartment. None of the confocal images show what area of the image is intra-versus extra-cellular. I suggest that if the authors wish to keep that point in the results sec tion text, they probably need to show some surface stain (perhaps using surface biotinylation) or an intracellular marker (perhaps calnexin could be used as Brechet et al us this in Supp. Fig. 3.) to demonstrate that FRRS1l has undergone a redistribution when C PT1c is co-expressed. It would also be helpful to show some evidence of quantitative image analysis to convince the reader of this point.
13. Figure 4c, left. Immunoblot. This is annotated on the left side with two black and one red line. Please describe these annotations in the figure legend.
14. Figure 4c, left. Immunnoblot. Please describe what is the "FRRS1lC -tag" in the third lane from the left in the figure legend.
15. The results section on the data in Figures 2 and 4 were quite difficul t to follow, especially in contrast to other parts of the text, which were very clear and concise.
16. Figure 5a,b. Why were not the same assays performed for C PT1c? This appears incomplete and the data would be valuable for their hypothesis about the mutual localization and interactions.
Other edits: 1. They should spell out the full name of FRRS1l and C PT1c and describe them in some more detail in the introduction, including information on their expression in other tissues etc.
2. They repeatedly refer to "profound" effects and findings. For this reviewer, none of these findings justify the use of this adjective. Many of their "profound" findings would be better described as expected, or in agreement with, or significant. I also note that some of these "profound" findings are not backed up with statistics and thus it remains for the authors to demonstrate that they are significant.
Reviewer #2 (Remarks to the Author): In this study Fakler and colleagues study AMPAR complexes associated with FRRS1l and C PT1c, constituents that they identified in an earlier proteomics screen (Schwenk at al., Neuro n 2012). They show that AMPARs associated with these components lack the 'classical' core auxiliary subunits and provide some evidence that they constitute early intermediates localizing to the ER. This study is well conducted and will clearly be of interest to the iGluR community. Although the study examines FRSS1l disease mutations it falls short of discussing/investigating mechanistic details of FRRSS1 action and ignores current literature on AMPAR biogenesis, which should be considered in this paper.
The experiments provide an interesting initial characterisation of these AMPAR associated proteins, but the investigation into AMPAR/FRRS1l/C PT1c function currently lacks any development of a mechanism beyond a "trafficking role". It would be of considerable interest to see at what stage in AMPAR biogenesis these proteins work, do they play a role in initial protein folding, and/or in the assembly of dimers or tetramers, or even during core subunit heteromerisation? Would QconC AT allow them to assess whether FRRS1l/C PT1c indeed associates with AMPAR tetramers or could they selectively complex with assembly intermediates (monomers/dimers)?
AMPAR processing in the ER has been investigated and this information should be integrated into the current study; e.g. individual core subunits traffic differently, e.g GluA3 traffics poorly compared to the other subunits (C oleman et al. 2010); RNA editing and alternative splicing of subunits determines ER exit rates (e.g. C oleman et al. JBC 2010, Penn et al. EMBOJ 2008) -do the current findings bear any relevance to this and might it play a role in selective action of FRRS1/C PT1c. For example, it appears from Figure 2a that GluA2 is a major component of FRRS1l/C PT1c ER complexes consistent with GluA2 retention in the ER due to Q/R editing (Greger et al. Neuron 2002). Is this the case? Also, have they probed blots with anti-GluA2 following anti-GluA1 AP to assess the proportion of AMPAR heteromers? One may expect that ER complexes are incompletely assembled AMPARs as 1Q/2R heteromers are export-competent.
Related to the text on p7 (middle), why is 'data not shown' for the ability C PT1c to retain GluA1 in the ER via interaction with FRRS1l? What happens in the presence of GluA2? One might expect FRRS1l/C PT1c to drive assembly of GluA1 with GluA2. Likewise, the neuronal results ( Figure 2a) show that all FRRS1l/C PT1c complexes are associated with GluA2 while being evenly distributed between GluA1 and GluA3, allowing them to recruit either subunit to unassembled GluA2.
The immunocytochemistry in Figs 4b and 5a does not include AMPA staining, GluA2 should be a nice marker for ER-localized AMPARs and should have been included Do sh-FRRS1l and sh-C PT1c affect rectification of evoked EPSC s? Loss of FRRS1l/C PT1c complexes may enable forward trafficking of GluA2-lacking receptors to the synapse before they have the chance to reassemble with GluA2. The loss of 60% is reminiscent of conditional GluA2 knockout (Lu et al., Neuron 2009). Figure 1a: is the first red box intentionally selecting C NIH-3 but not C NIH-2? It appears that these two proteins are found to similar levels in these samples. It also appears that there is a significant amount in the ER complexes unlike the TARPs and other core subunits.

Minors
P4: 'FRRS1l effectively associates with C PT1c and Sac1 independent of GluA1-4' is not completely true as there is more of these proteins in the AMPAR-containing samples (comparing brown bars against red bars in figure 1b). P5: 'effective (close to stoichiometric) co-assembly' of FRRS1l and C PT1c is somewhat an overstatement as although every FRRS1l interacts with one C PT1c on average (top panel of Figure  2a), there is an additional population of C PT1c that does not interact with FRRS1l (bottom panel). Accordingly, C PT1c shows an FRRS1l-like pattern of preferences for GluA subunits but with reduced amounts.
P5 again: One would also argue that SAC 1 and PORC N are barely associated with the FRRS1l -or C PT1c-containing AMPARs based on Figure 2a. Figure 4a also shows limited pull-down of SAC 1 by AMPARs although it is pulled down by FRRS1l. The mention of these is probably fine however as they are not followed up later on. P10: The sentence starting 'the decrease and increase in EPSC amplitude observed with sh -FRRS1l and sh-C PT1c' is confusing because FRRS1l overexpression, which causes the increase, is not mentioned. P11: 'Knockdown or exogenous (over-)expression of FRRS1l or C PT1c' suggests both manipulations were applied to both genes/proteins but no data is shown for C PT 1c overexpression. Please change to make this clear.
P12: C PT1c knockout mice are mentioned in relation to physiological relevance but the phenotype is not stated. Please briefly summarise. More generally, I'd be nice if the authors could briefly introdu ce what's currently known about FRRS1 and C PT1c function. P17 and Figure 8b: there is no evidence that AMPARs assembling with C PT1c and FRRS1l are tetramers.
Supplementary table 2 is missing.

Reviewer #3 (Remarks to the Author):
This is an interesting and generally well-done/convincing study showing that FRRS1i and C PT1c assemble with the pore-forming subunits of AMPA receptors (GluA1-4) in a protein complex distinct from that of the auxiliary subunits including TARPs, cornichons and GSG1L. The au thors also show that perturbations of FRRS1L impact synaptic signaling by AMPA receptors. As mutations in FRRS1L cause severe defects in intellectual ability and epilepsy, this work has relevance for both basic and clinical sciences.
The authors first perform proteomic analyses to identify distinct AMPAR assemblies. This work is rigorous, interesting and convincing. The authors next characterize families in which mutations in FRRS1L cause severe intellectual ability. This work is also well done and contri butes to an already existing genetic literature concerning mutations in FRRS1L and profound neurodevelopmental disabilities. The authors go on to show that AMPA receptors associated with FRRS1i and C PT1c localize to the ER and that assembly of this complex is disrupted by disease mutations. These are important findings that provide mechanistic insight regarding the pathogenesis of FRRS1i -linked human disease. Finally the authors study the functional effects on AMPA receptors of manipulating FRRS1i and C PT1c protein levels. This work shows that knockdown of FRRS1i or C PT1c reduces the amplitude but not the kinetics of AMPR-mediated currents. Surprisingly, knockdown of C NIH-2, TARP-2 or TARP-8 had no effect on AMPAR current amplitude. These latter results conf lict with previous publications. Also, I could not find in either the main text or figure 7 legend any mention of the data in figure 7b concerning sh-C NIH-2, sh-TARP-2 or sh-TARP-8. This issue must be addressed in detail.
I have additional specific comments below. 1. The immunofluorescence and EM data in figures 4b and 5 must be objectively quantitated. 2. The AMPA/NMDA receptor ratios referred to on page 10 are crucial controls and the traces should be shown in the main figures. 3. The methodology for collection the interpretation of the correlations in figure 8a were not clear to me. For example, do the authors find it interesting that GluA3 and TARP -8 show correlation with FRRS1i but GluA2 and TARP-2 do not?
Reviewer #4 (Remarks to the Author): This paper describes identification of AMPA -type glutamate receptor (AMPAR) complexes that transiently form in the endoplasmic reticulum (ER) through proteomic approaches. The authors describe FRRS1l and C PT1c as critical components o f these complexes and show that they cooperatively bind to the GluA1-4 proteins. They also describe three families with intellectual disability and epilepsy and show that biallelic mutations of the FRRS1L gene are responsible for their condition. Finally, the authors show that virus-mediated suppression or overexpression of FRRS1l leads to alteration of synaptic transmission by changing the number of AMPARs in synapses and extra -synaptic sites.
This is a generally well-designed and well-conducted study. The manuscript is concisely written and the data are clearly presented with well-made figures. The strengths of this paper include meticulously conducted proteomic and electrophysiological studies, which beautifully highlight the roles of FRRS1l 2] The effect of FRRS1L knockdown on AMPAR-mediated current has previously been shown by Madeo et al., though the current manuscript includes much more detailed electrophysiological studies. What is not known is how FRRS1l regulates the AMPARs. The authors state that this is to be elucidated in the future (page 7), but additional data on molecular effects of FRRS1l (or mutation thereof) would be desirable. For example, what happens to G luA tetramers in ER if there is no FRRS1l (or C PT1c)? Do they remain in ER or leave ER but not get trafficked correctly to synapses? 3] As a minor point, the following statement is not entirely clear: "…all membrane proteins with different topology and suggested localization to the ER…" (page 5). How was ER localization of FRRS1l and C PT1c suspected?

Reviewer 1
We thank the reviewer for the positive comments on our work and his suggestions for further improvement that were incorporated into the revised manuscript.

General comment(s):
Overall, this is an intriguing and thorough piece of research, using a remarkably diverse array of methods. The evidence of the importance of FRRS1l for AMPAR receptor biogenesis is important and will be of general interest. Brechet showing the apparent deleterious effect of the intellectual disability mutations is also very important.
There are two main strengths of the manuscript. One is in documenting the association of these proteins in complexes and the other is in providing genetic and knockdown evidence that they are involved with synaptic transmission and behavior. While we appreciate the positive judgement on the overall concept and validity of our work, we firmly disagree with the statement that the ms does not provide insights into 'the nature of the interaction' (between GluAs and FRRS1l) and into its mode of operation (suggested involvement of potential enzymatic activities). Thus: (a) The structural requirements of the interaction have in fact been investigated in detail and the hydrophobic domain in the C-terminus of FRRS1l (per se or by guiding GPI-anchoring at S317) has been identified as the decisive domain for interaction (Figure 4c, d); deletion of this domain by two disease mutations (Q321*. V195*) abolishes FRRS1l binding (Fig. 4d)  Statistics have been included into the ms where meaningful. In particular, we have added statistics for the proteomic data in Figures 1, 2 and 8 (as requested by the reviewer). The data on the two-step APs (Figure 1b) containing combined data for rats and mice (to show the close resemblance) in the original ms, were separated in the revised ms (rat data in Fig. 1b, mouse data in Supplementary Figure 1b); EM data were quantified (evaluation of 848 immunoparticles). Data on genetics, as well as on all functional measurements had already been indicated in the original ms.

Figure 1a. Shows a heat map of protein abundance from multiple APs. It is not clear in the figure legend or methods how the numbers were generated to produce this heat map. The units (mol. Abundance [log a.u.]) require clear explanation.
Generation of the heat map (Figure 1a) is detailed in the Methods section and has been introduced in previous work (in particular, the QConCAT technique for absolute quantification, see also Schwenk et al., Neuron 2012). The data from proteomic analyses have been summarized and added to the revised ms as Supplementary Table 3; the original raw data are deposited to public databases as suggested by the reviewer.

Figure 1a,b. Why were not the same assays performed with antibodies to CPT1c? This appears incomplete and the data would be valuable for their hypothesis about the mutual localization and interactions.
Experiments as in Figures 1a, b were not performed for CPT1c, as the anti-CPT1c antibodies (ABs) in hand did not allow for depleting affinity-purifications (APs) thus precluding clear separation of AMPAR pools.

Figure 2. Refers to 4 different anti-FRRS1l proteins and two different anti-CPT1c antibodies and presents mean +-SEM of 5 and three Aps. This is confusing: what do the error mean? How did the different antibodies perform? The key with red/blue is also unclear.
The discrepancy results from one AP in either case using a mixture of the different ABs; red bars refer to the respective AB target protein in either of the two APs (FRRS1l (upper panel), CPT1c (lower panel)) that were used for normalization (of protein amounts). This has been clarified in the revised legend to Figure 2a.
6. The abstract says "FRRS1l and CPT1c …specifically and cooperatively bind to the GluA1-4 proteins of AMPA receptor". Can they clarify if this protein is really specific to AMPA receptors or is it that the AMPA receptor is the major binding partner in the brain. Expression databases indicate that FRRS1l and CPT1c are expressed in many non-neuronal tissues where there are not thought to be functional AMPA receptors. Could it be that the FRRS1l and CPT1c are more general rather than AMPAR-specific? Further to this point see the next two comments.

Searching public databases (including EMBL-EBI Expression Atlas, Human
Proteome Map, Human Protein Atlas) indicated that the only tissue displaying consistent and high expression of both FRRS1l and CPT1c is the brain (adult, less in fetal brain). Moreover, several papers report focused expression in brain (e.g. Price et al, 2002;Madeo et al, 2016); in fact, the only non-neuronal tissue displaying some expression outside the brain was testis. In this context, the statement of FRRS1l/CPT1c being specific interactors of AMPARs appears justified, in particular as there is no indication for FRRS1l and CPT1c being 'more general'.

They human phenotypes. While it is possible that the phenotypes are due to altered expression of AMPA receptors, I would advise caution in how this is explained (until mouse genetic experiments show some related phenotypes with electrophysiology and rescue). The FRRS1l protein can have other functions that could lead to the intellectual disability phenotype and the authors have not separated these possibilities.
Detailed proteomic analyses (multi-epitope APs with target-knockout controls, Figure  2a) revealed that the vast majority of interaction partners -identified for both (!) FRRS1l and CPT1c -are AMPAR constituents (with protein amounts totaling to more than 99% of all partner proteins identified). Based on these findings and the aforementioned observations (see point 6), it appears rather unlikely that FRRS1l is serving a function that is both independent from AMPARs and relevant for the presented disease. The disease mutations reported here were identified in two independent systematic studies analyzing cohorts of consanguineous families for genes causative for autosomal recessive intellectual disability. In line with a 'recessive disorder' all affected patients are homozygous for the respective mutation, while the healthy siblings and parents are either heterozygous for the mutation or homozygous for the wildtype allele. Genotypes have been added to the revised Figure 3a, as suggested by the reviewer.

Figure 4a. Sac1 immunoblot of anti-GluA AP samples (second panel from left). This immunoblot looks under-exposed compared to the other anti-GluA AP panels. The Sac1 band in the GluA1/2-CPT1/FRRS1l lane is faint. There are even fainter bands in the GluA1/2-Sac1 lane and the GluA1/2-Sac1-FRRS1l lane
Equivalent amounts of input and eluates of all APs were loaded, separated (on the same gel) and Western-probed under the same conditions to guarantee comparability (and avoid signal distortion by over-exposure). Although faint, the dynamic range of the Sac1-staining clearly indicated more effective binding to GluAs in the presence of both FRRS1l and CPT1c (over the individual co-assemblies). This result from heterologous expression is perfectly in line with the AP results from native tissue that show Sac1 as a specific but low abundant partner of AMPAR assemblies (Figures  2a, b) Page 5. Sentence beginning "Interestingly, these AMPAR subunits displayed…" is confusing. Please re-write.
The sentence was re-phrased in the revised ms.
11. Page 7. First two lines. "And suggest the importance of FRRS1l-containing AMPARs for normal brain development and function". Brechet shows that FRRS1l is found in at least two distinct complexes: a complex that lacks AMPARs and complex that contains AMPARs. It is not known if the mutations in FRRS1l affect the AMPAR containing or the AMPAR-lacking complexes? I recommend removing this statement from the results section.
As pointed out above (point 7), AMPARs are (by far) the predominant interactors of FRRS1l and it appears therefore reasonable to assume that the entirety of FRRS1l protein is present in an equilibrium between AMPAR-associated and AMPAR-free pools. Moreover, all mutations are finally shown to affect the assembly of FRRS1l with AMPARs. In this sense, we would prefer to leave the sentence as it stands. Supp. Fig. 3.) to demonstrate that FRRS1l has undergone a redistribution when CPT1c is co-expressed. It would also be helpful to show some evidence of quantitative image analysis to convince the reader of this point.

Page 7. Paragraph beginning "Further analysis by confocal…". The authors suggest the confocal data in figure 7b shows CPT1c expression causes a redistribution of FRRS1l from the surface to an intracellular membrane compartment. None of the confocal images show what area of the image is intra-versus extra-cellular. I suggest that if the authors wish to keep that point in the results section text, they probably need to show some surface stain (perhaps using surface biotinylation) or an intracellular marker (perhaps calnexin could be used as Brechet et al us this in
Re-distribution of FRRS1l (from plasma membrane to ER) by the ER-resident CPT1c is shown in Figure 4b, in direct contrast to the failure (to induce re-distribution) by CPT1a.
To confirm this re-distribution, we have added two additional experiments: A surface biotinylation assay (as suggested by the reviewer) and electrophysiological recordings. These results were included into the revised ms as new Supplementary The color-coding of the distinct MW bands of the FRRS1l protein, as well as the Cterminal tag used were added to the revised caption of Figure 4c (and had been indicated in the text, supplementary figure and methods part of the original ms). Figures 2 and 4 were quite difficult to follow, especially in contrast to other parts of the text, which were very clear and concise.

The results section on the data in
We tried to elaborate on this as far as possible and in line with the reports and requests from the other reviewers.

Figure 5a,b. Why were not the same assays performed for CPT1c? This appears incomplete and the data would be valuable for their hypothesis about the mutual localization and interactions.
These experiments were not possible due to the unfavorable properties of the anti-CPT1c ABs in hand.
Other edits: 1. They should spell out the full name of FRRS1l and CPT1c and describe them in some more detail in the introduction, including information on their expression in other tissues etc.
The names of both proteins were explicitly given in the introduction section (although, these names are misleading with respect to their (primary) function, see response to the general comment).
2. They repeatedly refer to "profound" effects and findings. For this reviewer, none of these findings justify the use of this adjective. Many of their "profound" findings would be better described as expected, or in agreement with, or significant. I also note that some of these "profound" findings are not backed up with statistics and thus it remains for the authors to demonstrate that they are significant.
This word was replaced for most parts.

Reviewer 2
We appreciate the reviewers' positive comments, as well as his criticism that prompted additional experiments included into the revised manuscript as detailed below in the responses to the reviewer's comments.
Specific comments: 1. The experiments provide an interesting initial characterisation of these AMPAR associated proteins, but the investigation into AMPAR/FRRS1l/CPT1c function currently lacks any development of a mechanism beyond a "trafficking role". It would be of considerable interest to see at what stage in AMPAR biogenesis these proteins work, do they play a role in initial protein folding, and/or in the assembly of dimers or tetramers, or even during core subunit heteromerisation? Would QconCAT allow them to assess whether FRRS1l/CPT1c indeed associates with AMPAR tetramers or could they selectively complex with assembly intermediates (monomers/dimers)?
As requested by the reviewer, we have added data on native gel-separations of AMPARs and their evaluation (for subunit composition) by both Western-blot and cryo-slicing BN-MS (using QConCAT for quantification). These data demonstrate that FRRS1l/CPT1c do only form complexes with GluA-tetramers, association with assembly intermediates were not detected in the membrane fractions from rodent brain (Supplementary Figure 5b, c). Consequently, FRRS1l/CPT1c are not expected to influence the folding and assembly of the AMPAR pores.
Moreover, we probed interaction of FRRS1l/CPT1c with GluA1-4 (separately) for potential preferences in assembly. The results of respective anti-FRRS1l APs, however, did not provide any indication for differences in FRRS1l-binding to the individual GluA proteins (Supplementary Figure 5a). Importantly, FRRS1l did not show any preference for GluA2 (used in R-edited form here). These additional data were added to the revised ms as Supplementary Figure 5.
AMPAR processing in the ER has been investigated and this information should be integrated into the current study; e.g. individual core subunits traffic differently, e.g GluA3 traffics poorly compared to the other subunits (Coleman et al. 2010); RNA editing and alternative splicing of subunits determines ER exit rates (e.g. Coleman et al. JBC 2010, Penn et al. EMBOJ 2008) -do the current findings bear any relevance to this and might it play a role in selective action of FRRS1/CPT1c. ER processing has been mostly investigated in heterologous expression experiments (e.g. Coleman et al) with appearance of functional receptors at the plasma membrane or mature glycosylation pattern as a read-out. While the relevance of these experiments (missing the essential ER components identified here) is not immediately clear in the context of this work, factors influencing ER-exit have not been studied in this ms, but may contribute to the 'trafficking' of AMPARs to the surface. Importantly though, these factors do not compensate for disturbances introduced into biogenesis by knock-down of FRRS1l and/or CPT1c. Both facts have been stated in the Discussion section of the revised ms (p. 13, lines 28-35); the suggested citations were added.

For example, it appears from Figure 2a that GluA2 is a major component of FRRS1l/CPT1c
ER complexes consistent with GluA2 retention in the ER due to Q/R editing (Greger et al. Neuron 2002). Is this the case? Also, have they probed blots with anti-GluA2 following anti-GluA1 AP to assess the proportion of AMPAR heteromers? One may expect that ER complexes are incompletely assembled AMPARs as 1Q/2R heteromers are exportcompetent.
As detailed in the new Supplementary Figure 5 (and also in Schwenk et al, 2012), GluA2 is the most abundant pore-forming subunit of AMPAR assemblies both in their 'plasma membrane' form (low molecular mass range, assembly with CNIH/TARPs), as well as in their FRRS1l/CPT1c-associated 'ER-form' (high molecular mass range, at the maximum of FRRS1l/CPT1c). In fact, there is no observable difference between plasma membrane-and FRRS1l-associated AMPARs with respect to the GluA-composition.
The proportion of GluA heteromers has been detailed in previous work by AB-shift assays with anti-GluA1 and anti-GluA2 ABs showing that the vast majority of detectable AMPARs (in all membranes) are heteromers (Schwenk et al, 2012; Figure  3).
Related to the text on p7 (middle), why is 'data not shown' for the ability CPT1c to retain GluA1 in the ER via interaction with FRRS1l? What happens in the presence of GluA2? One might expect FRRS1l/CPT1c to drive assembly of GluA1 with GluA2. Likewise, the neuronal results (Figure 2a) show that all FRRS1l/CPT1c complexes are associated with GluA2 while being evenly distributed between GluA1 and GluA3, allowing them to recruit either subunit to unassembled GluA2.
There appears to be some misunderstanding here: The data (from biotinylation assays and patch-clamp recordings) demonstrating redistribution of GluA1-associated FRRS1l by CPT1c (cited as not shown in the original ms on p.7) were added to the revised ms as new Supplementary Figure 4.
There is no effective ER-retention of GluA tetramers as stated by reviewer, rather transient binding of FRRS1l/CPT1c to GluA tetramers catalyses assembly of the latter with CNIH/TARP thus rendering them competent for ER-exit and transport to the plasma membrane (see Figure8b).

The immunocytochemistry in Figs 4b and 5a does not include AMPA staining, GluA2 should be a nice marker for ER-localized AMPARs and should have been included.
The goal of Figure 5a was probing absence of FRRS1l from the synaptic compartment in brain neurons (contained in hippocampal slice preparations), as suggested by the ER-staining obtained in culture cells. GluA2 (showing both, synaptic, extrasynaptic and intracellular staining) would not have been a suited marker for these experiments. Nonetheless, we added experiments with heterologously expressed GluA2(R) to the revised ms (new Supplementary Figure  3b) that show close co-localization of GluA2 and FRRS1l/CPT1c in the ER.
Do sh-FRRS1l and sh-CPT1c affect rectification of evoked EPSCs? Loss of FRRS1l/CPT1c complexes may enable forward trafficking of GluA2-lacking receptors to the synapse before they have the chance to reassemble with GluA2. The loss of 60% is reminiscent of conditional GluA2 knockout (Lu et al., Neuron 2009).
Rectification was not tested, but the gating characteristics (also suitable to reflect the respective changes) did not reveal any differences (see Figure 6, 7); a reduction to (not of) 60% may not be a strong indicator in this sense. Figure 1a: is the first red box intentionally selecting CNIH-3 but not CNIH-2? It appears that these two proteins are found to similar levels in these samples. It also appears that there is a significant amount in the ER complexes unlike the TARPs and other core subunits.

Minor comments
The framing of the red box in the heat map (with a dynamic range of 5 orders of magnitude) for CNIH2/3 was somewhat empirical and drawn to be consistent with the AP results shown in Figure 2a -where CNIH2 was co-purified consistently, albeit at very small amounts (less than 1%) with ER-resident (FRRS1l-containing) AMPARs. This small amount may reflect the transitional priming complex forming during biogenesis (Figure 8b). P4: 'FRRS1l effectively associates with CPT1c and Sac1 independent of GluA1-4' is not completely true as there is more of these proteins in the AMPAR-containing samples (comparing brown bars against red bars in figure 1b).
This sentence was meant to emphasize robust assembly even in the absence of GluAs, rather than making a comparative statement on the equilibrium between GluAfree and GluA-associated CPT1c or Sac1 (see also Figure 4a). This sentence was changed to emphasize that 'close to stoichiometric' refers to anti-FRRS1l APs only.
P5 again: One would also argue that SAC1 and PORCN are barely associated with the FRRS1l-or CPT1c-containing AMPARs based on Figure 2a. Figure 4a also shows limited pull-down of SAC1 by AMPARs although it is pulled down by FRRS1l. The mention of these is probably fine however as they are not followed up later on.
This was left as in the original ms.
P10: The sentence starting 'the decrease and increase in EPSC amplitude observed with sh-FRRS1l and sh-CPT1c' is confusing because FRRS1l overexpression, which causes the increase, is not mentioned.
This has been rephrased to avoid confusion. P11: 'Knockdown or exogenous (over-)expression of FRRS1l or CPT1c' suggests both manipulations were applied to both genes/proteins but no data is shown for CPT1c overexpression. Please change to make this clear.
This has been rephrased to avoid confusion. The information on FRRS1l and CPT1c as available through the current literature has been added to the revised ms (Introduction, Results and Discussion). Figure 8b: there is no evidence that AMPARs assembling with CPT1c and FRRS1l are tetramers.

P17 and
As pointed out above, these data are explicitly shown in the new Supplementary Figure 5 (but have also been shown in Schwenk et al., 2012 (Fig. 2)).