LARGE, an AMPA receptor interactor, plays a large role in long-term memory formation by driving homeostatic scaling-down

Dynamic trafficking of AMPA-type glutamate receptor (AMPA-R) in neuronal cells is a key cellular mechanism for learning and memory in the brain, which is regulated by AMPA-R interacting proteins. LARGE, a protein associated with intellectual disability, was found to be a novel component of the AMPA-R protein complex in our proteomic study. Here, our functional study of LARGE showed that during homeostatic scaling-down, increased LARGE expression at the Golgi apparatus (Golgi) negatively controlled AMPA-R trafficking from the Golgi to the plasma membrane, leading to downregulated surface and synaptic AMPA-R targeting. In LARGE knockdown mice, long-term potentiation (LTP) was occluded by synaptic AMPA-R overloading, resulting in impaired long-term memory formation. These findings indicate that the fine-tuning of AMPA-R trafficking by LARGE at the Golgi is critical for memory stability in the brain. Our study thus provides novel insights into the pathophysiology of brain disorders associated with intellectual disability.


36
LARGE is expressed strongly in the brain (particularly the hippocampus), relative to other 37 tissues (Peyrard et al., 1999). In humans, mutations in LARGE are associated with 38 congenital muscular dystrophy type 1D, which is characterized by clinical features including

45
These human and mouse studies suggest that abnormal synaptic function may be 46 responsible for intellectual disabilities in human patients with LARGE mutations.

47
In our previous proteomic analysis, we found that LARGE forms a protein complex with 48 the AMPA-type glutamate receptor (AMPA-R) (Kang et al., 2012). Excitatory glutamatergic 49 synaptic transmission within the central nervous system is primarily mediated by AMPA-R, 50 as well as NMDA-type glutamate receptor (NMDA-R), and increasing numbers of proteins 51 have been found to form complexes with and thus regulate the dynamic trafficking of AMPA-8 Next, we used confocal imaging to assess the subcellular localization of LARGE and thus 153 understand the mechanism by which LARGE downregulates AMPA-R surface targeting.

154
LARGE is known to localize at the Golgi in heterologous cells (Brockington et al., 2005) 155 and to function at the Golgi in myocytes (Kanagawa et al., 2004). Similarly, in cultured 156 hippocampal neurons, we observed a major pool of LARGE at the Golgi and Golgi outposts 157 ( Figure 3A). Accordingly, we hypothesized that LARGE downregulates AMPA-R trafficking 158 from the Golgi to cell surface by increasing AMPA-R localization at the Golgi.

159
To test this concept, we first analyzed the co-localization of GluA1 with the Golgi marker

184
Both co-IP strategies consistently showed that LARGE could specifically bind to AMPA-R in 185 non-neuronal cells that do not express other known AMPA-R-binding proteins ( Figure 3E).

186
Interestingly, in a co-IP of LARGE with GluA2, LARGE bind only to the GluA2 correspond to   Figure 3F). Notably, GluA1 bound 195 to LARGE with a higher affinity relative to that exhibited by GluA2 or GluA4 (Figure 3G).

196
The interactions of LARGE with GluA1, 2, and 4 indicated that LARGE binding to AMPA-R 197 is not subunit-specific. Similarly, the binding of the other AMPA-R interacting proteins, such

218
Together, these data strongly suggest that the co-localization of GluA1 and LARGE at the 219 Golgi increases during synaptic scaling-down. Indeed, subcellular fractionation confirmed 220 increases in both GluA1 and LARGE at the Golgi during scaling-down ( Figure 4D). Finally, 221 co-IP confirmed a significant increase in the association of LARGE with GluA1 during 222 bicuculline-induced synaptic scaling-down ( Figure 4E). Although the amount of GluA1 223 immunoprecipitated by a GluA1 antibody decreased after bicuculline treatment, probably 224 11 because of increased protein (e.g., LARGE) binding and consequently reduced epitope 225 exposure, the amount of LARGE that co-immunoprecipitated with GluA1 remained 226 significantly elevated. Together, our results strongly support our working model, wherein  short-term plasticity, which were evaluated using the paired-pulse ratio (PPR). The PPRs at 12 inter-pulse intervals of 200, 100, 75, 50, and 25 ms did not differ between control and LARGE 249 KD mice (Figure 5-figure supplementary 1B), suggesting that the synaptic changes 250 observed in the latter mice are not presynaptic events.

251
The above results strongly suggest that LARGE KD increases the synaptic current by 252 increasing the number of AMPA-R molecules at the postsynapses. We therefore further

265
Second, we subjected cultured hippocampal neurons to confocal imaging to demonstrate 266 that the number of GluA1 molecules within the dendritic spine increased significantly with 267 LARGE KD relative to control neurons, and this increase was reversed by LARGE rescue 268 ( Figure 5C). Moreover, LARGE KD neurons had significantly larger spine heads but similar 269 spine densities relative to control neurons; again, this was reversed by LARGE rescue

291
We note, however, that KO mice are constitutive mutants. Therefore, the observed 292 memory deficits may be attributable to abnormal brain development. Accordingly, we

306
We next subjected LARGE KD mice to various memory tests to identify the specific memory-307 associated role of LARGE in the brain. An initial open field test revealed no significant 308 differences between KD and control groups ( Figure 7A). In other words, LARGE KD mice 309 exhibit normal locomotion and anxiety levels. To test spatial working memory, we used a Y-

313
Next, we used a simple novelty preference test that has used in previous studies of GluA1

317
Although both groups exhibited a similar degree of preference for the novel arm at 1 min, 318 the LARGE KD group failed to exhibit a preference for the novel arm at 24 h, compared with 319 the control group ( Figure 7C).

15
Finally, we performed the novel object recognition test, a popular testing paradigm for 321 hippocampal function as a relay point of recognition memory (Stilling et al., 2014). During 322 the training session, both groups displayed similar degrees of preference for two equal 323 objects, with no inter-group difference in the short-term (5 min) preference for a novel object.

324
Over the long term (24 h), however, the LARGE KD group failed to display a preference for 325 the novel object relative to the control group ( Figure 7D). Taken together, these findings 326 suggest that hippocampal LARGE KD specifically impairs long-term, but not short-term, 327 spatial and recognition memory.

377
Next, we applied simple novelty preference and novel objective recognition tests to 378 respectively examine spatial and recognition memory in LARGE KD mice. The former test 379 identified an impairment in long-but not short-term spatial memory ( Figure 7C). In simple 380 novelty preference test, spatial long-term memory is formed through repetitive training over

394
In consistent with our results, LTP occlusion due to elevation in postsynaptic AMPAR surface

459
The AAV was packaged and purified as follows. The shRNA designed from the selected

563
The novel object preference test was performed as described previously (Stilling et al., 2014).

564
Mice were habituated to a white acryl box for 5 min on each of 2 consecutive days. Mice 565 were then habituated to the same two objects placed in corners of the box for 5 min on each 566 of the 2 consecutive days. The following day, the objects were exchanged for two new, 567 identical objects (A + A), and the mice were allowed to explore the objects for 5 min. Next, 568 the mice were placed in their home cages for 5 min (short-term memory task) and re-569 exposed to the arena in which one object had been exchanged (A + B). After 24 h, B was 570 exchanged for C (long-term memory task). The durations of object contacts were measured.

573
In vivo field EPSP recordings 574 Three to 4 weeks prior to in vivo field recording, AAV expressing LARGE shRNA with GFP 575 was infused into the hippocampal CA1 regions of adult C57BL/6 mice (8-9 weeks old) as 576 described in Stereotaxic injection of virus into animals. The stereotaxic unilateral injection of 577 0.5 μl of higher-titer AAV (1  10 11 TU/ml) was performed using a stereotaxic, motorized 578 nano-injector (World Precision Instruments) at a rate of 0.1 μl/min via a Hamilton syringe 579 connected to a microinjection pump.

582
Sigma) and placed into a stereotaxic frame. Rectal temperature was maintained 583 intraoperatively at 36.5ºC ± 0.5ºC using a temperature controller (Harvard Instruments).

584
The scalp was opened and separated. Trephine holes were drilled into the skull, and

625
Changes in frequency and amplitude were analyzed, quantified, and presented using traces, 626 cumulative plots, and scatter plots. We confirmed that the frequency and amplitude did not

665
Sigma #D1556), and optimized for two scales (5-ml volume gradient and 0.7-ml volume 666 gradient). Briefly, cells or brain lysates were centrifuged for 10 min at 3000  g, and the 667 pellet was discarded. The supernatant was then centrifuged for 1 h at 100,000  g to remove

771
To quantify the confocal images presented in Figure 3C, the GluA1 intensity in the spine 772 and integrated intensity of individual endogenous GluA1 puncta in the dendritic spine were 773 measured. Images were analyzed using NIS-Elements AR (Nikon). To quantify the confocal 774 images presented in Figure 5C and 5E, fluorescence signals in the soma and dendrites were 775 quantified by measuring the area containing signals above a certain threshold. After

798
The images were analyzed using NIS-Elements AR (Nikon).            in Golgi was increased by LARGE overexpression (GluA1 + LRG), compared with a control (GluA1) (n = 30, 30 cells; Two tailed t-test, *P < 0.001). Co-localization of GluA1 with GM130 in HEK293T cells analyzed by 3D reconstruction of a series of z-stack confocal images. Complete (yellow) and partial co-localization (orange). Co-localization of GluA1 and LARGE was quantified using the co-localization analysis tool in the NIS-Elements software (Nikon). In the analysis, Manders overlap coefficients were given and used to obtain the relative co-localization values between those two proteins. Scale bar = 10 μm. (B) Fractionation of subcellular organelles from the hippocampal CA1 of mice. Representative data showing the fractionation of organelle markers and LARGE. P-cadherin, plasma membrane marker; GM130, Golgi markers.

GluA1
GluA1 + LRG HEK293T cells were transfected with myc-GluA1, myc-GluA2, and/or LARGE-GFP. IP with anti-GFP antibody. Both GluA1 and GluA2 were co-immunoprecipitated with LARGE-GFP. As see in Input, GluA2 yield two bands (upper and lower bands). Most GluA2 co-immunoprecipitated with LARGE was GluA2 correspond to intracellular GluA2, judging from its molecular weight.  Spine head area (μm 2 ) Cum. Prob. Figure 5-figure supplementary 2. Confocal image analyses demonstrated the effect of LARGE KD on structural synaptic plasticity. Confocal images of cultured hippocampal neurons showed that LARGE KD (shRNA) significantly increase not the number of spines but the size of spines compared to that of control neurons, which was reversed by LARGE rescue (n = 7, 7, 7 neurons; One-way ANOVA, *P < 0.01, **P < 0.005). In the cultured hippocampal neurons, neuronal dendrites and spines was visualized by GFP. PSD95 is synaptic marker. MAP2 is dendrite marker. Scale bar = 10μm.  Confocal images showed the location and diffusion range of AAV microinjected into CA1. Endogenous LARGE mRNA expression in CA1 was significantly knocked down by infection of AAV expressing LARGE shRNA with GFP (shRNA) (n = 3, 3 mice; Two tailed t-test, *P < 0.001). Scale bar = 100 μm (left), 50 μm (right). (B) Digital image and Western blot analysis of hippocampal CA1 region of rat brain infected with AAV. Brain slices were imaged by digital imaging under a blue LED light. Strong and specific expression of GFP in hippocampi indicated specific delivery and expression of LARGE shRNA with GFP by AAV injection. Western blot analyses confirmed knockdown of LARGE in GFPexpressing hippocampi (n = 3, 3 rats; Two tailed t-test, *P < 0.005) The absence of a significant difference in shock threshold among groups demonstrated that LARGE expression status did not affect fear conditioning. (A) Schema of shock threshold tests of animals. Shocks (gray blocks) were delivered every 30 s, with intensity increasing from 0.1 mA to 1.0 mA (4:28, 4:58) and decreasing back to 0.1 mA (10:30). (B) Regardless of genotype or injected AAV, all animals responded to shocks in a similar way. The shock intensity thresholds for jumping, vocalization, and flinching were measured for wild type (+/+), heterozygous (+/-), and knockout (-/-) animals and animals injected with AAV expressing scrambled shRNA with GFP (Cont) or LARGE shRNA with GFP (shRNA).