Gga3 deletion and a GGA3 rare variant associated with late onset Alzheimer’s disease trigger BACE1 accumulation in axonal swellings

Abstract

Axonal dystrophy, indicative of perturbed axonal transport, occurs early during Alzheimer’s disease (AD) pathogenesis. Little is known about the mechanisms underlying this initial sign of the pathology. This study proves that Golgi-localized γ-ear-containing ARF binding protein 3 (GGA3) loss of function due to Gga3 genetic deletion or a GGA3 rare variant that co-segregates with late-onset AD (LOAD), disrupt the axonal trafficking of the β-site APP-cleaving enzyme 1 (BACE1) resulting in its accumulation in axonal swellings in cultured neurons and in vivo. We show that BACE pharmacological inhibition ameliorates BACE1 axonal trafficking and diminishes axonal dystrophies in Gga3 null neurons in vitro and in vivo. These data indicate that axonal accumulation of BACE1 engendered by GGA3 loss of function results in local toxicity leading to axonopathy. Gga3 deletion exacerbates axonal dystrophies in a mouse model of AD prior to amyloid-beta (Aβ) deposition. Our study strongly supports a role for GGA3 in AD pathogenesis, where GGA3 loss of function triggers BACE1 axonal accumulation independently of extracellular Aβ, and initiates a cascade of events leading to the axonal damage distinctive of the early stage of AD.

One sentence summary

GGA3-mediated BACE1 accumulation in dystrophic neurites is an early pathogenic event and is independent of Aβ extracellular deposition.

Introduction

Dysfunction of axonal transport has been implicated in the pathogenesis of Alzheimer’s disease (AD) (1–3). Axonal swellings indicative of axonal transport disruption have been found in postmortem brains from early stage AD patients (4). It has also been shown that in vivo impairment of axonal transport and decreased axonal transport rates might have an impact on AD pathogenesis early in the disease process (5–7). Moreover, increased plasma concentration of neurofilament light chain (NfL) indicative of axonal damage has been detected in AD presymptomatic stages (8, 9). However, the mechanisms leading to axonal damage in the early phase of pathology remain unclear.

Dystrophic neurites surround amyloid-beta (Aβ) deposits forming neuritic plaques (10). The aspartyl protease BACE1 is the β-secretase that catalyzes the rate limiting step in Aβ generation (11). Under normal conditions, BACE1 localizes to presynaptic terminals whereas in AD brains it accumulates in dystrophic neurites (12–15). Likewise, defects in axon transport leading to endogenous BACE1 build-up in axonal swellings have been reported (16). Although we do not know precisely when BACE1 pathology occurs, other studies suggest it to occur early in AD mouse models (17). Nevertheless, the specific mechanism(s) of BACE1 accumulation remains unknown.

We previously demonstrated that the Golgi-localized γ-ear-containing ARF binding protein 3 (GGA3), a monomeric clathrin adaptor highly expressed in the brain (18), regulates BACE1 lysosomal degradation (19). GGA3 consists of four segments: a VHS domain that binds the acidic di-leucine sorting signal, DXXLL; a GAT domain that binds Arf:GTP and ubiquitin; a hinge region that recruits clathrin; and a GAE domain that exhibits sequence similarity to the ear region of γ-adaptin and recruits a number of accessory proteins (20). We have shown that in non-neuronal cells GGA3 depletion induces BACE1 accumulation in early endosomes by preventing its delivery to lysosomes (19, 21, 22), and that Gga3 deletion increases BACE1 amount in vivo (22) and exacerbates Aβ pathology in a mouse model of familial AD (23). Moreover, we established that GGA3 is decreased and inversely correlated with BACE1 in the temporal cortices of AD patients (19) and in a mouse model of familial AD (23).

Despite compelling evidence indicating that GGA3 regulates BACE1 trafficking, the role of GGA3 in BACE1 neuronal polarized sorting is unknown.

Here we characterized the neuronal localization of GGA3 in murine hippocampal neurons and found that the protein is distributed, along with BACE1, in both dendrites and axons. We determined that Gga3 deletion triggers BACE1 elevation and axonal trafficking disruption resulting in BACE1 accumulation in axonal swellings. BACE pharmacological inhibition prevented axonal swelling formation, in vitro and in vivo, and improved BACE1 axonal motility. These data indicate that axonal accumulation of BACE1 triggered by Gga3 deletion results in localized toxicity leading to the disruption of the trafficking of other presynaptic proteins and axonopathy.

In support of a role of GGA3 in AD pathogenesis, we identified a GGA3 rare variant, Ins545T, that co-segregated with late-onset AD (LOAD). Ins545T results in a loss of function as demonstrated by its inability to rescue BACE1 axonal trafficking and accumulation in axonal swellings in Gga3^−/− neurons. Finally, we demonstrated that in 2-month-old 5XFAD mice, Gga3 deletion exacerbates axonal pathology and triggers BACE1 accumulation in axonal swellings prior to senile plaques formation.

Our findings demonstrate that GGA3 loss of function triggers BACE1 axonal buildup and initiates a cascade of events leading to the axonal damage characteristic of the early stage of AD in the absence of extracellular Aβ deposition in mice.

Results

GGA3 is sorted to both dendrites and axon

Despite the abundance of the clathrin adaptor GGA3 in the brain, little is known about its distribution in neurons or its role in somatodendritic and axonal sorting of cargoes. Confocal microscopy of in vitro day 8 (DIV8) mouse hippocampal neurons expressing exogenous GGA3 tagged at its N-terminus with a green fluorescent protein (GFP-GGA3) showed that GGA3 localizes to both dendrites and axons (MAP2-negative neurites) (Fig. 1A). Quantification of mean gray fluorescence intensity in dendrites versus axons yielded a polarity index (PI) of 1.85±0.12 (mean±SEM), confirming that GGA3 is sorted to the somatodendritic and axonal compartments with a slight preference for dendritic targeting (Fig. 1C).

Figure 1. GGA3 is sorted to both dendrites and axon.

(A) Representative neuron expressing GFP-GGA3 (green - grayscale inverted left) immunostained for MAP2B (red). Arrowheads point to the axon. Stitched confocal z-stacks. Magnification: 63x Oil. Scale bar: 100 μm. (B) Representative neuron expressing AP1(μ1A)-GFP (grayscale inverted). Vector-mCherry (red) and MAP2B (blue). Arrowheads point to the axon. Stitched confocal z-stacks. Magnification: 40x Oil. Scale bar: 50 μm. (C) PI analysis: GFP-GGA3 (n=18), AP1(μ1A)-GFP (n=31). (D) DIV10 neurons grown in a microfluidics device stained for MAP2B (red), NFH (green), DAPI (blue). Stitched confocal z-stacks. Magnification: 20x Oil. Scale bar: 50 μm. (D) Western blot of cell and axonal proteins probed for MAP2B, TubBIII, GGA3.

Increasing evidence is accumulating for a role of the adaptor protein complex AP1 in the sorting of cargoes to the somatodendritic domain (24). Furthermore, the μ1A isoform, the only isoform expressed in the brain (25), is highly polarized in the soma/dendrites and is excluded from the axon. To assess the reliability of our experimental system, we investigated AP1(μ1A)-GFP polarization in our cultures. As expected, AP1(μ1A)-GFP (Fig. 1B) exhibited a PI of 10.92±1.18, distinctive of somatodendritic proteins (Fig. 1C).

Next, we evaluated endogenous GGA3 neuronal distribution by using a commercially available anti-GGA3 antibody. The specificity of this antibody has been validated by western blot of Gga3^−/− brain extracts (22). However, after immunostaining Gga3^−/− neuronal cultures, non-specific signal was detected, preventing the use of this antibody for immunostaining. Thus, we cultured hippocampal neurons from postnatal pups in polydimethylsiloxane (PDMS) microfluidic devices. These devices allowed us to culture neurons in a polarized manner, and to directly isolate/analyze axons. Neurons cultured in one of these chambers project their axons across the barrier into the other chamber (Fig. 1D). To determine the purity of the axonal fraction, we first performed immunofluorescence analysis. Two markers were used (Fig. 1D): microtubule associated protein 2B (MAP2B), which is excluded from the axon, and neurofilament heavy chain (NFH). The signal difference between these two markers was used to distinguish axons from dendrites and demonstrate that solely axons were able to cross the barrier and extend into the axonal chamber after 10 days in culture. We then collected protein lysates from both the cell and the axonal chambers (Fig. 1D). Due to the small volume and low concentration of our axonal lysates, the entire sample was loaded into the gel without protein concentration analysis. For this reason, different amounts of protein lysates from the cell chamber were loaded and used as a reference to estimate the concentration of our target proteins in the axons (Fig. 1D). Western blot showed a MAP2B positive band only in the cell chamber lysates, whereas a band for TubβIII was detected both in the cell and axonal lysates (Fig. 1D). This approach confirmed the presence of endogenous GGA3 in axons (axonal side/cell side densitometry: 0.10±0.02, n=4) (Fig. 1D), recapitulating the distribution of exogenous GFP-GGA3. Although other clathrin adaptor proteins, like AP1 and AP4 (24, 26), are present only in the somatodendritic compartment, we demonstrated the presence of GGA3 in the axon.

BACE1 and GGA3 are co-transported in axons

BACE1 localizes to the presynaptic compartment (13) and it is targeted to the axon via transcytosis (27, 28). We assessed BACE1 polarity and colocalization with GGA3 in our model. Exogenous BACE1-mCherry was transfected in low-density hippocampal neurons, and a PI calculated: the PI for BACE1-mCherry (2.04±0.08) was comparable to GFP-GGA3 (1.85±0.12), indicating that the two proteins distribute and are targeted similarly to the somatodendritic and axonal compartments, where they colocalize (Manders’ Coefficient: 0.87±0.04 and 0.78±0.06, respectively) (Fig. 2A).

Figure 2. BACE1 and GGA3 are co-transported in axons.

(A) BACE1-mCherry (red) and GFP-GGA3 (green) colocalization analysis in wild-type neurons. Z-stack confocal images. Magnification: 63x Oil. Scale bar: 100 μm. Manders’ Coefficient: n=8. Single frames from Movie S1A–B (top, TIRF Microscope 60x Oil, 2 fps) and corresponding kymographs (bottom) of (B) GFP-GGA3- (C) and BACE1-mCherry-positive vesicles (n_v=397 and 897) moving for 3 min along 100 μm axons in wild-type neurons (n=14 and 15). In kymographs, lines with negative/positive slopes represent particles moving anterograde/retrograde. Vertical lines represent stationary vesicles. Percentage of anterograde (Antero), retrograde (Retro) and stationary (Stat) particles were calculated, along with the time spent by the vesicles moving/pausing and their velocity/run length. For Percentage of Frequency and Time (%), values are mean±SEM of the number of neurons analyzed (n). Statistical analysis: One-way ANOVA, Tukey’s post-hoc test. For Velocity/Run Length, only moving vesicles were analyzed. Black line: median of the number of vesicles analyzed (n_v). Statistical analysis: Two-tailed Mann-Whitney test. (D) Single frame from Movie S1C (top, TIRF Microscope, 60x Oil, 2 fps) and kymograph (bottom) of a neuron co-expressing GFP-GGA3 and BACE1-mCherry. (E) Analysis of the frequency distribution of particle velocities, n=13, n_v=195. Only moving vesicles were analyzed. Statistical analysis: Chi-square test.

We also investigated the dynamics of GGA3 and BACE1 axonal trafficking. Live cell imaging experiments (Movie S1A–B, Fig. 2B–C) showed that both proteins are capable of bidirectional movements. Whereas GGA3 is a primarily retrograde-directed protein (26% anterograde vs. 37% retrograde, p=0.0359) that displays slow (median velocity: 0.22 μm/s) and stop-and-start motility (Fig. 2B), BACE1 vesicles move quickly (median velocity: 0.53 μm/s) and robustly in both directions (Fig. 2C). The preference of GGA3 for retrograde events was further supported by the fact that motile GGA3-positive vesicles spend a greater percentage of time moving retrograde (17%) than anterograde (9%, p=0.0496) with the residual 74% pausing or remaining stationary. BACE1-positive vesicles, however, show elevated motility (motile 67% of the time) and no bias for anterograde or retrograde events. In addition, GGA3 exhibited a medium axonal density (27 vesicles/100 μm), whereas BACE1 exhibited a higher axonal density (62 vesicles/100 m). Velocity and run length frequency distribution for both BACE1 and GGA3 vesicles revealed faster (p=0.0048 and p<0.0001, respectively) retrograde events capable of covering a lengthier distance (p<0.0001 and p=0.0030, respectively) while moving toward the soma (Fig. 2B–C).

We next analyzed the axonal trafficking of vesicles co-expressing GFP-GGA3 and BACE1-mCherry (Movie S1C, Fig. 2D). 37% of BACE1-positive vesicles were co-transported with GGA3 along the axon (Fig. 2D). Kymograph analysis revealed that this subset of vesicles behaved similarly to the vesicles expressing solely GGA3 (Fig. 2E, Fig. S1): pronounced bidirectional motility accompanied by a low median velocity (0.26 μm/s). These data indicate that not only GGA3 and BACE1 colocalize, but also that a subset of the vesicles containing both proteins is co-transported along the axon.

Gga3 deletion induces BACE1 axonal accumulation

We reported that GGA3 deletion/depletion results in BACE1 elevation (19, 22, 23). In order to investigate where this elevation occurs in neurons, BACE1-mCherry was overexpressed in Gga3^+/+ and Gga3^−/− hippocampal neurons (Fig. 3A) and its intensity fluorescence was measured, revealing an increase of BACE1 in both dendrites (Gga3^+/+ 46.46±2.51, Gga3^−/− 59.13±3.62, n=19) and axons (Gga3^+/+ 22.45±1.22, Gga3^−/− 49.68±3.47, n=19) in Gga3^−/− neurons. The higher accumulation of BACE1 in axons compared with dendrites produced a shift of the BACE1 PI from 2.04 in Gga3^+/+ neurons to 1.22 in Gga3^−/− (p<0.0001) neurons.

Figure 3. Gga3 deletion induces BACE1 axonal accumulation.

(A) Representative images of BACE1-mCherry in Gga3^+/+ and Gga3^−/− neurons. Arrowheads indicate axons. PI analysis n=19. Stitched confocal z-stacks. Magnification: 63x Oil. Scale bar: 50 μm. (B) Gga3^+/+ and Gga3^−/− neurons expressing Syn-GFP (green - grayscale inverted left). MAP2B (blue). Arrowheads indicate axons. Stitched confocal z-stacks. Magnification: 40x Oil. Scale bar: 50 μm. PI summary table. Representative images are in Fig. S2. (C) BACE1 (green, D10E5), DAPI staining (blue). Mean Gray Intensity Fluorescence analysis: Gga3^+/+ n=5, Gga3^−/− n=6 mice. At least two coronal sections/animal (4 hippocampi) were analyzed. Stitched epifluorescent 20x z-stacks. Scale bar: 250 μm. Statistical analysis: Two-tailed unpaired t test.

To assess whether Gga3 deletion was exclusively affecting BACE1 distribution additional proteins were investigated. The PIs for synaptophysin, a protein localized with BACE1 at presynaptic terminals (13, 27) and the axonal cell adhesion molecule NgCAM, the chicken homolog of L1, a BACE1 substrate (29, 30), were not changed in Gga3^−/− neurons (Fig. 3B, Fig. S2B).

To rule out potential compensatory mechanisms, the polarization of exogenous GGA3, its homolog GGA1, and the somatodendritic adaptor complex AP1(μ1A) was assessed and no change was observed (Fig. 3B, Fig. S2A, C). Finally, transferrin receptor (TfR), a somatodendritic protein largely excluded from the axon (31), also showed no difference in its distribution following Gga3 deletion (Fig. 3B, Fig. S2D).

Our previous studies in non-neuronal cells have shown that GGA3 depletion induces BACE1 accumulation in early endosomes by preventing its trafficking to the lysosomes (21, 22). Thus, BACE1 subcellular localization was assessed in soma and dendrites. LAMP2 was used as a marker for late endosomes/lysosomes, whereas EEA1 was used to identify early endosomes (Fig. S3). In agreement with our previous findings, we observed a reduction in the rate of colocalization (p<0.0001) between BACE1 and LAMP2-positive structures in Gga3^−/− neurons (Fig. S3), together with an accumulation of BACE1 in early endosomal compartment (p<0.0001, Fig. S3).

Previous studies (13, 32) have established that BACE1 amount is highest within the presynaptic terminals of the CA3 hippocampal mossy fibers. To determine whether BACE1 accumulates in axons in vivo, we performed BACE1 immunofluorescence experiments in the hippocampus of 4-month-old Gga3^+/+ and Gga3^−/− mice (Fig. 3C) and found that BACE1 accumulates (+44%) at the mossy fibers in Gga3^−/− (50.56±2.14) mice compared with wild-type littermates (28.24±1.56, p<0.0001).

Lack of BACE1 axonal motility in Gga3^−/− neurons

Given that roughly 40% of BACE1-positive vesicles are co-transported with GGA3 in the axon, the impact of Gga3 deletion on BACE1 axonal motility was investigated. Live cell imaging (Movie S2A) of BACE1-mCherry positive vesicles in Gga3^−/− neurons showed a striking reduction of BACE1 rapid motility, a significant decrease of both BACE1 anterograde (−66%, p<0.0001) and retrograde (−60%, p<0.0001) vesicles (Fig. 4A), together with a decreased median velocity of motile vesicles (from 0.53 in Gga3^+/+ to 0.30 μm/s in Gga3^−/− neurons). Axonal trafficking of synaptophysin was not affected by Gga3 deletion (Movie S2B) (Fig. 4B). However, the co-expression of BACE1 and synaptophysin in Gga3^−/−, but not in wild-type neurons, triggered an alteration of synaptophysin trafficking (stationary vesicles from 36% in Gga3^+/+ neurons to 69% in Gga3^−/− neurons, p=0.0002) (Movie S2C) that resembled the motionless phenotype observed for BACE1 vesicles in Gga3^−/− axons (Fig. 4C). These data strongly support a key role of GGA3 in regulating BACE1 axonal transport, but also highlight the possibility that BACE1 axonal accumulation triggers a detrimental cascade leading to alteration of synaptophysin trafficking.

Figure 4. Lack of BACE1 axonal motility in Gga3^−/− neurons.

(A) Single frames from Movie S3A (top, TIRF Microscope, 60x Oil, 2 fps) and corresponding kymographs (bottom) of BACE1-mCherry axonal particles in Gga3^+/+ (n=15), Gga3^−/− (n=17) neurons. Syn-GFP motility (B) TIRF Microscope, 60x Oil, 2 fps): Gga3^+/+ and Gga3^−/− (n=13) neurons (Movie S3B). (C) Co-expression of BACE1-mCherry (red) and Syn-GFP (green, TIRF Microscope, 60x Oil, 2 fps) in Gga3^−/− (n=13), Gga3^+/+ (n=12) neurons (Movie S3C). Statistical analysis: Two-way ANOVA, Bonferroni’s Multiple Comparison test.

BACE1 accumulates in axonal swellings in Gga3^−/− neurons

BACE1-mCherry overexpression in Gga3^−/− cultures induced accumulation of BACE1 stationary vesicles in enlarged axonal regions resembling dystrophic neurites (Fig. 5A). Almost 90% of the Gga3^−/− neurons analyzed exhibited BACE1-positive axonal swellings (enlargements > 2.5 μm) with a linear density significantly higher than in wild-type axons (18.83±2.26 vs. 0.86±0.48, p<0.0001). A closer observation of the dystrophies showed the presence of trapped vesicles capable of repetitive bouncy movements (Movie S4A, Fig. 5A). Previous work demonstrated that lysosomes accumulate in axonal dystrophies in AD brains (13, 16, 17). Accordingly, we identified LAMP2 positive structures in Gga3^−/− axonal swellings (Fig. S4, Movie S3). However, we also observed that in these enlargements BACE1 and LAMP2 were spatially separated and almost no overlap between the two signals was identified (Manders’ Coefficient: 0.34±0.06, n=5) (Fig. S4, Movie S3). We have already demonstrated that GGA3 plays a fundamental role in trafficking BACE1 to the lysosome where is normally degraded. In line with this, our data prove that BACE1 does not co-localize with LAMP2 positive compartment in Gga3^−/− neurons, suggesting that the axonal accumulation of BACE1 is due to a defect in lysosomal trafficking, which requires BACE1 transport to the soma where mature lysosomes reside (33, 34).

Figure 5. BACE1 accumulates in axonal swellings in Gga3^−/− neurons.

(A) Single frame images from Movie S4A showing Gga3^−/− BACE1-positive axonal swellings (arrowheads). Neurons: Gga3^−/− n=30, Gga3^+/+ n=35. Statistical analysis: Two-tailed unpaired t-test; two-way ANOVA, Bonferroni’s Multiple Comparison test. (B) Syn-GFP expression in Gga3^+/+ (n=25), Gga3^−/− (n=21) axons (Movie S4B). BACE1 and synaptophysin co-expression in Gga3^−/− (n=20), Gga3^+/+ (n=21) neurons. Arrowheads indicate spheroids. Statistical analysis: One-way ANOVA, Tukey’s post-hoc test; two-way ANOVA, Bonferroni’s Multiple Comparison test. (A-B) Pseudo-colored, TIRF Microscope, 60x Oil. Scale bar: 10 μm. Insets are consecutive frames from corresponding movie. Axonal swelling linear density and frequency of neurons with/without swellings were calculated. (C) Representative images of cerebella from Gga3^+/+ and Gga3^−/− mice stained for calbindin. Purkinje cell (PcL), Molecular Layer (ML), Internal Granular Layer (IGL). Purkinje cells axonal spheroids in Gga3^−/− neurons (insets, arrowheads). Confocal z-stacks. Magnification: 40x Oil. Scale bar: 100 μm. Gga3^−/− n=6, Gga3^+/+ n=7 mice. Three sections/mouse were analyzed. (D) Image adapted from Allen Mouse Brain Connectivity Atlas: GFP-positive afferents from the medial entorhinal cortex to CA1 stratum lacunosum-moleculare (Slm). Axonal spheroids (arrowheads) in Gga3^−/− mice (Bielschowsky’s stain). Brightfield images. Magnification: 20x. Scale bar: 100 μm. n=6. At least two sections/mouse were analyzed (4 hippocampi). (C-D) Statistical analysis: Two-tailed unpaired t-test.

No sign of axonal dystrophy was observed after synaptophysin overexpression (Fig. 5B) in any condition unless BACE1 was co-expressed in Gga3^−/− neurons. The axonal enlargements detected after co-expression of the two proteins had a mean linear density of 14±2.48 and were present in 80% of the neurons analyzed. Swellings were populated by both synaptophysin- and BACE1-positive vesicles (Movie S4B).

Next, we investigated whether axonal pathology was present in the brains of Gga3^−/− mice. We first analyzed the cerebellum, a paradigmatic model to study axonal abnormalities (Fig. 5C). Purkinje cell bodies form a single cell layer (PcL) and extend an elaborate dendritic arborization into the molecular layer (ML), whereas the axon protrudes from the opposite side of the soma, descending through the granular layer (IGL) into one of the cerebellar/vestibular nuclei. Adult Purkinje neurons are unable to regenerate their axons and can survive axotomy for long periods (35). A recent study demonstrated that BACE1 is expressed in Purkinje cells (36).

After staining for calbindin, a specific marker for Purkinje neurons, the average caliber of Purkinje cell axons was measured (~1.4 μm), and enlargements greater than 2.5 μm were classified as axonal swellings (Fig. 5C). 4-month-old Gga3^−/− mice exhibited nearly a three-fold increase in the mean number of swellings per section, as compared to wild-type littermates (6.69±0.57 vs. 2.19±0.34, p<0.0001). We also assessed axonal morphology in the hippocampus by Bielschowsky’s stain (Fig. 5D) and found that axons populating the CA1 area – most likely afferents from the entorhinal cortex to the stratum lacunosum-moleculare (Allen Mouse Brain Connectivity Atlas, 2011 – http://connectivity.brain-map.org/) exhibit alterations, fragmentation and spheroids in 4-month-old Gga3^−/− animals (Fig. 5D). Axonal density was reduced by ~25% in Gga3^−/− mice compared with wild-type littermates (35.71±1.66 vs. 45.8±0.65, p=0.0002). These CA1 findings were confirmed by neurofilament staining in the same Gga3^−/− mice (Fig. S5A).

We have already demonstrated BACE1 elevation in the hippocampus and cortex of Gga3^−/− mice (23) and western blot analysis confirmed increased BACE1 also in Gga3^−/− cerebella (Fig. S5B). Furthermore, immunohistochemistry showed a darker chromogenic signal for BACE1 in the ML, PcL, and at the glomeruli in the IGL of the cerebellum (Fig. S5C) and in the hippocampi, including the CA1 region, of Gga3^−/− animals (Fig. S5C). In agreement with our in vitro findings, these results suggest that Gga3 deletion and consequent BACE1 elevation are sufficient to induce the formation of axonal swellings in Gga3^−/− mice.

β- and γ-secretase inhibition prevents axonal dystrophies in Gga3^−/− neurons

To dissect the mechanism underlying the formation of BACE1-positive axonal swellings in Gga3^−/− neurons, we used well-characterized molecules to inhibit β- (C3) or γ-secretase (DAPT) activity in Gga3^−/− primary neurons overexpressing BACE1 (Fig. 6A–B). Inhibition of either β- or γ-secretase prevented the formation of BACE1-positive spheroids, as measured both by linear density (Gga3^−/−: 16.67±2.00, C3: 1.42±0.62, DAPT: 0.37±0.26) and the proportion of neurons bearing axonal swellings (Gga3^−/−: 95%, C3: 20%, DAPT: 5%) (Fig. 6A). Moreover, the protective effect achieved following inhibition was comparable to the one obtained by exogenous expression of GGA3 in Gga3^−/− neurons (linear density: 1.53±0.61, neurons with swellings: 23%) (Fig. 6A, Fig. S6).

Figure 6. β- and γ-secretase inhibition prevents axonal dystrophies in Gga3^−/− neurons.

(A) Representative axonal pseudo-colored images from live cell imaging movies. Analyzed conditions: Gga3^+/+ (n=29), Gga3^−/− (n=24), Gga3^−/−_GFP-GGA3 (n=27), Gga3^−/−_C3 (n=26), Gga3^−/−_DAPT (n=32). TIRF Microscope, 60x Oil. Scale bar: 10 μm. Statistical analysis: One-way ANOVA, Tukey’s post-hoc test; two-way ANOVA, Bonferroni’s Multiple Comparison test. (B) Single frames from Movie S5 (top, TIRF Microscope, 60x Oil, 2 fps) and corresponding kymographs (bottom) of BACE1-mCherry particles. Analyzed conditions: Gga3^+/+ (n=19), Gga3^−/− (n=19), Gga3^−/−_GFP-GGA3 (n=11), Gga3^−/−_C3 (n=20), Gga3^−/−_DAPT (n=23). Table reports statistical analysis/significance: Two-way ANOVA, Bonferroni’s Multiple Comparison test. Analysis of dystrophic phenotype after pharmacological inhibition of β-secretase in the cerebellum (C) and hippocampal CA1 (D). Number of mice analyzed is indicated in the graphs. Representative images in Fig. S6. Statistical analysis: One-way ANOVA, Tukey’s post-hoc test.

We next analyzed BACE1 axonal motility (Fig. 6B). We observed that GGA3 reintroduction in Gga3^−/− neurons restored BACE1 trafficking by specifically rescuing BACE1 retrograde movements when compared to Gga3 null condition. No rescue was observed for anterograde vesicles. This observation is in agreement with our findings indicating that GGA3 is a primarily retrograde-directed protein. We also found that Gga3^−/− neurons treated with C3 exhibited significantly fewer stationary BACE1-positive vesicles compared with those treated with DMSO alone (Gga3^−/−: 66% vs. C3: 49%, p=0.0184) (Movie S5). DAPT-mediated γ-secretase inhibition had a greater effect in preserving BACE1 motility compared with BACE1 inhibition (stationary vesicles: DAPT: 32%, Fig. 6B). We compared the degree of inhibition in Aβ X-40 (hereinafter referred to as Aβ40) production between the two compounds and found that DAPT treatment induced a greater reduction of Aβ40 compared with C3 treatment (C3: −82%, DAPT: −93%) (Fig. S7A). This effect might, in part, explain the higher efficiency of DAPT treatment in preserving BACE1 axonal trafficking and preventing its accumulation in axonal swellings.

Inhibiting the processing of other γ-secretase substrates could also explain the neurotrophic properties of DAPT treatment. Recent studies have demonstrated that γ-secretase-mediated activation or inhibition of the Notch pathway plays a major role in microtubule stabilization (37). In fact, whereas activation of the Notch pathway causes a reduction in neurite branching and loss of varicosity, its inhibition reverts such morphological effects, leading to an increase in cytoskeleton plasticity with intense neurite remodeling. This evidence might explain the potentiated effect observed on the swelling and trafficking phenotypes after DAPT treatment.

We then treated 2-month old Gga3^−/− mice with the potent and selective BACE inhibitor, MBi-3 (38) in-diet, and analyzed CA1 and cerebellar axonal morphology after an 8-week treatment period. Animals were treated with either vehicle control diet or three different doses (3, 10 and 30 mg/kg/day) to test a range of BACE inhibition (16% to 78%) using Aβ40 as a pharmacodynamic biomarker (Fig. S7B–I). Body weight and food consumption (Fig. S7C–D) were monitored twice per week and were used to calculate the actual delivered dose of MBi-3 (Fig. S7B). All animals consumed a similar amount of food for the duration of the study (Fig. S7D).

Although all the three doses were able to prevent Purkinje cell swelling formation in the cerebellum of Gga3^−/− mice (Fig. 6C, Fig. S8A), in the hippocampus only the 30 mg/kg/day treatment, that produced >75% reduction of Aβ40, preserved the typical axonal morphology of the CA1 area (Fig. 6D, Fig. S8B).

MBi-3 plasma, cerebellum and cortex concentrations are listed in Table S1. The differences in activity of MBi-3 treatment between cerebellum and hippocampus could not be explained by differences in compound exposure. Correlation analysis between MBi-3 concentration and phenotypes analyzed showed a strong dose-dependent correlation between the treatment and the axonal density in the CA1 area of the hippocampus (Fig. S7E–F), where only the higher dose prevented axonal dystrophies, but not in the cerebellum (Fig. S7G–H), where even the lowest dose of MBi-3 was able to avoid axonal swelling formation. Cortical concentration of Aβ40 in MBi-3-treated mice were reduced by ~16%, ~66% and ~78% (3, 10, 30 mg/kg/day respectively, Fig. S7I) when compare with vehicle diet-treated animals. Using non-linear regression analysis, an oral ED₅₀ of 5.1 mg/kg and EC_{50 Cortex} of 74 nM were calculated and are both in line with previous studies in mice (38, 39) (Fig. S7J).

In vivo and in vitro pharmacological inhibition data strongly support the hypothesis that accumulated BACE1 in Gga3^−/− axons retains its proteolytic activity, and that such accumulation of active BACE1 leads to axonal swelling formation and disruption of axonal trafficking.

An AD-linked GGA3 rare insertion results in loss of function

To determine the impact of GGA3 genetic variants in AD, we analyzed deep (>40x) whole genome sequencing (WGS) data from the 1393 subjects (446 families) in the National Institute of Mental Health (NIMH) AD Genetics Initiative with affected and unaffected siblings as described in Blacker et al. (40). Additionally, we evaluated three other AD datasets: two WGS and one whole exome sequencing (WES) (Table S2). We sought to find functionally relevant single nucleotide variants (SNVs) or small insertions/deletions (INDELs) showing some evidence of association with AD. We focused on potentially functional variants or INDELs with low, medium and high impact as annotated in SNPEff (41).

Several variants showed nominal significance for protection in NIMH but showed an opposite signal direction in other datasets (Table S3). There was also one synonymous variant (rs34008167, Val586Val), which showed consistent signal direction among three datasets and had a combined meta-analysis p=0.004. In order to find additional sources of replication we downloaded the summary statistics from Jansen et al. (42) that is the largest AD GWAS to date. Rs117805695 and rs146877619 were nominally significant, but the direction was opposite to the one in NIMH. Rs34008167 with a p=0.474 didn’t achieve nominal significance. We next assessed small insertions and deletions in NIMH. We found that rs150787028, a G/GGGT in-frame indel resulting in the insertion of Thr at position 545 of the GGA3 long isoform (NP_619525.1, Ins545T), was present in six out of eight affected individuals from four families with LOAD. All of these families are of African American ethnicity belonging to the African population (AFR). The minor allele frequency (MAF) of rs150787028 is 0.007 in ExAC, rendering it a rare variant (minor allele frequency ≤ .01) in aggregate populations. However, the minor allele frequency in the AFR population is listed as 0.05 (1000 genomes project (43)). The NIMH data set includes 23 AFR families with a total of 46 affected and 8 unaffected individuals. Four out of 23 AFR families carry this SNP. Given that the minor allele frequency of rs150787028 in the AFR population is 0.05, the number of proband carriers expected would be approximately two out of 23. However, we observed 4 proband carriers out of 23 probands, resulting in a minor allele frequency of 0.09 in the affected subjects. Thus, this rare variant appears to strongly concentrate and co-segregate with LOAD in AFR patients. Unfortunately, the number of informative families was not sufficient to carry out a formal family-based association test on these results. The only other dataset, which had INDELs available, was the ADNI dataset. There were three carriers of rs150787028, all of which were African American and two had mild cognitive impairment. The combined information from two datasets makes Ins545T a plausible candidate for functional studies.

Site-directed mutagenesis was performed to introduce the Ins545T variant in a GFP-GGA3 cDNA (NM_138619.3) (Fig. 7A). We first overexpressed GFP-Ins545T in wild-type hippocampal neurons to assess its functionality in basal conditions. Mutated GGA3 showed a typical distribution across neuronal compartments similar to the wild-type protein (PI: 1.98±0.13, n=18) (Fig. 7B). Next, we investigated the ability of the variant to rescue BACE1 accumulation in axonal swellings. As expected, wild-type GGA3 reintroduction was able to significantly reduce the number of BACE1-positive swellings (Gga3^−/−: 12.22±1.57 vs. GFP-GGA3 rescue: 1.22±0.51, p<0.0001). However, the mutant protein was not as effective in rescuing the swelling phenotype (Gga3^−/−: 12.22±1.57 vs. GFP-Ins545T rescue: 8.50±0.99, p=0.0597). Likewise, whereas wild-type GGA3 dramatically reduced the number of neurons with swellings (Gga3^−/−: 91% vs. GFP-GGA3 rescue: 18%, p<0.0001), no significant rescue was observed after GFP-Ins545T expression (Fig. 7C, Fig. S9). Finally, in contrast to wild-type GGA3, reintroduction of GFP-Ins545T did not result in a substantial improvement in BACE1 motility (stationary vesicles: Gga3^−/−: 66% vs. GFP-Ins545T rescue: 51%) (Fig. 7D, Movie S6).

Figure 7. An AD-linked GGA3 rare insertion results in loss of function.

(A) Schematic of human GGA3 domains. Hinge domain sequence alignment of human (NM_138619-3) wild-type and mutated (Ins545T) GGA3. (B) Neuron expressing GFP-Ins545T (green - grayscale inverted left) and stained for MAP2B (red). Arrowheads point to the axon. Confocal z-stacks. Magnification: 40x Oil. Scale bar: 100 μm. (C) Representative axonal pseudo-colored images from live cell imaging movies. BACE1 axonal swelling phenotype was analyzed in different conditions: Gga3^−/− (n=24), Gga3^−/−_GFP-GGA3 (n=28), Gga3^−/−_GFP-Ins545T (n=31), Gga3^−/−_GFP-Ins545T_C3 (n=28). TIRF Microscope, 60x Oil. Statistical analysis: One-way ANOVA, Tukey’s post-hoc test; two-way ANOVA, Bonferroni’s Multiple Comparison test. (D) Single frames from Movie S6 (top, TIRF Microscope, 60x Oil, 2 fps) and corresponding kymographs (bottom) of BACE1-mCherry particles. Analyzed conditions: Gga3^−/− (n=19), Gga3^−/−_GFP-GGA3 (n=11), Gga3^−/−_GFP-Ins545T (n=19), Gga3^−/−_GFP-Ins545T_C3 (n=13). Table reports statistical analysis/significance: Two-way ANOVA, Bonferroni’s Multiple Comparison test.

We next asked whether β-secretase inhibition improved the swellings and trafficking phenotype following Ins545T expression. Interestingly, C3 treatment ameliorated BACE1 axonal trafficking and accumulation in dystrophic axons of Gga3^−/− neurons overexpressing GFP-Ins545T (Fig. 7C–D, Fig. S9). Collectively, these data demonstrate that Ins545T was unable to rescue the axonal pathology associated with BACE1 elevation in Gga3 null neurons, suggesting that this variant results in a loss of GGA3 function.

Gga3 deletion exacerbates axonal pathology in 5XFAD brains and induces endogenous BACE1 accumulation in axonal swellings

Aβ accumulates intracellularly in AD brains, possibly before extracellular amyloid deposition (44). Early axonal defects have been observed in mouse AD models and human AD and occurred long before detectable amyloid deposition (4). Based on these evidences, we decided to investigate the effect of Gga3 deletion on early axonal pathology in 5XFAD mice, a mouse model of familial AD. We analyzed the number of swellings in 5XFAD mice crossed with Gga3^−/− mice (5X/Gga3^−/−) (23) following Bielschowsky’s stain (Fig. 8A) and found that Gga3 deletion in 5XFAD mice exacerbates the swelling pathology in the cortex of 2-month-old mice when compared to control littermates (1.68±0.31 vs. 3.75±0.20, p<0.0001). The swellings analyzed presented a large caliber with some of them lacking staining in their center (“ghost” swellings) as previously reported by others (4). It is well established (45) that at 2 months of age 5XFAD mice display almost no extracellular Aβ in the cortex, specifically in the outer layers of the primary somatosensory cortex, the area analyzed for this study. Thus, the swellings and dystrophies that we identified are independent of extracellular Aβ deposition. Furthermore, we performed IF staining for BACE1 using the 3D5 BACE1 antibody (validated in the brain of BACE1^−/− mice, (12)) and found that in 5X/Gga3^−/− mice endogenous BACE1 is present in dystrophic neurites/axonal spheroids, identified by NFH staining (Fig. 8B). These data support the role of GGA3 loss of function during the initial stage of AD pathogenesis, owing to BACE1 axonal trafficking perturbation and consequent generation of intracellular amyloid beta toxic species.

Figure 8. Gga3 deletion exacerbates axonal pathology in 5XFAD brains and induces endogenous BACE1 accumulation in axonal swellings.

(A) Representative images of the somatosensory cortex from 5X/Gga3^+/+ and 5X/Gga3^−/− mice (Bielschowsky’s stain). White boxes highlight axonal swellings. Brightfield images. Magnification: 40x. Scale bar: 10 μm. 5X/Gga3^+/+ n=7, 5X/Gga3^−/− mice: n=9. At least two sections/mouse were analyzed. Statistical analysis: Two-tailed unpaired t-test. (B) IF staining for NFH (purple) and BACE1 3D5 (green) in 5X/Gga3^−/− mice. DAPI in blue. Confocal z-stacks. Magnification: 63x Oil. Scale bar: 10 μm.

Discussion

This study proposes a mechanism underlying the axonal damage observed in the early stage of AD pathology (4, 8, 9). Our previous and current studies demonstrate that GGA3 is a key regulator of BACE1 trafficking. Our new data elucidate a specific role for GGA3 in coordinating BACE1 axonal trafficking. Our findings indicate that GGA3 loss of function, due to genetic deletion or to an AD-linked GGA3 variant, induces an impairment of BACE1 retrograde trafficking, most likely by affecting the subset of vesicles that cotransport both proteins and are mainly retrograde-directed. As a consequence, BACE1 cannot be trafficked back to the soma, where it is normally degraded in mature lysosomes, and starts accumulating in the axon. BACE1 dysfunctional anterograde transport is most likely secondary to increased axonal production of Aβ, which in turns causes axonal swellings both in vitro and in vivo. Accordingly, β- or γ-secretase pharmacological inhibition rescues BACE1 defective trafficking and axonal buildup in vitro. Moreover, in vivo BACE inhibition prevents hippocampal and cerebellar axonopathy in Gga3^−/− mice. Our data support previous work showing that axonal pathology is associated with intraneuronal Aβ accumulation and represents an early event in AD (46, 47). Several studies have reported the presence of intraneuronal Aβ in dystrophic neurites in AD brains (48) and in AD mouse models (49). Similarly, various studies have shown that BACE1 accumulates in dystrophic neurites around senile plaques (12–15, 50). Our findings are in line with the previously proposed idea that axonal swellings could form because of impaired axonal transport and promote aberrant Aβ generation (4). In fact, if aberrant Aβ generation occurs locally at sites of blockage (for instance BACE1 accumulation in Gga3^−/− swellings), then amyloid deposition may occur as a result of focally increased secretion of Aβ toxic species or lysis of Aβ-enriched axonal swellings (4). Although BACE1 processes a diverse array of substrates, many of which appear to play a critical role in axonal guidance and myelination (30, 32, 51–54), so far none of these substrates has been directly associated with axonopathy.

In genetic analyses, we found a GGA3 in-del, Ins545T, that co-segregated with LOAD in African American families in the NIMH AD data set and was also present three African American individuals in the ADNI dataset. This is a rare variant (MAF = 0.007 in ExAC). As such, it is highly unlikely to yield a significant p-value for association owing to low informativeness. Late-onset AD is a polygenic disease and is caused by a combined action of multiple mutations. Most associated loci have a relatively small genetic effect size (with APOE being an exception) and contain multiple variants, which are in linkage disequilibrium with each other (correlated). Among these, exonic variants, particularly those that affect the corresponding amino acid sequence of the encoded protein, are the most likely candidates to confer a direct effect on biological function. Those variants are usually rare, for instance rs150787028. We used a large collection of AD WGS datasets to investigate functional variants associated with AD. The identified rare in-del variant had a higher MAF in African American families as compared to reference population. This variant results in a clear functional effect on GGA3 function, as demonstrated by the inability of GGA3 Ins545T to rescue BACE1 accumulation in axonal swellings and its axonal trafficking in Gga3^−/− neurons. We also note that we have reported other rare functional variants to be associated with AD, which also could contribute to AD heritability (55).

This study also demonstrates that the lack of GGA3 induces an increased number of axonal swellings not associated with Aβ plaques in the cortex of 5XFAD mice where endogenous BACE1 accumulates during the pre-plaque phase of the pathology. Given that our experiments were carried out in mice that do not develop Aβ plaques or in young familial AD mice, the observed effects on BACE1 are independent of extracellular Aβ deposition and plaque-associated BACE1. However, accumulation of BACE1 in the axon would be predictive of enhanced amyloidogenic cleavage of APP and increased Aβ accumulation, further disrupting axonal trafficking. Accordingly, we showed that genetic deletion of Gga3 increases Aβ concentration in 5XFAD mice (23). Collectively, these data indicate that axonal accumulation of BACE1 owing to GGA3 loss of function causes disruption of BACE1 axonal trafficking and axonopathy, which eventually could lead to, neuronal dysfunction, neuronal loss, and Aβ deposition, observed in AD. Thus, loss of function of GGA3 may be an initiating event in the etiology of AD. Our study also provides a possible explanation for the prevalence of axonopathy during AD early stages (4, 8, 9), in which axonal BACE1 elevation is not triggered by senile plaques formation but is an upstream event that occurs during the pre-plaque phase of the pathology. Finally, our study suggests that BACE inhibition can prevent axonal damage in the absence of extracellular Aβ deposition.

Our study has some limitations and some questions remain to be addressed. We demonstrated that pharmacological inhibition of BACE prevented axonopathy in Gga3^−/− mice. It remains to be determined the extent to which BACE inhibition prevents axonal damage in mouse models of AD prior to Aβ deposition. Clinical trials based on BACE inhibitors verubecestat and atabecestat were discontinued (56–58) due to a modest worsening of cognitive function in AD subjects. More recently, Biogen and Eisai discontinued two phase-III trials (with BACE inhibitor elenbecestat) due to unfavorable risk-benefit ratio. Most likely these adverse effects are due to impaired processing of BACE1 substrates other than APP when BACE activity is inhibited by 60–90%. Moreover, in these studies Aβ deposition was already present in AD patients. It remains to be determined if lower doses of BACE inhibitors will be able to avoid mechanism-based side effects and prevent axonal pathology when administered to human subjects prior to extracellular Aβ deposition. In this view, the identification of plasma NfL dynamics as a biomarker of axonal damage during the presymptomatic stages of AD (8, 9, 59) could help identifying the exact time window when BACE inhibitors can be effective in preventing AD progression.

Materials and Methods

Study design

This study aimed to demonstrate that GGA3 loss of function triggers BACE1 accumulation in dystrophic neurites independently of extracellular Aβ deposition. BACE1 axonal trafficking disruption and its axonal buildup was demonstrated by live cell imaging and confocal microscopy in Gga3^+/+ and Gga3^−/− neuronal cultures. In vivo the induction/exacerbation of axonal pathology by genetic deletion of Gga3 was assessed in Gga3^+/+, Gga3^−/−, 5X/Gga3^+/+ and 5X/Gga3^−/− mice by confocal and brightfield microscopy. Pharmacological inhibition of BACE was performed both in vitro and in vivo to demonstrate that axonal pathology is caused by increased BACE1 activity. Genetic association analysis was performed in the WGS dataset from the AD NIMH dataset and three other AD datasets (two WGS and one WES) to identify GGA3 variants associated with LOAD. Live cell imaging and confocal microscopy experiments in Gga3^−/− neurons demonstrated that GGA3 in-del, Ins545T is unable to rescue BACE1 axonal trafficking and axonal accumulation and thus it is a loss of function mutation. No sample-size calculation was performed. For BACE pharmacological inhibition in mice sample-size is in line with Merck guidelines and previous experience. Mice of both sexes were assigned randomly to the experimental groups. For in vivo studies at least two to three separate groups of mice from different litters were analyzed in independent experiments. For in vitro studies two to four separate sets of cultures from different litters were analyzed in independent experiments. In some cases, selected samples were excluded from specific analyses because of technical flaws during sample processing or data acquisition. Standardized experimental/analysis protocols have been used across the entire study. To verify the solidity of the data and analysis reproducibility datasets have been blindly analyzed by multiple investigators with the exception of Fig. 1A–D, Fig. 2A–B–D, Fig. 7B, Fig. S4. All experimental findings were reliably reproducible. Trafficking experiments from Figures 6 and 7 were performed simultaneously and share the same set of controls. Trafficking experiments for BACE1-mCherry in wild-type neurons were run together with trafficking experiments for GFP-GGA3 in wild-type neurons and BACE1-mCherry in Gga3^−/− neurons. For this reason, BACE1-mCherry trafficking data in wild-type neurons are the same in Figure 2 and 4.

Statistical analysis

Data are expressed as the mean±standard error of the mean (SEM), represented as error bars. The number of biological/technical replicates is reported in the figure legend. Curve fitting, non-linear regression and statistical analyses were performed using GraphPad Prism. Statistical tests used are reported in figure legends. ROUT method was applied to identify outliers. A p value of 0.05 was used as the significance threshold throughout this study. In all figures, p values are illustrated as followed for all tests used: ^n.s. p > 0.05, * p < 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001.

Supplementary Material

Movie 1

Movie S1. BACE1-mCherry and GFP-GGA3 axonal trafficking in wild-type neurons

Movie 2

Movie S2. BACE1-mCherry and Synaptophysin-GFP axonal trafficking in Gga3^+/+ and Gga3^−/− neurons

Movie 3

Movie S3. 3D reconstruction of BACE1-mCherry and LAMP2 in Gga3^−/− axonal swelling

Movie 4

Movie S4. BACE1-mCherry and Synaptophysin-GFP accumulation in axonal swellings in Gga3^−/− neurons

Movie 5

Movie S5. Representative movies for BACE1-mCherry trafficking experiments in Figure 6

Movie 6

Movie S6. Representative movies for BACE1-mCherry trafficking experiments in Figure 7

Data file S1

Datafile S1. Individual-level Data File n<20 (provided as separate Excel file)

Supplementary Material

Figure S1. Representative kymographs for BACE1-mCherry and GFP-GGA3 cotransport experiments in Figure 2D

Figure S2. Representative images for polarity index analysis in Figure 3B

Figure S3. Colocalization analysis for BACE1-mCherry with EEA1 and LAMP2

Figure S4. Representative image for BACE1-mCherry and LAMP2 in Gga3^−/− axonal swelling

Figure S5. NFH staining of hippocampal CA1 axonal bundles and BACE1 accumulation in cerebellum and hippocampus of Gga3^−/− mice

Figure S6. Images for BACE1-mcherry and GFP-GGA3 channels corresponding to axon in Figure 6A

Figure S7. Supplementary data and analysis related to BACE inhibition experiments

Figure S8. Representative images of cerebella and hippocampi from MBi-3 treated mice

Figure S9. Images for BACE1-mcherry and GFP-GGA3 channels corresponding to axons in Figure 7C

Table S1. [MBi-3] μM - PK analysis summary

Table S2. Genetics dataset description

Table S3. Selected single variants, showing nominal association with AD in one of the datasets.

Table S4. Summary of the antibodies used throughout this study

Data, code and material availability:

The MBi-3 compound provided by MSD/Merck&Co., Inc., and the GFP-GGA3, GFP-GGA1, TfR-GFP and AP1(μ1A)-GFP constructs provided by J.S. Bonifacino, PhD are covered by a materials transfer agreement. Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact (Giuseppina Tesco, MD, PhD - giuseppina.tesco@tufts.edu). All the remaining data are present in the main text or supplementary material.