AP-1/σ1A and AP-1/σ1B adaptor-proteins differentially regulate neuronal early endosome maturation via the Rab5/Vps34-pathway

The σ1 subunit of the AP-1 clathrin-coated-vesicle adaptor-protein complex is expressed as three isoforms. Tissues express σ1A and one of the σ1B and σ1C isoforms. Brain is the tissue with the highest σ1A and σ1B expression. σ1B-deficiency leads to severe mental retardation, accumulation of early endosomes in synapses and fewer synaptic vesicles, whose recycling is slowed down. AP-1/σ1A and AP-1/σ1B regulate maturation of these early endosomes into multivesicular body late endosomes, thereby controlling synaptic vesicle protein transport into a degradative pathway. σ1A binds ArfGAP1, and with higher affinity brain-specific ArfGAP1, which bind Rabex-5. AP-1/σ1A-ArfGAP1-Rabex-5 complex formation leads to more endosomal Rabex-5 and enhanced, Rab5GTP-stimulated Vps34 PI3-kinase activity, which is essential for multivesicular body endosome formation. Formation of AP-1/σ1A-ArfGAP1-Rabex-5 complexes is prevented by σ1B binding of Rabex-5 and the amount of endosomal Rabex-5 is reduced. AP-1 complexes differentially regulate endosome maturation and coordinate protein recycling and degradation, revealing a novel molecular mechanism by which they regulate protein transport besides their established function in clathrin-coated-vesicle formation.


Results
Vps34 is hyperactivated on σ1B −/− endosomes. Previous data show that synapses of σ 1B − /− mice have enlarged endosomes and contain two times more PI-3-P than endosomes from wt mice. Also the late endosome MVB pathway, which depends on the PI 3-kinase Vps34, is upregulated in synapses of σ 1B − /− mice 7,16 . These data indicate, that AP-1/σ 1A and AP-1/σ 1B may take part in the regulation of early endosome maturation into MVB endosomes via the regulation of Vps34-activity. In order to analyse such a molecular mechanism in-vitro, we first tested whether the enlarged endosomes seen in EM images in-vivo can still be found after the 5 h density centrifugation required for their isolation. Vesicles with tubules are readily detected in the early endosome (low-density, ld) density gradient fraction from σ 1B − /− , but not in fractions from wt brains (Fig. 1A). Vps34 is bound to these early and late endosomes (high-density, hd) (Fig. 1B), but endosomes from σ 1B − /− mice do not contain more Vps34 than wt endosomes (Fig.1C), indicating increased Vps34 specific activity. To test whether Vps34 activity is indeed up-regulated, we added the specific inhibitor 3-methyl adenine   20 . Endosomal PI-3-P levels of wt and σ 1B − /− endosomes were compared by determining the amount of EEA1 bound to these endosomes via its PI-3-P-specific FYVE-domain (Fig.1D). EEA1 is not a presynaptic, or apical, protein and will bind to these endosomes only after cell lysis 16,[21][22][23][24] . In the Kratzke et al. 16 publication, we published the data showing that 23.2 ± 4.4 and 16.7 ± 3.1% of the loaded EEA1 is associated with the early endosome (ld) and late endosome (hd) enriched density gradient fractions. Endosomes from wt mice contain only 6.2 ± 2.5 and 7 ± 2.5% of the loaded EEA1. We performed experiments in parallel at the time adding 3-MA (5 mM) to the extracts from σ 1B − /− neurons. Addition of the Vps34 PI 3-kinase inhibitor normalised EEA1 binding and reduced it even below wt levels ( Fig. 1D; see also final result chapter and Kratzke et al.) 16 . Thus the increase in PI-3-P in σ 1B − /− endosomes is indeed due to hyperactive Vps34.
Rab5/Vps34-pathway is activated by σ1A. Vps34 PI 3-kinase activity 18 is stimulated by complex formation with the protein kinase Vps15 25 . The Vps34 C-terminal helical domain blocks the Vps34 active site. Vps15 binding to this helix and the preceding domain stabilises the active-site open conformation of Vps34 20 . Because σ 1A and σ 1B differ in their C-terminal helical domains, a differential Vps34 binding via these domains could cause the differences in Vps34 activity. σ 1A binding could stimulate, or σ 1B binding could inhibit Vps34 activity. Neither the Vps34 C-terminal nor the catalytic domain bound σ 1-adaptins in γ 1/σ 1 hemicomplex Y3H assays, suggesting an indirect regulation by σ 1 adaptins (Fig. S1). Membrane-cytoplasm recycling of Vps34/Vps15 is a mechanism of Vps34-pathway regulation, but not in neurons, where Vps34/Vps15 remains membrane bound 17 . Rab5 GTP binds Vps15, enhancing Vps34 activity by an unknown molecular mechanism 17 , but the amount of Rab5 on σ 1B − /− endosomes was not increased 16 . Rab5 cytoplasma-endosome recycling could be reduced in these synaptic endosomes compared to fibroblastoid endosomes, as is the Vps34/Vps15 recycling. Alternatively, other Rab5 functions and the respective GEFs could be reduced, whereas the Vps34-pathway is activated and thus the amount of endosome bound Rab5 would not be altered. Indeed, σ 1B − /− early endosomes do contain more of the Rab5-GEF Rabex-5, whereas the neighbouring high density, late endosome fraction contains less 16 . Thus σ 1A binding to Rabex-5 could be a mechanism for Rab5-pathway activation. We tested the Rabex-5 C-terminal aa 396-491 for σ 1-binding in γ 1/σ 1 hemicomplex Y3H binding assays, because these mediate Rabex-5 early endosome binding independent of the Rab5/Rabex-5 linker Rabaptin-5α 26,27 . Unexpectedly, Rabex-5 binds σ 1B, not σ 1A ( Fig. 2A). In addition, we tested the Rab5-pathway-inactivating RabGAP5 for σ 1 isoform binding, because its inhibition could contribute to Rab5-pathway activation. RabGAP5 consists of the catalytic TBC domain, a SH3-domain and a C-terminal RUN-domain, present in proteins linked to GTPase functions 28 . Also the RabGAP5 RUN-domain (aa 578-760) binds σ 1B, not σ 1A ( Fig. 2A). Therefore both proteins might use similar motifs and sequence alignment indicated two related motifs in both: P_E_A:E_C:L_L and P_L_Q:K_P:E_Q:G_V. We replaced in Rabex-5 the E and L residues following the P in both putative motifs by A. This abolished σ 1B binding in Y3H assays (Fig. S2). Rabex-5 proteins with one of the two motifs mutated gave highly variable results in in-vitro AP-1 pulldown experiments, indicating that they are unstable proteins in-vitro and that both motifs might act in concert to enable σ 1B binding. However, wt Rabex-5 and RabGAP5 domains pulled down AP-1 out of wt and σ 1B − /− synaptosome extracts, confirming the Y3H protein interaction data. Rabex-5 isolates 160% more AP-1/σ 1A complexes out of σ 1B − /− synaptosome extracts and RabGAP5 30% more, which was not expected due to the absence of σ 1A-Rabex-5 and RabGAP5 binding in the Y3H experiment (Fig. 2B). Because of these contradictory results from the in-vitro experiments, we tested for AP-1 colocalisation with these two proteins and with Rab5 on membranes in-vivo with the proximity-ligation-assay (PLA). We used wt, σ 1B − /− and AP-1-deficient (μ 1A − /− ) MEF cell lines. Their endosomes are larger than synaptic endosomes, thus providing a higher spatial resolution. Indeed, in σ 1B − /− cells more AP-1/Rabex-5 and also more AP-1/ Rab5 colocalised. Moreover, these complexes are confined to smaller areas in σ 1B − /− cells compared to wt cells, indicating maturing membrane subdomains (Fig. 2C,D). Also more RabGAP5 colocalises with AP-1/σ 1A, but these complexes are more dispersed, which should result in reduced RabGAP5 interaction with Rab5, limiting Rab5-pathway inactivation (Fig. 2C,D). Due to the difference in the AP-1/Rabex-5 and AP-1/RabGAP5 complex distributions, their compositions would be expected to be different. We did not investigate the AP-1/RabGAP5 distribution further at this point, because pulldown experiments and the increased Rabex-5 early endosome binding in σ 1B − /− neurons demonstrate that it is the most relevant mechanism. All these data are in line with Rab5 GTP -mediated activation of Vps34 activity and with our previous biochemical analysis, which showed increased Rabex-5 association with early endosomes 16 . There is, however, a discrepancy with the Y3H data, which show σ 1B, but not σ 1A binding of Rabex-5 ( Fig. 2A).
ArfGAP1 links σ1A and Rabex-5. The discrepancy between Y3H and pulldown experiments can be explained by the formation of AP-1/σ 1A-Rabex-5 complexes by a linker protein, like the Rabex-5/Rab5 linker Rabaptin-5α . Rabaptin-5α is also bound by the C-terminal 'ear'-domain of AP-1/γ 1. Altering the protein ratio of  both alters early endosome protein recycling 29 . Thus, we expressed Rabex-5 for pulldown experiments in E. coli and incubated the Rabex-5-loaded beads with brain extracts from wt and σ 1B − /− mice. Again, more AP-1 complexes were isolated from σ 1B − /− brains than from wt brains by Rabex-5 ( Fig. 3A). However, Rabex-5 isolated less Rabaptin-5α out of σ 1B − /− synaptic extracts than from wt extracts (Fig. 3A), excluding a linker function in this pathway. Thus we tested ArfGAP1 for a σ 1A/Rabex-5 linker function for several reasons. Like Rabaptin-5α , it binds the γ 1 'ear'-domain and it binds to endosomes. Also, it exists as a brain-specific isoform of unkown function 30,31 . ArfGAP1 has functions independent of its N-terminal GAP domain activity and the brain-specific isoform has an altered C-terminal domain [32][33][34][35][36] . ArfGAP1 stimulates Arf1 GTP GTP-hydrolysis by a conformational change induced by binding of its C-terminal lipid-sensor domains, ALPS1 and 2, to highly curved membranes 37 . Arf1 GTP recruits AP-1 on membranes and releases it upon GTP-hydrolysis and thus ArfGAP1 regulates AP-1 membrane binding. The brain-specific ArfGAP1 is generated by a deletion/insertion modification altering the ALPS2 motif 31 . Indeed, ubiquitous ArfGAP1 U and brain-specific ArfGAP1 B were enriched in Rabex-5 pulldowns from σ 1B − /− synaptosome extracts compared to wt extracts (Fig. 3A). Therefore we tested for ArfGAP1 binding by σ 1 adaptins. It has been shown that the C-terminal ArfGAP1 domain of the ubiquitous isoform binds the γ 1 'ear'-domain 30 . This interaction did not interfere with our analysis of σ 1 isoform binding, because in the Y3H-assays γ 1 is expressed without its C-terminal 'ear' domains ( Fig. 3B). The ArfGAP1 U C-terminal domain showed comparably weak, whereas ArfGAP1 B showed stronger binding towards σ 1A. As expected from the efficient isolation of AP-1/σ 1A by Rabex-5 in pulldown experiments (Fig. 3A), both ArfGAP1 C-terminal domains showed significantly lower affinity towards σ 1B (Fig. 3B). Thus, we tested, whether C-terminal ArfGAP1 domains, like the Rabex-5 domain, isolate more AP-1/σ 1A out of σ 1B − /− than from wt synaptosomes. Both isolated AP-1, and specifically ArfGAP1 B isolated more AP-1/σ 1A out of σ 1B − /− synaptosomes (Fig. 3C). This differential isolation would not be possible, if the C-terminal ArfGAP1 domains would bind the γ 1 'ear' domain with high affinity, because γ 1 is present in both AP-1 complexes. Collectively, these data indicated the formation of AP-1/ArfGAP1/Rabex-5 complexes. To test whether Rabex-5 binds ArfGAP1 proteins and whether such complexes are able to bind AP-1 complexes, E. coli-expressed Rabex-5 was first tested for isolating E. coli-expressed ArfGAP1 and then those complexes were tested for AP-1 isolation out of brain extracts. Rabex-5 bound both ArfGAP1 proteins (Fig. 4A) and indeed, Rabex-5/ArfGAP1 U and Rabex-5/ArfGAP1 B complexes isolate more AP-1/σ 1A from σ 1B − /− synaptosomes (Fig. 4A,B). Even the difference between the ArfGAP1 isoforms appears to be reproducible in these demanding consecutive pull-down experiments, but this is not statistically significant (χ 2 -test distribution 0.8). Thus tripartite AP-1/ArfGAP1/Rabex-5 complexes can form and therefore we tested whether the ArfGAP1 distribution on membranes depends on AP-1 complexes as does the distribution of Rabex-5/Rab5 (Fig. 2C,D). We expressed GFP-tagged ArfGAP1 B and ArfGAP1 U in wt, σ 1B − /− and in MEF cells lacking both AP-1 complexes (μ 1A − /− , Δ AP-1) 10 and determined their distribution on peripheral endosomes and the peri-nuclear trans-Golgi-network (Figs 4C and S3). AP-1 and ArfGAP1 localise to neighbouring domains with only limited colocalisation in wt cells, best visible on the larger trans-Golgi network (TGN), in line with transient interactions (Fig. S3). There was no qualitative difference in the AP-1-dependence of their distributions on both compartments, but the effect was more pronounced on the larger TGN. Over 85% of ArfGAP1 proteins bound endosomes (Fig. S3). Indeed, concentrations of ArfGAP1 U and ArfGAP1 B increased in the absence of AP-1/σ 1B compared to wt cells. They also decreased in Δ AP-1 (μ 1A − /− ) cells lacking both AP-1 complexes (Fig. 4C). The ArfGAP1 B distribution appears to depend more on the presence of AP-1 complexes than ArfGAP1 U , because its . The difference between ArfGAP1 isoforms appears to reproduce the differences in σ 1 isoform affinities also in the consecutive pull-down experiments, however the χ 2 test gave p = 0.8 or 55% and more experiments would have to be done. (C) AP-1 dependence of the endosomal concentrations of ArfGAP1 isoforms. ArfGAP1 U (ubiquitous) and ArfGAP1 B (brain) were expressed as GFPtagged proteins in the MEF cell lines (wt = wild-type; σ 1B − /− ; Δ AP-1 = μ 1A − /− (see TGN concentrations in suppl.) (cells n = 10) Box-plot diagram shows the median and the values outside the 50% percentile as lines; black dot indicates an experiment not included for the statistics. Comparing the significance of the data sets between wt and the mutant cell lines by χ 2 tests gave distributions ≤ 1 × 10 −16 (> 99%).
concentration fell below wt levels in Δ AP-1 cells. The ArfGAP1 U concentration decreased only to wt levels, when both AP-1 complexes were absent. The second of the two lipid sensor motifs of ArfGAP1 U , which is modified in ArfGAP1 B , thus appears to mediate either a lipid-domain dependent distribution 38 or the association with a yet unknown protein, making its concentration less AP-1-dependent. Nevertheless, ArfGAP1 U is also more concentrated, when AP-1/σ 1B is absent.
That ArfGAP1 performs this linker function was not expected, because σ 1B − /− endosomes contain less ArfGAP1 than wt endosomes 16 . ArfGAP1 has low binding affinity towards low curvature membranes 39 and thus less ArfGAP1 might bind constitutively to these enlarged endosomes 7 , whereas the remaining ArfGAP1 pool is retained in these complexes with AP-1. Its Arf1 GTP GAP-activity is expected to be inhibited in this tripartite endosomal complex 37 , preventing AP-1 membrane dissociation. This explains the increase in AP-1/σ 1A complexes bound to σ 1B − /− endosomes relative to the reduced amounts in synapses and in the synaptic CCV pool 16 . There is also a marked increase of ArfGAP1 proteins in the endocytic AP-2 CCV pool of σ 1B − /− synapses, indicating that their redistribution between these two membrane pools might also be regulated 16 . σ1B inhibits Rab5/Vps34-pathway activation by σ1A. The formation of AP-1/σ 1A/ArfGAP1/ Rabex-5 complexes (Figs 2-4) and the increase in endosomal Rabex-5 in σ 1B − /− synapses 16 indicates that AP-1/σ 1B binding to Rabex-5 inhibits formation of the AP-1/σ 1A/ArfGAP1/Rabex-5 complexes and thus Rabex-5 recruitment to endosomes. As a consequence the Rab5/Vps34 pathway is less active in wt than in σ 1B − /− synapses. To test whether we can rescue the Vps34 hyperactivation on these endosomes by AP-1/σ 1B, we added wt cytosols to the density centrifugation. Adding cytosolic proteins to the density gradient solutions enables their interaction with membrane proteins during the entire 5 h long centrifugation, during which synaptic early endosomes are separated from the majority of the endosomes (and normally cytosolic proteins) found in fractions 1 and 2. Brain cytosol was prepared from isogenic wt animals and MEF cytosol was prepared from an isogenic wt cell line. MEF cells contain much less AP-1/σ 1B and thus served as negative control 7 . 7 mg protein of σ 1B − /− brain extract was loaded on top of the optiprep density gradient (fraction 1). A total of 1.4 mg cytosolic proteins were added to the density gradient solutions before the density gradient was formed. Fraction 4, enriched in synaptic early endosomes, contains 4% of the loaded σ 1B − /− brain proteins 16 . The amount of wt cytosolic proteins added also corresponds to 4% (280 μ g/gradient fraction) of loaded σ 1B − /− proteins per gradient fraction. Vps34 activity and PI-3-P formation was again analysed by EEA1 binding to the endosomal fractions 16 . The increase in PI-3-P seen in σ 1B − /− endosomes was inhibited by the addition of wt brain (Fig. 5), but not by wt MEF cytosol (not shown). Collectively, these data strongly suggest that AP-1/σ 1B inhibits Vps34 activation by reducing the amount of Rabex-5 bound to these early endosomes.

Discussion
In synapses of AP-1/σ 1B 'knock-out' mice SV recycling is slower than in wt synapses, and large early endosomes are formed on which the remaining second AP-1 complex, AP-1/σ 1A, accumulates. This indicates that AP-1 complexes take part in SV reformation from these endosomes 7,16 . Such a role would be in line with their protein sorting function in CCV formation, but both, the mechanism and its function in SV recycling have yet to be demonstrated. Here we reveal a novel molecular mechanism and a novel AP-1 function in protein sorting and transport, that is not linked to CCV formation. AP-1/σ 1A and AP-1/σ 1B complexes bound to synaptic early endosomes regulate their maturation into late endosomes, a process required for SV protein degradation via endolysosomes. These late endosomes are MVB late endosomes 7,16 into which membrane and membrane-associated proteins are targeted after they have been modified with ubiquitin 40 .
Endosomal AP-1/σ 1A activates the Rab5/Vps34-pathway essential for early endosome maturation into MVB late endosomes via binding of ArfGAP1/Rabex-5 complexes and it does so preferentially with the brain-specific isoform of ArfGAP1. AP-1/σ 1A -ArfGAP1 -Rabex-5 complex formation leads to more Rabex-5 bound to these endosomes 16 and to the concentration of Rabex-5 and Rab5 in AP-1-coated endosomal subdomains. ArfGAP1 stimulates Arf1 GTP hydrolysis, leading to membrane dissociation of AP-1. GAP-activity is stimulated by conformational changes in the C-terminal domain 37 . It is thus reasonable to assume that simultaneous binding of σ 1A and Rabex-5 to the C-terminal domain prevents activation of the GAP-activity, leading to stable AP-1/σ 1A membrane binding. This would explain why endosomes of σ 1B − /− synapses contain more AP-1/σ 1A complexes than wt endosomes 16 . AP-1/σ 1B interferes with formation of this stable complex by binding Rabex-5 directly (Figs 2,3 and 4). Early endosomes from wt cells contain less Rabex-5 and less AP-1/σ 1A complexes than σ 1B − /− early endosomes 16 (Fig. 6). Thus, without trapping and inhibiting ArfGAP1, an AP-1/σ 1B/Rabex-5 endosomal complex is not stable. In vesicular protein sorting high affinity AP-1 membrane binding requires PI-4-P, a membrane-bound cargo protein and Arf1 GTP . The instability can be due to low membrane binding affinity of this bipartite complex and to the Arf1 GTP inactivation by ArfGAP1. Importantly, σ 1B − /− endosomes contain less ArfGAP1 than wt endosomes 16 and thus ArfGAP1 might be limiting for Arf1 GTP activation, but regulatory protein modifications might be involved as well (see also below).
Clathrin is not only involved in CCV formation, but also in the MVB pathway. EM images of σ 1B − /− synapses indicated an increase of clathrin on early endosomes, which would be in line with the upregulation of the MVB pathway 7 . However, this early endosomal clathrin pool is not recruited by the early endosomal AP-1 Scientific RepoRts | 6:29950 | DOI: 10.1038/srep29950 complexes. The ESCRT-0 component Hrs recruits these clathrin molecules and this interaction is essential for the formation of the internal vesicles of MVB endosomes [41][42][43] .
Besides having GDP-GTP exchange activity, Rabex-5 is also a ubiquitin E3-ligase and thus could also ubiquitinate AP-1/σ 1A-bound cargo proteins. This would send those proteins immediately into the Vps34-dependent MVB pathway, which should increase the fidelity and speed of the transport of synaptic proteins into the degradation pathway. Rabex-5 is not the only ubiquitin ligase in synapses and all depend on the activation of the Rab5/ Vps34 pathway for delivery of their specific target proteins to MVB endosomes, and not only proteins bound to AP-1/σ 1A will be sent into this pathway 44 . By microscopy, Vps34 and its product PI-3-P (FYVE-GFP staining) have been detected in dendrites, but in axons and synapses only trace amounts of Vps34 were detected and PI-3-P was not detected at all 45,46 . This indicates that synaptic PI-3-P turnover is especially fast and thus coupled to SV recycling or that the Vps34/MVB-pathway has a highly selective function in synapses, transporting only a limited subset of proteins into MVB endolysosomes for degradation. This is in line with our data demonstrating that not all SV proteins are reduced in σ 1B − /− synapses 16 .
We propose a model in which both AP-1 complexes regulate SV reformation as well as SV protein degradation (Fig. 6). The next question we have to answer is the regulation of the balance between SV protein recycling and endosomal transport, and in which context SV protein degradation is induced. We did not succeed in generating σ 1-isoform specific antibodies and can only estimate σ 1A:σ 1B ratios based on the 25% reduction in AP-1 CCV in σ 1B − /− synapses 16 . A 4:1 ratio indicates that endosomal AP-1/σ 1A regulates also the basal rate of synaptic early endosome maturation, which can be upregulated by the inhibition of σ 1B-Rabex5 binding. All proteins identified to be involved are modified by post-translational modifications, and it is likely that complex formation is regulated by modification of one or more of these proteins. Rabex-5 is phosphorylated at its C-terminus on Ser/Thr and Tyr residues (summarised on PhosphoSitePlus), which might prevent binding by ArfGAP1 and thus its stable endosome association. Also ArfGAP1 is modified by phosphorylation. LRRK2 (leucine-rich repeat kinase 2) activity, whose mutations are the most common cause for Parkinson's Disease, inhibits ArfGAP1 GAP-activity and ArfGAP1 is able to stimulate LRRK2 activity 47,48 . LRRK2 has also been shown to be involved in the regulation of SV recycling as well as in late endosome transport 49,50 . Therefore LRRK2 is part of the regulatory network, that coordinates protein and membrane traffic in synapses and is thus a candidate kinase for taking part in the regulation of this AP-1 function. We are currently working to identify protein modifications and the respective kinases involved in the regulation of this pathway. In Alzheimer's disease, like in Parkinson's disease, disturbed endosomal protein sorting contributes to disease development, and the novel AP-1 function presented here might also play a role in the development of this disease 51 . Figure 5. Rescue of the Rab5/Vps34-pathway activation in σ1B −/− brain endosomes. Inhibition of Vps34 activity in σ 1B − /− early (low-density; density gradient fraction 4) and late (high-density; density gradient fraction 5) endosomes (as in Fig. 1) by the addition of AP-1/σ 1B cytosol to σ 1B − /− membranes. PI-3-P was determined by membrane association of the PI-3-P binder EEA1. wt black circles, dotted line (n = 4, s.d.) , σ 1B − /− red triangles, solid line (n = 6; s.d.), σ 1B − /− plus AP-1/σ 1B red circles, dotted line (n = 4, s.d.). Gradient fraction 1 corresponds to the load and it is not included. Representative western-blots of fractions 4, (lowdensity, early endosomes) are shown. In all experiments wt and σ 1B − /− extracts were processed in parallel and compared and thus EEA1 of wt fraction 4 (ld E) is shown twice. χ 2 test distributions for the data obtained from σ 1B − /− extracts for this fraction is 1 × 10 −10 (> 99%).
In σ 1B − /− synapses only a subset of SV proteins is degraded, presumably by their selective modification by E3-ligases, and thus the protein composition of the SV formed should be altered. This should affect properties such as their mobility and their targeting to and anchoring at active zones, as well as their fusion kinetics with the plasma membrane [52][53][54] . In contrast, the homologous AP-3 complex involved in SV protein sorting during SV biogenesis facilitates direct fusion and degradation of entire SV with Rab7 late endosomes and their degradation 55,56 . This pathway bypasses the Rab5 early endosomes and gives a hint why synapses contain more Rab7 than Rab5 57 . The degradation of only a subset of SV proteins in σ 1B − /− synapses indicates that the alteration in the composition of the pool of SV proteins are part of an adaptation mechanism. We will test how these changes in σ 1B − /− synapses affect SV properties and synapse functions using high resolution microscopy and electrophysiology. The mouse 'knock-out' phenotype demonstrates that the AP-1/σ 1A -AP-1/σ 1B competition in the regulation of early endosome maturation and SV protein degradation is part of a network regulating synaptic plasticity and synaptic silencing essential for the regulation of learning, memory formation and memory recall.
The AP-1/σ 1A and AP-1/σ 1B complexes regulate the maturation of synaptic early endosomes into late, MVB endosomes. They do so by complex formation with or without ArfGAP1 and Rabex-5. This novel function of the clathrin AP-1 adaptor-protein complexes raises their role to one of the major coordinators of protein export routes out of early endosomes.

Methods
Endosome isolation, western blotting, EM images, data analysis. The σ 1B mouse 'knock-out' model has been described in Glyvuk et al. 2010 and has a SV129/BL6 genetic background. Animals used for tissue isolations were − /− animals and isogenic + /+ animals derived from + /− matings. Animals are kept in the central animal facility of the Faculty of Medicine of the Georg-August-University Göttingen in accordance with the appropriate guidelines. Animals were killed with CO 2 and cervical dislocation in accordance with the appropriate guidelines. Animal housing and the protocol for killing the animals were approved by the 'Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit' (LAVES). Brains were immediately isolated and frozen in liquid nitrogen and stored at − 80 °C. Differential and OptiPrep TM density centrifugations were done as established and described in detail in Kratzke et al. 16 . Membrane bound and soluble proteins were separated by differential centrifugation. Mouse cortex was homogenised with a 1 mL glass homogeniser (10 strokes loose, 10 tight plunger) and centrifuged at 1000 g, 10 min. Supernatant (S1) is named cortex fraction. Centrifugation of S1 at 13000 g for 15 min yielded supernatant S13 and pellet P13. Centrifugation of S13 at 100000 g for 45 min yielded a supernatant S100 and a pellet P100. All steps were performed in 38 mM potassium aspartate, 38 mM potassium glutamate, 38 mM potassium gluconic acid, 20 mM MOPS, 5 mM reduced glutathione, 10 mM potassium carbonate, 0,5 mM magnesium carbonate, 1 mM EDTA, 1 mM EGTA, 1:5000 protease inhibitor cocktail (Sigma, München, Ger), 10 nM Calyculin A, 1 mM Na 3 VO 4 , pH 7,1) at 4 °C. Continuous OptiPrep ™ (Axis-Shield, Heidelberg, Ger) gradients were prepared using a Gradient Station ip (BioComp Instruments, Fredericton, CA). Cortex fractions were prepared as described above. Synaptosomes were lysed by 20 passages through a ball homogeniser (Isobiotec, Heidelberg, Ger) with a clearance of 12 μ m. This extract was layered over 0-30% OptiPrep ™ gradients, centrifuged at 65000 g for 5 h and six fractions were collected. Protein concentration was Figure 6. Model for the regulation of SV protein recycling versus SV protein degradation via the multivesicular body late endosome (MVB) pathway by AP-1 complexes AP-1/σ1A and AP-1/σ1B. The Rab5/Vps34-dependent MVB pathway is activated by the ArfGAP1-mediated recruitment of Rabex-5 into a stable AP-1/σ 1A -ArfGAP1 complex. Rabex-5 binding to σ 1B prevents complex formation and thus stable Rabex-5 recruitment onto endosomes. Post-translational modifications like protein phosphorylation are likely involved in the differential regulation of the pathway by both AP-1 complexes (see also discussion). ArfGAP1 binding to the γ 1 (subunit shown in red) 'ear'-domain may support ArfGAP1 recruitment. σ 1A shown in yellow, σ 1B in gold, μ 1A in dark grey and β 1 in light grey.
Scientific RepoRts | 6:29950 | DOI: 10.1038/srep29950 determined according to the Bradford assay (BioRad, Munich, Ger). Fractions of the velocity centrifugation were mixed with 60% OptiPrep ™ to a final concentration of 32.5% or greater, laid under a continuous 0-30% OptiPrep ™ gradient and centrifuged at 100000 g for 18 h. For the complementation experiment 7 mg of σ 1B − /− brain lysate was loaded on top of the preformed gradient (fraction 1). 1.4 mg of wt cytosolic proteins (100.000 × g supernatant) was mixed in the gradient density buffers before the gradient was formed (fractions 2-6). The EEA1 antibody was sc-6414 (Santa Cruz, USA) and the Vps34 antibody was #3811 (Cell Signaling, USA). Secondary, HRP-conjugated antibodies were from Dianova (Hamburg, Germany). Western blots were developed using luminescence reagents (Millipore) and images were recorded using the Fuji LAS 1000 (Fujifilm Corp., Düsseldorf, Germany). The protein content in each fraction is expressed in % of the total. Diagrams were generated using DataGraph (Visual Data Tools, USA). Box-plot diagrams show the distribution of all the data and their median and the values outside the 50% percentile as lines. Data not included for the calculation of the median are indicated by a black dot. Each data set is based on a minimum of 3 independently performed experiments. The significance of the differences detected in experiments comparing data from wt and σ 1B − /− mice as well as data obtained with different protein isoforms were verified by the χ 2 (chi square, Excel) test, because it is most appropriate for smaller sample numbers. In addition, it allows to compare 'knock-out' with wt data sets, which were defined as 100% signal intensity in each experiment. χ 2 distributions ≤ 0.004 indicate a ≥ 95% probability (m = 1) that two data sets are different. EM images of the gradient fraction enriched in synaptic endosomes were prepared after concentration of the gradient fractions from 400 μ l to 50 μ l and glutaraldehyde fixation and uranyl-actetate staining following standard EM procedures.
Yeast-3-hybrid assays. γ 1 N-terminal core domain (1-550 aa) encoding murine cDNA was cloned into pGADT7 (GAL4 DNA-AD-fusion). Murine σ 1-adaptin ORFs were cloned in the MSC-II of pBridge and the indicated proteins and protein domains to be tested for σ 1-binding were cloned into the MSC-I site (GAL4 DNA BD-fusion) of pBridge (InVitrogen, Karlsruhe, Germany). Plasmids were transformed for growth and interaction assays into the yeast strain AH109.
In-vivo distribution of proteins. MEF cell lines used have been established in the literature cited in the results. Cells were fixed with 4% PFA, AP-1 was labelled with anti-γ 1 antibody (BD 610386) and Rab5 (Abcam ab18211) or Rabex-5 or RabGAP5 (Proteintech 12735-1-AP, 20825-1-AP) antibodies. Subsequent steps of the 'proximity-ligation-assay' were done according to the manufacturer instructions (Sigma). pEGFP-C2 ArfGAP1 U and ArfGAP1 B plasmids were transiently transfected into MEF cell lines. Cells were fixed in 4% PFA, AP-1 was labelled with anti-γ 1 (BD biosciences) and secondary Alexa 633 antibodies (life technologies). Confocal images were recorded with a Leica SP2 microscope (Leitz, Germany) and quantified using the microscopes software package and ImageJ software (NIH, USA).