Alpha adaptins show isoform-specific association with neurofibrillary tangles in Alzheimer’s disease

Abstract

Aims:

The heterotetrameric assembly protein complex 2 (AP-2) is a central hub for clathrin-dependent endocytosis. The AP-2 α-adaptin subunit has two major isoforms, encoded by two separate genes: AP2A1 and AP2A2. Endocytosis has been implicated in the pathogenesis of neurodegenerative disease, and recent studies linked α-adaptins (gene variants, splicing defects, and altered expression) with late-onset Alzheimer’s disease (LOAD) risk. Here, we used multiple antibodies to investigate α-adaptin isoforms and their localization in human brains.

Methods:

The specificities of ten different α-adaptin antibodies were evaluated using immunoblots after human AP2A1 and AP2A2 plasmid transfection in cultured cells. Additional immunoblot analyses were then performed on protein homogenates from control and LOAD subjects. Formalin-fixed, paraffin-embedded brain sections from control and LOAD subjects were immunohistochemically stained, and immunofluorescence experiments were performred for quantitation of colocalization with digital image analysis.

Results:

Eight of the ten evaluated antibodies recognized transfected α-adaptin proteins on immunoblots. The α-adaptin subspecies were relatively uniformly expressed in five different human brain regions. The α-adaptins were present in the detergent-insoluble fraction from cognitively impaired, but less so in control, brains. Immunohistochemical analyses showed colocalization of AP2A1 with tau pathology in LOAD brains. By contrast, AP2A2 colocalized with microglial cells.

Conclusions:

These observations provide evidence of isoform-specific changes of α-adaptins in the brains of LOAD subjects. Antibodies that were verified to recognize AP2A1, but not AP2A2, labelled neurofibrillary tangles of LOAD patients. The findings extend our understanding of AP-2 proteins in the human brain in healthy and diseased states.

Graphical Abstract

The AP-2 α-adaptin proteins (AP2A1 & AP2A2), which serve as a central hub for clathrin-dependent endocytosis, are implicated in neurodegenerative disease pathogenesis. Here we show that α-adaptins are present in the insoluble fractions in the brain of LOAD subjects. Importantly, isoform-specific changes of α-adaptins are detected in LOAD brains, with AP2A1 but not AP2A2 colocalizing with NFTs in LOAD brains.

INTRODUCTION

Late-onset Alzheimer’s disease (LOAD) is a devastating neurodegenerative disorder. The neuropathological hallmarks of LOAD are extracellular amyloid-β (Aβ) plaques, and neurofibrillary tangles (NFTs) composed of intracellular tau protein aggregates [1]. Although much has been learnt about Aβ plaques and NFTs, we lack a complete understanding of the cellular pathways that contribute to and react to those protein aggregates.

Endocytosis has been strongly implicated in the pathogenesis of LOAD. Genome-wide association studies (GWAS) have identified highly impactful LOAD-linked risk gene loci, including BIN1 and PICALM (both clathrin adaptor proteins, integral to clathrin-mediated endocytosis (CME)) [2]. Other gene variants associated with LOAD risk, including CD2AP, SORL1, EPHA, RIN3, and CLU also implicate endolysosomal pathways in AD pathogenesis [2], [3], [4]. Further, the internalization, propagation, and clearance of aggregation-prone proteins occur via endocytic mechanisms [5], [6], [7], [3].

The assembly protein complex 2 (AP-2) is a central hub that links clathrin, accessory clathrin factors, membrane lipids, and membrane-associated cargo proteins [8, 9]. The AP-2 protein complex is a heterotetrameric complex present predominantly at the plasma membrane but has also been implicated in various intracellular processes, including autophagy and BDNF neurochemistry [10, 11]. The AP-2 complex comprises four different proteins: large α- and β2- adaptin subunits, a medium-sized μ2- adaptin subunit, and a small σ2- adaptin subunit (Fig.1A) [12, 13].

Figure 1. Schematic representation of the AP-2 heterotetramer complex and the α-adaptins.

A. AP-2 core consists of four adaptin subunits, with N-terminal domain of α and β2 subunits along with μ2 and σ2 subunits forming the trunk, while the C-terminal domain of α and β2 protruding out from the trunk/core like ear appendages. The trunk and appendage domains are connected via a flexible linker which forms the hinge [37–39]. B. Cartoon depicting the domain organization of the three α-adaptin isoforms showing the positions of the sequences used to raise antibodies (a-f, orange lines, AP2A1; j–g, blue lines AP2A2). AP2A1 contains a region in the proximal part of the hinge (grid pattern) that is conserved between the long and short isoforms of the protein. However, this region is not conserved in AP2A2 (checkerboard pattern); thus, antibodies against this sequence (i, j, blue lines) react with AP2A2 but not with AP2A1^L and AP2A1^S. Additionally, AP2A1 contains a region (diagonal pattern) that corresponds to the exon spliced in in the brain but spliced out in most other tissues. Hence, antibodies against this sequence (f, orange line) react only with AP2A1^L but not with AP2A1^S and AP2A2.

The focus of the present study is the AP-2 α-adaptin proteins – comprising at least three isoforms, all encoded by two separate genes, AP2A1 and AP2A2. The human AP2A1/α_A and AP2A2/α_C proteins (product of the AP2A1 & AP2A2 genes respectively) share ~80% identity at the amino acid level [14]. The two protein sequences are most divergent in the hinge domain as shown in Fig.1B [14]. The third α-adaptin isoform results from alternative splicing of the AP2A1 transcript and is missing a 22 amino acid sequence in the hinge region [14, 15]. The long AP2A1 isoform (AP2A1^L) has a predicted molecular weight of 108 kDa, while the short AP2A1 isoform (AP2A1^S) and AP2A2 isoform are ~104 kDa (Fig.1B, 2). Most known α-adaptin antibodies cross-react with all isoforms of the α-adaptin proteins. Without the use of isoform-specific antibodies, AP2A1^S may escape detection on an SDS-gel, as it has similar mobility to AP2A2 (Fig.1B, 2). While the AP2A1^S and AP2A2 transcripts are robustly expressed in most tissues, the AP2A1^L transcript is brain-enriched [14, 15].

Figure 2. Immunoblots showing the specificity of the ten different α-adaptin antibodies used in this study.

Homogenates of HeLa cells transiently transfected with either an empty construct (Mock) or FLAG-tagged AP2A1^L and AP2A2 constructs were subjected to SDS-PAGE and immunoblot analysis using ten different α-adaptin antibodies. The immunoblots in each panel were probed with the individual antibodies labelled according to code in Fig.1B depicting the respective epitopes. Immunoblot using an antibody against the FLAG-tag was included as a positive control, indicating the overexpressed plasmid’s transfection efficiencies, and β-actin serves as a loading control (bottom panel, far right). The arrows and corresponding labels on the right side of the immunoblots represent either the individual α-adaptin isoform recognized by the antibody, or where the band is supposed to be. Arrows are not included for the 2 blots probed with antibodies that show inconsistent labelling patterns. The molecular weight marker is depicted on the left side of the immunoblots.

Multiple credible reports link AP-2 with AD pathogenesis. AP2A1 was recently identified as a hub protein in AD proteome networks and had lower protein levels detected in brains from patients with AD than those from patients with amyotrophic lateral sclerosis, Parkinson’s disease, or control subjects [16]. Multiple GWAS have implicated genetic variation in AP2A2 and also gene products (BIN1 & PICALM) that interact directly with the AP2A2 protein as risk factors for LOAD risk [17]. A recent transcriptome-wide association study identified alternative splicing events in both AP2A1 and AP2A2 to be associated with AD, further reinforcing the association of AP-2 adaptors with AD susceptibility [18]. Notably, neurodegeneration-associated increase in calpain activation was shown to promote proteolysis of α-adaptins and other endocytic adaptors with apparent deleterious impacts on synaptic plasticity [19, 20]. We previously found intriguing staining patterns when assessing human brains with an antibody that recognizes both AP2A1 and AP2A2 [21].

Although these prior studies indicate associations between α-adaptins and AD aetiology, relatively little is known about the specific α-adaptin isoforms (AP2A1 and AP2A2) and their localization in human AD brains. Notably, most prior studies did not consider the different α-adaptin subtypes or antibody-specific staining patterns. Here, starting with ten different anti-α-adaptin antibodies, we assessed α-adaptin isoforms in post-mortem brains from control and LOAD subjects.

MATERIALS AND METHODS

Cell Culture experiments to analyse α-adaptin antibodies and expression levels

Plasmids expressing human AP2A1 and AP2A2 proteins were used to test the antibodies via immunoblotting. The following plasmids were used: pCMV6-XL5 empty vector; pCMV6-XL5-MYC/DDK-AP2A2 (Accession No: NM_012305), and pCMV6-XL5-MYC/DDK-AP2A1^L (Accession No: NM_014203)—all from Origene (Rockville, MD). HeLa cells were grown in 6-well plates as previously described [21]. According to the manufacturer’s protocol, transfections were performed using Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA, Cat # L3000008), using 3.0 μg total plasmid in each well. Each transfection was performed with two biological replicates. The cells were harvested 28 hours post-transfection and subsequently lysed for 10 min on ice using RIPA buffer (Thermo Fisher Scientific, Waltham, MA, Cat # 89901) containing protease & phosphatase inhibitors (Thermo Fisher Scientific, Waltham, MA, Cat # 78438 & 78427, 1:100 dilution). The homogenates were then centrifuged at 15,000 × g for 10 min @ 4 °C to remove the insoluble cell debris. The supernatant (total lysate) was transferred to a clean tube and used for immunoblot analysis.

The samples were separated using 10% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto a nitrocellulose membrane for immunoblotting. Following blocking with 5 % non-fat dry milk in PBS/T (1 mM KH₂PO₄, 155 mM NaCl, 2.7 mM Na₂HPO₄-7H₂O containing 0.05 % [v/v] Tween-20, pH 7.2), the blots were incubated overnight with primary antibodies diluted in PBS/T and then washed four times in PBS/T for 10 minutes. The blots were then incubated for 1 hr in respective horseradish peroxidase-conjugated secondary antibodies diluted in PBS/T (Jackson ImmunoResearch Laboratories; Cat # 115-035-166, 111-035-144 and 305-035-003; 1:10,000 dilution). The chemiluminescent signals on the blots were visualized with Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific, Waltham, MA, Cat # 32106). The primary antibodies used are listed in Table 1 & Suppl Table 1. Mouse anti-DDK antibody was from Genscript (Cat # A00187, 1:1000 dilution) and rabbit anti- β-actin antibody from Rockland Immunochemicals (Cat # 18309, 1:5000 dilution).

Table 1.

List of α-adaptin antibodies used in this study and the results of immunoblotting and bright field immunohistochemistry experiments.

Code	Catalogue #	Host	Epitope	AP2A1L (αAL)	AP2A1S (αAS)	AP2A2 (αC)	Labelling of NFTs
a	AC1-M11	mouse	A 1–977	+	+	+	−
b	LSB15729	rabbit	A 1–260	+	+	−	−
c	BD610501	mouse	A 38–215	+	+	−	+
d	LSC178945	rabbit	A 22–199	−	−	−	N/A
e	sc55497	mouse	A 678–977	+	+	−	−
f	A706–727	goat	A 706–727	+	−	−	+
g	LSC482433	rabbit	C 1–939	+	+	+	+
h	157221-AP	rabbit	C 530–939	−	−	−	N/A
i	LSC156387	rabbit	C 610–637	−	−	+	−
j	C619–656	rabbit	C 619–656	−	−	+	−

Preparation of brain homogenates for analysing α-adaptin expression levels

Human brain autopsies were performed at the University of Kentucky with Institutional Review Board (IRB) approval as described previously [22, 23]. A convenience sample of cases (Suppl Table 2) was selected from among autopsies with a post-mortem interval (PMI) of <5 hrs (selected samples were recent LOAD cases with Braak NFT stage V or VI). Specific brain regions were dissected and snap-frozen in liquid nitrogen at autopsy. For α-adaptin immunoblots, 150–200 mg of frozen substantia nigra, sensory cortex, motor cortex, frontal cortex, temporal neocortex and anterior cingulate cortex tissue were each homogenized in 10 volumes of ice-cold lysis buffer (50 mM Tris-Cl, pH 8 containing 150 mM NaCl, 0.5% NP-40, 0.5 % Triton-X-100, 0.5 mM EDTA), containing protease and phosphatase inhibitors (Thermo Fisher Scientific, Waltham, MA, Cat # 78438 & 78427, 1:100 dilution) and incubated for 10 min at 4 °C. The homogenate was centrifuged (15,000 × g for 10 min at 4 °C), the supernatant was carefully transferred to a new tube, and protein concentrations were determined by Biorad DC assay kit (Cat # 5000112). 25 ug/lane of homogenates were applied to SDS-PAGE and analysed by immunoblotting using the appropriate α-adaptin antibodies.

Biochemical analysis on fractionated proteins from the amygdala of LOAD subjects

Dissected amygdalae were snap-frozen in liquid nitrogen at autopsy and then stored at −80°C until these experiments were performed. Information on the biosamples used in these experiments is presented in Suppl Table 2. Protein extraction and serial fractionation were carried out using the already published methodology [24]. Briefly, each amygdala was weighed and homogenized in low salt (LS) buffer (10 mM Tris-Cl, pH 7.5, 5 mM EDTA, and 10 % (w/v) sucrose, 5 ml buffer per gram tissue) using tissue grinders (VWR # 47732–446). The homogenates were centrifuged at 25,000 × g, 4 °C for 30 min, and the supernatants were saved as LS fractions. The pellets were re-extracted with LS buffer, centrifuged as above, and the supernatants were discarded. Then, the pellets were extracted with 5 mL TX buffer (LS buffer supplemented with 1 % (v/v) Triton-X-100 and 0.5 M NaCl) per gram original tissue, centrifuged at 180, 000 × g, 4 °C for 30 min, and the supernatants were saved as TX fractions. The pellets were re-extracted with TX buffer, centrifuged as above, and the supernatants were discarded. The pellets were extracted with 5mL myelin flotation buffer (TX buffer containing 30 % [w/v] sucrose) per gram of original tissue, centrifuged at 180,000 × g, 4 °C for 30 min, and the supernatants were discarded.

Subsequently, the pellets were extracted with 5 mL SARC buffer (LS buffer supplemented with 1 % [w/v] N-lauroylsarcosine and 0.5 M NaCl) per gram original tissue, incubated at 22 °C for 2 hr on an end-over-end shaker, centrifuged at 180,000 × g, 22 °C for 30 min, and the supernatants were saved as the SARC fractions. Finally, the pellets were extracted with 0.75 mL urea buffer (7 M urea, 2 M thiourea, 4 % (w/v) CHAPS, and 30 mM Tris-HCl, pH 8.5) per gram of original tissue at 22 °C, centrifuged at 25,000 × g, 22 °C for 30 min, and the supernatants were saved as the urea soluble fractions. The LS, TX, SARC, and myelin flotation buffers were supplemented with protease inhibitor cocktail (Millipore Sigma P8340, 1:300), phenylmethylsulfonylfluoride ([PMSF], 0.2 mM), N-ethylmaleimide ([NEM], 5 mM), nicotinamide (20 mM), Trichostatin A (1.5 M), sodium orthovanadate (1 mM), and PhosSTOP phosphatase inhibitor tablets (1 tablet per 10 mL, Millipore Sigma 4906837001).

The fractions were subjected to SDS-PAGE, on 3–8% TRIS-acetate gradient protein gels (Cat # EA0378BOX, Thermo Fisher Scientific, Waltham, MA) using TRIS-acetate SDS running buffer (Cat # LA0041, Thermo Fisher Scientific, Waltham, MA); the gels were run at 50V constant. The proteins were transferred to nitrocellulose membranes (Cat # 1704158, Bio-Rad, Hercules, CA), followed by blocking with 5 % non-fat dry milk in TBST (50 mM Tris-HCl, 0.85 % [w/v] NaCl, 0.1 % [v/v] Tween-20, pH 7.5). The primary and secondary antibodies were applied in blocking buffer. The blots were probed with the primary antibodies goat anti-AP2A1^L (f. A706–727, a gift from Dr Margaret Robinson), mouse anti-AP2A1 (Cat # BD610501, BD Biosciences, Franklin Lakes, NJ), mouse anti-AP2A1 (Cat # sc55497, Santa Cruz Biotechnology, Dallas, TX), rabbit anti-AP2A2 (Cat # LSC156387, LSBio, Seattle, WA), rabbit anti-GAPDH (Cat # G9545, Sigma-Aldrich, St. Louis, MO), mouse anti-phospho-Tau (PHF-1, a gift from Dr Peter Davies), all in 1:1,000 dilution. The secondary antibodies were IRDye 800CW goat anti-rabbit (Cat # 926–32211, Li-Cor, Lincoln, NE), Alexa Fluor 680 goat anti-mouse (Cat # A21058, Thermo Fisher Scientific, Waltham, MA), IRDye 800CW donkey anti-goat (Cat # 925–32214, Li-Cor, Lincoln, NE), and Alexa Fluor 680 goat anti-rabbit (Cat # A21109, Thermo Fisher Scientific, Waltham, MA). The images were acquired on the Li-Cor Odyssey CLx Imaging System.

Immunohistochemistry (IHC)

Before IHC, brain tissue was immersion-fixed in 10% neutral-buffered formalin (Cat # C4320–105, Cardinal Health, Dublin, OH) for >2 weeks before paraffin embedding. Immunohistochemical stains were performed as previously described [25], except that formic acid pre-treatment was used for all slides. Brain tissue sections of the temporal neocortex (Brodmann areas 21/22) and frontal cortex (Brodmann area 9) were used for the IHC experiments. Briefly, sections cut at an 8-μm thickness from formalin-fixed-paraffin embedded tissue blocks were deparaffinized before microwave antigen retrieval for 6 min (power 8) using citrate buffer (Declere buffer, Cell Marque; Rocklin, CA). See Suppl Table 1 for antigen retrieval methods for each antibody. The sections were then placed in 100% formic acid (Cat. # A-119 P, Fisher Scientific) for 3 min, which is a standard antigen retrieval step in our lab. Sections were incubated in H₂O₂ to inhibit endogenous peroxidase, rinsed in water, and then incubated in a blocking solution containing 5% normal goat or rabbit serum in TRIS-buffered saline (TBS) for 1 hour at room temperature. Sections were then incubated in primary antibodies diluted in 5% (goat or rabbit serum, TBS) for 48 hours at 4 °C. Secondary antibodies were biotinylated IgG (Cat # BA-1000, # BA-2000, # BA-4000, and # BA-5000 from Vector Labs, Burlingame, CA) diluted at 1:200 in 5% (goat or rabbit serum, TBS) for 1 hr at room temperature. After washing, the Vectastain ABC kit (Vector Labs, Burlingame, CA) was used with Vector Nova Red (Vector Labs), followed by counterstain with haematoxylin rinsing, dehydration, clearing, and mounting in coverslip media.

Immunofluorescence (IF)

Double and triple-label IF experiments were performed using methods previously described [26], with some modifications. Brain tissue sections of the temporal neocortex (Brodmann areas 21/22) and frontal cortex (Brodmann area 9) were used for the triple-label and double-label IF experiments. Sections were cut from formalin-fixed, paraffin-embedded tissue blocks at 8-μm thickness and deparaffinized before microwave antigen retrieval for 6 min (power 8) using citrate buffer (pH 6, Declere buffer, Cell Marque, Rocklin, CA). The mounted sections were placed in 100 % formic acid (Thermo Fisher, Waltham, MA) for precisely 3 min. Next, sections were incubated for 45 sec at room temperature in a solution of TrueBlack (Cat # 23007, Biotium, Fremont, CA) prepared in 70% ethanol, to reduce lipofuscin auto fluorescence.

For triple-label IF experiments, sections were blocked in 5 % (donkey serum, TBS) for 1 hr at room temperature, then incubated with primary antibodies anti-AP2A1^L (goat monoclonal, 1:200 dilution; a gift from Dr Margaret Robinson) anti-AP2A2 (rabbit monoclonal, 1:200 dilution; a gift from Dr Margaret Robinson) and PHF1 (mouse monoclonal, 1:10000 dilution, a gift from Dr Peter Davies), diluted in 5% (donkey serum, TBS), for 48 hours at 4 °C. One section each from two experimental cases (both Braak NFT stage VI), were stained with the three primary antibodies. After washing four times, sections were incubated in secondary antibodies conjugated to Alexa Fluor probes 568, 488 (diluted 1:500, Cat #s A11057 and A21206, Life Technologies), and 647 (Cat # A31571, Life Technologies) diluted 1:5000 in 5% (donkey serum, TBS) for 1 hr at room temperature. Slides were cover slipped using Invitrogen ProLong Gold mounting medium with DAPI (Cat # P36935, Thermo Fisher Scientific, Waltham, MA). Separate controls were run lacking one primary antibody (separately for PHF1 and α-adaptins), but with the three secondary antibodies (all antibodies at same concentrations, all other steps the same), to ensure that neither auto fluorescence nor antibody cross-reaction was leading to false-positive colocalization. Images were acquired on a Nikon C2Plus Confocal Microscope, using a Plan Fluor 20× Oil (1.3 NA) objective, a pinhole of 1.2 airy unit (AU), step size of 2.5 μm, and an image size of 1024×1024, with a 2x zoom, using a channel series scan. Five non-overlapping images of the temporal cortex were acquired for each section for each of the two cases.

For double-label IF experiments, sections were blocked in 5 % (donkey serum, TBS) for 1 hr at room temperature, then incubated with primary antibodies anti-AP2A2 (rabbit monoclonal, 1:100 dilution; a gift from Dr Margaret Robinson) and Iba-1 (mouse monoclonal, 1:1000 dilution, Abcam # ab15690), diluted in 5% (donkey serum, TBS), for 48 hours at 4 °C. After washing four times, sections were incubated in secondary antibodies conjugated to Alexa Fluor probes 488 (diluted 1:500, A21206, Life Technologies), and 647 (Cat # A31571, Life Technologies) diluted 1:1000 in 5% (donkey serum, TBS) for 1 hr at room temperature. Slides were cover slipped using Invitrogen ProLong Gold mounting medium with DAPI (Cat # P36935, Thermo Fisher Scientific, Waltham, MA). Separate controls were run lacking one primary antibody (separately for Iba-1 and AP2A2), but with the two secondary antibodies (all antibodies at same concentrations, all other steps the same), to ensure that neither auto fluorescence nor antibody cross-reaction was leading to false-positive colocalization.

Imaging colocalization quantification

To determine the % overlap volumes of α-adaptins and PHF-1 surfaces (colocalization), Imaris 3D microscopic data analysis software was used (version 8.3 Bitplane USA Concord, MA) and quantification was performed as described previously [26]. An intensity based 3D segmentation and visualization tool was used to create surfaces and all settings for the surface reconstruction were held constant for all the images acquired. Briefly, a smoothing feature as well as an adaptive thresholding with a local background subtraction was applied to each image. Using images acquired with slides stained for PHF-1 only, or the α-adaptins only, we determined the threshold where positive staining could be detected without detecting background. Using these settings, the intensity threshold for segmentation was adjusted. Finally, a filtering step was applied to remove surfaces less than 65 μm³ in volume thereby ensuring small neurites would not be included in the count of the number of positive surfaces. After the surfaces were rendered, the statistical test results were obtained for each individual object. This analysis provided an unbiased digital quantification of the number of PHF-1+ and AP2A1+ NFTs, and the overlap volumes of the individual surfaces.

RESULTS

Testing the specificity of anti-α-adaptin antibodies

The ten different anti-α-adaptin antibodies that were initially gathered are shown in Table 1 and Fig. 1B. The antibodies used in this study recognized a) the long AP2A1 isoform-AP2A1^L; b) the long and short AP2A1 isoforms - AP2A1 ^(L&S); and/or, c) the AP2A2 isoform. In the absence of AP2A1^S specific antibodies, the interpretation of this isoform comes from immunoblotting interpretation of the banding pattern. The specificity of all the anti-α-adaptin antibodies was first analysed by immunoblotting using lysates of HeLa cells transfected with either AP2A1^L or AP2A2 (Fig. 2). RNA-seq data from public access portals suggests that HeLa cells intrinsically express AP2A1^S and AP2A2 transcripts but not AP2A1^L. Correspondingly, following transfection of empty vector, faint ~104 kDa bands were detected on the immunoblots revealing endogenous AP2A1^S and AP2A2. As expected, this signal was augmented following AP2A2 transfection, whereas a ~108 kDa band was detected following AP2A1^L transfection (Fig. 2). An immunoblot using an antibody against the FLAG-tag was included as a positive control, indicating the plasmid’s transfection efficiencies, and β-actin served as a loading control (Fig. 2).

The mouse monoclonal antibody raised against full-length AP2A1^L (a. AC1-M11, Fig. 1B, 2) is known to react with all α-adaptin isoforms [14], and as expected, it labelled both AP2A1^L and AP2A2 bands with approximately equal intensity. This antibody, which should recognize the two AP2A1 isoforms, highlighted both the upper and lower bands, although the upper band was more intense than the lower band. Several other antibodies were also thought to detect all α-adaptin isoforms, as they were raised against a fragment of AP2A1 that is homologous with AP2A2 (b. LSB15729, c. BD610501, and e. sc55497, Fig. 1B). However, these antibodies in our hands seemed to predominantly detect AP2A1 without labelling the AP2A2 construct, indicating specificity for AP2A1.These antibodies were subsequently used to detect both the long and short AP2A1 isoforms - AP2A1 ^(L&S).

Two of the antibodies were produced in the laboratory of Dr Margaret Robinson. One was raised against the brain-enriched AP2A1^L protein, and designed to not recognize either the AP2A1^S isoform or AP2A2. Thus, as expected, this antibody only recognized the upper AP2A1^L band and not any other α-adaptin isoforms (f. A706–727, Fig. 1B, 2). The other antibody from Dr Robinson’s laboratory was raised against a fragment in AP2A2 with the least homology with AP2A1, and as expected, this antibody labelled only AP2A2 and not the AP2A1 isoforms (j. C619–656, Fig. 1B, 2). We also studied a commercially available AP2A2 antibody that was raised against an epitope very similar to that of the Margaret Robinson lab antibody and is consequently specific for AP2A2 (i. LSC156387, j. C619–656, Fig. 1B, 2).

We previously reported the use of a rabbit polyclonal antibody raised against full-length AP2A2 [21]. Based on our immunoblots from transfected cells, this antibody was not specific for AP2A2 in addition to AP2A2, it cross-reacted with both AP2A1 isoforms (g. LSC482433, Fig. 1B, 2). We also note that two out of the ten anti-α-adaptin antibodies we evaluated did not work under our experimental conditions, as they showed a labelling pattern on immunoblots that indicated a lack of recognition of transfected α-adaptin proteins (d. LSC178945, h. 15722–1-AP, Fig. 1B, 2).

We performed immunoblots on protein extracts from control human brains (n = 4; Suppl. Table 2) to analyse the normal expression of α-adaptin isoforms in different brain regions. Between the AP2A1 isoforms, AP2A1^L was the predominantly expressed variant, and it was comparably expressed in all the brain regions investigated (Suppl. Fig. 1A–C). AP2A2 and AP2A1^L were also expressed at similar levels in these brain areas (Suppl. Fig. 1C–D). Additionally, we compared the expression levels of α-adaptin isoforms in the temporal cortex tissues (Brodmann area 21 & 22) of control (n = 5) and AD brains (n = 5) (Suppl. Table 2). In this small convenience sample, we did not find a statistically significant difference in α-adaptin expression levels between control and AD cases (Suppl. Fig. 6).

Enrichment of α-adaptin isoforms in the insoluble fractions of LOAD brain

We previously identified detergent-insoluble, urea soluble proteins in the amygdala of demented subjects following protein extraction and Liquid Chromatography- Electrospray Ionization- Tandem Mass Spectrometry (LC-ESI-MS/MS) analysis [24]. We followed a simple and robust technique for separating proteins with different solubilities, allowing them to be serially extracted using progressively harsher extraction reagents, each containing more potent solubilizing chaotropes and detergents.

Using this approach, we asked whether the α-adaptin isoforms in the amygdala of a LOAD subject (n = 1) fractionated into less soluble fractions than those in a control subject (n = 1). The LC-ESI-MS/MS analysis revealed several AP2A1 and AP2A2 polypeptides in the insoluble fraction from cognitively impaired subject but not among the control subject. Although the analysis showed four peptides common to both AP2A1 and AP2A2, several unique peptides that map to the individual isoforms were also identified (data not shown).

Immunoblotting of the low salt, triton X-100, sarcosyl, and urea fractions with an antiserum against AP2A1^L detected bands at the expected size (108 kDa) in control and AD subjects. While AP2A1^L was primarily present in the more soluble low salt and triton X-100 fractions of the control, it significantly shifted towards the less soluble sarcosyl and urea fractions in the AD case (Fig. 3A). Similar to AP2A1^L, the control case had an AP2A2-immunoreactive band (~104 kDa) predominantly in the soluble fractions, and this signal shifted towards the insoluble fractions in the AD case (Fig. 3B). Interestingly, several lower molecular weight species (~60–108 kDa), likely corresponding to cleaved forms of the α-adaptins, were identified in the AD brain’s insoluble fractions.

Figure 3. Immunoblots showing the enrichment of AP2A1^L and AP2A2 signal in the detergent-insoluble protein fractions of LOAD patient amygdala.

Amygdalae from a control subject (case 19) and a LOAD subject (case 16) were subjected to fractionation based on solubility, as described in the main text. The fractions were subjected to denaturing gel electrophoresis in the order of solubility, with the “LS” fraction being the most soluble and the “urea” fraction (the detergent-insoluble fraction) the least soluble. The immunoblots were probed with AP2A1^L antibody (A, f. A706–727) and AP2A2 antibody (B, i. LS156387). For both α-adaptins, there is enrichment in the relatively insoluble fractions. For an internal negative control, the blot in A was re-probed with anti-GAPDH antiserum (C), showing no enrichment in the relatively insoluble fractions. For an internal positive control, the blot in B was re-probed with anti-PHF1 antiserum (D), showing enrichment of phosphorylated Tau signal in the insoluble fraction in the form of a continuous high M.W smear. The M.W marker is depicted on the left side of the immunoblots. All of the antibodies recognize proteins at the expected M.Ws (shown as red stars and font).

Both the control and AD brain showed enrichment of GAPDH in the more soluble low salt and triton X-100 fractions (Fig. 3C). By contrast, the Tau protein, a bona fide pathological marker that is frequently identified in the detergent-insoluble fractions of the diseased brain [24], served as our positive control. Immunoblotting of the control and AD fractions with an anti-PHF1 antibody detected the expected full-length phosphorylated Tau (~55–70 kDa) in the AD case, but the corresponding bands were barely detectable in the control case (Fig. 3D). In the AD case, the Tau signal was seen as high-MW smears in the insoluble urea fraction.

We also found that AP2A1^S (expressed at lower levels than AP2A1^L) was more enriched in the less soluble fractions of the AD brain rather than in the control (Fig. 4). Although AP2A1^L but not AP2A1^S was the predominantly expressed isoform in the detergent-soluble fractions of control subject, the levels of AP2A1^L and AP2A1^S were comparable in the insoluble fractions (Fig. 4). Taken together, these findings indicate that the α-adaptin isoforms fractionate into less soluble fractions in dementia.

Figure 4. Immunoblots showing the enrichment of AP2A1^L&S signal in the detergent-insoluble protein fractions of LOAD patient amygdala.

Amygdalae from a control subject (case 19) and a LOAD subject (case 16) were subjected to fractionation based on solubility, as described in the main text and Fig 3. The immunoblots were probed with antisera that can detect both long and short isoforms of AP2A1 (A, c. BD610501 and B, c. sc55497). The arrows and corresponding labels on the right side of the immunoblots represent the individual α-adaptin isoform recognized by the antibody. The M.W marker is depicted on the left side of the immunoblots.

Colocalization of AP2A1 with neurofibrillary pathology in LOAD brains

We previously reported that in the temporal neocortical regions of LOAD subjects, AP2A2 was frequently associated with neurofibrillary pathology [21]. To investigate the pattern of expression of α-adaptin isoforms in the temporal neocortex (Brodmann areas 21 & 22) and frontal cortex (Brodmann area 9), we performed brightfield IHC using antibodies that either recognize only the long AP2A1 isoform- AP2A1^L (f. A706–727, Fig. 5A, B and LSB9566, Fig. 5C) or the long and short AP2A1 isoforms - AP2A1 ^(L&S) (c. BD610501, Fig. 5D). Here, the commercially available AP2A1^L specific antibody LSB9566 is identical to the Margaret Robinson lab antibody (f. A706–727). In LOAD patients, AP2A1 immunoreactivity was seen in cells with morphologic features of NFTs. Although we previously reported that AP2A2 co-localized with NFTs in LOAD brains [21], here we provide evidence that the antibody used in our prior study cross-reacts with all α-adaptin isoforms (g. LSC482433, Fig. 5E, F). When IHC was performed using antisera specific for AP2A2, it intensely stained some cells that resembled microglia, but did not specifically immunolabel NFTs (j. C619–656, Suppl Fig. 2A, B). Table 1 shows the α-adaptin antibodies described in this study and the results from the IHC analyses performed on a convenience sample of 16 subjects including control (n = 4, Braak 0 & I) and LOAD (n = 12, Braak ≥ IV) cases. Suppl Table 2 shows the cases and controls used in this study.

Figure 5. Brightfield immunohistochemical staining of AP2A1 in human brain sections using four different antisera.

Shown are representative results from staining of temporal neocortex from demented subjects with severe Alzheimer’s disease pathology (panels A-E) and a nondemented subject with minimal pathology (panel F). Three antisera raised against epitopes within AP2A1 showed immunostaining of what appeared to be neurofibrillary tangle-like structures: f. A706–727 (A and B, inset magnification); LSB9566 (C); and c. BD610501 (D). These results represent a tendency to stain neurofibrillary tangles (black arrowheads) in dementia cases. Panel E shows staining of neurofibrillary tangles by the antiserum that labels both AP2A1 and AP2A2 (g. LSC482433). By contrast, the nondemented case in panel F showed even neuronal staining of AP2A1 and AP2A2. Scale bars = 300 μm (A); 60 μm (B); 80 μm (C); 150 μm (D); 100 μm (E); and 200 μm (F).

In order to confirm the colocalization of α-adaptin isoforms with pTau in NFTs, triple label IF was carried out in temporal neocortical tissues (Brodmann areas 21 & 22) from LOAD brains (n = 2, Braak VI) using antibodies specific for AP2A1^L (f. A706–727), AP2A2 (j. C619–656) and pTau (PHF-1). These α-adaptin isoform specific antibodies were considered the most robust for the triple-label IF experiments. As seen in Fig. 6A–D and Suppl. Fig 5A–D, pTau in LOAD brains was often colocalized with AP2A1^L but not with AP2A2. The majority of AP2A1^L immunoreactivity was diffuse cytoplasmic staining -- often colocalized with pTau in NFTs. We also observed bright AP2A2-immunoreactive structures that were immunonegative for pTau or AP2A1^L. Adjacent sections of single antibody-stained control slides confirmed the staining specificity (Suppl. Fig. 3).

Figure 6. Triple-label immunofluorescence staining & quantification: pTau (PHF-1), AP2A1^L, and AP2A2.

Representative photomicrographs from an AD case showing the localization of PHF1 (cyan; A); AP2A1^L (red; B); and AP2A2 (green; C) proteins. Panel D shows the merged photomicrographs with light pink indicating signal overlap and colocalization. These photomicrographs depict the temporal neocortex (Brodmann areas 21/22) of a Braak VI case (Suppl. Table 2). The experiment was performed using antibodies specific for AP2A1^L (f. A706–727), AP2A2 (j. C619–656) and pTau (PHF-1). In (A-D), bottom panels are enlarged images of boxed areas in corresponding top panels. PHF-1 was seen as expected in neurofibrillary tangles (arrows, panel A), neuritic threads, and colocalized substantially with AP2A1^L (arrows, panels B & D), but not with AP2A2. Note that overlap between PHF-1 and AP2A1^L is only partial - some PHF-1 and some AP2A1^L signals are not colocalized (bottom panels A, B & D). Scale bars = 20 μm (Top panels, A-D) and 3 μm (Bottom panels, A-D). E) Box and whiskers plot showing the quantification of overlap volumes between PHF-1 [+] AP2A1^L [+] and PHF-1 [+] AP2A2 [+] double immunopositive surfaces from 5 fields per sample. The plot shows the median value (bar) and quartiles (box), the whiskers show the minimum and maximum values of the range. The circles show the mean overlap volume of surfaces from 5 non-overlapping field of views (*** indicates p-values <0.0002 from an unpaired two-tailed t- test). The mean overlap volume between PHF-1 and AP2A1^L was 50.7 ± 8.5 % and between PHF-1 and AP2A2 was 0.3 ± 0.4 %. F) Box and whiskers plot showing the quantification of overlap volumes between AP2A1^L [+] PHF-1 [+] and AP2A1^L [+] AP2A2 [+] double immunopositive surfaces from 5 fields per sample. The circles show the mean overlap volume of surfaces from 5 non-overlapping fields of view (**** indicates p-values <0.0001 from an unpaired two-tailed t- test). The mean overlap volume between AP2A1^L and AP2A2 is 1.2 ± 2.1 % and between AP2A1^L and PHF-1 was 66.4 ± 8.7 %.

For the colocalization analysis, Imaris software was used for 3D segmentation and surface reconstruction as shown in Suppl. Fig. 4. The analysis included an assessment of five non-overlapping images of one section from Braak VI, LOAD cases (n=2). Here we identified n=32 PHF1[+] AP2A1[+] AP2A2[−], n=1 PHF1[+] AP2A1[−] AP2A2[+], and n=1 PHF1[+] AP2A1[+] AP2A2[+] surfaces from a single LOAD case. Importantly, our analyses provided a relatively unbiased quantification of the volume colocalized between PHF-1, AP2A1^L and AP2A2 immunopositive profiles. We found a high degree of overlap between PHF-1 and AP2A1^L immunopositive profiles, as opposed to PHF-1 and AP2A2 immunopositive profiles (Fig. 6E and Suppl. Fig 5E). The overlap between AP2A1^L and AP2A2 immunopositive profiles was also poor (Fig. 6F and Suppl. Fig 5F).Taken together, our results indicate that AP2A1^L, but not AP2A2 colocalizes with pTau in LOAD brains.

As a follow-up to our IHC findings, and to analyse the expression of AP2A2 within microglial cells, we also performed double- label IF in frontal cortex tissues (Brodmann area 9) from control (n = 1, Braak I) and LOAD (n = 1, Braak V) cases using antibodies specific for AP2A2 (j. C619–656) and the microglial marker Iba-1. As seen in Fig. 7, Iba-1 was often colocalized with AP2A2 detected with the j. C619–656 antibody in particular, suggesting that the two proteins are expressed in the same cells.

Figure 7. Double-label immunofluorescence staining: Iba-1 and AP2A2.

Representative photomicrographs from human brain sections showing the localization of Iba-1 (red; A, D) and AP2A2 (green; B, E) proteins. Panel C, F show the merged photomicrographs with yellow indicating signal overlap and colocalization. These photomicrographs depict the frontal cortex (Brodmann area 9) of one control (Braak I) and one AD case (Braak V, Suppl. Table 2). The experiment was performed using antibodies specific for AP2A2 (j. C619–656) and Iba-1. Iba-1 was seen as expected in cells with microglial morphology (white arrow, panel A, D), and colocalized substantially with AP2A2 (white arrow, panels B, C, E & F). Note that overlap between Iba-1 and AP2A2 is only partial - some Iba-1 and some AP2A2 signals are not colocalized (yellow arrow heads, panels A-C). Scale bars = 50 μm.

DISCUSSION

We assessed α-adaptin proteins in the human brain, using multiple antibodies. We applied several different methods, including extensive immunoblotting and immunohistochemistry. Our data indicate that AP2A1^L, but not AP2A2 immunoreactivity, was colocalized with NFTs of LOAD patients. These findings help extend our understanding of α-adaptin proteins in the human brain and their association with neuropathology.

In the heterotetrameric AP-2 complex, the N-terminal domains of α and β2 adaptins, along with μ2 and σ2, are arranged into a core region (the ‘trunk’), while the C-terminal domains of the adaptins project from the core as two symmetrical arms (the ‘appendages’ or ‘ears’) connected via flexible linkers (the ‘hinges’) [27, 28]. The core mediates recruitment to membranes, cargos, and sorting-signal recognition. The hinge-appendage extensions interact with clathrin, other adaptors, and various accessory proteins [29, 30]. The autonomous α-adaptin appendage binds at sequence motifs found in dozens of ligands, many of which are AD risk genes [31, 32].

The neurobiology of AP-2 in healthy and diseased states is incompletely understood. The homozygous deletion of AP-2μ (AP2M1) is lethal in mice [33], while neuron-specific AP2M1 KO mice show reduced neuronal complexity and severe neurodegeneration of the thalamus and cortex [10]. AP-2 also was shown to promote neuronal branching and survival by controlling the retrograde transport of signalling endosomes containing BDNF/TrkB complexes [10].

Multiple prior studies implicate AP-2 and related proteins in AD pathogenesis. Among the cargo proteins that AP-2 has been implicated in transporting at the plasma membrane are APP, Tau, BACE1, ApoE, and CLU gene products [17]. Further, an endocytosis‐independent function of AP‐2 was described in regulating intracellular trafficking and degradation of BACE-1 and subsequent reduction in amyloidogenic processing of APP [34]. Increased amyloidogenic processing of APP in the AD brain was partially attributed to a reduction in α-adaptin levels observed in iPSC‐derived neurons from LOAD patients [34]. AP-2 and PICALM are also involved in the trafficking of amyloid precursor protein cleaved C-terminal fragment (APP-CTF) from the endocytic pathway to the LC3 marked autophagic degradation pathway [11]. AP-2 (and genetic variation at the MUC6/AP2A2 locus) has also been implicated in the brain’s vulnerability to herpesvirus infection linked to AD pathogenesis [35].

We previously reported that genetic variation in a large variable number of tandem repeat (VNTR) region in the nearby MUC6 gene was associated with altered AP2A2 expression in the brain and was associated with variation in severity of AD-type tau neurofibrillary pathology [21]. We complemented that work with IHC studies using an antibody raised against full- length AP2A2 and found that it immunolabeled NFTs in LOAD, but not tauopathy in progressive supranuclear palsy. However, in the current work, we report evidence that the same antibody we used previously actually labelled both AP2A1 and AP2A2 proteins on immunoblots -- the epitope recognized by that antibody is shared by the two different α-adaptin proteins. We now report that (to our surprise) when ten different α-adaptin antibodies were applied, only a subset of the antibodies were isoform specific and among these, two that recognized AP2A1 specifically on immunoblots also labelled the NFTs in brain sections. By contrast, the antibodies that specifically recognized AP2A2 on immunoblots hardly labelled any NFTs. A potential limitation of this study is that antibody cross reactivity and cleavage products on the immunoblots may have negatively influenced our interpretation of antibody specificity.

Interestingly, of the two AP2A2 specific antibodies, one of them distinctly labelled microglial cells in aged brains. Although antibody cross-reactivity is possible, we speculate that abnormal accumulation of AP2A2 in microglial cells might be related to dysregulation of uptake and clearance of pathogenic protein aggregates in the diseased brain. It is anticipated that future studies will involve larger cohorts of control and LOAD brains to compare the associations of AP2A2 with microglia (or a population of microglia), and AP2A1 with NFTs. These investigations will help determine if higher proportions of AP2A2 in microglia and AP2A1 in NFTs is associated with a more advanced diseased stage. Nevertheless, these results raise interesting questions regarding α-adaptin isoforms’ functions and dysfunctions in human brain.

We also found that AP2A1 & AP2A2 proteins were relatively uniformly expressed in five different brain regions according to protein immunoblots. AP2A1 & AP2A2 were both present in the insoluble fraction from cognitively impaired subject but less so in a control (non-demented, low-pathology) brain. This finding may implicate AP-2 insolubility as a pathogenic mechanism. Intriguingly, a study of brain aging in mice found that mRNA levels of endocytic proteins, including α-adaptins, are often altered in the aging brain [36]. Although AP2A1^L but not AP2A1^S was the predominantly expressed isoform in the detergent-soluble fractions of control subjects, the levels of AP2A1^L and AP2A1^S were comparable in the insoluble fractions of AD subjects. These changes may correspond to variations in translational activity, post-translational modifications, and/or an increase in the proteolytic cleavage of full-length AP2A1^L isoform.

In conclusion, these findings extend our understanding of α-adaptins -- endolysosomal hub proteins -- in the human brain in healthy and diseased states. However, there remain many unanswered questions. We speculate that our results may indicate a pathway relevant to AD pathogenesis wherein genetic variation that affects one protein (MUC6 VNTR variation that affects AP2A2 expression) may promote the incorporation of a related protein in human pathology (AP2A1 in tau tangles of AD). However, more work is required to test this hypothesis. We also need to test the related possibility that the AP2A1 localization in tau tangles is a secondary event, unrelated to tangle formation. Since α-adaptins have been implicated in autophagy and p62/SQSTM1 pathways [10, 11], it is possible that the presence of α-adaptins in NFTs are analogous to ubiquitination.

Key points:

• AP-2 is a heterotetrameric complex involved in endolysosomal trafficking and implicated in the pathogenesis of LOAD.

• The AP-2 α-adaptin subunit has two major isoforms: AP2A1 & AP2A2

• The α-adaptins are present in the insoluble fractions in the brain of LOAD subjects.

• AP2A1 but not AP2A2 colocalizes with neurofibrillary tangles in the brain of LOAD subjects.

Abbreviations:

AD

Alzheimer’s disease

AP-2

Assembly protein complex 2

APP

Amyloid precursor protein

BDNF

Brain derived neurotropic growth factor

CME

Clathrin-mediated endocytosis

CERAD

Consortium to establish a registry for Alzheimer’s disease

DIPPs

Detergent-insoluble pathogenic proteins

GWAS

Genome-wide association studies

High MW

High molecular weight

LC-ESI-MS/MS

Liquid chromatography- Electrospray ionization- Tandem Mass spectrometry

LOAD

Late-onset Alzheimer’s disease

MW

Molecular weight

ND

Neurodegenerative diseases

NFT

Neurofibrillary tangle

PHF

Paired Helical Filament

SNPs

Single nucleotide polymorphisms

VNTR

Variable number of tandem repeat