Dementia-linked amyloidosis is associated with brain protein deamidation as revealed by proteomic profiling of human brain tissues

Amyloidosis is the process by which malformed proteins aggregate and form insoluble plaques in body tissues, eventually leading to impaired organ function. While amyloidosis has been implicated in the pathology of more than 20 major human diseases, including several neurodegenerative disorders, the variable composition and solubility of amyloid deposits have so far prevented detailed analysis of their role in human cognitive impairment. We therefore sought to develop robust methods to facilitate proteomic analyses of human brain tissues and identify disease-associated changes in the brain proteome that might shed new light on the process of amyloidosis in human neurodegeneration. In order to accurately identify brain proteins that undergo aggregation in vivo, it is essential to prevent artifactual protein aggregation during sample processing in vitro. Since protein aggregation and structural modifications are pH-sensitive events that can be triggered by inappropriate buffer use, we first tested the efficiency of protein extraction from brain tissue samples using a range of mildly acidic and alkaline buffers (AAB, ABB and TEAB; see methods) at various different temperatures (10, 22 and 37 °C) for variable duration (1, 3 or 7 days). The extracted soluble and aggregated proteins in all preparations were measured and analyzed by LC-MS/MS. We observed that optimal protein extraction was achieved using mild detergent (2 % SDS) together with a weak acid pH buffer (AAB, pH 6.0), which generated a high total protein yield, restricted sample aggregation during processing, and protected against structural changes in vitro. These data demonstrated that via extensive protocol optimization, it is possible to conduct robust proteomic analyses of primary human brain tissues, thereby facilitating the identification of specific amyloids, changes in plaque composition, and structural modifications of brain proteins that might promote amyloidosis and/or disease progression in human neurodegenerative disorders.

Protein aggregation into amyloid plaques is a pathologic hallmark of multiple neurodegenerative diseases, but therapies directed against amyloid proteins have shown little clinical benefit in human patients so far, suggesting a critical role for non-amyloid proteins in the formation of plaques. Having optimized our protocol for extraction of both soluble and aggregated proteins from post-mortem samples of human brain, we next conducted a proteome-wide analysis of amyloid structure, relative abundance and plaque composition in human brain tissues affected by dementia. To do this, we extracted both soluble and aggregated amyloidal proteins from brain using our UC-based method, subjected the aggregated proteins to further extraction and solubilization in formic acid (FA), and separated the deamidated peptides by ERLIC fractionation and structural analysis by LC-MS/MS. Using this approach, we identified a total of 5225 ± 248 brain tissue proteins, of which 4339 ± 104 individual proteins remained after application of a stringent false-discovery rate (FDR; cutoff value <1.0 %) and the characterized proteins have been identified with at least 2 unique peptides. Overlap analysis revealed that while 1697 proteins were present in both the soluble and aggregated fractions, 159 individual brain proteins were specific to the aggregated (pellet) fraction only (Fig. 1). We observed significant enrichment of S100A9, ferritin, mitochondrial creatine kinase (U-type), and hemoglobin subunits (α, β, ε and ζ) within the protein aggregate (Fig. 2a, b), as well as increased abundance of collagen, palmitoyl-protein thioesterase 1, laminin, coronin-1A, S100-B, S100-A8, Syntaxin-binding protein 2, cathepsin G, grancalcin, and syntaxin-binding protein 2 (Additional file 1: Table S1A–D). While protein S100A9 was present in both sample types, the levels detected in the pellet fraction (protein score 5083.50 ± 172.50, emPAI 300.74 ± 0.00) were far higher than those detected in the soluble fraction (1315.50 ± 119.50; 8.81 ± 0.00), suggesting substantial enrichment of S100A9 in brain amyloid aggregates (Fig. 2, Additional file 2: Table S2 and Additional file 1: Table S1C). Similarly, the iron-storing protein ferritin was also enriched in the pelleted protein aggregate (emPAI 450.45 ± 0.00) relative to the levels detected in the soluble fraction (15.20 ± 0.00; Fig. 2b). Also enriched in the pellet fraction were the hemoglobin subunits α, β, ε and ζ, consistent with reports that elevated hemoglobin levels are associated with increased risk of neurodegenerative disease and rapid cognitive decline (Additional file 2: Table S2 and Additional file 1: Table S1D). Conversely, a distinct subset of brain proteins was more abundant in the soluble fraction than in the pellet, including neuromodulin, glial fibrillary acidic protein, myelin basic protein isoform 4, synaptosomal associated protein 25, brain acid soluble protein 1, and components of β-tubulin (Fig. 3a and b). After further separation of the soluble fraction by UC-ERLIC, we were also able to detect several different isoforms of APBB2 and amyloid-like protein APLP2, as well as APBB1, APBB1IP and SAA4. These data indicated that multiple amyloid-like protein isoforms displayed only intermediate levels of aggregation and were therefore largely absent from the pellet fraction (Fig. 4).

Accumulation of structural modifications that impair protein function is a common feature of multiple age-related pathologies including major neurodegenerative diseases. Alterations including deamidation, racemerization and glycation have all been reported to disrupt normal protein function, and deamidation in particular is known to induce major perturbations in protein structure, folding/topology and overall stability. In the current study, we observed extensive deamidation of brain proteins in the pelleted insoluble aggregate, including S100A9, ferritin, hemoglobin, creatine kinase (U-type), S100-B, collagen α-2(IV) chain, collagen α-2(I) chain, laminin subunit β-2, dystonin (isoform 3), and serine/threonine-protein kinase (isoform 2) (Additional file 2: Table S3 and Additional file 1: Table S1E). We also detected deamidation of proteins coronin-1A and syntaxin-binding protein 2, which have previously been implicated in neurodegeneration of the hippocampus. Importantly, the presence of deamidated proteins was not restricted to the aggregated pellet, since the soluble brain proteome also exhibited extensive deamidation of creatine kinase B. In parallel, we observed that mitochondrial creatine kinase was highly enriched in the pellet and deamidated exclusively within the aggregate, suggesting that organelle dysfunction and compromised energy metabolism may be a feature of brain plaque formation and neurodegeneration.

Many specific deamidation sites were exclusively identified in the aggregated proteins, including modifications of S100A9, ferritin, hemoglobin subunits, S100-A8, S100-B, collagens, mitochondrial creatine kinase (U-type), β-tubulin and laminin. The deamidation sites of protein S100A9 that were commonly identified in Mascot and Maxqunat were listed in Table 1. These specific structural modifications may have functional consequences that influence the progression of neurodegenerative diseases. For example, the inflammation-associated calcium binding protein S100A9 incorporates two EF-hand motifs that exhibit differential affinity for Ca⁺⁺ binding, and deamidation of both motifs was detected exclusively in the aggregated fraction (Fig. 5). In contrast, only low levels of S100A9 were detected in the soluble fraction and the protein was not deamidated in this sample. These data suggest that deamidation and aggregation of brain proteins that regulate inflammatory processes may contribute to the pathology of human neurodegeneration. While these data suggested that protein deamidation and aggregation are key features of brain plaque formation, we also detected the restorative enzyme PIMT (ProteinL-isoaspartate O-methyltransferase) within the pellet aggregate, perhaps indicating a failed response to repair disrupted protein functions. In vitro incubation of protein samples at various different temperatures (10, 22 and 37 °C) for variable duration (1, 3 or 7 days) in different buffers like AAB (pH 6), ABB (pH 8) and TEAB (pH 8.5) revealed precipitation of protein except acetate buffer (pH 6). As alkaline condition is known to induce protein deamidation, this observation may suggest that deamidation causes aggregation of proteins.

Sequencing-grade modified trypsin (V5111) was from Promega (Madison, WI, USA). Protease inhibitor cocktail tablets (cOmplete) were from Roche (Basel, Switzerland). All other chemicals and reagents were from Sigma–Aldrich (St. Louis, MO, USA) unless otherwise stated. Post-mortem brain tissue samples of a 69-year-old male diagnosed with subarachnoid hemorrhage and dementia issues were obtained from the brain bank of the Choju Medical Institute of the Fukushimura hospital (Toyohashi, Aichi, Japan), and the protocols utilized were approved by the local ethics committee of the Fukushimura hospital. The scientific use of human material was conducted in accordance with the Declaration of Helsinki, and informed consent was obtained from the guardians of the patients [48]. All procedures were approved and performed in accordance with the ethical guidelines of the Nanyang Technological University ethics board. The tissue was first dissected, the large blood vessels removed, and the remainder cut into small pieces and homogenized. In initial experiments, we used 12.5 ml ice-cold homogenization buffer (2 % w/v SDS, 20 mM Tris–HCl, pH 7.4) supplemented with Roche cOmplete protease inhibitor cocktail. In later experiments, protein sample processing was performed under mildly acidic conditions (2 % SDS, 50 mM ammonium acetate buffer, pH 6.0) in order to minimize deamidation artifacts and eliminate false positive identification [49]. After tissue homogenization, cell debris was removed by centrifugation at 3000 × g, 10 °C for 10 min, then the supernatant was ultracentrifuged at 112000 × g, 10 °C for 1 h. The resultant supernatant was collected and subjected to two further rounds of ultracentrifugation at 112000 × g, 10 °C for 1 h. The amyloid-enriched pellets generated by these two ultracentrifugation steps were combined and re-suspended in 12.5 ml mild acid buffer (2 % SDS, 50 mM ammonium acetate buffer (AAB), pH 6.0) and then mixed and ultracentrifuged at 112000 × g, 10 °C for 1 h. The supernatants generated by each of the ultracentrifugation steps were pooled for later analysis. The pellet containing the insoluble amyloidal plaque was re-suspended in 1 ml formic acid (FA; 70 %), then vortexed for 2 min, and centrifuged for 15 min at 18000 × g, 4 °C. To ensure complete dissolution, the final sample pellet was re-suspended in 1 ml FA (99 %), vortexed for 2 min, and centrifuged at 18,000 × g, 4 °C for 10 min. To support protein deamidation causes protein aggregation, in vitro extracted protein were incubated in buffers like ammonium acetate buffer (AAB, 50 mM, pH 6.0), triethylammonium bicarbonate (TEAB, 50 mM, pH8.5), ammonium bicarbonate buffer (ABB, 50 mM, pH 8.0), at different temperatures (10, RT (22 °C) and 37 °C) and different durations (day 1, day 3 and day 7).

Proteins were reduced with dithiothreitol (10 mM), alkylated using iodoacetamide (55 mM) and then subjected to overnight digestion in sequencing-grade modified trypsin at 37 °C. Tryptic peptides were fractionated by ERLIC separation on an amine-column using a Shimadzu Prominence UFLC system. Mobile phase A (85 % acetonitrile, 0.1 % acetic acid) and phase B (10 % acetonitrile, 0.1 % FA) were used to establish a 60 min gradient. The eluted fractions were collected at 1 min intervals, pooled into 26 separate fractions for each sample, then dried and reconstituted in 0.1 % FA.

Peptides were separated and analyzed on a Dionex Ultimate 3000 RSLC nanoLC system coupled to a Q-Exactive apparatus (Thermo Fisher, MA). Approximately 5 μl sample was injected into an acclaim peptide trap column via the autosampler of the Dionex RSLC nanoLC system. The flow rate was set at 300 nl/min. Mobile phase A (0.1 % FA in 5 % acetonitrile) and mobile phase B (0.1 % FA in acetonitrile) were used to establish a 60 min gradient. Peptides were then analyzed on a Dionex EASY-spray column (PepMap® C18, 3um, 100A) using an EASY nanospray source at an electrospray potential of 1.5 kV. A full MS scan (350–1600 m/z range) was acquired at a resolution of 70,000 at m/z 200, with a maximum ion accumulation time of 100 ms. Dynamic exclusion was set to 30 s. Resolution for MS/MS spectra was set to 35,000 at m/z 200. The AGC setting was 1E6 for the full MS scan and 2E5 for the MS2 scan. The 10 most intense ions above a 1000 count threshold were selected for HCD fragmentation, with a maximum ion accumulation time of 120 ms. An isolation width of 2 Da was used for the MS2 scan. Single and unassigned charged ions were excluded from MS/MS. For HCD, normalized collision energy was set to 28. The underfill ratio was defined as 0.1 %.

Raw data files were converted into the mascot generic file format using Proteome Discoverer version 1.4 (Thermo Electron, Bremen, Germany) with the MS2 spectrum processor for de-isotoping the MS/MS spectra. The concatenated target-decoy UniProt human database (sequence 88 473, downloaded on 29 November 2013) was used for data searches. The database search was performed using an in-house Mascot server (version 2.4.1, Matrix Science, Boston, MA) with MS tolerance of 5.1 ppm and MS/MS tolerance of 0.02 Da. Two missed trypsin cleavage sites per peptide were tolerated. Carbamidomethylation (C) was set as a fixed modification, while oxidation (M) and deamidation (N and Q) were variable modifications. Annotated MS/MS spectra of Mascot-detected deamidated peptides were exported and showed in Supplementary data (Annotated MS_MS spectra _Mascot). Data were then searched again using the same criteria in MaxQuant (version 1.5.2.8) [50] to ensure maximum confidence protein identification, quantification and assessment of deamidation status. In Mascot, label-free quantitation (LFQ) of proteins uses emPAI [51], which is based on spectral counting, whereas LFQ in Maxquant is based on the ion intensities of the extracted ion chromatogram [52]. For the first search, peptide mass tolerances were 20 ppm and 4.5 ppm respectively, whereas for MS/MS the threshold was 20 ppm with FTMS de-isotoping enabled. The search was performed in revert decoy mode with PSM FDR, protein FDR and site decoy fraction set at 0.01. Annotated MS/MS spectra of Andromeda-detected deamidated peptides were exported and showed in Supplementary data (Annotated MS_MS spectra_Maxqunat). Proteins were considered for further analysis only when identified by both software packages as displaying differential abundance in the soluble and aggregated fractions, and featured deamidation sites that were consistently identified by both search engines.

Post-mortem brain tissue samples were obtained from the brain bank of the Choju Medical Institute of the Fukushimura hospital (Toyohashi, Aichi, Japan), and the protocols utilized were approved by the local ethics committee of the Fukushimura hospital. The scientific use of human material was conducted in accordance with the Declaration of Helsinki, and informed consent was obtained from the guardians of the patients. All procedures were approved and performed in accordance with the ethical guidelines of the Nanyang Technological University ethics board.

Not Applicable.

Raw mass spectrometry data and quantification results from MaxQuant (including protein and peptide identification) have been deposited to the ProteomeXchange Consortium [53] via the PRIDE partner repository with the dataset identifier PXD002516.