Rapid isolation of postsynaptic fractions from small PM samples

Our initial objectives were to: i) enhance sample preparation speed to improve preservation of protein complexes and allow greater numbers of samples to be analysed, and ii) increase protein yield enabling studies on small amounts of tissue. We reduced the duration of the isolation of postsynapse-enriched fractions from 6–8 h [22] to 2 hours by removing the density gradient fractionation step and maintaining the well-known non-ionic detergent insolubility of the PSD of excitatory synapses (Figure 1A, see methods). The high protein yield (2.8 ± 0.13 μg SEM, n = 35 protein in P2 fraction/mg of tissue) of this protocol enabled a 10-fold reduction in the amount of starting material permitting routine isolation of synaptic-enriched structures from as little as 100 mg of human brain tissue. Enrichment of synapse proteins in the P2 fraction was confirmed (Figure 1B,C).

A simple screening method to measure PM synapse proteome integrity

Inspection of these 20 immunoblots (Additional file 2: Figure S1) revealed that PM samples in GluN2B (2B subunit of the N-methyl-D-aspartate glutamate receptor) presented two low-molecular weight bands (see bands 2 and 3 on Figure 1D) that corresponded with GluN2B proteolysis products due to PM conditions as described previously [23]. These degradation products, which are absent from biopsy samples (NSB, Figure 1D), are already apparent 2 h after death [23]. Different PM samples showed different intensities for these three bands; for instance, some samples lacked all three bands (sample 18/07) indicating complete degradation of the receptor and some lacked only the uppermost Band 1 (samples 19/06 or 19/08) indicating partial degradation. To estimate the extent of degradation of each sample, the ratio of Band 1/Band 2 intensity was measured for all 28 samples (median ratio: 0.83). We next correlated these values with the counts of present immunoblotted PSD proteins (Additional file 2: Figure S1) and in contrast to what was observed with brain bank indicators, we found significant positive correlation (Spearman’s rho = 0.59, P <0.001, Figure 1E). These results suggest this ratio can be used to estimate synaptic proteome preservation in PM brain samples and hereafter we refer to this ratio as HUSPIR (HUman Synapse Proteome Integrity Ratio).

To validate the usefulness of HUSPIR we prospectively screened another group of nine PM samples (Selected PM Group) that were presumed to be of high quality since they were obtained from individuals who died suddenly, before the age of 60, had a short agonal state and PM interval, and showed no detectable neuropathological abnormalities. The median HUSPIR value for this group (2.43, Figure 1F) was significantly higher (p = 0.001; Mann Whitney Test) than the median HUSPIR for the initial set (Unselected PM Group, Figure 1F). As predicted by HUSPIR the selected group presented all PSD components analysed by immunoblotting (Figure 1G and Additional file 2: Figure S2). These results confirm that the simple and rapid measurement of HUSPIR, based on GluN2B band ratios, reliably predicts synapse proteome integrity. Based on these results, we propose using PM samples with a HUSPIR >1 for biochemical evaluation.

Synaptic proteome integrity varies between cortical and other brain regions

We were also interested in evaluating if postsynaptic proteome integrity was homogenous between different areas of the same brain. To address this issue we isolated postsynaptic densities (PSD) from 26 different brain regions in 4 different post-mortem brains from the ‘sudden-death brain bank collection’ (Edinburgh MRC-Brain Bank). These regions included 13 cortical areas, 5 limbic structures and 8 sub-cortical regions (Figure 2). Each isolated PSD fraction was subjected to GluN2B immunoblot (Figure 2A,B) to determine its corresponding HUSPIR value.

As summarized in Figure 2C HUSPIR values varied between cortical and other brain regions. While most HUSPIR values for cortical regions were above 1, suggesting a good preservation of its postsynaptic proteome, other brain regions showed a HUSPIR <1 in more than half of the samples analysed (34/52; Figure 2C). The fact that limbic areas anatomically close to the cortex (cingulate gyrus: BA23 and BA24) show high HUSPIR values in the 4 brains analysed gives further evidence for a better preservation of the synaptic proteome in more anterior brain regions. Finally, this data also indicates that when working with non-cortical brain areas HUSPIR should be measured for each brain sample included in the study. The HUSPIR value of one cortical region has a good predictive value for other cortical regions of the same brain, as shown in Figure 2C.

Isolation and proteomics of human MAGUK associated signalling complexes

In the next phase of our study we have combined the new methods for screening samples with well preserved synaptic proteomes with affinity methods to isolate MAGUK Associated Signalling Complexes (MASC). This affinity method relies on a 6 amino acid immobilized peptide (pep6, SIESDV) that binds to PDZ domains of MAGUK proteins and a control peptide (pep6δV, SSIESD) that cannot bind to these domains [24]. We used this protocol, which involves the use of stringent biochemical methods, to ask if it was possible to isolate MASC from 3 PM human frontal cortex samples with HUSPIR >1 (Additional file 1). As a positive control we used the same protocol on 3 NSB samples and as a further positive control, we also isolated MASCs from mouse cortex. As shown in the coomassie stained SDS-PAGE gel (Figure 3A), all samples showed strong recovery of multiple proteins with pep6, and as expected, a low background with pep6δV purifications. Immunoblotting of these samples with antibodies to three MAGUK proteins (DLG2/PSD93, DLG3/SAP102, DLG4/PSD-95) showed robust signals in pep6 samples and no signal in negative controls (Figure 3B). These results show that it is possible to affinity isolate MASCs from NSB as well as PM human brain and we hereafter refer to these isolates as hMASC.

We next identified all proteins in hMASC samples using mass spectrometry and used bioinformatics to compare PM and NSB components (Additional file 3). See Additional file 2 for detailed bioinformatics functional analysis of hMASC. In total we found 239 and 227 proteins in NSB and PM hMASC respectively. A very large proportion (>75%) of NSB hMASC proteins were found in PM hMASC complexes, demonstrating for the first time that it is possible to isolate bona fide synaptic multiprotein complexes from PM samples if these have a HUSPIR value >1. These results also indicate the applicability and robustness of the affinity isolation methods developed here to analyse human PM brain tissue. Comparison of the functional types of proteins in PM and NSB hMASC showed a high similarity (Figure 3C and Additional file 4).

Despite the high degree of similarity between NSB and PM hMASCs we reasoned that some protein types are likely more vulnerable to PM conditions than others. If this is the case it is important to identify them, as these will be underrepresented in PM hMASCs. We analysed several protein features, including protein length, isoelectric point, presence of transmembrane regions, and protein turn-over [25], in proteins isolated from both tissue sources. Overall, proteins found in NSB and PM hMASCs showed only minor differences among these variables. Only protein length was slightly biased, with proteins found in two or three triplicate samples somewhat longer in NSB hMASC compared to PM hMASC (median length 558, compared to 428 residues, respectively; p = 0.033, Mann Whitney Test). Next we compiled a list of the specific proteins with differential expression and noted this included proteins involved with activity-dependent reorganization of the synaptic proteome, including calcium-calmodulin dependent kinases (CAMK2A, B and G), scaffold proteins of the DLGAP family, G-proteins and elements of the tubulin cytoskeleton (Additional file 2: Figure S3). In contrast to these unstable components, the more stable groups included the MAGUK scaffold proteins, their associated NMDA receptors, signalling and cytoskeletal proteins, among others (Additional file 2: Figure S4).

Human MASC proteins in disease

To understand the relevance of hMASC to human disease we searched online genetics databases and the published human genetics literature for evidence of hMASC genes involved in brain disorders (see methods). Our approach involved: i) identification of relevant genes and relevant diseases, ii) allocation of diseases into specific categories according to the International classification of Disease System (ICD-10), iii) statistical analysis of the representation of disease categories, and iv) comparison of hMASC with hPSD to examine the relative importance of hMASC, as compared with the whole postsynaptic density.

According to OMIM database of inherited diseases [26] 66 hMASC genes are involved with 93 genetic diseases. Fifty-one of these are nervous system diseases with 47 specific to the CNS. Literature text-mining and curation identified CNS diseases for which hMASC genes were positively associated, bringing the total number of CNS diseases related to the hMASC to 76. Classification with ICD-10 of these CNS diseases showed they include rare monogenic disorders as well as complex, more common psychiatric and neurologic disorders, including Schizophrenia and Autism spectrum disorder (Additional file 5). We next asked whether particular disease categories were over- or under-represented in hMASC compared to the postsynaptic proteome (hPSD, see Bayés et al. [2]). Looking into inherited diseases, we found a two-fold greater proportion of hMASC genes (both NSB and PM) involved with CNS compared with hPSD genes (Figure 4A). As shown in Figure 4B, the ‘Disorders of the Nervous System’ and ‘Mental and Behavioural Disorders’ ICD-10 categories were over-represented relative to the hPSD, whereas the hPSD was more enriched in Endocrine, Nutritional and Metabolic Diseases than hMASC (Figure 4B). As hMASC was most enriched for Mental & behavioural Disorders, we examined Intellectual Disability disorders in more detail. We created a database of genes causing any form of ID (ID) and non-syndromic ID (NSID) (Additional file 6) and found a significant over-representation of hMASC and hPSD genes in both these types of ID, with NSID highest (Figure 4C,D).

Human neocortex samples

NSB samples were obtained from frontal cortex by Prof. Ian Whittle (Edinburgh University) in a study approved by the Lothian Region Ethics Committee (/2004/4/16). Informed consent was obtained from all donors. All PM tissue was obtained from the Medical Research Council Edinburgh Brain Bank (Additional file 1). The three PM samples used to purify MASC were from the frontal cortex (Additional file 1).

Rapid isolation of postsynaptic fractions from PM tissue

To facilitate characterization of the human postsynaptic proteome we developed biochemical methods based on those used in rodents [28],[29]. Approximately 300 mg of grey matter was routinely used for isolation of synaptic fractions; nevertheless, the protocol can be scaled down to 100 mg of starting material. Prior to tissue processing, white matter was removed from frozen cortex blocks. Grey matter was homogenized with a glass-teflon dounce (tissue weight: volume ratio of 100 mg:1 ml) in a buffer containing 0.32 M sucrose, 10 mM HEPES pH 7.4, 2 mM EDTA, 5 mM sodium o-vanadate, 30 mM NaF and protease inhibitors cocktail (Roche). Homogenized tissue was centrifuged at 800 × g for 15 min at 4°C; supernatant removed and centrifuged at 10,000 × g for 15 min. The pellet was resuspended in Triton buffer (50 mM HEPES pH 7.4, 2 mM EDTA, 5 mM EGTA, 5 mM sodium-O-vanadate, 30 mM NaF, 1% Triton X-100 and protease inhibitors cocktail (Roche)) with half the volume used for homogenization. Sample was then centrifuged at 30,000 × g for 30 min at 4°C and the final pellet was resuspended in 200 μl of SDS buffer (50 mM Tris pH 7.4, 1% SDS) and used for immunoblotting.

Isolation of postsynaptic densities from human post-mortem brain samples

Postsynaptic densities were isolated using standard differential ultracentrifugation methods. Briefly, dissected tissues was homogenized in buffer A (0.32 M sucrose, 2 mM HEPES, pH 7.4, EDTA-free protease inhibitor cocktail (Roche), phosphatase inhibitor cocktail set II (Calbiochem) using a Dounce homogenizer. The resulting homogenate was centrifuged at 1000 × g for 10 min at 4°C. The pellet was rehomogenized in the same buffer and centrifuged at 1000 × g for 10 min at 4°C. The two supernatants were pooled and centrifuged at 18600 × g for 15 min at 4°C. The next pellet was resuspended in buffer B (1.5 M sucrose and 50 mM Tris–HCl, pH 7.4 containing the same protease and phosphatase inhibitors as in buffer A). A discontinuous sucrose gradient was built by layering equal volumes of the resuspended pellet (bottom), 0.85 M and 0.3 M sucrose in 50 mM Tris–HCl, pH 7.4. This gradient was centrifuged at 60000 × g for 42 min using a MLA-55 rotor (Beckman-Coulter). The interface between 1.5 M and 0.85 M sucrose, corresponding to the synaptosomal fraction, was collected, diluted in 2 volumes of 50 mM Tris–HCl, pH 7.4 and centrifuged at 48000 × g for 30 min at 4°C. The resulting pellet was resuspended in 1.5% Triton X-100, 25 mM Tris–HCl, pH 7.4, incubated on ice for 30 min, layered on top of a 0.85 M sucrose solution in 50 mM Tris–HCl, pH 7.4 and centrifuged at 104000 × g for 30 min at 4°C. The resulting postsynaptic density pellet was resuspended in 50 mM Tris–HCl, pH 7.4 and 10% glycerol and mixed with SDS-PAGE electrophoresis loading buffer.

Isolation of human MAGUK Associated Signalling Complexes (MASC)

MASCs were isolated from human NSB or PM frontal cortex using an improved version of an affinity purification protocol developed with mouse brain samples [12]. Briefly, this method used a peptide corresponding to the six carboxyl terminal residues of GluN2A and GluN2B, which bind PDZ domains of Membrane Associated Guanylate Kinase (MAGUK) proteins (DLG4/PSD95, DLG2/PSD93, DLG3/SAP102 & DLG1/SAP97) [30],[31]. A negative control for the affinity purification uses a peptide lacking the C-terminal valine residue [24]. Brain tissue was homogenized (tissue weight:volume ratio of 100 mg:1.85 ml) with DOC Buffer (Tris 50 mM pH 9.0, 1% sodium deoxycholate, 50 mM NaF, 20 μM ZnCl₂, 1 mM Sodium o-vanadate and protease inhibitors cocktail (Roche). The homogenized sample was centrifuged at 50,000 g for 30 min and pellet discarded. For affinity purification, peptides were first bound to Affi-Gel 10 resin (Bio-Rad) at a 5 mg/ml concentration following manufacturers indications. Brain homogenate and peptide-bound resin were mixed at 1:100 (v/v) ratio and incubated over-night at 4°C with agitation. Prior to elution, resin was washed five times with 5 column volumes of DOC Buffer. Elution was performed with DOC Buffer containing free peptide at 5 mg/ml. Eluted sample was reduced by adding fresh DTT (SIGMA) to a final concentration of 1 mM and heating at 70°C for 10 min. Reduced sample was carboxymethylated by adding fresh iodoacetamide (SIGMA) to final concentration of 2 mM and incubated in the dark at room temperature for 30 min.

Sample preparation for quantitative MS

In gel digestion was performed as reported previously [2]. Extracted peptides were isotope labelled using a triplex dimethylation strategy [32],[33] and labelled peptides pooled for LC-MS/MS analysis. This was done as follows: Light labelling: Neg. Control (50% from biopsy and 50% from PM Pep6δV), Medium labelling: Biopsy Pep6, Heavy labelling: PM Pep6. Pooled peptide samples from each of the 12 gel bands were analysed separately by nanoLC-MS/MS on a LTQ Orbitrap Velos (Thermo Fisher) hybrid mass spectrometer equipped with a nanospray source, coupled with an Ultimate 3000 Nano/Capillary LC System (Dionex). The system was controlled by Xcalibur 2.1 (Thermo Fisher) and DCMSLink 2.08 (Dionex). Peptides were desalted on-line using a micro-Precolumn cartridge (C18 Pepmap 100, LC Packings) and then separated using a 60 min RP gradient (4-32% acetonitrile/0.1% formic acid) on a BEH C18 analytical column (1.7 μm, 75 μm id × 10 cm,) (Waters). The LTQ-Orbitrap Velos was operated with a cycle of one MS (in the Orbitrap) acquired at a resolution of 60,000 at m/z 400, with the top 10 most abundant multiply-charged (2+ and higher) ions in a given chromatographic window subjected to MS/MS fragmentation in the linear ion trap. An FTMS target values of 1e⁶ and an ion trap MSn target value of 1e⁴ was used and with the lock mass (445.120025) enabled. Maximum FTMS scan accumulation time of 250 ms and maximum ion trap MSn scan accumulation time of 100 ms were used. Dynamic exclusion was enabled with a repeat duration of 45 s with an exclusion list of 500 and exclusion duration of 30 s.

MS data was analysed using MaxQuant [34] version 1.1.1.36. Data was searched against a Human IPI sequence database (v3.68) using following search parameters: trypsin with a maximum of 2 missed cleavages, 7 ppm for MS mass tolerance, 0.5 Da for MS/MS mass tolerance, with Acetyl (Protein N-term) and Oxidation (M) set as variable modifications and carbamidomethyl (C) as a fixed modification. A protein FDR of 0.01 and a peptide FDR of 0.01 were used for identification level cut offs. Protein quantification based on calculation of medium/light, heavy/light and heavy/medium dimethylated peptide ratios was performed using razor and unique peptides. The ratio of peptide abundances between Pep6 and PepδV affinity purifications was calculated independently for each protein and replicate. For a protein to be considered as a true positive in a given replicate its enrichment ratio had to be above threshold which was the sum of the median ratio and its absolute deviation (Median + MAD) [35]. Proteins with an abundance ratio above the threshold in 2 or 3 replicates were considered true components regardless of the number of peptides they have been identified with. For proteins found only in one replicate with a ratio above the threshold, to be considered a true positive, it had to be identified with at least three peptides.

Analysis of MASC genes causing genetic nervous system diseases

Gene and phenotype entries were automatically extracted from OMIM [26] using the mim2gene file linking via their Entrez Gene identifiers. Diseases caused by MASC proteins were further classified, using information in the OMIM entry (Additional file 5), into central nervous system (CNS), peripheral nervous system (PNS), both CNS and PNS, or other. All diseases affecting the CNS, regardless of presence or absence of PNS involvement, were then grouped into chapters of the International Classification of Disease (ICD-10) developed by the World Health Organization (WHO). Those OMIM diseases that could not readily be found in ICD-10 were assigned to one of the ICD-10 chapters based on their symptoms, if possible, or classified as No-ICD 10 correspondence. Diseases in chapter VI (Diseases of the Nervous System) were divided into more specific blocks within that chapter.

To analyse the relevance of postsynaptic genes in syndromic or non-syndromic intellectual disability (ID or NSID) a list of genes known to cause these disorders was generated (Additional file 6). To create this database all genes linked to the terms “Mental Retardation” or “Intellectual Disability” were extracted from GeneCards [36] and added to a list of genes causing NSID curated from the scientific literature. Binomial statistics were used to calculate enrichment of postsynaptic components in ID or NSID.

Text-mining and expert curation of MASC gene mutation associations to human nervous system disorders

PubMed abstracts up to August 2014 were searched for Human MASC gene mutations associated to disease, using a natural language processing approach to information extraction provided in the Linguamatics I2E platform [37]. Disease and disorder ontology terms from the Medical Dictionary for Regulatory Activities dictionary (MedDRA,www.meddra.org) were combined with a flexible pattern capturing genetic mutation terms, such as mutation, polymorphism, deletion, variation, and amino acid substitution patterns such as ‘K483E’, requiring mutation terms to be local to the gene, but allowing the disease term to appear anywhere in the sentence. The resulting sentences were examined by a biologist, checking each for correct gene identity (disambiguation), relevance to human genetics and limiting to disorders of nervous system origin, using a simple accept/reject criterion. In the cases where many sentence ‘hits’ were obtained to a particular gene and disease term combination, a maximum of 10 documents were examined. Redundant disease terms, such as symptomatic terms that describe a part of a larger clinical syndrome or disease were manually removed, thus we report a single representative PubMed identifier for each gene mutation-disease association combination. We then combined the curated text-mined PubMed search results with the OMIM extraction results, to produce an overview of the relationship between reported Human MASC gene mutations and nervous system disorders.