Differentially charged isoforms of apolipoprotein E from human blood are potential biomarkers of Alzheimer’s disease

Informed consent from the patients and/or their care providers for using the samples described in this study was obtained by the Bryan Alzheimer’s Disease Research Center (BADRC) of Duke University Medical Center, the Knight ADRC (KADRC) of Washington University School of Medicine in St. Louis (MO), and the Instituto Neurologico de Colombia (Colombian Neurological Institute, INDEC). Samples from AD patients are indicated with the letter D, while non-AD samples are indicated with the letter N. Samples from APOE 3/3 or APOE 4/4 diseased subjects are indicated 33D and 44D, respectively; non-AD APOE 3/3 samples are indicated 33 N (Figure 1).

The general method is summarized in Figure 1. (A) Hippocampi from AD patients and from non-AD control subjects were analyzed to determine differential protein-expression levels. From this analysis, we found sets of proteins unique to each disease state and unique to each APOE genotype. (B) The latter was also observed in the panels of proteins that displayed differential protein expression in the patient’s CSF compared with matched controls; (C) from the distribution of CSF proteins, it was found that for some proteins there were changes in the distributions of differentially charged protein isoforms. Therefore, the distribution of differentially-charged isoforms in the blood of AD patients was analyzed by using the method shown in Figure 2, which is explained in detail under section Two-dimensional multiplexed Western blotting (2D mxWb). (D) A general pattern for the differentially charged isoform distribution of ApoE was then sought in the blood of AD patients compared with age- and sex-matched controls. (E) the distribution of charge-isoforms of ApoE was determined in a “blind” group of unknown blood samples from several patients for whom no information was provided. (F) The distribution patterns of ApoE charge-isoforms from the unknown subjects were compared with the ApoE distribution patterns from the known AD patients; and (G) the matching of the charge-state distribution to the reference patterns was used to predict the disease state of the unknown samples.

Sample procurement, protein isolation, and proteomic analysis of human brain tissues have been previously reported [19]. CSF from deceased AD patients was acquired by the Kathleen Price Bryan Brain Bank of the BADRC, by following regulations of the Duke University Medical Center, as previously described [20], and was provided to us as de-identified samples (IRB exemption 6878-05-1R0ER, from Duke University Health System). These samples were centrifuged at 14,000 rpm to remove particulates, and the supernatant was filtered with a 20-μm filter. The CSF samples were prefractionated to enrich low-abundance proteins by removing high-abundance proteins by using the Multiple Affinity Removal System (MARS; Agilent Technologies, Wilmington, DE, USA), an antibody-based 100-mm affinity column, by following the manufacturer’s instructions. CSF (30 µL) was mixed with 120 μl of buffer A and injected into the column at a flow rate of 0.5 ml/min, by using the auto-sampler included with the 1100 Multidimensional Liquid Chromatography (MDLC) system (Agilent Technologies). The flow-through, containing the low-abundance proteins, was collected, and the high-abundance proteins that remained bound to the column were eluted with buffer B at a flow rate of 1 ml/min (buffers A and B are proprietary buffers included with the MARS depletion kit, and their composition is not disclosed by the manufacturer).

Proteins fractionated by affinity chromatography were reconcentrated with Agilent concentrator spin columns (5 kDa MWCO, 4 ml). Purified proteins were prepared as previously described [19, 21]. Each de-identified sample, from patients whose ages were between 78 and 85 years, was classified only according to age, sex, APOE genotype (3/3 versus 4/4), and cognitive status (demented (D) versus nondemented (N)).

An additional group of samples used in the present study was obtained from a larger sample group previously used to study the correlation between type 2 diabetes mellitus and AD in the province of Antioquia, Colombia [22]. The samples were randomly selected from a group of 20 samples, all of which had been determined to be homozygous for APOE3[22]. The samples were collected by following norms and regulations approved by the ethics committees of the INDEC and the Universidad Pontificia Bolivariana (approved by the UPB research committee, 21/10/2010, as described in [22]). Because the methods used for this analysis involved working with human biologic specimens (that is, blood serum), an IRB exemption was filed and approved by the Office of Human Research Ethics of the University of North Carolina at Chapel Hill. The researchers involved in the present study received de-identified samples that were matched only by age and gender, and characterized as being from AD patients or from non-AD control subjects.

The selected samples were from six individuals (n = 6) who met the DSM-IVR (Diagnostic and Statistical Manual of Mental Disorders, 4th edition, American Psychiatric Association) criteria for AD diagnosis, and six control subjects who did not meet any DSM-IVR criteria for AD [22]. All AD patients were 65 years of age or older, with no relatives with an AD diagnosis or any type of dementia, and had no medical history of Parkinson disease, stroke, Huntington diseases, chronic subdural hematoma, and other conditions that could invalidate the diagnosis of AD [22]. The control group consisted of individuals older than 65 years of age of both genders, without a medical history of dementia syndrome or any other neurodegenerative disease. The individuals were matched by sex and age between the experimental and control groups.

Serum (40 μl) was aliquoted from each of the samples for protein purification. Sample clean-up used a chloroform-methanol extraction procedure [23], performed by mixing the serum sample with a 3:4:1 vol/vol mixture of d_iiH₂O, chilled methanol, and chloroform, and centrifuged for 15 minutes at 13,500 rpm. The samples were then suspended in lysis buffer [19, 21], and the protein concentration of each sample was determined by using a BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). Protein quality was determined by 1D Western blot for all samples.

Eighteen blood plasma samples, each containing 100 to 200 μl, from AD and control patients of the KADRC, were provided to us completely de-identified; the ethics committee of the KADRC, Washington University, approved all the methods involved in sample collection, preparation, and use for research. Because the samples were provided to us de-identified, an IRB exception was approved by the University of North Carolina. These samples were prepared as indicated earlier, and 10 μg of total protein from each sample was used for 2D multiplexed Western blotting (2D mxWb), as explained later.

Differential protein expression in CSF from deceased AD patients and non-AD matched controls was determined by 2D Difference Gel Electrophoresis (2D DIGE), comparing all samples and determining protein-expression differences that correlated with APOE genotype and disease state [19, 24]. Samples from 33D and 44D patients were used as the experimental groups and were compared with the 33 N control group. An internal control was created by pooling equal amounts of protein from all of the samples. This internal control was labeled with Cy2 and was applied to each gel, as previously described [19, 24]. 2D DIGE was performed and analyzed, as previously described [21].

The statistical significance of differences in protein expression was determined with GE Healthcare DeCyder 2D 7.0 software, as described [21], based on the Student t test by using P < 0.05 as threshold. Separate tests were performed comparing 33 N with 33D, 33 N with 44D, and 33D with 44D. The variability of the data and the Principal Component Analysis (PCA) were determined with DeCyder Extended Data Analysis (EDA) module [25]. The samples were compared as follows: (a) 33 N, containing six “non-AD” samples, all APOE 3/3 genotype; (b) 33D, containing six “AD” patients, all genotype APOE 3/3; and (c) 44D, containing six “AD” patients, all APOE 4/4 genotype. PCA plots of each spot map were based on the expression profile of the selected proteins.

Spots showing significant (P < 0.05) differential expression were excised from the gels, digested with trypsin [26], and identified by using MS and MS/MS on an Applied Biosystems AB 4800 MALDI TOF/TOF mass spectrometer at the Proteomics Center of the University of North Carolina at Chapel Hill, as previously described [27]. The proteins were identified by using the in-house GPS Explorer Mascot database [28].

We developed this approach to determine variability in protein isoform distributions. Protein posttranslational modifications (PTMs) indicate addition, subtraction, and/or modifications of functional groups [29, 30]. It is commonly found that such modifications alter the protein’s isoelectric point, the molecular weight, or both, making 2D gel electrophoresis an efficient method for analyzing PTMs [31]. Changes in the isoelectric point result in protein isoforms with different charges. The procedure for 2D mxWb is shown in Figure 2. Proteins from multiple samples were simultaneously separated by isoelectric focusing, as explained previously [19, 31], by using one strip per sample (Figure 2A). For the experiments described here, we used an Ettan IPGphor II (GE Healthcare) with a capacity of up to 12 strips. After protein separation by isoelectric focusing, the strips from each sample are successively loaded onto a polyacrylamide gel to run the second dimension (Figure 2B). The first strip is run for approximately 20 minutes; after this time, this strip is removed, the gel surface is rinsed with d_iiH₂O, and a second strip is loaded and run for another 20 minutes. The total number of strips that can be run in a single 2D PAGE gel depends on several variables, including (a) the apparent molecular weight of the target protein, (b) the size and the density of the gel, and (c) the potential difference used to displace the proteins. To determine specific changes in a protein’s pI distribution, we used human recombinant ApoE protein (hr-ApoE; Fitzgerald Industries International, Acton, MS, USA; cat. 30R-AA016) as a reference. This protein is separated by IEF along with the other strips, and is loaded on the polyacrylamide gel by using the same procedures as for the target proteins.After running the second dimension, proteins are transferred to a PVDF membrane, and the proteins are probed with specific antibodies (Figure 2C). For detection of ApoE, ApoE-goat Ab (CalBiochem, La Jolla, CA, USA; cat. 178479) was used as the primary Ab, and donkey anti-goat IgG-HRP (Santa Cruz Biotechnology; cat. SC-2020) was used as the secondary Ab. From the Western blot analysis, it is possible to determine the migration patterns of the charge-isoforms for multiple samples in a single experiment.

Solubilized protein samples from human brain tissue were separated with 1D SDS (12%) PAGE electrophoresis with pre-cast gels (Bio-Rad, Hercules, CA, USA). The apparent protein molecular weight was determined with “Kaleidoscope” prestained protein standards (Bio-Rad). Gel strips for each sample were excised from the Coomassie-stained gel, and each gel strip was cut into 1 × 1-mm gel pieces and transferred into Axygen 1.7-ml tubes. Manual “in-gel” protein digestion was done as previously explained [26].

Mass spectrometric data acquisition was performed in a data-dependent manner on a hybrid LTQ-Orbitrap mass spectrometer. A full-scan mass analysis on an Orbitrap (externally calibrated to a mass accuracy of <1 ppm, and a resolution of 60,000 at m/z 400) was followed by intensity-dependent MS/MS of the 10 most abundant peptide ions. Collision-induced dissociation (CID)-MS/MS was used to dissociate peptides with a normalized collision energy of 35 eV, in the presence of He bath gas at a pressure of 1 mTorr. The MS/MS acquisition of each precursor m/z was repeated for 30 seconds and subsequently excluded for 60 seconds. Monoisotopic precursor ion selection (MIPS) and charge-state screening were enabled for triggering data-dependent MS/MS scans. Mass spectra were processed, and peptide identification was performed by using Mascot ver. 2.3 (Matrix Science Inc.) implemented on Proteome Discoverer ver 1.3 software (Thermo-Fisher Scientific). All searches were performed against a curated Human data base [32]. Peptides were identified by using a target-decoy approach with a false discovery rate (FDR) of 1% [33]. A precursor ion mass tolerance of 200 ppm and a product ion mass tolerance 0.5 Da were used, with a maximum of two missed tryptic cleavages [34]. Methionine oxidation was selected as a variable modification.

Spectral counting was performed on the Mascot DAT files by using ProteoIQ: ver 2.3.02 (NuSep Inc., Athens, GA, USA). Proteins identified with LC-MS/MS were subjected to “probability-based” confidence measurements by using an independent implementation of the statistical models Peptide and Protein Prophet deployed in Proteo IQ [35, 36]. Protein hits were filtered with a probability of 0.5 and a Mascot identity with a significant score cut-off greater than 26. Identified proteins were analyzed for potential posttranslational modifications.

One hundred thirty-one spots were differentially expressed in at least one comparison as follows: 15 in the 33N versus 33D comparison; 37 in 33N versus 44D; 21 in 33D versus 44D; and 58 in N versus D (see Additional file 1). Based on spot size and intensity, 40 of these spots contained a sufficient amount of protein to be identified with MS. In total, 32 of the 40 spots were successfully identified (Table 1, Figure 3B, and Additional file 1). Protein spots 541 and 614 are pI variants (that is, charge-isoforms) of transferrin variant protein (PIR: Q53H26); spots 575 and 579 are pI variants of α1-B glycoprotein (PIR: P04217); spots 1332 and 1363 correspond to pI variants of aspartate aminotransferase (PIR: P17174); spots 1451, 1454, 1469, and 1470 represent pI variants of glyceraldehyde-3-phosphate dehydrogenase (PIR: P04406); spot numbers 1472 and 1523 are charge-isoforms of apolipoprotein J (also known as clusterin (PIR: P10909)); spots 1521 and 1527 are pI variants of ApoE4 (PIR: P02649); and spots 1912, 2067, and 2068 are pI variants of apolipoprotein A-1 (PIR: P02647). Changes in the expression levels of all of these proteins are shown in Table 1. Please note that other charged isoforms of these proteins may be present, in addition to those indicated here, but these were not selected, as they did not show changes in protein-expression levels.

Analysis of hippocampus from AD patients indicates that (a) the sets of proteins characterizing the diseased state and the APOE genotype of the AD patients represent distinctive experimental groups (PCA analysis, Figure 3C); (b) charge-isoforms of certain proteins can be differentially expressed without overall changes to the expression level of the whole protein [19]; (c) a pI variant of ApoE at pH 5.2 is found in the brains of APOE 4/4 diseased patients and not found in APOE 3/3 AD or in the non-diseased controls (Figure 4A); and (d) the ApoE charge-isoform distribution in the non-AD controls is different from the distribution found in the sample from subjects with AD. This is a significant observation about the ApoE charge-isoform distribution in the human brain, but is, unfortunately, of low value for clinical biomarker research, considering that these are brain tissues. However, further analysis examining human blood from patients with diagnoses of AD (Figure 4B) shows that observations in hippocampal specimens are closely mirrored in the blood of the AD patients, which shows a unique distribution of differentially charged isoforms found only in AD patients, and not in the non-AD controls (Red arrows in Figure 4B point out spots that are unique to each sample). These observations were confirmed in six controls and six AD patients, resulting in a distinctive pI distribution that is summarized in Figure 4B.

To determine the ability of our isoelectric isoform-distribution panel to predict AD, we performed 2D mxWb analysis of 18 de-identified samples (Figure 4C). The results of this analysis are shown in Table 2, which demonstrates that we were able to predict accurately the AD state in 16 of the 18 samples (88.9%), with only two of the samples, both of the APOE 4/4 genotype, not predicting the diagnosis. The latter could be explained because we did not have any ApoE ϵ4/4 samples in our reference panel for comparison; the possibility also exists that our prediction was correct, but dementia was not yet apparent clinically.