Comprehensive Quantitative Profiling of Tau and Phosphorylated Tau Peptides in Cerebrospinal Fluid by Mass Spectrometry Provides New Biomarker Candidates

INTRODUCTION

Protein structure and function is modulated by post-translational modifications, the most common modification being protein phosphorylation at serine, threonine, and tyrosine residues. It is estimated that one third of all cellular proteins are phosphorylated to some degree, and these phosphorylation events are involved in many regulatory functions including, but not exclusive to, structural integrity, signal transduction, differentiation, and proliferation, metabolism, and degradation [1]. The reversible phosphorylation of a protein substrate is continuously modulated by protein kinases and phosphatases, and dysregulation of this tight control is a feature of many diseases. In cancer, where signaling pathways involved in cell cycle control, such as apoptosis and proliferation, are disrupted, there is a long history of understanding the role played by abnormal phosphorylation, which has led to many new drug targets [2]. The modulation of phosphorylation events in disease is therefore a key area of research.

Aberrant phosphorylation of a number of proteins is also characteristic of Alzheimer’s disease (AD) molecular pathogenesis. AD is a progressive neurodegenerative disorder accounting for more than 60% of all dementia cases, and affecting an estimated 47 million individuals worldwide [3]. The pathological hallmarks of AD include plaques, comprised of the amyloid-beta (amyloid-β) protein [4], and neurofibrillary tangles (NFTs) containing hyperphosphorylated forms of the microtubule associated protein tau in specific brain regions [5]. Tau exemplifies the aberrant phosphorylation in AD as the protein molecules become hyperphosphorylated and in this hyperphosphorylated state, tau aggregates into paired helical filaments (PHF), the main constituent of the pathological NFTs [5]. This altered phosphorylation is likely to be the result of dysregulation of the kinases and phosphatases that modulate the phosphoproteome [6–8]. The analysis of the global phosphoproteome alterations in AD (and other diseases) is therefore an important area of research that needs to be explored.

Mass spectrometry (MS) based approaches are ideal for detecting phosphorylation events. Liquid chromatography–tandem MS (LC-MS/MS) based workflows have been employed to identify proteins and their phosphopeptides from a range of sample types. Phosphopeptides only represent a small proportion of peptides generated by tryptic digestion of a complex sample, therefore pre-fractionation and enrichment techniques are required to reduce the complexity and enhance the detection of low abundant proteins and phosphorylation events alongside the more easily detectable high abundant proteins.

The guidelines published by the National Institute of Health [9] indicate that ideally biomarkers should be present in readily accessible samples such as plasma or CSF, and also should be directly derived from the diseased tissue. A complicating issue is that such potential biomarkers are present in very low abundance in CSF and even lower amounts in plasma. Furthermore, the broad concentration span of plasma and CSF proteins and the predominance of plasma albumin and other high-abundant proteins limits the detection of novel biomarker candidates that may aid diagnosis of disease, help monitor effects of treatment in trials and track disease progression. Some studies using MS [10] or HPLC separation of CSF proteins followed by SDS-PAGE and western blotting using different tau antibodies [11], indicate that tau in CSF is processed to shorter fragments before being released to the CSF, with N-terminus and mid-domain of the protein being most abundant. However, further studies using approaches with high analytical sensitivity are needed to fully characterize CSF tau, especially its phosphorylated forms, to fully explore the biomarker potential of phosphorylated tau species in the CSF. Currently, the CSF proteins widely accepted as the core biomarkers for AD include total tau levels and tau phosphorylated at threonine (Thr) residue 181 (pThr181) [12]. Despite the wide use of pThr181 as a diagnostic tool, the phosphorylation of tau in the AD brain is known to extend far beyond this one site of interest, with hyperphosphorylation of tau in the brain characteristic of disease. There are also neurodegenerative diseases other than AD (e.g., progressive supranuclear palsy and some forms of frontotemporal dementia) that are characterized by NFT pathology but intriguingly, these are most often normal in regards to pThr181 tau concentrations in CSF [13]. Potentially, other tau phosphoforms are released into the CSF in these diseases. Previous attempts to use MS workflows to detect tau species in the CSF have proven problematic, with limited observations of phosphorylated tau peptides detected [10, 14]. The challenge is how to find these low abundant peptides in a biofluid with a high protein dynamic range, and hyper-abundant proteins such as albumin. To answer this question, we have applied a novel Tandem Mass Tag^® (TMT^®) MS approach. TMTcalibratortrademark is a tissue-enhanced workflow, combining tissue and peripheral fluidproteomics to allow for the broadest peptide coverage of analytical samples. It was developed to detect and effectively quantify the expression of low abundant peptides and proteins in complex biological samples, including CSF. The approach combines high pH reverse phase chromatography to fractionate the sample, TMT^® isobaric labeling to allow six samples to be analyzed simultaneously [15] alongside a 4-point tissue calibrator, two separate phosphopeptide enrichment techniques in parallel, and increased liquid chromatography separation using a 50 cm column and long MS gradient (Russell CL, under review). In addition to analyzing phosphopeptides enriched by two separate techniques, we also analyze the un-enriched (predominately non-phosphorylated) peptides for comprehensive profiling of a tissue or peripheral fluid sample. In order to enhance the MS data acquisition of the low abundant peptides in the analytical samples of interest, the 4-point calibrator is incorporated into each analysis, made possible by the use of TMT^® labels. TMTcalibratortrademark has been designed to ensure the calibrator proteins dominate the overall protein content of the TMT^®10plex sample. As we are limited by the amount of protein that can be injected onto the analytical LC column, a balance between dominant calibrator and analytical samples must be achieved. We designed the 4-point TMTcalibratortrademark to be spiked into the total combined analytical samples at a ratio of 1:4:6:10. These are arbitrary values allowing us to determine a suitable protein load from the six samples for analysis. The six individual CSF samples are then combined, after separate TMT^® labeling, at a 2x:2x:2x:2x:2x:2x ratio, falling within the 4-point calibrator range. The overall dominance of the calibrator signals in the TMT^® 10plex ensures the vast majority of peptides detected are from tissue-derived proteins. During MS acquisition, where a peptide is present both in the calibrator tissue and the body fluids we obtain both the peptide sequence and quantitative TMT^® reporter ion signals from all 10 samples within the same MS/MS spectrum. This allows us to detect peptides that would otherwise be lower than the limit of detection for standard data-dependent acquisition, and results in a list of peptides that are common to the calibrator and the analytical samples.

To investigate the tau species present in the CSF, we included a TMTcalibratortrademark approach with phosphopeptide enrichment. For this study, we used postmortem brain tissue from AD patients with known Braak V/VI classification, and analyzed three CSF samples from non-AD controls and three CSF samples from biochemically confirmed AD patients. The aim was to use the proteins present in the brain tissue (the calibrator) to trigger the MS/MS acquisition allowing detection of any tau peptides present in the CSF samples that are common to the AD brain material. In addition to detection, the use of TMT^® labeling also allowed us to determine any regulation of the tau peptides detected in the CSF between AD patients and control individuals, albeit in a small cohort.

MATERIALS AND METHODS

RESULTS

TMTcalibratortrademark with phosphopeptide enrichment provides broad coverage of the analytical samples of choice with the unique ability to match tissue and body fluid expression of identical peptides and proteins. In this study, over 34,000 unique un-phosphorylated peptides were observed. In addition, more than 11,000 unique phosphopeptides were detected and quantified in our dataset, mapping to more than 11,500 phosphosites. These peptides correspond to over 19,000 proteins, which match in excess of 4,000 protein groups.

We explored this extensive dataset for the presence of peptides aligned to the sequence of human tau 441 (accession number P10636-8). Considering all tau peptides we achieved 86% coverage of the protein. These peptides matched across the entire sequence of the protein, including peptides at both the C and N-termini (Fig. 1). In total, we detected 168 unique tau peptides including 115 phosphopeptides, present in at least one of the CSF samples or in the brain tissue analyzed. To determine which of these are common to both the brain material and the CSF samples, the dataset was filtered based on the presence of a signal in the individual TMT^® reporter ion channels. When missing values were removed, we found 87 tau peptides covering 76.8% of tau 441, that were present in all 10 TMT^® channels, meaning these peptides were detectable in all six CSF samples analyzed, and in the 4-point brain lysate calibrator. Peptides were detected spanning the majority of the sequence, including the C- and N-termini, and the middle region. Residues 127–155 and 291–321 were the main regions under-represented in the dataset, but this is likely due to a high number of tryptic cleavage sites resulting in very small peptides that are not detectable. We also identify a large peptide that is specific to isoform 1 of tau (758 residues, accession number P10636-1)indicating the presence of this tau isoform in the CSF. 47 of the 87 peptides detected in all 10 channels were phosphorylated (Table 1), representing the most comprehensive detection of phosphorylated tau in CSF to date. Phosphorylation was seen at multiple sites (Table 1), with singly and multiply phosphorylated tau peptides present. In total we observed 31 different tau phosphosites that were annotated with a high confidence score (>75%) by the PhosphoRS tool embedded within the Proteome Discoverer software, 20 of which were present in all of the samples analyzed (as shown in Table 1 alongside peptide sequence). Phosphosites annotated with a question mark (?) could not be confidently assigned to a single amino acid residue within that peptide sequence. Site localization within the phosphopeptide is therefore ambiguous in these instances.

A benefit of using TMT^® to analyze multiple samples simultaneously is that the relative levels of peptides and proteins detected between sample groups can be accurately determined. This allows us to compare the levels of the tau peptides and phosphopeptides detected in the CSF of non-AD individuals with the levels seen in the AD CSF patients. Following bioinformatics processing, a log2 fold change is attributed to each peptide and protein detected (see methods) and using this parameter to filter the data, we found, as expected, an overall increase in total tau levels in the CSF of AD patients compared to the non-AD control samples analyzed. Overall, a 40% increase was seen, based on the log2 fold change of all un-phosphorylated tau peptides (n = 40) (Fig. 2A) We also include the levels of tau measured by ELISA from these samples, that were used (alongside other measurements) to designate patients as AD or non-AD (Fig. 2B). Additionally, 11 phosphopeptides were upregulated by at least 40% in the CSF of AD patients compared to controls. The significance value of this regulation is also incorporated into the data sheet produced, and this can also be used to further filter the data. The most significant of the tau phosphopeptides regulated in AD compared to controls was a doubly phosphorylated peptide, phosphorylated at Ser231 and Thr235 (Fig. 3A, B) and a singly phosphorylated peptide, phosphorylated at Ser64 (Fig. 3C, D) (numbering according to tau441).

The AD biomarker literature largely focuses on tau phosphorylation at Thr181, and measurement of this phosphorylation site by ELISA is one of the three well-established CSF biomarkers. We detect five different peptides phosphorylated at Thr181 in the whole dataset, including singly (Fig. 4A, B) and multiply phosphorylated peptides (Fig. 4C, D). Summing the overall change of all unique pThr181 peptides observed results in the expected increase in AD CSF levels compared to non-AD controls (Fig. 4E). We also include the levels of pThr181 ELISA measurements from these samples (Fig. 4F), that were used (alongside other measurements) to biochemically classify the patients. Taking the same approach of summing the signal intensities of all of the peptides at each of the phosphosites, we can generate an overall tau phosphorylation site map, and see phosphosites that are regulated in AD CSF(Fig. 5).

DISCUSSION

Tau in the CSF is believed to be of diagnostic importance, with regard to AD and neurodegeneration,with several studies suggesting it may be able to predict the rate of disease progression [21]. Extensive efforts have been made to further characterize the nature of CSF tau, with previous studies largely utilizing immuno-detection methods, relying on anti-tau antibodies [11, 12]. Immuno-detection approaches have had limited success in characterizing the tau in CSF, as the methods are limited by the availability, specificity, and sensitivity of the antibodies. Furthermore, tau is known to be heavily modified by post translational modifications (PTMs), with acetylation, phosphorylation, and glycosylation events on numerous residues reported within the protein sequence [22]. These PTMs can interfere with the detection of tau by anti-tau antibodies and therefore impact on the success of immuno-detection based approaches. MS provides an ideal analytical tool to overcome the specificity limitations of antibody-based studies, but lacks sufficient sensitivity for comprehensive profiling in a complex sample such as CSF. Here we have designed and implemented a new MS based method to extensively characterize tau in both AD brain tissue and CSF simultaneously to determine which tau species are present in the CSF. Using this innovative approach, we have detected tau and phospho-tau peptides in the CSF derived from both central and N- and C- terminal regions and, in addition, by utilizing TMT^® labeling, observations of differential regulation of tau peptides in AD compared to non-AD controls have been made. Here we used TMT^® labeled peptides derived from postmortem brain material to enhance the intensity of the equivalent CSF peptides and drive the MS/MS acquisition, allowing the detection of low abundant peptides common to the brain and CSF. Although this TMTcalibratortrademark study was not specifically targeting tau in the AD brain or the CSF, nor enriching for tau peptides upstream of MS analysis, peptides covering 86% of the full tau 441 protein were detectable. 87 different tau peptides common to the brain and the CSF were observed, almost half of which were found to be phosphorylated (n = 47) representing the most extensive characterization of CSF tau reported to date. These peptides covered multiple phosphosites, including but not limited to, the widely studied Thr181 residue. Phosphorylation of tau at any of the sites reported here could be useful diagnostic biomarkers, and could also be used to monitor the efficacy of novel drugs targeting aberrant tau phosphorylation, for example kinase inhibitors. Additional tau peptides were also measured but were either exclusive to the brain material, or not present in all six CSF samples.

Previous work by groups investigating tau in the CSF has had mixed results, largely due to the differences in sample preparation and up-front processing. Studies using MS to detect CSF tau have demonstrated some success [10, 14], and the most recently reported work identified 18 non-phosphorylated peptides in the CSF using a two-step pre-fractionation workflow and monitoring peptides using a spiked heavy labeled recombinant tau protein [23]. Until now, this is the most successful attempt at characterizing tau in the CSF. TMTcalibratortrademark has been designed to avoid immuno-enrichment steps, and does not require depletion or filtering of the samples. The combination of fractionation, enrichment for phosphopeptides, analysis of both enriched and un-enriched material and a long LC gradient has resulted in an extensive coverage of the peptides and proteins present in the AD brain and CSF.

A further benefit of this approach is the use of isobaric TMT^® labeling. This has allowed for the simultaneous analysis of six CSF samples in one experiment, and allows relative quantitation of tau peptides and phosphopeptides in the CSF of AD compared to non-AD controls to be made. As expected, the overall tau levels were elevated in the CSF from AD patients relative to control, non-AD individuals. We observed a 40% increase in total tau levels, and this is in line with previous reports using MS to analyze tau levels [14] but lower than the increase found using immunoassays [24]. The levels of 11 tau phosphopeptides were also seen to significantly increase in AD by at least 40%, including phosphorylation at Thr181. The detection of multiply phosphorylated peptides around the Thr181 residue leads us to question the accuracy of the reported immuno-based approaches commonly used to measure phospho-tau levels, as phosphorylation at multiple sites is likely to interfere with antibody-antigen binding. Although this work is preliminary, and the results need to be replicated in a larger study, it may be worth developing sensitive assays for fragments of tau with phosphorylation sites other than p181 and p231, both of which have been extensively studied usingELISA.

The regulation of the phosphorylation events on tau also showed a relative decrease at some phosphosites, most pronounced on Ser285 and Ser289, and down regulation at phosphosites has recently been reported in another study [23]. We hypothesize that this could be the result of extensive aggregation of the tau proteins in AD resulting in reduced levels in the CSF. Alternatively, it is known that O-GlcNAcylation and glycosylation of tau results in a reciprocal down regulation of phosphorylation [25], so it may be that modifications to the tau peptide sequence that are not captured in this study, such as glycosylation, are either directly or indirectly affecting the levels of phosphorylation at certain sites.

Monitoring an extensive array of tau phosphorylation sites in CSF will likely gain importance, particularly as our understanding of their individual and combined roles in disease etiology increases and drug development activities shift away from anti-amyloid strategies to focus on tau and ways to prevent tangle formation where there are a number of experimental molecules currently in development [26].

The ability to characterize tau in the CSF is not only important in AD biomarker research, but could also have significant impact in the discovery of biomarkers for other tauopathies. The TMTcalibratortrademark approach could be used to detect the tau species present in the CSF of patients with other neurodegenerative tauopathies, as designed here, or also designed to directly compare CSF from different diseases, for example, AD CSF compared to frontotemporal dementia (FTD) CSF. There is a need for distinct CSF biomarkers of different tauopathies as the ELISA CSF total tau measurements cannot distinguish AD from some other neurodegenerative diseases; the specificity is only approximately 50% when comparing AD to FTD or vascular dementia [27], and it is considered a general marker of axonal damage, as seen for example in stroke [28], Creutzfeldt-Jakob disease [29], and traumatic brain injury [30]. The ability to detect and relatively quantify phosphorylated tau in the CSF is of great benefit as it is likely to be differences in the PTMs of tau that help to better distinguish AD and other dementias. In contrast to the poor specificity of total CSF tau, phosphorylation at the widely reported Thr181 residue is able to differentiate between AD and FTD [12, 27], and it is reasonable to think that alternative phosphosites will be of equal importance in tauopathy and neurodegenerative disease biomarker discovery.

One limitation of this study is the small sample set used. Confirmation of these findings will require analysis in a larger cohort of CSF samples, and experimental work is already underway to validate the preliminary findings presented here.

In conclusion, we present here the most comprehensive characterization of CSF tau in AD using a new MS technique designed to enhance the detection of low abundant proteins in peripheral bodily fluids using an AD brain tissue trigger and numerous phosphorylation sites on tau could be detected and quantified in CSF, many of which have not previously been observed. Furthermore, the unique experimental design has allowed for differential regulation of these phosphosites between AD and non-AD CSF to be investigated. The results indicate that tau in the CSF is extensively modified by phosphorylation events and further studies are warranted to determine the value of these modifications as potential biomarkers of disease. Analysis of further samples is necessary to confirm disease association, and these measurements are ongoing in a more extensive cohort at the time of submission of this manuscript.

While this report has focused exclusively on tau, this was an unbiased study and there were over 100 other proteins representing multiple facets of AD pathology that were regulated in the disease group CSF samples. Based on these preliminary findings, the same TMTcalibratortrademark approach can be used to provide important pharmacodynamic information pertaining to new therapeutic compounds during pre-clinical development and to validate translational biomarkers in early human studies. The deployment of TMTcalibratortrademark as a systems biology tool could therefore offer significant benefits as we strive to find better therapeutic strategies to combat dementia.