Identification of novel CSF biomarkers for neurodegeneration and their validation by a high-throughput multiplexed targeted proteomic assay

There is a growing need for new and more specific biomarkers in neurodegenerative diseases. The development of new tests will allow us to diagnose these conditions more specifically, and preferentially pre-symptomatically, if we are going to develop new therapies and test their efficacy. For this to occur, there is a need for a more efficient process of taking any potential biomarkers discovered in hypothesis generating experiments and developing them into translational tests suitable for rapid clinical evaluation. We believe this work describes an effective and streamlined process of improving the bottle neck of validating and getting new potential biomarkers to the clinic.

Markers for LBD are in need due to the difficulty in discriminating LBD from AD clinically. Using Label Free quantitative proteomics on LBD CSF we were able to identify several biochemical processes and pathways that were disrupted or perturbed in LBD CSF. Proteins with the potential for diagnostic use were developed into a 10 min, multiplexed, targeted MRM-LC-MS/MS test using a triple quadrupole mass spectrometry based platform (UPLC-MS/MS). The rationale behind the basing of translational tests onto a UPLC-MS/MS as a platform, is that this multiplex targeted proteomic methodology has been proven to be previously beneficial in dementia research [17, 18]. Targeted MRM based proteomic tests are gaining interest in the clinical research field [19, 20] due to its potential to measure multiple markers in one assay and does not have the cost or problems often encountered with the use of antibodies [19, 21]. The assay described in this study had been developed to streamline this method considerably creating a single a ‘one pot’ test with a 10 min LC gradient, with consequent high throughput capability and potential for clinical translation. We validated the test and markers on a larger multicentre cohort of samples including AD, LBD and controls. We were also able to include a cohort of PD CSF samples to compare with the groups and identify markers for overall neurodegeneration. Lumbar puncture is not part of the typical clinical workup for the diagnosis of PD and is usually carried out on research purposes only, hence samples are harder to obtain but were included to highlight the importance of these results in comparison with the other neurodegenerative diseases. Fig. 8 summaries all the CSF biomarkers that were specific for each neurodegenerative condition and those common to each. Only 4 biomarkers detected were specific for AD (Fig. 3) with ENPP2 (autotaxin) having not been described previously as a potential CSF marker of neurodegeneration. Using genetic studies, ENPP2/autotaxin has been described by Umemura et al. [22] as being expressed differentially in the frontal cortex of Alzheimer-type dementia (ATD) patients compared with those of non-AD controls but not the actual protein levels in brain or CSF. ENPP2/autotaxin is a lysophospholipase that acts on lysophosphatidylcholine converting it into lysophosphatidic acid (LPA). LPA is a potent mitogen and has been described previously as having a potential contributory role in AD pathology including Aβ42 formation [23] increased Tau-phosphorylation and neurite retraction in neuronal cells [24]. Indeed Hwang et al. have shown that LPA metabolism maybe a potential target for AD therapy by the use of gintonin as a LPA receptor activating ligand. Gintonin, an extract from ginseng, was shown to attenuate amyloid plaque formation and memory impairment in a transgenic mouse model of AD [25]. The other AD specific markers orexin, and LAMP1 have been either described previously or postulated as being potential CSF biomarkers of AD and thus provide confirmation and validation of the methodology developed in this work [10, 26]. We also included the recently described AD related microglia and macrophage membrane-bound receptor TREM2. Several TREM2 mutations have been identified recently that increase the risk of AD, frontotemporal dementia, PD, and amyotrophic lateral sclerosis [27]. Whilst a lot of research has been performed on the effect of TREM2 variants little has been done to investigate the normal function of TREM2. This study shows that CSF levels of TREM2 are increased in LBD and AD. CSF TREM2 has not been described elevated in dementia. However it has been described in the brain pathology of AD patients [28] and in the CSF of patients with multiple sclerosis and CNS inflammation [29]. This indicates that TREM2 may be an overall neuro-inflammatory marker. ENPP2 (autotaxin) and transthyretin were the most statistically significantly elevated biomarkers (Table 1) and provided complete discrimination between the AD and LBD cohorts. The use of these two biomarkers may provide significant diagnostic value for help in deciding treatment regimens by allowing better discrimination of AD from LBD. Transthyretin has been shown previously to be present in amyloid plaques and our data confirms previous findings of a correlation with Aβ42 levels [16]. However transthyretin is a protein known to be susceptible to post-translational modifications due its free thiol group (particularly in CSF) [30]. This makes reliable measurement of intact transthyretin using antibodies a potential problem and may be a reason as to why some studies have reported reduced levels of transthyretin in AD CSF [16, 31] whilst our study and another targeted proteomic study [17] have reported increased levels. Apolipoprotein E has been studied as a potential biomarker previously in dementia but due to only small changes or ambiguity between studies, the levels of apolipoprotein E have not been considered diagnostically useful. The differences in methodologies between mass spec based assays and immunoassays may explain why our study finds higher levels of apolipoprotein E in dementia compared to others [32, 33].

In addition to identifying 4 potential AD specific biomarkers, a further 19 biomarkers demonstrated a statistically significant elevation in the LBD and AD cohorts versus the control groups. Of these 19 biomarkers, carboxypeptidase E, malate dehydrogenase, serum amyloid A4, GM₂ activator protein, UCHL1 and prosaposin, had never been described previously as potential CSF biomarkers for neurodegeneration. The proteins prosaposin and GM₂ activator protein are both chaperone proteins critical for the lysosomal catabolism of glycosphingolipids. A deficiency in GM₂ Activator Protein has also been associated with the neurological conditions of Tay Sachs and Sandhoff disease. Although, the reason for their elevation in the CSF of LBD and AD patients is unclear, it is probably as a result of generalised lysosomal dysfunction as both these proteins have been detected by the authors in the urine of many different lysosomal storage diseases [34]. The proteins carboxypeptidase E and UCHL1 are both processing enzymes involved in the correct production of neuropeptides/neurohormones and degradation of ubiquitin monomers in the proteasome, respectively. UCHL1 is highly specific to neurons and its expression has been associated with AD, though it has also been shown to interact with α-synuclein another protein implicated in the pathophysiology of PD. Although our data did not show a significant elevation of UCHL1 in the CSF of PD patients compared to the controls.

Little is known about the protein serum amyloid A4 other than it is from the serum amyloid A class of proteins which are a family of apolipoproteins which are known to be involved in the acute inflammatory response. A chronic inflammatory status of neurons is a known factor associated with neurodegeneration but the increased levels of serum amyloid A4, a protein synthesised mainly in the liver and adipocytes is unclear. Finally, malate dehydrogenase was found to be elevated in LBD, PD and AD groups though only statistically significantly elevated in the LBD and AD groups. The role of cytosolic malate dehydrogenase 1 and its elevation in neurodegeneration is unclear as it is an enzyme involved in many metabolic pathways and catalyses the oxidation of malate to oxaloacetate using NAD+. Malate dehydrogenase is an important protein in the citric acid cycle and has a prominent role in gluconeogenesis to create glucose molecules from smaller intermediates. It is known that most neurodegenerative conditions have a mitochondrial component, whether primary or secondary is unclear. The elevation of malate dehydrogenase in the CSF may also be a general marker of mitochondrial dysfunction in its role in the shuttling of malate between the inner mitochondrial membrane and cystosol. Although malate dehydrogenase was >2-fold elevated in the PD group it was not statistically significantly elevated. Only the protein DJ1, also known as PARK7, and a peptidase that is known to have protective role for neurons against cell death, was found to be both >3-fold increased and statistically significantly elevated in the PD group. DJ1 has been shown to be raised in CSF of PD patients previously but the levels were not significant compared with LBD and AD samples and the usefulness of this protein as a diagnostic maker is still to be confirmed [11, 35]. This study shows DJ1 can be measured by targeted proteomics and potentially incorporated into a multiplex test. α-Synuclein was included in the targeted assay but it was not significantly altered in any of the groups tested using this method, which is in agreement with earlier studies showing only small or no changes in the CSF levels of the protein when measured using immunochemical techniques in synucleinopathies like LBD and PD [36].

The correlation analysis of the markers with the established markers given in Additional file 5: Figure S3, Additional file 6: Figure S4, Additional file 7: Figure S5 and Fig. 7 gives an insight into the changes occurring from control to disease and indicate the pathology behind the markers. No correlation with any of the markers were observed for the LBD group. Only specific associations could be observed in the AD group for Aβ42. Transthyretin has been shown previously to have a linear relationship with Aβ42 CSF levels [16]. Our data did demonstrate a correlation just with AD and not with LBD and controls (Fig. 7i) confirming this relationship. Transthyretin also revealed a strong linear relationship with p-tau and a weak relationship with total tau levels. IGF2 also showed an AD specific relationship with Aβ42 and p-tau levels. Many of the markers revealed correlations with both AD and LBD (Additional file 4: Table S3) for p-tau and total tau. Those with the strongest correlations were CNDP1, apoE, malate dehydrogenase, osteopontin, cystatin C and ubiquitin. The latter three also showed association of p-tau levels in the control group. This indicates these markers are more directly associated with p-tau and may only reflect p-tau levels in CSF and therefore have limited use diagnostically. Many markers did not show a relationship with either Aβ42, p-tau and total tau levels and are likely unrelated to the direct pathology of Aβ42 and tau plaques in AD and LBD. These markers may indicate other processes ongoing in neurodegeneration and warrant further investigation.

One limitation of this study is that some patients with LBD pathology can also exhibit amyloid plaques [33]. As with other CSF dementia biomarker studies we acknowledge that we cannot exclude these comorbidities at this point. Also although there were no statistically significant differences between age of patients and controls, we acknowledge that age ranges between patients and controls are heterogeneous, and patients are older than controls. Nevertheless, controls have been followed up for 1 to 4 years and so far none of them developed dementia. In addition, CSF amyloid, total tau and p-tau levels in these subjects are within the normal values (Table 2), suggesting that neither AD-related pathogenic events nor neurodegenerative ones are occurring.