An update on blood-based biomarkers for non-Alzheimer neurodegenerative disorders

Cerebrospinal fluid analyses and neuroimaging can identify the underlying pathophysiology at the earliest stage of some neurodegenerative disorders, but do not have the scalability needed for population screening. Therefore, a blood-based marker for such pathophysiology would have greater utility in a primary care setting and in eligibility screening for clinical trials. Rapid advances in ultra-sensitive assays have enabled the levels of pathological proteins to be measured in blood samples, but research has been predominantly focused on Alzheimer disease (AD). Nonetheless, proteins that were identified as potential blood-based biomarkers for AD, for example, amyloid-β, tau, phosphorylated tau and neurofilament light chain, are likely to be relevant to other neurodegenerative disorders that involve similar pathological processes and could also be useful for the differential diagnosis of clinical symptoms. This Review outlines the neuropathological, clinical, molecular imaging and cerebrospinal fluid features of the most common neurodegenerative disorders outside the AD continuum and gives an overview of the current status of blood-based biomarkers for these disorders. Most research into blood-based biomarkers for neurodegenerative disorders has so far focused on Alzheimer disease. In this Review, Aarsland and colleagues give an overview of the current status of blood-based biomarkers for the non-Alzheimer disease neurodegenerative disorders. Neurodegenerative disorders are characterized by protein aggregation and other pathological processes, which can affect the composition of biofluids such as blood and cerebrospinal fluid (CSF). Analysis of CSF and molecular imaging of the brain enable the stratification of patient populations on the basis of underlying pathology, but are limited as population screening tools. Advances in ultra-sensitive immunoassays for the measurement of amyloid-β, neurofilament light chain, total tau and phosphorylated tau, as well as mass spectrometry-based methods for the measurement of amyloid-β, have demonstrated that a blood-based screening tool for Alzheimer disease is a realistic and plausible possibility. Evidence now suggests that blood-based biomarkers could also be important for other common neurodegenerative disorders: for example, Lewy body dementia, atypical parkinsonian disorders and frontotemporal dementia. Neurodegenerative disorders are characterized by protein aggregation and other pathological processes, which can affect the composition of biofluids such as blood and cerebrospinal fluid (CSF). Analysis of CSF and molecular imaging of the brain enable the stratification of patient populations on the basis of underlying pathology, but are limited as population screening tools. Advances in ultra-sensitive immunoassays for the measurement of amyloid-β, neurofilament light chain, total tau and phosphorylated tau, as well as mass spectrometry-based methods for the measurement of amyloid-β, have demonstrated that a blood-based screening tool for Alzheimer disease is a realistic and plausible possibility. Evidence now suggests that blood-based biomarkers could also be important for other common neurodegenerative disorders: for example, Lewy body dementia, atypical parkinsonian disorders and frontotemporal dementia.

of neurodegenerative disorders 2 . However, many of these disorders include a combination of different proteinopathies and the delay between clinical onset and post-mortem diagnosis has made it difficult to establish correlations between brain pathology and clinical symptoms. Therefore, in vivo biomarkers, such as those derived from brain imaging or proteomic analysis of biofluids, will be crucial for accurate diagnosis during life and might allow diagnosis during the prodromal or even pre-clinical stages of disease. MRI provides regional measures of brain atrophy, which is reflective of neurodegenerative processes such as dendritic pruning, synaptic loss and neuronal depletion. MRI-based markers for atrophy are included in the diagnostic criteria for non-AD neurodegenerative disorders [3][4][5] , as different neurodegenerative disorders are associated with spatially distinct patterns of volume loss 6 . The introduction of in vivo PET has transformed our understanding of neurodegenerative disorders by helping refine models of disease progression, as well as providing a powerful diagnostic tool to complement clinical evaluation (Fig. 1). The first PET technique used for this purpose was metabolic imaging with 18 F-fluorodeoxyglucose ( 18 F-FDG), which is thought to provide a measure of neuronal or synaptic integrity 7 . Many new PET ligands have since been developed, including ligands specific for fibrillar Aβ 8 , paired-helical filament tau 9,10 , and synaptic vesicle protein 2A 11 . However, molecular imaging is costly and is only available at specialized centres.
The clearance of abnormal proteins via the cerebrospinal fluid (CSF) is an endogenous neuroprotective mechanism of the brain. In neurodegenerative disorders, abnormal extracellular, intracellular and synaptic proteins can leak into the CSF 12 , and the levels of these proteins in CSF samples can be used as biomarkers for disease, or for particular stages of disease progression. CSF analysis is more affordable and accessible than PET imaging, but lumbar puncture can be perceived as invasive, which might limit its use in clinical practice, depending on the health-care system. Therefore, blood-based markers for the pathophysiology underlying non-AD neurodegenerative disorders would be extremely useful as an initial triage step in a multi-stage assessment for cognitive complaints. These biomarkers could also enable the selection of participants for secondary prevention trials and could help monitor the response of participants to therapeutic interventions.
Excellent molecular imaging 13 (Fig. 1) and CSF 14 ( Table 2) biomarkers already exist for AD, and the most recent diagnostic criteria [15][16][17][18] and research framework 19 for the disease include the use of biomarkers to identify the typical amyloid proteinopathy. Promising blood-based biomarkers are also being developed for AD 20 (Table 2). By contrast, the search for fluid biomarkers for non-AD neurodegenerative disorders has only recently begun, but will greatly benefit from the developments that have taken place in the AD field. In this Review, we briefly summarize the clinical and neuropathological features of the most common non-AD neurodegenerative disorders and discuss the main findings from neuroimaging and CSF studies of these diseases. We then discuss how the recent progress made in the AD field can aid the search for disease-specific blood-based biomarkers for other neurodegenerative conditions. The numerous triplet disease disorders, for example, spinocerebellar ataxias and Huntington disease, are not discussed in this Review because they form, to some extent, a separate group of genetically defined movement disorders.
Non-AD neurodegenerative disorders PD, PDD and DLB PD is the second most common age-related brain disorder after AD 21 and is characterized by the accumulation of α-synuclein in intracellular inclusions, known as Lewy bodies and Lewy neurites. The risk of developing PD

Key points
• Neurodegenerative disorders are characterized by protein aggregation and other pathological processes, which can affect the composition of biofluids such as blood and cerebrospinal fluid (CSF). • analysis of CSF and molecular imaging of the brain enable the stratification of patient populations on the basis of underlying pathology, but are limited as population screening tools. • advances in ultra-sensitive immunoassays for the measurement of amyloid-β, neurofilament light chain, total tau and phosphorylated tau, as well as mass spectrometrybased methods for the measurement of amyloid-β, have demonstrated that a blood-based screening tool for alzheimer disease is a realistic and plausible possibility. • evidence now suggests that blood-based biomarkers could also be important for other common neurodegenerative disorders: for example, lewy body dementia, atypical parkinsonian disorders and frontotemporal dementia.
Author addresses 1 increases with age (the mean age of onset is ~60 years) and the lifetime risk is slightly higher for men than for women 21 . Although most cases of PD are sporadic, some rare cases are familial 22 . The pathological hallmark of PD is the progressive loss of nigrostriatal dopaminergic neurons in the substantia nigra pars compacta. As a result of this dopaminergic pathology, individuals with PD typically have parkinsonism, which is a clinical syndrome defined by rest tremor, rigidity, bradykinesia and gait dysfunction with postural instability 23 . PD is the most common cause of parkinsonism, but the syndrome can also be caused by other neurodegenerative disorders, for example, progressive supranuclear palsy (PSP), corticobasal syndrome (CBS) or FTD, or secondary causes, for example, metabolic, toxic, drug-induced or vascular insults. Individuals with PD can also have non-motor symptoms, including hyposmia, sleep disorders, autonomic dysfunction, pain, behavioural disturbances and cognitive impairment. Remarkably, a considerable number of individuals with PD will eventually develop cognitive impairment and dementia over the course of their illness, that is PDD 24,25 . However, the timing of the onset of PDD is highly variable, which means that some individuals develop dementia soon after being diagnosed with PD, whereas other individuals display no signs of cognitive impairment for many years 26,27 . Individuals in whom dementia precedes or arises concomitantly with the motor symptoms are diagnosed as having DLB 28 . Together, PD, PDD and DLB constitute the Lewy body diseases and a considerable clinical and pathological overlap exists between them. In particular, PDD and DLB are distinguished solely on the basis of the relative timing of parkinsonism and dementia; that is, individuals in whom dementia occurs more than one year after the diagnosis of PD are diagnosed clinically as having PDD, whereas individuals in whom dementia occurs before or simultaneously with parkinsonism are diagnosed as having DLB 4 . This distinction is arbitrary, and the disease in many patients is difficult to classify because the timing of cognitive decline and parkinsonism can be difficult to establish. PDD and DLB show different profiles of cognitive impairment, but both diseases are characterized by relatively more executive, attentional and visuospatial impairment than is observed in AD 4 , although memory is usually impaired and is often the first symptom to be reported. Interestingly, individuals with PD can have lesions outside the brain, which most commonly affect the autonomic nervous system, leading to symptoms such as orthostatic hypotension and constipation 29 . Neuropsychiatric symptoms that are typical of Lewy body diseases include visual hallucinations and REM-sleep behaviour disorder 30 .
In addition to α-synuclein pathology, DLB and PDD often show varying degrees of AD co-pathology, that is, aggregates of Aβ and tau 26 . The correlation between clinical symptoms and the extent and severity of α-synuclein patho logy is often complicated by the co-existing AD pathology. This co-existing pathology must be considered when assessing the probability of α-synuclein being the cause of specific clinical symptoms 4 . Less frequently, co-existing TDP43 pathology is detectable in individuals with PD 31 .
Beyond the exclusion of secondary causes of parkinsonism -such as vascular, demyelinating or space-occupying lesions within the brainstem or basal ganglia -conventional T1-weighted and T2-weighted MRI sequences are considered to be of limited use in the diagnosis of PD 32,33 . However, individuals with DLB show reductions in brain volume, which can be detected with conventional MRI 4 . The degree and regional distribution of this volumetric loss varies between individuals, but absent or minimal atrophy of the medial temporal lobe has been identified as a consistent feature of DLB 4 . Recent advances in MRI methodology, including ironsensitive techniques such as susceptibility-weighted imaging and quantitative susceptibility mapping, have enabled the identification of abnormalities within the substantia nigra and nigrostriatal system in individuals with PD 34 . On 18 F-FDG PET, a pattern of hypometabolism in temporal, parietal and occipital regions is typically observed in individuals with DLB, PD or PDD [35][36][37][38] . Individuals with PDD also show hypermetabolism in the motor cortex, striatum, thalamus and cerebellum 39 . Dopaminergic function, whether measured by NaTuRe RevIeWS | NEuROLOgy single-photon emission CT (SPECT) or PET, is markedly decreased in individuals with DLB, PD or PDD as compared with healthy age-matched controls 40,41 , which is in keeping with the degeneration of nigrostriatal dopamine neurons that can be observed in these individuals post mortem. Studies that quantified Aβ deposition with 11 C-PiB-PET found that levels of Aβ are unchanged in individuals with PD, slightly increased in individuals with PDD and elevated in individuals with DLB, as compared with healthy aged-matched controls [42][43][44][45] . Furthermore, individuals with cognitive decline were more likely to have elevated levels of Aβ deposition 43,44,46 . The contribution of tau pathology to the development of Lewy body diseases is not yet clear 47,48 . Tau PET studies of DLB, PD and PDD have so far yielded inconsistent results, with some studies finding no difference in ligand binding between healthy controls and individuals with the Lewy body diseases [49][50][51] . The tau PET ligand 18 F-flortaucipir has been shown to bind to neuromelanin in the substantia nigra 52,53 , and evidence suggests that several novel tau ligands with a similar structure to 18 F-flortaucipir can also bind to this site 54 . Therefore, tau PET might be able to detect the characteristic loss of neuromelanin-rich neurons that occurs in this region in individuals with PD or PDD.

FTD
FTD includes a group of neurodegenerative disorders that are clinically and pathologically heterogeneous, but predominantly affect the frontal and/or temporal lobes of the brain. The two main clinical presentations of FTD are the behavioural variant (bvFTD), which leads mainly to personality alterations and behavioural problems, and primary progressive aphasia (PPA), which is less common than bvFTD and causes progressive deterioration of speech and/or language 55 . PPA can be further divided Detection of protein pathology in the brain with molecular imaging is considered the 'gold standard' by which the accuracy of blood-based biomarkers is tested. Therefore, to understand how amyloid-β (Aβ) deposition, tau deposition and glucose metabolism are visualized with molecular imaging techniques and how these measures differ between different neurodegenerative disorders is important for fluid biomarker research. The top row shows axial slices of 11 C-Pittsburgh compound B ( 11 C-PiB) scans, which reflect neuritic Aβ plaque density , from four patients. Higher levels of tracer retention are indicated by a higher standardized uptake value ratio (SUVR). The scan of the individual with Alzheimer disease (AD) shows considerable tracer retention throughout the cortex, especially when compared with the non-specific tracer retention in the white matter, and is considered 'Aβ-positive'. The 11 C-PiB-PET scans in the individuals with behavioural variant FTD (bvFTD), corticobasal syndrome (CBS) or progressive supranuclear palsy (PSP) only show non-specific tracer retention, which is localized to the white matter, and are therefore considered 'Aβ-negative'. The second row shows PET images acquired using the tracer 18 F-flortaucipir ( 18 F-FTP), which binds to intracellular aggregates of abnormally phosphorylated tau. Tracer binding in the individual with AD is high in temporoparietal areas, including the posterior cingulate and precuneus, as well as the dorsal prefrontal cortex.
In the individuals with bvFTD, CBS or PSP, the arrowheads highlight areas of mild-moderate tracer binding. Asterisks indicate brain regions of non-specific tracer retention ('off-target' binding), which overlap with known patterns of off-target binding in healthy individuals, including binding in the basal ganglia, midbrain regions and cerebellum. The third row shows PET images acquired using the tracer 18 F-fluorodeoxyglucose ( 18 F-FDG), the retention of which is a marker for glucose uptake by the tissue. Across all individuals, regions of low 18 F-FDG retention overlap with regions of high 18 F-FTP retention. The bottom row shows structural T1-weighted MRI scans that correspond to the slices shown for the 18 F-FTP and 18 F-FDG scans. The PiB-PET scans were acquired 50-70 min after tracer injection, and SUVRs were created using a cerebellar reference region; the 18 F-FTP scans were acquired at 80-100 min after tracer injection and SUVR created using an inferior cerebellar reference region; the 18 F-FDG scans were acquired at 30-60 min after tracer injection and SUVR created using the pons as the reference region.
www.nature.com/nrneurol into three subtypes: semantic variant PPA, non-fluent variant PPA and logopenic variant PPA; the last of these variants is usually associated with AD pathology 56 . According to international consensus criteria 57 , probable bvFTD is defined by the persistence or recurrence of at least three of the following symptoms: early onset (<65 years) of behavioural disinhibition; apathy; loss of empathy; perseverative, stereotyped, compulsive or ritualistic behaviours; hyperorality and dietary changes; executive deficits with relative preservation of memory and visuospatial functions; and progressive deterioration of behaviour and/or cognition 57 . Around one-third of cases of FTD are familial 55 and the most common genes involved are MAPT, GRN and C9orf72 (reF. 56 ). A probable diagnosis of FTD is made when an individual with symptoms of FTD either has a causative genetic mutation or shows neuroimaging signs of disproportionate involvement of the frontal and/or temporal lobes 56 . The term 'frontotemporal lobar degeneration' (FTLD) is used to describe the neuro pathological changes observed in individuals with FTD. FTLD involves pathological inclusions of either TDP43, tau or the FET proteins (that is, FUS, EWS and TAF15). Consequently, FTLD is classified as FTLD-TDP, FTLD-tau or FTLD-FET 58 . FTD can overlap clinically and pathologically with motor neuron disease (MND) or some extrapyramidal syndromes, such as CBS and PSP. For example, inclusions of TDP43 are the most common underlying pathology in MND. Furthermore, some cases of amyotrophic lateral sclerosis (ALS), the most common MND, are caused by mutations in C9orf72 or FUS (known as ALS-FUS) 59 MRI studies have shown prominent, usually symmetric atrophy of the frontal lobes in individuals with bvFTD 3 . In AD, glucose hypometabolism can be observed in the posterior temporal and parietal lobes, and the posterior cingulate and precuneus [63][64][65][66][67] . However, in FTD, three main spatial patterns of glucose hypometabolism can be observed, affecting precentral and inferior frontal regions in individuals with non-fluent PPA, the anterior temporal lobes in those with semantic PPA (usually with marked leftward asymmetry) 68,69 , High P-tau reflects phosphorylation state of tau and thus probably tau pathology in AD; P-tau is more specific for AD than T-tau; CSF assays for P-tau 181  High plasma NfL is a general biomarker for neurodegeneration and not specific for AD Candidate screening tool for global neurodegeneration; potential outcome measure for intervention trials Aβ, amyloid-β; AD, Alzheimer disease; CSF, cerebrospinal fluid; NfL , neurofilament light chain; PET, positron emission tomography; P-tau, phosphorylated tau; T-tau, total tau. NaTuRe RevIeWS | NEuROLOgy and frontal and temporal-limbic regions in those with bvFTD 70 . In a case series in which PET was used to examine Aβ deposition in the brain, only 0-15% of individuals with FTD reached the threshold for Aβ positivity, which agrees with neuropathological studies that did not identify Aβ plaques as a principle characteristic of FTLD [71][72][73] .
One study found minimally elevated tau binding on PET in individuals with non-fluent variant PPA, CBS or bvFTD as compared with healthy controls 74,75 . This elevated tau binding was localized to brain regions already known to have a role in these diseases, that is, inferior frontal areas in individuals with non-fluent variant PPA, the precentral gyrus and frontal white matter in some individuals with CBS, and frontotemporal regions in individuals with bvFTD 74,75 . However, these in vivo PET findings should be interpreted cautiously as autoradiography studies have suggested that tau PET tracers do not bind to the isoforms of tau present in the non-AD tauopathies [76][77][78] . Owing to clinical heterogeneity and the lack of reliable peripheral biomarkers, FTD continues to pose major diagnostic challenges 79 .

CSF biomarkers
For brain disorders, principally AD, the development of blood-based biomarkers has been informed and guided by knowledge derived from studies of CSF-based biomarkers. Studies have consistently shown that CSF levels of Aβ 42 , total tau (T-tau) and phosphorylated tau (P-tau) can be used to distinguish between individuals with AD and healthy controls 80 , and these core AD biomarkers are now key components of the research diagnostic criteria for AD 18,19 . The CSF Aβ 42 to Aβ 40 ratio has also been used as a biomarker for AD and has repeatedly been shown to be more reliable than CSF Aβ 42 alone because the ratio is not affected by inter-individual differences in amyloidogenic amyloid precursor protein (APP) processing 81 .
The CSF concentrations of the core AD biomarkers (Aβ 42 , T-tau and P-tau) are usually normal in individuals with non-AD dementias 82,83 , which is in agreement with molecular imaging data ( Fig. 1) and can be useful in the differential diagnosis of individuals with cognitive symptoms . However, in 50% of individuals with DLB, and in many individuals with PDD, CSF Aβ 42 concentrations are lower than in healthy individuals 84,85 , which highlights the AD pathology that can be present in these disorders. Furthermore, in individuals with Creutzfeldt-Jakob disease (CJD), CSF T-tau concentrations are markedly increased 86 , whereas CSF P-tau concentrations are normal or only marginally increased 87 . In individuals with FTD or other primary tauopathies (for example, PSP), CSF tau concentrations -including those of T-tau, specific epitopes of P-tau and some N-terminal tau fragments -are largely the same as the concentrations in healthy individuals 83,[88][89][90][91][92] . This finding is surprising, but agrees with the results of tau PET studies ( Fig. 1) and might suggest that, in individuals with these tauopathies, secretion of tau proteins to the extracellular space and the CSF is lower than in individuals with AD, or that alternative processing of tau occurs, resulting in tau fragments that are not detected by the conventional tau immunological assays.
Neurofilament light chain (NfL) is the smallest of the neurofilament triplet proteins that are the structural components of axons. During healthy ageing, NfL is released from axons and enters the extracellular space, which results in increased CSF NfL concentrations. However, this NfL release is accelerated in response to axonal damage, which can be caused by neurodegeneration, inflammation, trauma or stroke. Several studies have shown that CSF NfL concentrations are highest in brain disorders with subcortical pathology, such as vascular dementia and normal pressure hydrocephalus 93,94 . Notably, CSF NfL concentrations were higher in individuals with FTD than in individuals with AD (age at disease onset was similar in both groups of participants), which suggests that this biomarker can be used to differentiate between these two diseases 95 . In addition, CSF NfL concentrations are substantially higher in individuals with CJD than in healthy individuals, which correlates with the changes in CSF T-tau concentrations observed in this disease 96 . Although CSF NfL concentrations are relatively normal in individuals with PD, several studies have shown a marked increase in CSF NfL concentrations in individuals with the atypical parkinsonian disorders, specifically CBS, multiple systemic atrophy (MSA) or PSP, when compared with those in healthy individuals 91,97,98 . Neurofilament heavy chain (NfH), which is also considered to be a biomarker for neurodegeneration, is the most extensively phosphorylated protein in the human brain and influences the regulation of cell structure, homeostasis and axonal transport 99 . CSF NfL and phosphorylated NfH (pNfH) concentrations are well correlated 100 and show similar changes when measured in individuals with a neurodegenerative disorder 101,102 .
A reduction in the CSF total monomeric α-synuclein concentration has been proposed as a biomarker for PD and DLB. However, most studies have found only a marginal (albeit statistically significant) reduction in concentrations of this marker in individuals with PD or DLB as compared with concentrations in healthy individuals 91,103,104 . A meta-analysis, which included >3,000 individuals enrolled in 17 different studies, also found significantly lower CSF total α-synuclein concentrations and significantly higher oligomeric and phosphorylated α-synuclein concentrations in individuals with PD than in healthy individuals 105 . However, the authors concluded that the sensitivity and specificity of CSF α-synuclein concentration for the diagnosis of PD and DLB is not sufficient for it to be considered a helpful biomarker for these diseases. This negative finding could be because α-synuclein concentrations are 10,000-fold higher in blood than in CSF, suggesting that contamination of CSF samples with blood might introduce peripherally derived α-synuclein (which is not related to neurodegeneration). Hence, identification of brain-specific pathological forms of α-synuclein will be crucial to advancing the search for disease-specific biomarkers. Another explanation for the negative finding could be that, in individuals with PD or DLB, α-synuclein is simultaneously incorporated into inclusions and released into the CSF by synapse breakdown, resulting in no overall change in CSF α-synuclein concentrations. Regardless, evidence www.nature.com/nrneurol suggests that CSF oligomeric α-synuclein concentrations can be used to differentiate DLB or PDD from AD 106 .
New real-time quaking-induced conversion (RT-QuIC) technology measures the self-replicating properties of proteinopathic proteins, and analysis of CSF α-synuclein with this technique was used to diagnose PD and DLB with a sensitivity and specificity of >90% [107][108][109] . Furthermore, RT-QuIC detected CSF α-synuclein aggregation in the majority of individuals with LRRK2-linked PD, although CSF α-synuclein aggregation was even more prevalent in individuals with idiopathic PD 110 . New evidence also suggests that RT-QuIC can identify α-synuclein pathology in the CSF of individuals with DLB 111 . Importantly, new-generation α-synuclein RT-QuIC assays can be performed more rapidly than original prototypes and can provide results in 1-2 days 107,112 .
High concentrations of the postsynaptic protein neurogranin have repeatedly been found in the CSF of individuals with AD [113][114][115][116] , whereas in a wide range of other neurodegenerative disorders, including FTD and DLB 117 , CSF neurogranin concentrations are the same as in healthy individuals. In one study, CSF neurogranin concentrations increased soon after individuals reached the threshold for Aβ PET positivity 118 . However, studies assessing CSF neurogranin concentrations in individuals with PD have produced contradictory findings 117,119,120 . CSF concentrations of the presynaptic proteins growthassociated protein 43 (GAP43), synaptotagmin 1 and synaptosomal nerve-associated protein 25 (SNAP25) were increased in individuals with AD as compared with healthy individuals [121][122][123] . Thus, CSF neurogranin, GAP43, synaptotagmin 1 and SNAP25 could be the latest additions to the toolbox for differentiating AD from non-AD neurodegenerative disorders.
Proteomic profiling has identified changes in the CSF concentrations of proteins such as apolipoprotein E (ApoE), APP, cystatin C 124 , chitinase 3-like protein 1 (reF. 124 ), neuronal pentraxin 1 (reF. 125 ), transthyretin 124 and ubiquitin 82,124,126 in individuals with PD or DLB as compared with healthy individuals. However, only one study compared samples of CSF from individuals with PD, DLB or AD using label-free (or isobaric) liquid chromatography tandem mass spectrometry (LC-MS/MS), in which a protease-digested peptide mixture is typically ionized and fragmented for simultaneous identification and quantification. This study found changes in the levels of 72 proteins, including ceruloplasmin and some apolipoproteins such as ApoH and ApoC1, that were associated with PD and not with AD or DLB 127 . Based on these findings, Zhang et al. identified and validated a panel of eight CSF proteins -tau, Aβ 42 , β 2 -microglobulin, IL-8, vitamin D-binding protein, ApoA, ApoE and BDNF -that were highly effective at differentiating PD from AD 128 .

Introduction to blood-based biomarkers
Although collecting blood samples is straightforward and non-invasive, analysing the levels of brain-derived molecules in blood (either plasma or serum) and interpreting the results is a complex process. For example, concentrations of brain-derived molecules will be lower in blood than in CSF. Molecules can move between the CSF and the brain tissue in a continuous and uninhibited way, whereas the exchange of molecules between the brain and the blood takes place across the blood-brain barrier, via lymph vessels and through the glymphatic system 129 . Once in the bloodstream, brain-derived molecules are diluted in a complex matrix of plasma proteins, such as albumin, IgG, transferrin, haptoglobin and fibrinogen. Plasma and serum have an extraordinary dynamic range in that the concentration of the most abundant protein (albumin) is more than ten orders of magnitude higher than that of the least abundant proteins to be measured clinically 130 . The composition and complexity of this matrix can have a large effect on the ability of an immunoassay to accurately quantify a specific target, which can result in misleading conclusions 131 .
Protein biomarkers can also undergo proteolytic degradation by various proteases in plasma 132 and this seems to be the case for non-modified tau, which is stable in CSF but has a short (∼10 h) half-life in blood 133 . Furthermore, biomarkers might be expressed at high levels outside the brain, including in blood cells such as platelets and erythrocytes, which would make it challenging to decide whether changes in blood-based biomarker concentrations reflect events in the CNS or systemic changes. This peripheral expression is of particular relevance to several AD-related biomarkers such as Aβ, which is expressed in blood platelets and several other tissues. Some biomarkers might exist as multiple isoforms or contain stable and/or dynamic post-translational modifications. Interference from heterophilic antibodies (endogenous antibodies that interfere with the results of antibody-based assays) 134 or variations in the methods used to collect, process and store blood samples, can also affect the measurement of analyte levels. These factors all result in a high degree of variability in analyte levels among individual patients and research cohorts, which is unrelated to disease and can be difficult to account for during the interpretation of results. The blood fraction used for analysis can also affect biomarker measurements; for example, the concentration of an analyte in plasma does not always correlate with the concentration of the same analyte in serum 135 . Although serum and plasma measures of NfL 136 and P-tau 181 (reF. 137 ) correlate well, the preferred matrix for measuring Aβ 42 , Aβ 40 and T-tau is plasma, as concentrations of these markers in serum are often below the limit of quantification even for ultra-sensitive assays (N.J.A., K.B. and H.Z., unpublished work).
An important consideration in the development of a blood-based biomarker for a neurodegenerative disorder is the intended context of use (for example, casecontrol comparisons, correlation with disease phenotypes or correlation with disease end points) and the translation from laboratory validation to clinical use. The Alzheimer Precision Medicine Initiative recently published guidelines describing a multi-tiered approach to biomarker evaluation 20 . These guidelines include suggested levels of sensitivity and specificity, as well as negative predictive values and positive predictive values for specific contexts of use.
Mass spectrometry a sensitive technique used to detect, identify and quantify molecules on the basis of their mass-to-charge ratio.

Negative predictive value
The probability that individuals with a negative test result do not have the disease of interest.

Positive predictive value
The probability that individuals with a positive test result have the disease of interest.

NaTuRe RevIeWS | NEuROLOgy
Blood-based biomarker discovery techniques Proteomic studies of blood-based biomarkers can be targeted or non-targeted. Targeted approaches use prior knowledge of which proteins are involved in the disease of interest to select candidate molecules for use as biomarkers. Immunocapture or mass spectrometry methods are then used to establish whether levels of the candidate molecules in the blood are associated with the disease. By contrast, non-targeted approaches involve unbiased analy sis of a large number of different molecules to identify those that are altered in individuals with a specific disease. A non-targeted approach is not limited by current knowledge of disease pathophysiology and can therefore identify novel candidate biomarkers, which can then be verified using a targeted approach (Fig. 2).
Immunocapture techniques, which typically involve paired antibodies in the sandwich immunoassay format, remain the most popular techniques for all biomarker analyses and involve the stabilization of the antigen on a solid surface followed by detection of the antigen with a specific, labelled antibody. In a sandwich immunoassay, a capture antibody is used to immobilize the target antigen on the solid surface. The combination of antibody capture followed by mass spectrometry (IP-MS) has been a popular tool for detailed characterization of a target of interest, for example, Aβ peptides in AD biomarker research. Most new-generation immunocapture assays follow the same workflow as a colorimetric enzyme-linked immunosorbent assay (ELISA). For example, assays that use electrochemiluminescent (ECL) reporters theoretically allow the simultaneous detection of ten (Meso Scale Discovery) to 100 (Luminex, xMAP) analytes, which is known as multiplexing. However, these assays are still affected by the typical limitations of antibody-based capture methods, that is, variability in dynamic range, limited antibody specificity and the presence of cross-reactivity, which can restrict multiplexing to fewer analytes than the assays are designed to detect. Therefore, multiplexed ECL assays might not be suitable for use as initial biomarker discovery screens, but are of tremendous value for the high-throughput validation of specific targets or pathways.
The latest variations in capture-based methodology include the Proximity Extension Assay (PEA; Olink), SomaScan (SomaLogic), Single Molecule Counting (SMCxPRO; Merck), single molecule array (Simoa; Quanterix), as well as fully automated immunoassays with ECL detection, for example, Elecsys. As the abundance of CNS-derived protein in the bloodstream is likely to be low, high-sensitivity precision analysis of a single target molecule is of far more value to the search for blood-based biomarkers for neurodegeneration

Fig. 2 | Current strategies for blood-based biomarker discovery in neurodegenerative disorder research.
Proteomic studies of blood-based biomarkers can be targeted or non-targeted. Targeted approaches use prior knowledge of targets involved in disease, for example, P-tau 181 in Alzheimer disease, and use immunocapture or mass spectrometry methods to establish whether blood levels of the candidate molecules are associated with disease surrogates such as changes in cognitive assessment scores, molecular imaging or CSF biomarkers. Non-targeted approaches involve unbiased analysis of a large number of different molecules to identify those that are altered in individuals with a specific disease. Results of a non-targeted approach often lead to the investigation of new candidate molecules with a targeted approach. ECL , electrochemiluminescence; ELISA , enzyme-linked immunosorbent assay; iTRAQ, isobaric tags for relative and absolute quantitation; LC-MS, liquid chromatography-mass spectrometry; LC-MS/MS, liquid chromatography with tandem mass spectrometry; qPCR , quantitative PCR; Rel.Int., relative intensity; TMT, tandem mass tagging.
Colorimetric enzyme-linked immunosorbent assay a common protein analysis technique, usually conducted in a 96-well plate format, in which the antigen is stabilized on a solid surface and probed with a specific enzymeconjugated antibody; the resulting enzymatic reaction is then measured with a chromogenic reporter.
www.nature.com/nrneurol than the simultaneous low-sensitivity measurement of multiple analytes. The SMCxPRO and Simoa platforms use traditional antibody sandwich immunocomplex technology, but can detect sub-femtomolar concentrations of analytes in blood. This high level of sensitivity is achieved by the use of novel microfluidics techniques to isolate individual immunocomplexes, followed by excitation of a fluorogenic reporter, which enables detection of single molecules of the target of interest 138 . The Simoa or 'digital ELISA' is now the preferred tool for measuring T-tau, P-tau, NfL, pNfH and glial fibrillary acidic protein (GFAP) in blood. A common non-targeted approach to identifying blood-based biomarkers is LC-MS/MS. For blood analysis, LC-MS/MS methods are usually used alongside upfront peptide or protein fractionation, or the immunodepletion of highly abundant plasma proteins, which not only improves the number of analytes detected but improves the protein sequence coverage 139 . LC-MS/MS can also be employed in a targeted manner if an analyte of interest has been identified. Selection reaction monitoring (SRM) methods enable higher precision, more accurate quantification and higher throughput than unbiased LC-MS/MS methods 140 .

Blood-based biomarkers for AD
The search for robust blood-based biomarkers for AD pathology has now entered a second decade, and the recent advances in this field will have a tremendous impact on research into biomarkers for non-AD neuro degenerative disorders. For example, lessons can be learned from the strategies used in AD biomarker studies, and some targets of interest might be applicable to both AD and non-AD neurodegenerative disorders. Until recently, the identification of candidate blood-based biomarkers for AD with non-targeted proteomics was focused on proteins that are expressed at relatively high concentrations in the blood and are known to be predominantly peripherally expressed 141-143 . However, technological advances in immunocapture and mass spectrometry-based detection methods, in combination with better characterization of clinical cohorts (including neuroimaging and CSF biomarker information on AD pathology) have led to a number of breakthroughs. The endophenotype approach has identified promising blood markers for brain atrophy 143,144 and cerebral amyloid pathology 139,145-149 . Several studies have found consistent and statistically significant differences in these markers between individuals with AD and healthy individuals, which is in accordance with genomic and in vitro evidence of the involvement of the target proteins in disease processes 150,151 . However, these markers are not sufficiently sensitive and specific for clinical use. Therefore, at the current time, the most promising blood-based biomarker candidates for AD are the markers that were initially identified in CSF assays (Table 2).
Aβ peptides can be readily measured in plasma using standard ELISA or ECL assays, but a large number of studies using these approaches have shown no clear difference in Aβ peptide concentrations between individuals with clinically diagnosed AD and cognitively unimpaired older individuals 80 . These negative results are thought to have resulted from the influence of peripheral expression of Aβ 152 and interference of matrix components in the measurements 153,154 . However, these earlier findings are now being challenged, as evidence from highly sensitive mass spectrometric 155-157 , Simoa 158 and fully automated immunoassays 159 -which allow the sample to be diluted to mitigate the matrix effects -suggests that Aβ peptide ratios in plasma are a highly sensitive and specific marker in individuals with Aβ-positive brain scans.
Plasma T-tau was assessed in large research cohorts, and significant increases were observed in individuals with AD, as compared with healthy individuals and individuals with mild cognitive impairment (MCI) 160-162 . However, the range of T-tau concentrations seen in the group of healthy participants and the group of participants with MCI substantially overlapped with the range seen in the group of participants with AD, and plasma and CSF T-tau concentrations were only weakly correlated. These observations suggest that the utility of cross-sectional plasma T-tau measurement for the diagnosis of AD is limited 160 . However, a prospective study performed as part of the Framingham Heart Study demonstrated that higher plasma T-tau concentrations were associated with a 35% higher risk of AD dementia, after adjustment for age and gender 161 , which suggests that plasma T-tau might help predict future dementia even in individuals who are cognitively unimpaired. In a similar study from the Mayo Clinical Study of Ageing, high plasma T-tau concentrations predicted a steeper decline in global cognition, memory, attention and visuospatial ability in individuals with MCI and cognitively unimpaired individuals 163 .
An ECL method for the accurate detection of P-tau 181 in plasma has been developed 164 . In two studies using the new ECL method, plasma P-tau 181 was used to accurately distinguish between healthy older individuals and individuals with MCI who had a positive Aβ PET scan 165 . Plasma P-tau 181 concentrations were correlated tightly with CSF P-tau 181 concentrations and with 18 F-flortaucipir binding in Aβ-positive individuals with AD, and were significantly increased in those with more severe Braak staging 165,166 . A third cohort-based study 137 , in which a Simoa-based methodology was used to detect plasma P-tau 181 in >1,100 individuals, also found statistically significant increases in plasma P-tau 181 concentration in Aβ-positive individuals with AD, as compared with Aβ-negative young adults and Aβ-negative cognitively unimpaired older individuals. Increased plasma P-tau 181 concentration was associated with tau PET positivity as well as Aβ PET positivity and predicted 1-year cognitive decline and hippocampal atrophy 137 . In a primary care setting, plasma P-tau 181 concentration was used to distinguish individuals with AD from young adults and cognitively unimpaired older individuals with a high level of accuracy. The Simoa assay has also enabled the measurement of serum P-tau 181 concentrations, which are lower than plasma concentrations 137 . One study found that amyloid deposition-related changes in plasma P-tau 181 concentrations are of a similar magnitude to changes in CSF P-tau 181 concentrations, whereas many other biomarkers show smaller magnitudes of change Selection reaction monitoring (SrM). a targeted mass spectrometry technique for the detection and quantification of specific predetermined analytes with known fragmentation properties.
Endophenotype any characteristic that is normally associated with a condition but is not a direct symptom of that condition. NaTuRe RevIeWS | NEuROLOgy in plasma than in CSF 118 , suggesting that blood P-tau could be used as an early indicator of AD pathology. A study using a third technique, immunomagnetic reduction (IMR), also showed higher plasma P-tau 181 concentrations in individuals with MCI AD or AD dementia than in healthy individuals 167 .
Studies have consistently found elevated blood NfL concentrations in individuals with AD 168-170 , prodromal AD 168 or familial AD 171,172 . Although not a specific marker for AD, blood NfL has the potential to identify neurodegeneration. Further observations in cohorts of individuals with AD have also shown that blood NfL concentrations correlate with cognitive performance 168,169 , CSF biomarkers, post-mortem pathology 173 , and structural and molecular imaging findings 168,174,175 . Interestingly, blood NfL measurements can be used to predict onset of AD in individuals with Down syndrome 176, 177 . The progress that has been made in identifying CSF biomarkers for synaptic integrity in AD (for example, neurogranin) has yet to translate to the field of blood-based biomarkers. Plasma neurogranin concentrations are detectable with conventional ELISAs but are unchanged in individuals with AD and do not correlate with CSF neurogranin concentrations, probably owing to the contribution of peripherally expressed neurogranin peptides to blood neurogranin measurements 116,178 . As new CSF markers for synaptic integrity (for example, GAP43) emerge and as technology continues to advance, the hope for a synaptic integrity-specific blood-based biomarker remains.

Blood-based biomarkers for non-AD disorders
As mentioned earlier, most by far of the research into blood-based biomarkers for neurodegenerative disorders has focused on AD (Table 2), which is mainly because AD is the most common of the neurodegenerative disorders and thus provides a large population of individuals for inclusion in research studies. Imaging and CSF biomarkers now provide a consensus on an accurate diagnosis of AD, which means that blood-based biomarkers are becoming increasingly accurate. We expect that the advances made in the field of bloodbased biomarkers for AD will continue to accelerate the search for biomarkers for non-AD neurodegenerative disorders.

Targeted studies
Amyloid-β. Validated CSF biomarkers for Aβ plaque pathology in AD, that is, a reduced CSF Aβ 42 concentration or Aβ 42 to Aβ 40 ratio, are highly accurate and reflect the selective retention of Aβ 42 in plaques 80,81 . Aβ plaque pathology is not a common feature of non-AD neurodegenerative disorders (Fig. 1), suggesting that Aβ is a potential tool for differential diagnosis. However, some CSF and imaging data suggest that Aβ pathology is present in individuals with DLB or PDD, and correlates with cognitive decline in these individuals [42][43][44][45][46]84,85 . Lin et al. measured plasma Aβ 42 concentrations in healthy individuals and in individuals with PD, DLB, PSP, CBS, FTD or FTD with parkinsonism 179 . Of all the groups included in the study, individuals with DLB had the lowest plasma Aβ 42 concentrations, although the difference was not statistically significant 179 . Individuals with FTD had significantly higher blood Aβ 42 concentrations than all the other groups. This finding is particularly interesting because PET studies have shown low levels of Aβ pathology in the brains of individuals with FTD 71-73 (Fig. 1). The Aβ 42 to Aβ 40 ratio, which is a more reliable measure of cerebral Aβ pathology than Aβ 42 alone, was not analysed in this study 179 . Therefore, a pressing need is to establish whether the plasma Aβ 42 concentration and/or the Aβ 42 to Aβ 40 ratio is a useful biomarker in distinguishing non-AD neurodegenerative disorders from AD or for monitoring cognitive decline in Lewy body dementias.

T-tau.
Tau messenger RNA (mRNA) and protein has been detected in multiple organs, including the kidney 180 and the salivary glands 181 , but expression is highest in the brain 180 . Tau protein can be detected in plasma, although studies of individuals with AD have found a poor correlation between plasma tau and CSF tau concentrations 182 . Individuals with CJD have higher plasma T-tau concentrations than healthy individuals, individuals with FTD or individuals with AD 183,184 . This observation is consistent with measurements of CSF T-tau in individuals with CJD 86 . Blood T-tau concentrations in individuals with CJD are correlated with disease progression 184 . An IMR study identified significantly higher plasma T-tau concentrations in individuals with a clinical diagnosis of PD, DLB or atypical parkinsonian disorders than in healthy controls 185 , which is contrary to the lack of change that was observed in CSF T-tau concentrations 82,83 . Another study found a two-fold increase in plasma T-tau concentrations in individuals with FTD without parkinsonism compared with those with parkinsonism 179 . Moreover, when measured with Simoa, plasma T-tau concentrations were increased in individuals with bvFTD, PPA (irrespective of subgroup) or certain genetic subtypes of FTD (C9orf72, MAPT and GRN) compared with healthy controls 186 . However, the range of plasma T-tau concentrations in the different diagnostic groups showed large overlaps, and levels of the marker did not correlate with cross-sectional or longitudinal brain volume changes or disease duration 186 .
Evidence from studies in acute hypoxic brain injury demonstrates a biphasic release of T-tau into the bloodstream. This release results in a primary peak of plasma T-tau during the first few hours after injury, and a secondary, broader peak that arises a few days after injury and is predictive of neurological outcome 187 . These rapid changes in plasma T-tau concentrations have also been observed in patients with concussion 188 and during anaesthesia 189 , and might explain the lack of correlation between plasma and CSF T-tau concentrations in studies of neurodegenerative disorders. Preliminary evidence suggests that plasma levels of the N terminus tau fragment tau  are higher in individuals with MCI or AD than in healthy individuals 190 . Whether this tau fragment can be used as a diagnostic or prognostic biomarker in primary tauopathies is not yet clear, but tau  shares commonality with the species measured by some commercial T-tau assays, which will facilitate investigation of this question.
Immunomagnetic reduction (iMr). an immunoassay in which magnetic particles are coated with antibody and the reduction in the spin of the particles correlates with the amount of ligand bound.
www.nature.com/nrneurol P-tau. An increasing number of studies have mea sured plasma P-tau 181 concentrations in individuals with AD 137,[164][165][166][167]191 and some of these studies also included groups of individuals with non-AD neurodegenerative disorders 137,165,166,179 . An IMR study found that plasma P-tau 181 concentrations were significantly higher in individuals with PD, DLB, CBS, MSA or PSP than in healthy controls 179 . When combined with measurements of plasma Aβ 42 , P-tau 181 concentrations could distinguish between individuals with FTD and those with PD, DLB or atypical parkinsonian disorders with a specificity of 88.9% and a sensitivity of 92.9% 179 , which is a promising result in need of replication. By contrast, recent findings showed no difference in plasma P-tau 181 concentrations between individuals with non-AD neurodegenerative disorders and healthy individuals 137,165,166 . Therefore, plasma P-tau 181 concentrations distinguished individuals with AD from those with FTD, vascular dementia, PSP and/or CBS and PD with high levels of accuracy 137 . Plasma P-tau 181 concentrations also accurately distinguished individuals with AD from those with non-AD neurodegenerative disorders as a whole, as indicated by a receiver operating characteristic (ROC) area under the curve (AUC) of >0.93 (reF. 166 ). Despite the high accuracy of plasma P-tau 181 concentrations for the diagnosis of AD, concentrations of the marker were also strongly associated with Aβ PET positivity and CSF P-tau 181 concentration independent of the clinical diagnosis, suggesting that P-tau 181 could detect the contribution of AD-related pathology to non-AD neurodegenerative disorders 165 . For example, individuals with logopenic variant PPA, which is usually associated with AD pathology 56 , had significantly higher plasma P-tau 181 concentrations than individuals with other variants of PPA 165 .

Neurofilaments.
A close correlation between blood and CSF NfL concentrations has been observed in many studies 168,[192][193][194][195][196][197] . In individuals on the AD continuum the correlation coefficients between blood and CSF NfL concentrations reported in the literature range from ~0.5 to ~0.75(reFS 168,169 ), whereas in disorders associated with higher CSF and plasma NfL concentrations (for example, PSP) this correlation is far weaker 98,198 . Therefore, blood NfL could be a surrogate of CSF NfL for the assessment of ongoing axonal injury in some neurodegenerative disorders. However, whether blood and CSF NfL concentrations change concurrently or whether a delay occurs is unclear. Further work is also needed to establish whether this correlation remains strong throughout the course of the disease, which is an important requirement if the blood NfL concentration is to be used as an early marker for neurodegeneration or to monitor response to therapeutic interventions. Another potential confounder is the contribution of peripheral nerve disorders to the blood NfL concentration 199,200 .
Elevated blood NfL concentrations, as measured by the Simoa platform, are observed in almost all neurodegenerative disorders, as well as in inflammatory 201 , traumatic 202 and vascular conditions 203 . Blood NfL concentrations can be used to distinguish between individuals with PD and those with atypical parkinsonian disorders with a level of accuracy that is similar to that achieved with CSF NfL concentrations 98 . One study found higher blood NfL concentrations in individuals with ALS than in those with other neurodegenerative disorders (FTD, AD and PD) and healthy controls 204 . However, in individuals with FTD 205,206 or CJD 96 blood NfL concentrations are close to the concentrations reported in individuals with ALS 204 , and in individuals with PSP, blood NfL concentrations were correlated with disease intensity, clinical deterioration and brain atrophy 207,208 .
The serum concentration of pNfH can also be accurately detected using the Simoa platform 209 . In individuals with FTD or ALS, serum and CSF pNfH concentrations correlated well [209][210][211] . This robust association suggests that peripheral blood pNfH concentrations are a potential biomarker for neuronal damage in non-AD neurodegenerative disorders. Indeed, serum pNfH concentrations can distinguish individuals with ALS from healthy controls with an ROC AUC of >0.90, and can distinguish between individuals with ALS and those with FTD with an ROC AUC of >0.85 (reF. 209 ). Some evidence suggests that, in comparison with measures of NfL, measures of pNfH are more robust 212 , less influenced by pre-analytical variables 213 and exhibit different release and/or clearance dynamics 209 .

Fatty acid-binding proteins.
Fatty acid-binding proteins (FABPs) are small intracellular proteins that facilitate the transport of fatty acids between the cell membrane and different organelles 214 . Brain-derived FABP (B-FABP), which has a similar amino acid sequence to heart-type FABP (H-FABP), is highly expressed in neurons 215 , and increased CSF B-FABP concentrations have been linked to axonal neurodegeneration in AD 216,217 . Furthermore, reduced concentrations of both H-FABP and B-FABP were observed in brain tissue from individuals with Down syndrome and individuals with AD when compared with controls 218 . A study by Teunissen et al. found increased serum B-FABP concentrations in individuals with AD, PD or other non-specified dementias, as compared with controls 219 ; no differences in H-FABP concentrations were observed between these groups of individuals 219 . In other studies, serum B-FABP concentrations were increased in individuals with CJD 220 , DLB [221][222][223] or PD 223 as compared with age-matched healthy controls, and these increases were more pronounced than those seen in the study by Teunissen et al.

α-Synuclein.
Levels of total, oligomeric and phosphorylated α-synuclein in the body fluids of individuals with PD have been evaluated in multiple studies 224 , most of which have had negative results. As mentioned previously, the concentration of α-synuclein is high in red blood cells, therefore the inconsistent results that have come from measuring total α-synuclein in the plasma of individuals with PD 225-227 are unsurprising. However, increases in oligomeric α-synuclein in serum 228 and red blood cells [229][230][231] were observed in individuals with PD and showed a moderate ability to differentiate between individuals with PD and healthy controls. Furthermore, plasma concentrations of phosphorylated forms of α-synuclein were higher in individuals with PD than in healthy individuals, enabling researchers to distinguish between the two diagnostic Receiver operating characteristic (rOC). The rOC curve is a plot of the true-positive rate against the false-positive rate for a diagnostic test. The area under the rOC curve indicates the accuracy of the test; values close to 1 mean that the test reliably distinguishes between the two conditions, whereas a value of 0.50 means that the test is no better than chance. NaTuRe RevIeWS | NEuROLOgy groups with an ROC AUC of 0.71 (reF. 225 ). Similarly, detecting the presence of several post-translational modifications to plasma α-synuclein, for example, Tyr125 phosphorylation and glycosylation, enabled researchers to identify individuals with PD with an ROC AUC of 0.84 (reF. 232 ). More recently, Lin et al. 233 found that plasma total α-synuclein concentrations (measured with IMR) were significantly higher in individuals with PD than in healthy individuals, and this difference was particularly pronounced in individuals with PDD or advanced PD. The results of a further study 179 supported the findings of Lin et al. and showed that plasma α-synuclein concentrations in individuals with PD did not differ from those in individuals with atypical parkinsonian disorders. However, among individuals with FTD, those with parkinsonism had significantly higher α-synuclein concentrations than those without parkinsonism.
GFAP. GFAP is a marker for astrogliosis and is increased in the brain of individuals with neurodegenerative disorders 234 , particularly FTD 235 . Blood GFAP concentrations rapidly increase following the acute structural disintegration of astroglial cells, for example, during intracerebral haemorrhage and traumatic brain injury 236,237 . However, age-related changes in CSF and serum GFAP concentrations have also been observed 238,239 . Subtle changes in serum GFAP concentrations can now be detected using the Simoa platform, and increases were observed in individuals with AD, although the concentrations in individuals with PD or bvFTD were the same as in healthy individuals 240 . One study found that serum GFAP concentrations could distinguish between individuals with AD and those with bvFTD with a sensitivity of 89% and a specificity of 79% 240 . However, data from the Genetic Frontotemporal Dementia Initiative has shown that serum GFAP concentrations are higher in individuals with genetic FTD caused by GRN mutations than in healthy controls, but only mild, non-significant increases in GFAP were observed in symptomatic individuals with MAPT mutations or C9orf72 expansions when compared with controls 238 . In the same study, higher serum GFAP concentrations at the pre-symptomatic stage correlated with lower Mini Mental State Examination scores and lower brain volumes 238 . Interestingly, serum GFAP concentrations were also higher in individuals with Lewy body dementia than in healthy individuals and correlated with cognitive decline; however, serum and CSF GFAP concentrations were only weakly correlated 240 .
TDP43. TDP43 + aggregates are seen in half of individuals with FTD and are common in individuals with semantic variant PPA, but less frequent in individuals with non-fluent variant PPA or CBS 241 . TDP43 can be measured in CSF, but shows substantial expression in blood and blood lymphocytes 242,243 and the CSF TDP43 concentration does not reflect neuropathology in individuals with FTD 243 . However, in up to 50% of individuals with FTD, plasma concentrations of total 244 and phosphorylated 245 TDP43 are higher than in individuals with AD and healthy individuals and are correlated with the severity of TDP43 pathology in the brain 245 . A more recent study found increased CSF and plasma concentrations of phosphorylated TDP43 in individuals with familial FTD caused by mutations in C9orf72 or GRN when compared with individuals with other types of FTD or healthy controls 246 . Increased plasma TDP43 concentrations have also been observed in individuals with ALS 247 . However, TDP43 is ubiquitously expressed in the peripheral tissues 180 , which means that these findings need to be carefully interpreted and further efforts are needed to enable the separation of peripheral TDP43 from CNS TDP43. A limitation of the studies that have investigated TDP43 in biofluids is the use of antibodies that are restricted to peptide regions or phosphorylation sites of TDP43 that are not present on the disease-specific truncated form of TDP43 (reF. 248 ).

Non-targeted proteomic studies
The LC-MS proteomic analysis of blood samples is challenging, although recent studies have successfully highlighted potential blood-based biomarkers for PD [249][250][251] . These studies used proteomic tools such as high-resolution LC-MS/MS and 2D electrophoresis to identify proteins that were present at different levels in individuals with PD and in healthy controls. The most promising candidate identified with this approach seems to be plasma ApoA1. O'Bryant and colleagues used the Meso Scale Discovery panel approach and selected key proteins from their previous cross-validated AD proteomic profile to test in individuals with DLB or PD 252 . One panel of proteins was able to distinguish between individuals with DLB and aged-matched controls with an accuracy of 91%. A second protein panel was able to distinguish between individuals with DLB and individuals with PD with an accuracy of 92%. The proteins detected with these two panels were related to inflammation (IL-5, IL-6 and eotaxin), metabolic dysfunction (adiponectin) and vascular dysfunction (soluble vascular cell adhesion protein 1) in the periphery; however, the specific composition of the two panels showed little overlap 252 . Using a similar approach, King et al. also demonstrated a strongly increased inflammatory component in the blood of individuals with DLB as compared with blood of healthy individuals, but interestingly this increase only occurred during the MCI stage of disease and was also observed in individuals with MCI associated with AD 253 .

Exosomes
Exosomes are cell-derived extracellular vesicles between 30 nm and 100 nm in size that are released into the extracellular space when multivesicular bodies fuse with the plasma membrane 254 . Although exosome studies of neuro degenerative disorders are a relatively new concept, the field has grown substantially over the past 10 years 255 . Exosomes were once primarily thought of as transporters of unwanted cellular debris; however, these vesicles are now known to transfer biomolecules and pathogenic entities across the blood-brain barrier 256,257 . Goetzl et al. have now pioneered the isolation of 'neuron-specific' exosomes with ExoQuick (a commercially available kit by System Biosciences) and immunoprecipitation with an antibody against neural cell adhesion molecule L1 (L1CAM). This methodology has been used in numerous studies of AD and PD and has enabled the detection www.nature.com/nrneurol of altered levels of proteins, including Aβ 42 (reFS 258-260 ), P-tau 258,259,261,262 , cathepsin D 259,260 , RE1-silencing transcription factor [258][259][260] , neurogranin 258,259,263 , DJ1 (reF. 264 ) and α-synuclein 265 , in exosomes from individuals with these neurodegenerative disorders, as compared with exosomes from healthy controls. Interestingly, levels of exosomal GAP43 and synapsin 1 are decreased in individuals with AD, whereas levels in individuals with FTD are the same as in healthy controls 266 .
The potential of using exosome-based biomarkers as objective measures of therapeutic target engagement has been demonstrated in a clinical trial of a glucagonlike peptide 1 (GLP1) receptor agonist in individuals with PD. The GLP1 receptor is known to act via the AKT pathway, and the activity of this pathway was higher in neuronally derived exosomes from participants who were treated with the GLP1 receptor agonist than in exosomes from participants who received placebo, indicating target engagement 267 . Interest in the potential of exosome-based biomarkers is growing 268 , but reliably and efficiently enriching vesicles from biofluids and identifying their contents is difficult 268 . For example, methods of isolating exosomes rely on physical characteristics, such as size, flotation density, cell surface markers and morphology 269 . These properties can be used in combination, and a lack of consensus within the field has led to variations in protocols being used by researchers in different laboratories 268 .

RNA
RNA, especially microrNa (miRNA), remains stable in the blood by being protein-bound or encapsulated within extracellular vesicles 270 . RNA can be detected and measured in all blood fractions using quantitative PCR, northern blotting, oligonucleotide probe fluorescence assays, gene expression microarrays or next-generation RNA sequencing. One hypothesis is that each neurodegenerative disorder has its own unique peripheral miRNA signature 271,272 . Several studies have looked for circulating RNA biomarkers for AD [273][274][275] , and changes in expression levels of specific miRNAs in blood distinguished individuals with AD from healthy controls with up to 93% accuracy 276,277 . However, systematic studies investigating blood-based RNA biomarkers for non-AD neurodegenerative disorders are few.
Total α-synuclein mRNA expression levels in leukocytes did not differ between healthy individuals and individuals with DLB 278 , but levels of an alternatively spliced isoform encoding α-synuclein-126 were increased in the individuals with DLB 278 . Moreover, levels of mRNA from the mitochondrial genes MT-ATP8, MT-CO2, MT-CO3 and MT-ND2 were lower in leukocytes from individuals with DLB than in leukocytes from healthy individuals 279 . The results of one study indicated that in individuals with idiopathic REM-sleep behaviour disorder, lower serum levels of miRNA 19b are associated with a higher risk of developing DLB 280 . Although interest in circulating RNA in individuals with PD is increasing 281 , systematic research into blood-based RNA biomarkers for PDD remains sparse. A small study investigated the blood miRNA profiles of a group of ten individuals with non-AD neurodegenerative disorders, including those with DLB, vascular dementia and FTD, and found a significant downregulation of miRNA 590-5p and miRNA 142-5p, and significant upregulation of miRNA 194-5p in these individuals compared with levels in individuals with AD 282 . Furthermore, in a study investigating brain-enriched miRNA in plasma, the expression levels of miRNA 7, miRNA 9*, let-7 miRNA precursor, miRNA 335-5p and miRNA 451 were able to distinguish individuals with FTD from healthy controls with 88% accuracy 271 . Further research into circulating RNA profiles, especially extracellular vesicle RNA profiles, in individuals with non-AD dementias is essential.

Metabolomics
Metabolomics can be defined as "the unbiased analysis of the composition of small molecule metabolites in a given biological tissue or fluid, under a specific set of environmental conditions" 283 . The sensitivity of the metabolome to environmental changes makes it an ideal source of biomarkers, but this sensitivity also makes analysis of the metabolome susceptible to confounding factors, such as age, gender and body mass index. Several studies have used metabolomics to identify peripheral biomarkers for neurodegenerative disorders; for example, plasma levels of metabolites including sphingolipids, acylcarnitines and amino acids were used distinguish between individuals with AD and healthy controls [284][285][286][287] . A small number of studies have looked for metabolites that could be used to discriminate between individuals with stable MCI and individuals with MCI that is likely to convert to AD. Mapstone et al. identified a panel of metabolite biomarkers that could be used to distinguish between converting MCI and stable MCI with a sensitivity and specificity of up to 90% 288 .
Studies have identified 92 candidate metabolite biomarkers for PD. A total of six of these candidates were measured in more than one study, and of these six metabolites, three (5-acetylamino-6-amino-3-methyl uracil, ala nine and glutamate) showed consistent results between studies [289][290][291][292][293][294] , and three (indole acetate, theophylline and uric acid) showed inconsistent results between studies 289,290,293,295 . Several studies have tested the accuracy of metabolite biomarkers in distinguishing between individuals with PD and healthy controls and found ROC AUC values between 0.83 and 0.95 (reFS 289,290,296,297 ). However, only Stoessel et al. 298 tested the predictive performance of their markers using a random forest model. This machine learning model was trained on 70% of the data to predict the diagnosis of the remaining samples and had an accuracy of 66%. Interestingly, 14 of the candidate biomarkers for PD are also candidate biomarkers for AD 298 , which is a greater overlap than between the individual PD studies. This observation suggests that these metabolites represent generic markers for neurodegenerative disorders as opposed to specific markers for PD or AD. In addition, studies have found that plasma levels of uric acid, which is a product of purine breakdown, are decreased in individuals with early PD 299 , and lower uric acid levels were associated with lower scores in tasks that assessed attention, executive and visuospatial functions 300 .
MicroRNA (mirNa). Non-coding rNa molecules, generally 21 to 24 nucleotides in length, which are usually cleaved from a larger hairpin-containing rNa.

Random forest model
a machine learning algorithm that uses a collection of decision tree data structures to perform regression or classification.
The kynurenine pathway is the main route of tryptophan breakdown in mammals and involves the breakdown of kynurenine by kynurenine-3-monooxygenase to produce 3-hydroxykynurenine (3-HK). Numerous studies have shown that 3-HK is neurotoxic 301,302 through its ability to produce highly reactive free radicals. Quinolinic acid is a downstream product of 3-HK in the kynurenine pathway and has been shown to be an endogenous excitotoxin 303,304 acting via NMDA receptors. Reduced plasma levels of kynurenine 296 and increased plasma levels of 3-HK 290 and quinolinic acid 296 were detected in individuals with PD as compared with the levels in healthy individuals. The kynurenine pathway is also modified in the CSF of individuals with PD, resulting in reduced levels of kynurenine 296 and increased levels of 3-HK 290,305 and quinolinic acid 296 . However, changes to the kynurenine pathway have also been observed in individuals with AD 306,307 and are therefore not unique to PD.
The methionine cycle describes the metabolic pathways involved in the cytosolic re-methylation of homocysteine to produce methionine. The maintenance of this cycle, which is dependent on the presence of vitamins B 9 and B 12 , is often disturbed in individuals with PD or other dementias 308 . In both cross-sectional and longitudinal cohorts of individuals with PD 308 , higher homocysteine plasma levels were associated with cognitive decline.

Conclusions and future directions
The rapid advance in highly sensitive quantitative technologies has led to promising developments in the field of blood-based biomarkers for AD. In the same manner, blood-based biomarkers have the potential to improve detection and diagnosis of non-AD neurodegenerative disorders by increasing accessibility, acceptability and ease of testing, as well as reducing the financial cost. However, compared with AD, far fewer blood-based biomarker studies have been conducted for the non-AD neurodegenerative disorders and discovery screens to identify new putative biomarkers are clearly needed. Nonetheless, clear examples are emerging of existing blood-based biomarkers that could be used in the differential diagnosis of neurodegenerative disorders (Table 3). First, despite blood NfL being a global marker for neurodegeneration, levels of this marker are lower in individuals with PD than in individuals with AD, FTD or atypical parkinsonian disorders. Furthermore, NfL has the potential to be a valuable non-specific neuronal injury marker for use as an outcome measure in clinical trials 309,310 of potential therapeutic interventions for non-AD neurodegenerative disorders. Second, although plasma Aβ species are being rigorously investigated as potential biomarkers for AD, plasma Aβ 42 could also have a role in predicting cognitive decline in individuals with other neurodegenerative disorders with amyloid pathology, for example, DLB and PDD. The plasma Aβ 42 to Aβ 40 ratio has not yet been studied in the non-AD neurodegenerative disorders, but could be particularly applicable to the Lewy body dementias. Last, strong evidence suggests that plasma P-tau 181 has an important role in distinguishing AD from the non-AD neurodegenerative disorders and in determining the contribution of AD-related amyloid and tau pathology in individuals with mixed pathologies. More studies, including longitudinal studies, are needed to test the validity of NfL, Aβ and tau species as blood-based biomarkers in non-AD neurodegenerative disorders. These new studies will  20 ). If these criteria are satisfied, aged-related concentration thresholds for the diagnostic accuracy of these biomarkers also need to be established and certificated reference materials produced, in the same manner as for CSF biomarkers 311 , which will enable normalization across different analytical platforms and laboratories.
In the AD biomarker field imaging methods have been used to detect Aβ and tau pathology in vivo, enabling the diagnosis of AD dementia and pre-clinical AD. This approach has enabled the accurate stratification of research cohorts, which maximizes the likelihood of identifying a robust blood-based biomarker that reflects pathology. However, this Aβ and tau pathology-based stratification is far more common in proteomic biomarker studies than in metabolomic or RNA studies, which remain largely dependent on clinical assessments. To identify and validate blood-based biomarkers for non-AD neurodegenerative disorders, our ability to measure other key protein pathologies, such as TDP43 or α-synuclein, in vivo (with CSF analysis or imaging) has to improve. In the future, the identification of new blood-based biomarkers is likely to follow a tiered approach where autopsy-confirmed pathologies guide the discovery of CSF biomarkers. These CSF biomarkers would then help identify novel bloodbased biomarkers by providing candidates for targeted studies in blood or aiding in the accurate stratification of participants in non-targeted discovery studies. Once validated, blood-based biomarkers for non-AD neurodegenerative disorders could be used to track the development and interaction of co-pathologies over time, as well as to characterize clinical syndromes according to underlying pathology, which would enable personalized treatment and clinical care.
Published online 22 April 2020