Biomarkers for neurodegenerative diseases

Biomarkers for neurodegenerative diseases are needed to improve the diagnostic workup in the clinic but also to facilitate the development and monitoring of effective disease-modifying therapies. Positron emission tomography methods detecting amyloid-β and tau pathology in Alzheimer’s disease have been increasingly used to improve the design of clinical trials and observational studies. In recent years, easily accessible and cost-effective blood-based biomarkers detecting the same Alzheimer’s disease pathologies have been developed, which might revolutionize the diagnostic workup of Alzheimer’s disease globally. Relevant biomarkers for α-synuclein pathology in Parkinson’s disease are also emerging, as well as blood-based markers of general neurodegeneration and glial activation. This review presents an overview of the latest advances in the field of biomarkers for neurodegenerative diseases. Future directions are discussed regarding implementation of novel biomarkers in clinical practice and trials. As the development of biomarkers for neurodegenerative diseases advances, new opportunities arise for their implementation in clinical practice and trials.

80%, but correctly diagnosing other parkinsonian disorders-for example, MSA, PSP and CBD-is more challenging 24,25 . Altogether, biomarkers are needed to improve the diagnostic workup of many neurodegenerative diseases in the clinic, especially during early disease stages. A correct diagnosis will be even more important when disease-modifying therapies become available. Easily accessible, cost-effective and accurate diagnostic biomarkers have the potential for widespread clinical use, including in primary care.
Clinical trials and research. Many potential disease-modifying therapies directed against Aβ, tau or α-synuclein are currently being developed or evaluated in clinical trials. Considering the low accuracy of clinical diagnosis to predict the underlying brain pathologies, there is a clear risk of including patients without the pathology of interest in trials if pathology-specific biomarkers are not used for screening. For example, clinical trials in mild AD dementia show that about 25% of patients fulfilling clinical criteria for AD do not exhibit Aβ pathology 26,27 . Therefore, many AD trials today require biomarker evidence of Aβ pathology for inclusion. However, elderly individuals often have several different brain pathologies 28 , and it is important to determine the primary cause of the symptomatology in each case. For example, an individual with widespread Aβ pathology and considerable TDP-43 (or α-synuclein) pathology might not be appropriate for an anti-Aβ therapy trial if the latter pathology is causing most of the symptoms.
All neurodegenerative diseases exhibit at least some heterogeneity, such as variations in the localization of disease pathology, the level and type of neuroinflammation or the intensity of neurodegeneration. Certain subgroups, potentially defined by biomarkers, might consequently respond better to certain therapies. Even more likely is that different disease-modifying therapies have different optimal time windows during the disease when they are most effective. When targeting upstream pathologies, such as Aβ in AD and α-synuclein in PD, therapies will likely be more effective during early pre-symptomatic or prodromal disease stages before manifest and irreversible neurodegeneration has occurred 29,30 . It is also possible, but not proven, that upstream events (for example, early Aβ pathology) might trigger downstream events (for example, spread Box 1 | Key proteins associated with neurodegenerative disease Aβ Aβ is derived from the transmembrane amyloid precursor protein (APP), which is expressed in many tissues and is concentrated in neuronal synapses. APP is cleaved by β-APP cleaving enzyme (BACE1) followed by γ-secretase to yield Aβ. The Aβ peptide typically contains 37-43 amino acids, depending on the cleavage by γ-secretase, where Aβ40 is the most common isoform. The longer isoforms (Aβ42 and Aβ43) are prone to aggregate, forming smaller aggregates (termed oligomers) and larger insoluble fibrils. Aβ fibrils are the main constituent of extracellular plaques found in AD, and they might catalyze the formation of oligomers-the Aβ variant thought to be most toxic 140 . According to the amyloid cascade hypothesis, Aβ pathology is an upstream event in AD, driving neocortical tau pathology and neurodegeneration 10 .

Tau
Tau is a microtubule-associated protein involved in the assembly and stabilization of microtubules, localized mainly in axons in the central nervous system. In adults, six tau isoforms are expressed from the MAPT gene by alternative mRNA splicing. Three isoforms have four microtubule-binding repeats (4R tau), and three isoforms have three repeats (3R tau). In AD, tau aggregates (filaments) contain a mixture of 3R tau and 4R tau. Similar tau filaments are observed in primary age-related tauopathy (PART), where tau pathology is usually restricted to the medial temporal lobe. By definition, cases with PART exhibit minimal to absent Aβ pathology, and they very seldom have dementia if no other primary co-pathology extists 141,142 . In Pick's disease, tau aggregates consist of 3R tau, and, in PSP and CBD, the aggregates contain 4R tau. Recently, electron cryo-microscopy showed that each disease is characterized by a unique tau filament folding 143,144 . Besides the tau isoform composition, post-translational modifications might determine which tau filament structures are formed. In AD, certain post-translational modifications, including phosphorylation of the proline-reach region, might promote aggregation 145 . Pathologic tau might undergo cell-to-cell transmission, resulting in the transformation of normal tau in the recipient cell into misfolded tau and the formation of tau aggregates 8 . Mutations in the MAPT gene can cause widespread tau pathology, resulting in FTD-like dementia, parkinsonism and/or memory impairment 15 .
α-synuclein α-synuclein is mainly expressed in pre-synaptic nerve terminals. Smaller amounts can also be found in non-neural tissues, including erythrocytes. α-synuclein affects synaptic vesicle release and recycling in nerve terminals. The hydrophobic non-amyloid component domain of α-synuclein is important for aggregation. Similarly to tau pathology, α-synuclein might show prion-like behavior-that is, capacity to self-propagate via templated misfolding and intercellular dissemination 8 . In Lewy body disease (that is, PD and DLB), neuronal α-synuclein aggregates might start in the olfactory bulb or the dorsal nucleus of the vagus nerve and then spread to other brain regions 8,144 . There is some evidence that the pathology sometimes starts in the peripheral nervous system (including the enteric nervous system) before spreading to the brain 144,146 . Several missense mutations, duplications and triplication of SNCA can cause PD and/or DLB, supporting the role of α-synuclein in Lewy body disease 14,17 . In MSA, the structure of α-synuclein filaments differs from those in Lewy body disease, and aggregates are primarily formed in oligodendroglia (glial cytoplasmic inclusions) 144 .

TAR DNA-binding protein 43
TAR DNA-binding protein 43 (TDP-43) is a ubiquitous protein involved in RNA regulation. Post-translational modifications, including cleavage and hyperphosphorylation, can lead to aggregation of TDP-43. Frontotemporal lobar degeneration associated with TDP-43 (FTLD-TDP) is, together with Pick's disease, the most common cause of behavioral variant FTD, a condition with early changes in behavior, personality, emotion and executive control 13 . Similar TDP-43 aggregates are seen in amyotrophic lateral sclerosis (ALS), a motor neuron disease leading to muscle weakness. FTLD-TDP and ALS might be extremes on the phenotypic spectrum of a single disease, because many patients exhibit symptoms overlapping with both diseases. Furthermore, genetic changes in the C9orf72 gene and the gene coding for TDP-43 are associated with both ALS and FTLD-TDP 13,147 . However, TDP-43 aggregates are commonly found in the medial temporal lobes of individualso older than 80 years without clinical ALS or FTD-a condition called limbic-predominant age-related TDP-43 encephalopathy (LATE) 148 . LATE is associated with memory impairment and can mimic symptomatic AD. of neocortical tau pathology), where the latter eventually becomes independent from the initiating event 31 . Therefore, diagnostic biomarkers identifying disease pathology before the onset of overt and disabling symptoms are needed to recruit suitable individuals who are asymptomatic or with only mild symptoms for intervention trials. Furthermore, surrogate biomarkers that can reveal positive effects of treatments on downstream events, such as neurodegeneration, are also likely to be important in such trials, considering that clinical readouts are more difficult in pre-symptomatic disease stages. Finally, disease-modifying therapies, where drug target engagement in the human brain has been firmly established in early phase 1-2 studies, are more likely to be shown as clinically effective in subsequent large-scale trials 32 . Consequently, development of pharmacodynamic biomarkers reflecting relevant drug targets in vivo are crucial.
Biomarkers are also important in clinical observational studies. Although neuropathological studies are very valuable, longitudinal clinical studies spanning from early pre-symptomatic disease through symptomatic stages are critical to understand how pathological events emerge, develop and interact with each other over time and how these relate to different clinical symptoms. Accurate markers of different disease pathologies are also important when establishing relevant risk factors and protective factors, including in studies of genetic, epigenetic, developmental and lifestyle factors 33 .

current state-of-the-art biomarkers
Biomarkers for Aβ pathology. Studies using positron emission tomography (PET) indicate that the accumulation of Aβ starts ~20 years before dementia onset in AD 5,6 . Initial Aβ deposits preferentially form in the medial parietal and frontal cortex, areas overlapping with the default mode network, before other neocortical regions are affected (Figs. 1 and 2c) 5,[34][35][36] . Three PET ligands-that is, Vizamyl (( 18 F) flutemetamol), Amyvid (( 18 F) flobetapir) and Neuraceq (( 18 F) florbetaben)-have been approved by regulatory authorities in several countries for the diagnostic workup of AD. Using postmortem pathology as a standard of truth, in vivo imaging studies have shown that these PET ligands can detect insoluble Aβ fibrils in plaques with very high accuracy [37][38][39] . Consequently, a negative Aβ-PET scan in a patient with cognitive impairment will essentially rule out AD as the underlying etiology. The specificity of Aβ-PET decreases with age, however, as cortical Aβ pathology occurs in ~10-15% of individuals with normal cognition at age 60 and in ~40% at age 90 40,41 . In patients with MCI, Aβ-PET can be used to identify those who are at increased risk of developing AD dementia 42 , and, in cognitively normal individuals, abnormal Aβ-PET scans are associated with subsequent cognitive decline 43 . Guidelines have already been established for the appropriate use of Aβ-PET in clinical practice 44 , and, importantly, clinical use of Aβ-PET has been shown to result in meaningful changes in the management and treatment of patients with either MCI or dementia of uncertain etiology 45 .
Different Aβ species can be reliably measured in cerebrospinal fluid (CSF). The levels of CSF Aβ42 (but not Aβ40) are decreased by ~50% in AD 46 , and the drop occurs even before Aβ-PET becomes abnormal (Fig. 2c) 47 . Using a ratio of Aβ42 to Aβ40 (Aβ42/Aβ40) results in high concordance with Aβ-PET, typically above 90%. The efficacy of the ratio might be due to control of inter-individual variations in: (1) overall production or secretion of Aβ from neurons; (2) production and clearance of CSF; and (3) pre-analytical handling of CSF samples that affects both Aβ isoforms 48 . CSF Aβ42, when used together with Aβ40 or phosphorylated tau (P-tau), predicts the subsequent development of AD dementia in patients with MCI with high accuracy 49 . To facilitate clinical implementation, fully automated methods with high analytical precision and stability have been developed, which can be implemented in ordinary chemistry laboratories 50,51 . Furthermore, certified reference materials and methods are now available to enable establishment of uniform thresholds for diagnosis and prognosis 52 . An international consensus protocol for the collection and handling of CSF has been established to avoid false abnormal values caused by pre-analytical confounders 53 . Importantly, appropriate use criteria guiding the clinical use of these tests have been published 54 . Even though CSF AD biomarkers are already used in clinical practice in many countries, the collection procedure (lumbar puncture) is invasive and not widely available, especially not outside of specialist centers. Therefore, substantial efforts have been made to develop blood-based assays for detecting cerebral Aβ pathology. Most attempts were unsuccessful until a study in 2016, using an ultrasensitive method quantifying plasma Aβ42/40, demonstrated prediction of abnormal Aβ-PET outcome with moderate accuracy 55 . Thereafter, several mass spectrometry-based methods were developed for determining Aβ in plasma, and some of these methods can now detect cerebral Aβ pathology with clearly higher accuracies than most immunoassays [56][57][58] . Fully automated methods for plasma Aβ42/40 have also been developed, which can facilitate clinical implementation 59 . However, the levels of plasma Aβ42/40 are decreased by only 10-20% in individuals with cerebral Aβ pathology, compared to 40-60% for CSF Aβ42/40, likely because the plasma levels are also affected by production of Aβ outside the brain 57-60 . Consequently, CSF Aβ42/40 exhibits a higher diagnostic accuracy than plasma Aβ and is less susceptible to variations in cutpoints for determining abnormal status 57 . Future combinations of different blood-based biomarkers-for example, Aβ42/40, P-tau and other proteins such as glial fibrillary acidic protein (GFAP)might potentially reach the same accuracy as CSF Aβ42/40 for detection of cerebral Aβ pathology 61,62 .
Biomarkers for tau pathology. Several tau-PET ligands, such as ( 18 F) flortaucipir, ( 18 F) MK6240 and ( 18 F) RO948, exhibit binding to insoluble tau fibrils in AD brain tissue but little or no binding to tau Aβ-PeT and the CSF or plasma Aβ42/Aβ40 ratio reflect Aβ pathology in the brain. Tau-PeT imaging and, to some degree, plasma and CSF P-tau reflect tau-tangle pathology in AD. However, CSF and plasma P-tau do probably also reflect the increased secretion and phosphorylation of tau caused by Aβ pathology. Misfolded α-synuclein (α-syn), which forms Lewy bodies and neurites in PD and DLB, can be detected in, for example, CSF and skin by α-syn seeding aggregation assays such as α-syn RT-QuIC. nfL is primarily a marker of the degeneration of myelinated axons. neurogranin is a marker of post-synaptic degeneration and dysfunction, and there are several emerging markers of pre-synaptic degeneration, including SV2A, SnAP-25 and GAP-43. Plasma GFAP is a biomarker of reactive astrocytes, and sPDGFR-β might be a marker of damage to capillary pericytes.
pathology in other neurodegenerative diseases 63 . ( 18 F) flortaucipir is the most studied ligand and has been approved by the U.S. Food and Drug Administration for clinical use in AD diagnostics. The regional in vivo uptake of ( 18 F) flortaucipir correlates with the density of tau pathology measured postmortem in cases with AD 64,65 , and visual or quantitative assessments of ( 18 F) flortaucipir PET can accurately identify neocortical AD-like tau pathology 66,67 . Large-scale clinical studies have shown that the diagnostic precision of tau-PET outperforms magnetic resonance imaging (MRI) when separating AD dementia from other neurodegenerative diseases, and the specificity of tau-PET is even higher than that of Aβ-PET and CSF Aβ42/40 (refs. 68,69 ). However, the prognostic value of tau-PET is still unclear. Cross-sectional studies show that tau-PET is more strongly associated with cognitive deficits than Aβ-PET in both cognitively normal and impaired individuals, which is in agreement with autopsy-based studies [70][71][72][73] . Preliminary evidence suggests that tau-PET can predict subsequent cognitive decline and brain atrophy across the entire AD continuum [74][75][76][77] , but the prognostic value of tau-PET should be compared with MRI, Aβ-PET and fluid biomarkers in future studies; preliminary results indicate that tau PET outperforms MRI and Aβ-PET in early AD in this context 78 . Cross-sectional tau-PET studies support the pathology-based 'Braak staging scheme' [79][80][81][82] , suggesting that AD-like cortical tau emerges in the (trans)entorhinal cortex, followed by other medial temporal lobe areas with increasing age. In Aβ-positive cases, tau pathology further spreads through the neocortex in a stereotypic fashion (Fig. 2b). The Braak model, however, does not account for inter-individual differences in tau deposition. Even though AD is typically dominated by memory impairment, certain rarer, atypical variants of AD are dominated by language, visual or executive dysfunctions, respectively. Differences in regional tau patterns seem to explain these different clinical phenotypes, with symmetric temporoparietal tau pathology in the amnestic-predominant variant, asymmetric left temporal tau in the language-predominant variant, posterior cortical tau in the visual-predominant variant and widespread neocortical tau with sparing of the medial temporal lobe in the dysexecutive-predominant variant of AD (Fig. 2d) 83 . A recent tau-PET study indicated that all four of these tau subtypes are actually relatively common (ranging in prevalence from 18% to 33%), with the atypical clinical phenotypes representing the extremes of the three latter tau-PET subtypes 84 . Future studies are needed to unravel whether these AD subtypes respond differentially to certain interventions.
Considering that tau is an important drug target in AD, studies focusing on modifiers of tau accumulation and spread are of interest. Several tau-PET studies now provide evidence that tau pathology might spread through neuronal communication pathways, supporting the idea of trans-synaptic transmission of misfolded tau in AD 80,85 . This process seems to be accelerated by regional Aβ pathology 80,86 , and longitudinal tau-PET studies confirm that neocortical tau accumulation occurs almost exclusively in Aβ-positive individuals 76,82,87 . Furthermore, faster tau accumulation is associated not only with higher baseline tau and Aβ load but also with female sex and younger onset of AD 76,88 .
Different soluble tau species can be reliably measured in CSF. Total tau (T-tau) and tau phosphorylated at threonine 181 (P-tau181) have been widely studied, where the latter is selectively increased in AD and not in other neurodegenerative diseases 89 . Fully automated methods have been developed for CSF T-tau and P-tau181, and both can be used together with CSF Aβ to predict cognitive decline in patients with MCI 90 . However, tau can be phosphorylated at more than 40 different positions, and tau phosphorylated at threonine 217 (P-tau217) exhibits a somewhat stronger association with tau-tangle load and disease severity than P-tau181 (ref. 91 ). Furthermore, CSF P-tau217 might distinguish AD dementia from other dementias with even higher accuracy than P-tau181 (refs. 91,92 ). Recent data also suggest that tau fragments in CSF containing the microtubule-binding region might be more associated with tau-tangle pathology than other tau species 93,94 .
Reliable assays have recently been developed that can detect P-tau in blood. Autopsy-based studies have shown that plasma P-tau181 and P-tau217 can accurately differentiate individuals with AD neuropathologic changes from those without, including those with non-AD tau pathology [95][96][97][98] . In clinic-based cohorts, plasma P-tau can also, with high accuracy, separate AD dementia from other neurodegenerative diseases 95,[97][98][99] . In this setting, plasma P-tau217 actually performed equivalently to CSF AD biomarkers and tau-PET imaging, with an accuracy of around 90% 97 . In patients with MCI, higher baseline P-tau levels in plasma are associated with subsequent cognitive decline and conversion to AD dementia 95,100 . Importantly, plasma P-tau-based diagnostic algorithms can be used for individualized prognosis of patients with MCI, and they perform as well as CSF-based algorithms 101 . Online tools have been developed to illustrate how such algorithms can guide physicians in the future 101 . Efforts to define optimal blood-based diagnostic algorithms for the detection of individuals with pre-symptomatic AD are ongoing, which could be used for screening of suitable participants for pre-clinical AD trials 102 . Finally, longitudinal studies have shown that plasma P-tau (especially P-tau217) exhibit clear increases over time in both the pre-symptomatic and prodromal phases of AD and could potentially be used to detect pharmacodynamic effects of disease-modifying therapies 103 .
Recent studies have investigated how soluble P-tau levels relate to the amount of insoluble aggregates of tau and Aβ, respectively, in the brain. Neuropathology-based and PET-based studies indicate that levels of plasma P-tau181 and P-tau217 are associated with both Aβ plaques (during early disease stages) and tau-tangles (during later disease stages) 104 . In autosomal dominant AD, plasma P-tau181 and P-tau217 increase 15-20 years before dementia onset and before widespread neocortical tau-tangle pathology 97,105 . Also in sporadic AD do the levels of P-tau181 and P-tau217 (in both plasma and CSF) start to increase during the pre-symptomatic phase when cortical Aβ fibrils emerge and before insoluble tau-tangles can be detected with tau-PET in the neocortex 95,[106][107][108][109] . Several studies have shown that soluble P-tau levels mediate the effect of Aβ fibrils on the formation of neocortical tau-tangles, congruent with the idea that Aβ pathology might increase the production, phosphorylation and secretion of tau, which, in turn, is associated with buildup of neocortical tau-tangles 107,109,110 . Interestingly, antibody treatment against Aβ fibrils has been shown to lower the levels of soluble P-tau in CSF 27 , and future studies are needed to unravel whether this effect is associated with subsequent reduction in the accumulation of neocortical tau-tangles and, ultimately, reduced neurodegeneration.
Biomarkers for α-synuclein pathology. α-synuclein can be reliably detected in CSF, and levels are decreased, on average, in PD, DLB and MSA. However, there is a substantial overlap with controls and other neurodegenerative diseases, thereby hindering its utility in clinical practice and trials 111 . Ultrasensitive cell-free aggregation assays have been developed that indirectly reveal misfolded prion protein and other prion-like proteins. Such seeding aggregation assays-for example, real-time quaking-induced conversion (RT-QuIC) and protein-misfolding cyclic amplification (PMCA)can be used to detect pathological forms of α-synuclein in CSF. These CSF-based assays exhibit high sensitivity (>90%) and specificity (>97%) for detection of Lewy body disease (that is, PD and DLB) in neuropathologically confirmed populations with parkinsonism or dementia 112,113 . Different seeding aggregation assays exhibit high concordance for the detection of misfolded CSF α-synuclein when evaluated in the same cohort, indicating high reproducibility among the assays 114 . α-synuclein neuropathology differs between MSA and PD/DLB (Box 1), and a CSF-based α-synuclein seeding assay can be used to differentiate between these disorders 115,116 . Interestingly, the properties of the aggregates in CSF were different between PD and MSA, which suggests that different conformational strains of α-synuclein exist in these diseases 116 .
Considering that CSF collection is invasive, detection of misfolded α-synuclein in more easily accessible fluids and tissues would be a great step forward. Preliminary results indicate that skin biopsies, which include nerve terminals, can be used in seeding aggregation assays to reliably detect misfolded α-synuclein in PD, DLB and MSA 117,118 . Other potential peripheral tissues for detection of misfolded α-synuclein include the olfactory mucosa and the submandibular gland.
Biomarkers reflecting neurodegeneration. Structural MRI, including T1-and T2-weighted imaging, is used to study patterns of brain atrophy as an estimate of regional neurodegeneration. Because MRI measures such as gray matter volumes and cortical thickness often exhibit low intra-individual variability over time, rather small differences in atrophy rates can be detected in longitudinal clinical trials 119 . However, MRI cannot provide information on loss of particular cell populations or cellular structures. Molecular imaging, including PET, can be used to, for example, determine the density of dopaminergic nerve terminals in the basal ganglia, which is reduced in PD as well as in PSP, MSA and CBD 120 . These methods are often applied in studies evaluating potential neuroprotective effects of treatments on dopaminergic neurons. Recently, PET ligands binding to a synaptic protein called synaptic vesicle protein 2A (SV2A) were shown to detect regionally decreased synaptic density in several neurodegenerative disorders, including AD and PD 121,122 . Changes in SV2A PET correlates quite well with ( 18 F) fluorodeoxyglucose (FDG) PET retention in AD 123 , where the latter assesses regional glucose metabolism. FDG PET has been used for decades in many clinics in the diagnostic workup of several neurodegenerative disorders, because certain patterns of regional hypometabolism can imply a specific neurodegenerative disease. For example, frontal and anterior temporal hypometabolism is typically observed in the behavioral variant of FTD 124 .
Several fluid biomarkers of neurodegeneration have recently emerged. Most promising is neurofilament light (NfL), which can be measured in both CSF and blood. This biomarker reflects axonal degeneration and injury, irrespective of cause, and the levels are especially increased in ALS, FTD and atypical parkinsonian disorders (that is, PSP, MSA and CBD) 125,126 . However, NfL levels are also increased in AD, and studies on autosomal dominant AD show that the rate of change in blood NfL increases already about 15 years before symptom onset 127 . Importantly, higher levels of NfL are associated with faster disease progression and higher brain atrophy rates in most neurodegenerative disorders 125,127 . As a result, NfL can be regarded as a measure of the intensity of ongoing neurodegeneration. In several brain diseases, including multiple sclerosis and spinal muscular atrophy, effective disease-modifying treatments can normalize NfL levels, and reductions in NfL levels are associated with the clinical effectiveness of the treatment 128,129 .
Neurogranin is a post-synaptic protein whose levels in CSF, but not blood, are selectively increased in patients with AD and can predict future cognitive decline 130,131 . The CSF levels start to increase already during the pre-symptomatic phase of AD when Aβ-containing plaques have emerged, indicating early dysfunction and degeneration of post-synaptic entities in AD 106 . Assays have also been established for the quantification of several pre-synaptic proteins in CSF, including GAP-43 and SNAP-25. Future studies will determine whether the novel CSF and PET biomarkers of synaptic integrity can improve diagnostic workup and clinical trials beyond more established biomarkers of neurodegeneration, including MRI, plasma NfL and FDG PET.
Biomarkers of glial activation and blood-brain barrier function. GFAP, which likely reflects reactive astrocytes, can be reliably measured in both blood and CSF. Plasma levels of GFAP are increased in individuals with Aβ pathology 61 and can predict both future cognitive decline and conversion to AD dementia in cognitively unimpaired individuals 132 as well as in patients with MCI 133 . Plasma GFAP is also increased in other neurodegenerative conditions, including FTD caused by progranulin mutations 134 . However, future studies are needed to evaluate whether antemortem levels of plasma GFAP correlate strongly with the number of reactive astrocytes determined postmortem using immunohistochemistry.
The integrity of the blood-brain barrier is disrupted in several neurodegenerative diseases and can be evaluated by certain MRI techniques, including dynamic contrast-enhanced and dynamic susceptibility contrast MRI, which both depend on leakage of gadolinium-based contrast agents into the brain 135 . Furthermore, the CSF-to-plasma ratio of albumin is also often used to reflect the integrity of the blood-brain barrier, and it has been shown to be elevated in most dementia disorders independent of AD pathology 136 . Recently, it was suggested that CSF levels of soluble platelet-derived growth factor receptor beta (sPDGFR-β) is a marker of damage to capillary pericytes, which is associated with worse cognitive function 137 .

conclusions and future directions
Outstanding progress has recently been made when it comes to blood-based biomarkers for AD (Aβ and P-tau) and neurodegeneration (NfL), as well as PET imaging of AD tau pathology. However, there are still major gaps in our arsenal of biomarkers for neurodegenerative diseases. Even though the seeding aggregation assays for CSF α-synuclein have been a great step forward, PET imaging ligands detecting α-synuclein aggregates will be crucial for studying the development and regional spread of α-synuclein pathology in PD, DLB and MSA and to determine drug target engagement of anti-synuclein therapies. Blood-based assays revealing brain α-synuclein pathology would considerably facilitate studies in larger populations, but skin biopsies might also provide an effective alternative. The research community also needs to develop PET and fluid biomarkers for non-AD tau pathologies, including 3R and 4R tau pathologies, as well as for TDP-43-related pathologies (Box 1). Analyses of CSF are likely to be central in this process, because the levels of brain-derived proteins are much higher in CSF than blood, where brain-derived molecules are diluted in a complex matrix of abundant plasma proteins, such as albumin and immunoglobulins. An important note is that such novel biomarkers will need to be validated against neuropathology, considering the rather low precision of clinical diagnosis.
Already today, CSF and PET measurements of Aβ and tau pathology can be used to improve the accuracy of clinical diagnostics of AD, and they are very valuable tools in research aiming to better understand the disease. When an anti-Aβ therapy is approved for AD, Aβ-PET or CSF Aβ42/Aβ40 is likely to be used to confirm the presence of Aβ pathology before initiating the therapy in the clinic, and, potentially, repeated PET scans could be performed to reduce the dosing of the anti-Aβ treatment once the Aβ-PET signal is reduced to normal levels, as was done in a recent phase 2 trial for donanemab 138 . Tau-PET is likely to be used in a similar fashion if an anti-tau treatment becomes available, but tau-PET could already today improve the diagnostic workup of selected patients with dementia, considering its high specificity for AD. However, PET and CSF AD biomarkers are not widely available, and they are either very costly (PET) or invasive (CSF), which is likely to restrict their use in clinical practice to a more limited number of specialized centers. On the other hand, the development of blood-based biomarkers for AD has the potential to truly revolutionize the diagnostic workup of AD globally. That said, several crucial next steps are needed before the implementation of these blood biomarkers in the clinic. First, high-precision in vitro diagnostic (IVD) tests need to be developed and approved by relevant regulatory authorities. Preferentially, such IVD tests are in the form of automated assays that can be easily used in standard or referral chemistry laboratories worldwide. Furthermore, establishment of universal cutoffs for abnormality would facilitate widespread clinical implementation. Second, standard operating procedures for collecting and handling blood samples must be established, and such international efforts are currently ongoing. Third, even though plasma P-tau alone might be good enough for improving the diagnosis of AD dementia 97 , optimized diagnostic algorithms combining different blood-based biomarkers, and potentially also other easily available and cost-effective tools, must be established for individualized prediction of AD dementia in patients with mild cognitive symptoms. Fourth, such diagnostic algorithms must be superior to the standard clinical assessments by dementia experts and should preferably be non-inferior to CSF-based or PET-based methods. Recently, it was shown that a cross-validated prognostic algorithm based on plasma P-tau, APOE4 genotype and a few brief cognitive tests of memory and executive function outperformed the assessment of dementia experts when predicting subsequent development of AD dementia in patients with mild cognitive symptoms 139 . Fifth, appropriate use criteria must be established to avoid misinterpretation and misuse of these tests in the clinic. Even though blood-based diagnostic algorithms for AD will first be implemented in specialized clinics, they are likely to also improve the diagnostic workup in primary care. However, prospective studies are needed in primary care to truly establish whether blood-based diagnostic algorithms can improve the diagnostic workup of AD in that setting as well and preferably also improve treatment and care. Furthermore, we need to develop easily accessible diagnostic algorithms for neurodegenerative diseases other than AD. Potentially, blood-based algorithms detecting ongoing neurodegeneration might be used to identify symptomatic patients with high risk of having a neurodegenerative disease for referral by primary care to specialist centers for further assessments (Fig. 3).
Biomarkers are crucial for the development of disease-modifying therapies. It should be a goal already in phase 1b-2 studies to demonstrate that novel disease-modifying therapies exhibit relevant drug target engagement in the human brain, as it is important for both ethical and economic reasons to avoid large-scale trials of drugs not likely to work 32 . In AD trials, Aβ-PET or tau-PET are often suitable for determining drug target engagement of anti-Aβ or anti-tau treatments. In trials including individuals with pre-symptomatic or prodromal disease, non-invasive and low-cost biomarkers are needed to identify potential study participants most likely to respond to treatment (Fig. 3). In such AD trials, screening using different combinations of plasma P-tau, Aβ42/40 and NfL are likely to identify individuals with pre-symptomatic and prodromal AD 62 , although, in at least some trials, it will be appropriate to confirm AD pathology using either CSF or PET in patients who screened positive using blood-based biomarkers (Fig. 3). In prodromal PD/DLB trials, individuals with REM sleep behavior disorder who exhibit a positive α-synuclein seeding aggregation test could be a target population of interest. However, these assays require additional standardization, including further validations in inter-laboratory settings, standardization of the recombinant substrate and large-scale studies using peripheral tissues, such as the skin.
Biomarkers that can measure relevant treatment effects downstream of the drug target are likely to be used more frequently to optimize go/no-go decisions on moving drugs forward from phase 2 trials to large-scale and extended phase 3 trials (Fig. 3). Biomarkers of neurodegeneration are likely to be particularly relevant in this setting. Decreased decline in measures reflecting the number of remaining brain cells (for example, MRI measures of cortical thickness or dopamine transporter imaging) are already used in many trials. However, markers reflecting the intensity of neurodegeneration might be even more promising. The fact that actual levels of NfL, and not only the rate of change of NfL, can be normalized by effective treatments in different brain disorders makes this marker a key candidate, but others are likely to emerge. In the future, such blood-based markers could potentially be used in clinical routine practice to monitor the efficacy of disease-modifying therapies in individual patients.  The upper row shows how diagnostic algorithms based on easily accessible and time-and cost-effective biomarkers might guide primary care physicians in the future when deciding which patients should be referred to specialized clinics for final diagnosis and initiation of relevant treatments. However, in certain countries, elderly patients with, for example, AD are often not referred to specialized clinics, and, in these situations, the diagnostic and prognostic algorithms could improve the diagnostic workup directly in primary care (not shown). The lower row shows how biomarkers can be used to identify relevant individuals for pre-clinical (pre-symptomatic) trials. These individuals should have biomarker evidence of the underlying pathology for which the disease-modifying treatment is directed, and they should be likely to deteriorate clinically in the coming years. Furthermore, biomarkers should be used to identify drug target engagement and to determine potential downstream treatment effects (such as reduced neurodegeneration (nD)).