Blood-based biomarkers for Alzheimer’s disease: mapping the road to the clinic

Alzheimer’s disease

Alzheimer’s disease (AD) is a clinically and pathophysiologically heterogeneous complex neurodegenerative disease (ND). AD is the most common cause of age-related ND, impacting millions of individuals worldwide; currently, one out of nine people over the age of 65 are living with AD¹ and the prevalence of AD is expected to grow exponentially over the next several decades¹.

The pathogenesis of AD involves interacting pathophysiological cascades, including core events, i.e. accumulation of the 42-amino acid-long amyloid beta peptide (Aβ_1–42) into amyloid plaques in the brain parenchyma and the formation of intraneuronal neurofibrillary tangles composed of hyperphosphorylated tau protein². Emerging evidence stresses the existence of additional molecular pathophysiological pathways, such as axonal disintegration³, synaptic dysfunction and degeneration⁴, innate immune response and neuroinflammation^5,6, vascular and cell membrane dysregulation⁷, and brain metabolic dysfunction⁸ – across the different stages of AD. Moreover, other proteinopathies and pathologies frequently co-exist in the ageing brain. These include TDP-43 or α-synuclein proteinopathies, non-AD tauopathies, vascular pathology, and hippocampal sclerosis^9–12. For these reasons, establishing a definitive diagnosis and developing effective treatments of AD is complicated.

At present, the aetiology and pathogenesis of AD is the subject of ongoing research and debate. The amyloid cascade hypothesis proposes that the brain accumulation of aggregated forms of Aβ is the trigger and/or driver of the disease process¹³. However, recent studies raised questions about this hypothesis as the exclusive cause and/or intervening link between the pathophysiology of AD and its clinical phenotype. The notion that biochemical and cellular mechanisms generate complex cognitive alterations has renewed AD research, leading to replace the first descriptive studies with a molecular, mechanistic view. The exponential increase in knowledge on interacting pathogenic mechanisms in individuals suffering from AD holds promise for the development of future biomarker-guided _targeted therapies and prevention strategies^14–17.

The potential impact of biomarkers on primary care and neurology for the detection and diagnosis of AD

Given the complex clinical phenomenology, clinical, neurological, and neuropsychological examinations are still an integral component of accurate late-stage detection of clinically symptomatic AD and other ND. However, waiting times for appointments with specialists in the U.S., U.K. and Ireland (and other countries) can be very long resulting in substantial and often critical delays for patients and providers¹⁸. Memory clinics or general neurology clinics in many countries receive a broad range of referrals covering many conditions and diseases, therefore the improved streamlining of referrals to specialty clinics can have a significant impact on health care utilization and costs¹⁸. As one example, recent U.S.-based legislation requires that elderly people aged 65 and older receive annual cognitive examinations as part of the Annual Wellness Visit (CMS.gov)¹⁹; however, older adults continue to be inadequately assessed for cognitive decline during primary care visits²⁰. Given that the average duration of primary care visits for geriatric patients is 21 minutes²¹, this is perhaps not surprising. Additionally, cognitive examinations are regularly administered, scored, and interpreted incorrectly in primary care due to lack of training and expertise^22,23 and there are significant differences between primary care and specialty care approaches to diagnosis, treatment, and social support²⁴. Therefore, a process that aids primary care practitioners in deciding which patients should receive a referral to a memory clinic would be of substantial advantage to both specialists and general practitioners. Such a system would reduce the overall clinic and medical system burden by decreasing the numbers of unnecessary referrals and diagnostic procedures^25,26. Biomarker-based diagnostics can greatly aid a multi-stage selection of patients into appropriate centres.

To date, the literature has focused on diagnostic biomarkers^27,28 for specialty clinic settings with little attention to screening instruments required for broad-based implementation in primary care settings. Additionally, diagnostic paradigms continue to rely substantially on clinical symptoms, with only relatively recent research guidelines taking biomarkers of AD pathophysiology into account^29,30. Up to the mid-2000s, diagnosis of AD had a “clinicopathological” basis, with no definitive positive diagnosis possible until post-mortem confirmation of the presence of Aβ plaques and neurofibrillary tangles, and clinical diagnosis being assigned after a patient had reached the late-stage syndromal dementia threshold, by exclusion of other aetiologies of dementia²⁹. This concept of late-stage diagnosis of exclusion focusing on clinical phenotype showed profound limitations, given that clinical symptoms are by nature heterogeneous (including an increasing number of atypical clinical phenotypes) and vary regarding onset, progression, and sequence of events. They have considerable overlap with other central nervous system (CNS) proteinopathies causing dementia as well as with dementia due to cerebrovascular disease³¹. More recently, with the advent of reliable radiotracers with affinity for cerebral amyloid deposits³², and the availability of clinically well-validated assays for cerebrospinal fluid (CSF) detection of Aβ peptides and tau proteins³, there has been progress towards a “clinicobiological” diagnostic approach in which pathophysiological and topographic biomarkers are integrated into the diagnostic paradigm²⁹ as adjuncts to core clinical criteria³⁰. These biomarkers are clinically available in specialty clinic settings for some countries, providing the means for dementia specialty clinicians to confirm diagnosis with much higher certainty through the presence of characteristic features of AD pathophysiology, thereby allowing for a more conclusive aetiological diagnosis^14,33,34. Nevertheless, use of these technologies is not yet the standard and they are restricted to specialty clinic settings. On the other hand, biomarkers are now being incorporated into the comprehensive clinical diagnostic process, which will result in improved overall accuracy in detection of pathophysiology. However, as was the case for more matured translational research areas of biomedicine, such as oncology and cardiovascular medicine, the field is in need of biomarker-based diagnostics that can accurately and reliably identify individuals at risk and patients as precise and early as possible in the disease process^35–39, preferentially in primary care settings. The availability of these biomarker-based diagnostics (e.g. CSF Aβ_1–42 and tau) is also expected to open the window for precision medicine and allow a shift away from the traditional “one-size-fits-all” approach with “magic bullet” drugs in neuroscience to the development of biomarker-guided _targeted therapies^25,40–43.

Facilitation of the precision medicine paradigm in ND such as AD requires an array of converging advanced analysis methods. Through systems theory, it is possible to conceptualize original and innovative models to explicate all systems levels – assessed via systems biology and systems neurophysiology⁴² – and different data dimensions in space and time of the non-linear, dynamic, and chronically progressive nature of the clinically and biologically heterogeneous construct of polygenic AD^35,44, historically developed after the clinical description of the first patients and the associated brain histopathological remarks⁴⁵.

The agnostic, hypothesis-free systems theory approach seems appropriate to explain the complex and heterogeneous origin and time course of the pathophysiological failure underlying the different forms of AD³⁵. For multifactorial diseases, comprehensive holistic systems-level approaches are required; this is the case of the systems biology model aiming at understanding the genotype-phenotype relationships and the mechanisms at the level of the genome/epigenome, transcriptome, microRNome, proteome/peptidome, metabolome/lipidome, microbiome, and environmental factors participating in complex cellular networks^35,44,46. Longitudinal investigations using systems biology-based strategies help characterize the complex molecular pathophysiology of the subforms of AD, all evolving through the convergence of alterations of homeostasis and/or failures in many systems, networks, signalling pathways, or pathophysiological processes⁴⁴.

One of the key objectives of precision medicine is to bring new models for prevention, early detection, differential diagnosis, and treatment of AD and other ND, according to individual biological differences reflected by multimodal biomarkers^33,35,37. Therefore, following the advanced models of oncology and cardiovascular medicine, innovative biomarker studies are expected to detect specific diagnostic, prognostic, and predictive biomarker signatures to adapt the therapy to individual patients⁴⁰. As reported by the Institute of Medicine (IOM) Committee Recommendations for Advancing Appropriate Use of Biomarker Tests (companion diagnostics) for Molecularly _targeted Therapies, the ultimate goal of precision medicine is to enhance both clinical outcomes and the quality of patient care⁴⁷.

Precision medicine is being used in AD and in a growing number of ND as a result of: 1) the development of large-scale biological databases and 2) the evolution of advanced high-throughput “omics” sciences. Novel methods developed within the “omics” disciplines have been successfully applied in the more advanced cancer area and are expected to transfertilize to AD and other ND. For instance, validated microarray expression profiling and current RNAseq technologies allow identifying differential expression of whole genomes in any specific sample at any given time point. Numerous reports exploiting these transcriptomics methods to recognize biomarkers in specific cancer subtypes are present⁴⁸. The participation of microRNAs (miRNAs) in crucial cellular processes including cell death and their role in negative control of the expression of various oncoproteins make them interesting candidate biomarkers for cancer. Cancer-specific markers are detected in the blood from the earlier stages of tumor development. Since their concentration increases as tumor advances over time, they are useful dynamic indicators of tumor growth⁴⁹. In proteomics, innovative methods of immuno-PCR, based on conjugations of specific antibodies and nucleic acids and exploiting ultrasensitive PCR signal amplification, can result into a 100 to 10,000-fold increase in sensitivity compared with the analogous enzyme-amplified immunoassays, hence considerably increasing the sensitivity of protein biomarker detection^50,51. Powerful computational and integrative network biology tools are needed to assimilate the large multimodal information generated by the “omics” exploratory analyses and to comprehensively understand the decompensations at systems level⁵². An advantage in the AD field is that many of the pathologies are well known. Hence hypothesis-free systems theory approaches may be advanced in synergy with the refinement of methods that directly reflect core pathologies, much like what has recently happened in the field of blood-based direct, Aβ-related biomarkers for amyloid pathology^53–55 and blood biomarkers for axonal degeneration^56,57.

Why are blood-based biomarkers needed for AD?

While advancements in positron emission tomography (PET) and CSF biomarker analyses have the potential to improve accuracy within the diagnostic process, these methods show limitations precluding their use as first-line diagnostic tools. These limitations would be countered by use of blood-based biomarkers²⁸. If a sufficiently accurate and standardized blood test can be developed, it is highly likely to be more cost effective than PET scans, the substantially high cost of which limits accessibility, generalizability and availability^58,59, particularly outside the U.S.A. Because of this high cost, PET testing in AD will likely only become reimbursed as an adjunct to other less expensive tests, as was previously the case with PET scans for cancer. In addition, blood testing is less invasive than CSF testing, which requires lumbar puncture^60,61. Furthermore, blood testing is already a well-established part of clinical routines globally, requiring no further introduction and training for health care professionals (HCPs), and can be easily performed in a variety of relevant settings (including primary care, in community-based medicine centres or even in a patient’s home) and at repeated intervals⁶². Blood sample handling infrastructure is also well established, allowing for the shipping of samples during the initial development stages. Therefore, blood-based screening biomarkers for AD can meet the scalability needs required for primary care settings and even for the broad population-based screening approach that may evolve with the advancing innovative precision medicine framework. Finally, the use of blood-based biomarkers offers the potential for testing of a huge range of comprehensive exploratory and candidate pathophysiological biomarkers, reflecting the full spectrum of disease triggering and driving molecular mechanisms underlying polygenic AD, beyond the conventional amyloid- and tau-based tests.

Blood-based biomarkers are an ideal choice as the first-step of the multi-stage diagnostic process beginning in primary care settings, and provide the means to determine which individuals or patients should receive referral to assessment by specialists, including diagnostic CSF analysis, magnetic resonance imaging (MRI) or amyloid PET diagnostics^26,63, which is analogous to the existing process for many other conditions. In addition to meeting the significant clinical need, the availability of these tools would provide a viable path towards regulatory⁶⁴ and reimbursement approval, using the cancer paradigm as a model.

Physiological challenges of developing blood-based biomarkers for AD

Identifying potential blood-based biomarkers for CNS diseases offers several challenges (Figure 1). First of all, the vast complexity of blood, which contains both cells and extracellular fluid, needs to be considered. Blood includes a range of different molecules (proteins, peptides, nucleic acids, lipids, metabolites) that can be detected in plasma, exosomes, and cellular compartments. The latter includes erythrocytes, leucocytes, and platelets, isolated into distinct cell subsets via flow cytometry or via buffy coat after density gradient centrifugation. Given the presence of different unique cellular compartments, each one is a potential source of biomarkers and may introduce variability to analyses^62,65. The diversity of blood candidate biomarkers is relevant and includes: 1) protein concentrations/activity/isoforms and post-translational modifications; 2) metabolic products, such as amino acids, carbohydrates, lipids, organic acids; 3) nucleic acids. In the latter, single nucleotide polymorphisms (SNPs) – DNA sequence variations occurring when a single nucleotide in the genome differs between members of a species – can act as biological markers, helping localize genes associated with the disease⁶⁶. When SNPs occur within a gene or in a regulatory region adjacent to a gene, they may affect the gene’s function, thus playing a key role in disease development. The growing application of high-throughput next-generation DNA sequencing technologies is playing a substantial role in screening whole genomes and, consequently, in identifying novel genetic variants affecting the risk of AD^67,68. The 1000 Genomes Project^69,70, supported by the U.S. National Human Genome Research Institute consortium, made significant progresses toward this aim.

Secondly, the CNS is an effectively contained environment, and potential biomarkers may be present at very low concentrations in blood once they have crossed the blood–brain barrier (if they cross it at all as intact molecules)^59,71. Indeed, there is minimal evidence of the peripheral effects of AD⁷², although there is some evidence that the blood–brain barrier may be increasingly compromised in normal aging and with increased AD progression^73,74. Also, physiological mechanisms occurring peripherally may hamper the clinical utility of blood-based AD biomarkers, e.g. acute-phase or inflammatory proteins and small molecules, and metabolites present also in peripheral organs⁵⁹. In particular, possible confounders include: significant biomarker dilution, considering the modest volume of the CSF and the extensive volume of the blood and extracellular fluid⁷⁵; degradation in the liver or directly in the blood by proteases; matrix effects caused by adherence to plasma proteins or even blood cells; and excretion from the kidneys. These factors may substantially lower the concentrations of the biomarker, as well as decrease the time window for testing. A further major issue is the frequent existence of overlapping ND and co-morbidities in patients with AD, including cardiovascular, respiratory, hepatic, renal, and rheumatic disease, all of which may affect protein profiles in plasma⁵⁹. On the other hand, since AD is a complex polygenic disease, amyloid and/or tau accumulation never occur in isolation of other relevant molecular/cellular pathophysiological mechanisms and brain systems failures in disease models. Elucidating this true complexity and heterogeneity, including through blood-based biomarker analyses, may substantially aid in a more comprehensive understanding of the disease for the generation of the precision medicine paradigm^25,40,44.

Gathering consensus on blood-based biomarker development in AD

Given the strong rationale yet current challenges for the development of a blood-based biomarker for AD, an international, interdisciplinary expert working group was convened in March 2016 to discuss the ideal development process. The meeting was supported by Roche Diagnostics International. The working group was selected based on profiling of global experts and working groups that was done within the extensive biomarker literature search. Experts were selected based on their publications and involvement in research in the field of neurodegenerative biomarkers in CSF and in blood. The nature of support of Roche Diagnostics International was the assistance of organizing of scientific meetings, video, and teleconferences based on the scientific chair guidance and the support of the scientific writing agency. The meeting room as well as refreshments according to the guidance of the compliance guidelines were provided.

Current landscape

A landscape analysis was conducted to review all blood-based biomarkers in development for AD globally, and then assess these biomarkers on the basis of pre-defined selection criteria in order to prioritize those with the greatest likelihood of successful implementation. The data were sourced from publications (MEDLINE® database; 2011–2015), patents (worldwide patents, including PCT, USA and China; 2008–2015); press releases and conference abstracts from eight major neuroscience or AD conferences; and ChinaBio® proprietary databases.

After screening more than 28,000 reports, 1,404 studies of blood-based biomarkers for AD were identified, 1,039 of which were from publications and conference abstracts. The 1,039 studies were categorized according to biomarker type; 18% of the studies were on the conventional AD biomarkers, Aβ and tau (or their associated peptides/proteins or variants), 19% were on genetic markers, 29% were on biomarker panels, and 34% were on markers related to emerging mechanisms such as inflammation, immune response, oxidative stress, DNA damage, mitochondrial dysfunction, and neuronal or microvascular injury. The identified biomarkers were then screened regarding intended use, asset type, technology platform, test type and analyte, and whether they included human/clinical data. A final screening was performed to focus only on high-quality candidates, according to novelty, cohort quality, presence and quality of validation study, diagnostic performance, and perceived quality of research. The final list contained 196 candidate biomarkers, with biomarker panels and emerging _targets being the most common categories (Figure 2A). Immunoassays and molecular assays were the most popular technology platforms utilized for high-quality AD blood-based biomarkers, accounting for more than 70% of studies (Figure 2B). A potential limitation of the search was the relatively narrow date range, which may have resulted in the exclusion of some potentially useful biomarkers that were published outside of this date range. The most promising biomarker candidates and the key studies, including those published since the original analysis, are summarized below.

Among the conventional AD biomarkers, the ratio of Aβ_1–42/Aβ_1–40 has shown potential as a screening/diagnostic marker in several studies. Most early studies on plasma Aβ_1–42, Aβ_1–40, and Aβ_1–42/Aβ_1–40 using enzyme-linked immunosorbent assay (ELISA) methods, found no or minor differences between AD and control groups^76–78. More recently, a large cohort study using a novel ultrasensitive immunoassay technique (single molecule array (Simoa)), showed weak but significant correlations between plasma and CSF levels of Aβ_1–42, Aβ_1–40, and Aβ_1–42/Aβ_1–40⁵⁵. In addition, plasma levels of Aβ_1–42, Aβ_1–40, and Aβ_1–42/Aβ_1–40 were lower in AD patients compared with cognitively healthy, mild cognitive impairment (MCI) and subjective cognitive decline (SCD) subjects⁵⁵. An alternative technique for measuring Aβ peptides in plasma involves immunoprecipitation and mass spectrometry^54,79. A pilot study found a trend for a reduction in both plasma Aβ_1–42 and Aβ_1–42/Aβ_1–40 in AD compared with controls⁷⁹. A separate study, using a modified method involving proteolytic digestion of Aβ peptides before mass spectrometry, found that plasma levels of Aβ_1–42 and Aβ_1–42/Aβ_1–40 were reduced in amyloid PET-positive subjects⁵⁴. Furthermore, plasma Aβ_1–42/Aβ_1–40 had a good diagnostic accuracy, as indicated by a receiver operating characteristic (ROC) area under the curve (AUC) value of 0.8865. A recent study further supports the use of immunoprecipitation–mass spectrometry methods to measure plasma Aβ peptides⁵³. Nakamura and colleagues reported high performance of the ratios of plasma Aβ_1–40/Aβ_1–42 and amyloid precursor protein (APP)_669–711/Aβ_1–42, and their composite, for the prediction of amyloid PET burden⁵³.

Results in the field have been mixed as Lewczuk and colleagues did not demonstrate plasma Aβ_1–40/Aβ_1–42 to be a useful diagnostic in a multi-center study⁸⁰. Prior work also suggests that plasma amyloid markers may be impacted by cardiovascular and cerebrovascular factors^81,82 and may impact the diagnostic and predictive of these markers⁸³. Therefore, while promising, additional work is needed to determine if plasma Aβ has utility as a screening tool for brain amyloidosis and AD, including large clinical and prospective cohorts as well as direct comparisons of different pre-analytic and analytic methodologies since immunoprecipitation–mass spectrometry could be difficult to generalize for clinical use⁸⁴.

Plasma levels of tau protein have also been successfully quantified using novel, highly sensitive immunoassay techniques, and different technologies have shown plasma tau to be increased in AD compared with controls^85,86. Plasma tau levels showed a strong association with AD in a meta-analysis (average ratio of levels in AD versus controls: 1.95, 95% confidence interval 1.12–3.38, p=0.02)³. A study based on the large Alzheimer’s Disease Neuroimaging Initiative (ADNI) and Biomarkers For Identifying Neurodegenerative Disorders Early and Reliably (BioFINDER) cohorts confirmed an increase in plasma tau in AD dementia, but with a substantial overlap in levels with cogitively unimpaired elderly subjects⁸⁷. Interestingly, longitudinal evaluations showed correlations between high baseline levels of plasma tau levels and future cognitive decline, increased atrophy rates (measured by MRI) and hypometabolism (measured by ¹⁸F-fluorodeoxyglucose-PET (¹⁸F-FDG-PET))⁸⁷. Thus, current data suggest a minor increase in plasma tau in AD, although with too large overlap with controls to be diagnostically useful. In a recent study, elevated plasma tau levels were associated with cognitive decline, independent of elevated brain Aβ⁸⁸, suggesting potential use as a non-disease specific screening tool.

Among the emerging _targets category is the axonal protein, neurofilament light (NF-L), which can be quantified using the ultrasensitive Simoa technique⁸⁹. Serum NF-L levels are highly correlated with CSF levels, suggesting that blood measurements reflect brain pathophysiology⁹⁰. A recent study on the ADNI cohort found a marked increase in plasma NF-L in AD cases with a ROC AUC value of 0.87⁵⁷, which is comparable to the plasma Aβ_1–42/Aβ_1–40 ratio results reviewed above. Plasma NF-L was highest in MCI cases with positive amyloid PET scans, and predicted faster cognitive deterioration, higher rate of both future brain atrophy (measured by MRI) and hypometabolism (measured by ¹⁸F-FDG-PET)⁵⁷. Interestingly, a study on familial AD (FAD) showed that blood NF-L was increased not only in symptomatic FAD cases, but also in pre-symptomatic mutation carriers with levels correlating with expected estimated year of symptom onset as well as cognitive and MRI measures of disease stage⁵⁶. These findings suggest that plasma NF-L detects neurodegeneration in the preclinical phase of AD. However, high plasma (or CSF) NF-L is not specific for AD, but found in several neurodegenerative disorders such as frontotemporal dementia, progressive supranuclear palsy and corticobasal syndrome^91–93. Thus, plasma NF-L might have an application as a screening test for neurodegeneration in the initial primary care evaluation of patients with cognitive disturbances.

Also in the emerging _targets category is β-site amyloid precursor protein cleaving enzyme 1 (BACE1), the β-secretase responsible for the first cleavage step required for the generation of Aβ peptides from APP⁹⁴. Studies using ELISA-based methods have shown increased BACE1 activity in the plasma of AD patients compared with controls^95,96. A recent study found that plasma BACE1 activity was increased in both MCI subjects and AD patients compared with healthy controls, and significantly higher in MCI subjects who progressed to AD (over 3 years’ follow-up) than in cognitively stable MCI subjects who did not progress⁹⁶. These results suggest that plasma BACE1 activity has potential as a biomarker to predict progression from MCI to AD dementia, which could be valuable in the primary care and clinical trials settings; however, further studies are needed to validate these findings.

These four candidate blood-based biomarkers (Aβ_1–42/Aβ_1–40, tau, NF-L, and BACE1) and their roles in AD pathogenesis are depicted in Figure 3.

Blood-based biomarker panels are another area of interest for AD as a combination of markers might show better separation between groups than single biomarkers. In recent years, several protein panels have shown diagnostic or prognostic potential^63,97–99, including a 21-protein panel for AD screening, which has demonstrated a positive predictive value (PPV) of 0.85 and a negative predictive value (NPV) of 0.94 in a preliminary validation⁶³. Non-protein analytes such as amino acid¹⁰⁰, miRNA^101,102, and lipid panels¹⁰³ have also shown promise but independent, large-scale validation studies are needed.

Historical difficulties with development

Assay reliability, and robust replication and validation of initial results are key issues for blood-based biomarker development^{26,71,104,105}. Historically, an important limitation has also been the analytical sensitivity of available technologies; standard immunochemical methods have often not been sensitive enough to allow for the reliable quantification of CNS-derived molecules in the blood, a situation that has changed with the advent of ultrasensitive assays¹⁰⁶. There is considerable variability between studies owing to inconsistencies in clinical cohorts (e.g. diagnostic evaluations, stage of disease), sample availability, difficulties in standardizing the samples themselves, and pre-analytical and analytical differences. Similar to CSF assays, consistent use of automated assays between studies may ameliorate analytical differences. However, the lack of standardization of pre-analytical protocols is also a major issue. For example, it is known that the majority of errors in proteomic analysis arise in the pre-analytical phase^65,107. Pre-analytical variables can be divided into several main categories (Table 1)^71,107. Initial steps towards standardization have already been made¹⁰⁷, with the formation of a professional interest area focused on Biofluid Based Biomarkers (BBB-PIA) (in association with the Alzheimer’s Association’s International Society to Advance Alzheimer's Research and Treatment [ISTAART]). This group will drive towards consensus on harmonization of pre-analytical and analytical protocols, and will lead the development of a repository of clinical reference samples to enable assay harmonization and clinical performance assessment. However, further action is needed and communication of these issues to wider audiences is also warranted.

The experimental design of a biomarker evaluation study (including obtaining relevant cohorts with appropriate sample availability) will be important, as this has proved difficult in previous studies. Whilst many cohorts exist and potentially have samples suitable for use to develop a blood-based biomarker, these cohorts may not match the patient profile encountered in the _target setting. Finally, any candidate biomarker will need to offer excellent scalability to allow for large-scale use.

Ideal characteristics of a blood-based biomarker

The first steps in development of a blood-based biomarker for AD diagnosis should be to: a) finalize the specific context of use (COU) for regulatory purposes; and b) determine the _target product profile, which allows for _targeted development of the biomarker to match specific requirements for the intended COU. This prospective approach may enhance success compared with current approaches based on a discovery model, where biomarkers are first identified in case-control studies before being retrospectively adapted to an intended COU. In cases where a COU is prospectively defined, the original case-control study may not be fit-for-purpose to adequately test the COU. During the working group meeting, the consensus was that the COU for a blood-based biomarker in AD should be as a tool to assess patients reporting cognitive deficits in the primary care setting, allowing for identification of the subset of patients demonstrating biological signs consistent with AD who require further specialist diagnostic testing. Ideally, the biomarker could also be used to rule out other ND with symptoms similar and/or overlapping to those of AD. Validation against current clinical and biomarker-based diagnostic paradigms would also be required (ideally the IWG-2 diagnostic criteria²⁹: evidence of cognitive impairment in conjunction with amyloid PET or the CSF “AD signature” [reduced Aβ and elevated tau CSF concentrations]). Alternatively, it may be possible to employ the recently proposed biomarker-guided A/T/N descriptive classification system¹⁰⁸ in the validation stage.

The predictive accuracy requirements of a screening tool (PPV and NPV) are dependent on COU, in addition to the to the follow-up diagnostic resources, treatments and associated costs. A primary care-based screening tool would not be intended as a “diagnostic”, but would serve as a gatekeeper to the confirmatory diagnostic procedures utilized by specialty physicians. As is the case with all primary care screening tools, the primary concern is with high NPV. This is due to the base rates of the conditions in the general population (which are typically very low) and, therefore, the screening tool serves to preclude the vast majority of patients who are not in need of more invasive and expensive diagnostic procedures. For example, in mammography screening for breast cancer and in depression screening in primary care, PPVs are <30%⁶³, but are counteracted by excellent NPVs (>90%). In primary care settings, it is commonplace to see tools with PPVs <40% (or even <20%) but NPVs >90% (or even >95%)⁶³. The overall diagnostic accuracy of the primary care screening tool must be taken into account alongside the full diagnostic procedures in specialty clinics where available that include CSF and/or PET biomarkers (which have excellent PPV) for confirmatory diagnostic purposes. Once disease-modifying therapeutics become available (for prevention or treatment of symptomatic patients¹⁰⁹), these primary care screening tools become required as there will be many more patients seeking services. Therefore, the primary care screen will be needed to cost-contain while simultaneously broadly opening up access to specialty diagnostics. The consensus in the working group was that a candidate biomarker for primary care screening should ideally have a PPV of >50%, with an aim for this to be as high as possible. However, given the ethical implications of false negatives, it would be more important to achieve a NPV of ≥90%, with 95–98% being an ideal _target⁶³. On the other hand, a performance with a high PPV and an acceptable NPV may be desirable in a pharmaceutical setting that focuses on identifying as many amyloid-positive patients as possible for either inclusion in to clinical trials or a possible future treatment.

In terms of assay technology, the consensus indicated a strong preference for panel-based assays (i.e. a combination of biomarkers, either developed together or individually) over single biomarkers. Panels may offer wider applications beyond AD diagnosis, allowing more generalized testing for other ND, with each pathological condition having a unique protein signature. Several studies established panels of biomarkers to discriminate between cognitively healthy controls and AD participants and assessed large arrays of differently combined proteins to yield high specificity and sensitivity. Indeed, since there is a definite need for a holistic approach to standardize blood biomarkers for AD diagnosis, it is crucial to understand the link among various individual biomarkers and overcome the outdated approach of examining single candidate biomarkers at a time¹¹⁰. Given that various methods exist to detect biomarkers in blood^28,110–112, it is crucial to standardize the technologies generating complex multidimensional data and the whole workflow. General consensus on both protocols and ultrasensitive analytical methods are needed across multicentre studies to set up a standardized panel of biomarkers for AD diagnosis^110,111. In this regard, recent technical advances have provided ultrasensitive immunoassay methods that allow detecting and quantifying biomarkers at concentrations that are several-fold lower than those accessible by existing immunoassays. This may result in the development of a new array of blood biomarkers covering the entire spectrum of AD molecular pathophysiological mechanisms. Moreover, these technologies may serve as the foundation enabling the diagnosis of AD and other ND earlier than ever before¹¹³.

However, the number of biomarkers in the various types of panels so far developed increases the complexity of manufacture and commercial development, complicates validation and standardization processes, and leads to increased costs. This is a significant concern when considering scalability to meet global primary care needs. Panel-based assays also offer an additional challenge, as there are risks associated with analyzing large numbers of candidate molecules in small numbers of patients, which may result in data over-fitting and misleading results so that a panel seems to perform much better than individual biomarkers, but in fact does not¹¹⁴. This can be partly offset by use of a training set to initially reduce the dimensionality before validation in an independent cohort¹¹⁵. Therefore, while the working group agreed that a multi-marker panel may be needed, it was also agreed that the number of biomarkers should be restricted as much as possible within the NPV and PPV requirements as this would help in meeting the validation and regulatory hurdles.

Refinement and validation

Following identification of a prioritized candidate biomarker, the biomarker will require refinement and validation. Refinement will involve agreeing on a best-practice protocol, including blood collection procedures, pre-analytical sample handling and procedures for harmonization between different laboratories. Validation steps are outlined below.

Does the landscape fit the _target product profile?

Following the landscape analysis, the 196 candidate biomarkers (Figures 2A and 2B) were further divided into tiers according to additional criteria (outlined in Callout Box 1). Of these 196 candidates, 19 were prioritized for further consideration by the working group (a table of all 19 assays is provided in the Supplementary Information); however, none were deemed to have met the agreed _target product profile.

Most of the candidate biomarkers were limited by a lack of validation in independent cohorts (only five biomarker studies identified in the search have been followed up with validation studies) (Figure 5)^{97,98,122–124}. This is regarded as a critical step in development and so is a conspicuous gap. Furthermore, where validation has been performed, it may not have been conducted by an independent group, and may not be based on a model published prior to the study¹⁰⁵. This results in a risk of hypothesizing after results are known (HARKing), which can lead to selective reporting and inflated predictive accuracy¹²⁵.

Validation studies may require additional cohorts or, potentially, they could be validated within subgroups of the original cohort. However, appropriate cohorts for the ideal intended COU do not currently exist; validation between cohorts will therefore be a further hurdle in the development of assays. As previously discussed, the appropriate cohorts (stemming from patients in primary care with CSF and blood samples, as well as PET data) are not available and therefore the _target product profile could not be met by any study.

Additional issues present with some biomarkers are PPV and NPV values lower than those specified in the _target product profile, as well as small sample sizes (Figure 5). Studies using large and well-established cohorts such as ADNI and the Australian Imaging, Biomarker and Lifestyle Study of Ageing (AIBL) are more likely to meet the initial requirements. As discussed previously, biomarker panels are thought to offer the greatest chances of success, although panels featuring high numbers of markers will be difficult and expensive to develop and validate. In addition, most available technology platforms have a low ceiling on the number of biomarkers that can be contained within the assay, which is usually around five biomarkers. There are many emerging areas of interest (e.g. exosomes) that offer promise, although they are too early for consideration at this time. Additionally, blood-based assays for well-known biomarkers of AD, such as Aβ_1–42/Aβ_1–40, tau, NF-L, and BACE1, may have potential, and should be monitored closely as more information becomes available and novel technologies emerge to address the hurdles currently facing use of these biomarkers.

The future: blood-based biomarker diagnostics

Blood-based biomarkers have the potential to improve detection and diagnosis for AD by increasing convenience, acceptability and ease of testing, as well as reducing costs. There are a number of key areas in which they are likely to provide most benefit. The first and most important of these is testing for AD in the primary care setting, which will potentially allow identification of patients at the earliest stages of the disease. Ideally, this would be prior to the development of any noticeable cognitive deficits, but this will depend on future developments. The advent of testing in primary care would lead to substantial changes in the current treatment paradigm by allowing patients to be guided towards specialist diagnostics (e.g. CSF or PET scans) and access to care earlier in the course of the disease. This will be particularly important once disease-modifying therapies (DMTs) become available, but may provide benefit to patients and caregivers even at the present time¹²⁶.

Once DMTs become available there will, undoubtedly, be a tremendous surge in numbers of patients in primary care settings seeking potential prescriptions. However, without a convenient primary care screen, confirmatory diagnostics will still be required (i.e. CSF, MRI or PET), which will act as a bottleneck in the diagnosis and treatment process, and will present a considerable burden in terms of costs. Availability of a multi-tiered diagnostic process commencing with a blood test in primary care would dramatically increase access to DMTs, while conveniently exclude from further diagnosis and treatment the vast majority of patients who would not need to undergo these expensive and invasive procedures. An additional potential use of blood-based AD biomarkers will be to provide a cost-effective and rapid test for AD pathology in order to determine the eligibility of patients for recruitment into clinical trials for DMTs^127–129. These tests might also have applicability in the monitoring of disease progression and drug effects on AD pathophysiology during a trial¹³⁰. As such, the availability of blood-based biomarker diagnostics may in fact accelerate the development of DMTs.

The development of blood-based biomarkers often stalls in the early stages due to a disconnection between academia, where biomarkers are identified, and industry, who have the resources to take biomarkers further and develop them commercially (Figure 6). This disconnect arises from the different needs of these different sectors and from mutual mistrust developed following a history of poor interactions. The current initiative by the working group and Roche Diagnostics International represents a step in an attempt to improve that process, both efficiently and economically.