Imaging and Fluid Biomarkers of Alzheimer Disease: Complementation Rather Than Competition

GENERAL DIFFERENCES BETWEEN IMAGING AND CSF/PLASMA BIOMARKERS

Traditionally, neuroimaging, CSF, and blood and plasma biomarkers for AD have been categorized by their ability to reflect one of the hallmark pathologies involved, that is, amyloid-β (Aβ) plaque aggregation pathology (A category), development of neurofibrillary tangles from aggregated tau protein (T category), and resulting neuronal injury (N category). According to the so-called A/T/N classification scheme, the presence of the 2 central pathologies (namely, A and T) has been required, complemented by evidence of ongoing neuronal injury (category N) to postulate a diagnosis of ongoing AD (2). Importantly, the different available methods of imaging, CSF analysis and, more recently, also plasma sampling (3), have been generally considered as interchangeable to allow a basic assessment of the A/T/N status.

In the case of neuroimaging markers, they can be further categorized into structural (MRI) or molecular imaging (PET) markers. In general, the imaging biomarkers allow a direct evaluation of the 3 categories of pathology with respect to their extent, localization, and intensity. This was convincingly demonstrated by numerous trials including in vivo and postmortem cross-correlation. With regard to the A category, several Food and Drug Administration (FDA) and European Medicines Agency–approved commercially available ¹⁸F-labeled PET tracers (florbetaben, florbetapir, flutemetamol) allow the detection of aggregated neuritic amyloid plaques noninvasively and in vivo. For more recently introduced tau PET imaging (T category), 1 FDA-approved tau PET tracer (flortaucipir) is available to date (several more are currently in the development pipeline), which has been demonstrated to reliably detect the presence, level, and localization of manifest deposition of tau neurofibrils (4), typical of AD. Again, importantly, both amyloid and tau PET biomarkers were required to demonstrate sensitivity and specificity toward the respective neuropathologies by means of in vivo versus postmortem cross-evaluation before approval, the highest evidence level of AD biomarker testing.

Neuronal injury (N category) can be assessed by well-established FDG PET, which allows detection of synaptic or neuronal dysfunction by measuring regional glucose metabolism, as well as by structural imaging (MRI) that measures resulting regional brain atrophy. Importantly, imaging biomarkers of molecular pathologies allow not only the judgment of the presence of different forms of pathology but also their quantitative levels, anatomic distribution or topography, and extent. The latter is of paramount importance with regard to detection of characteristic patterns of neurodegeneration which have high specificity for different neurodegenerative disorders.

Fluid AD biomarker diagnostics are based on a fundamentally different measurement principle. Here, peripheral indicators are used to draw indirect conclusions about brain pathologies. In CSF analysis and, similarly, in plasma analysis, a decrease in the Aβ42 concentration (or in the Aβ42/Aβ40 ratio) in the respective fluids is considered an indirect indicator of ongoing increased aggregation of the Aβ42 protein in the brain (A category). Thus, this measure does not provide direct evidence of already manifest amyloid plaque aggregation but rather represents a momentary snapshot of a pathologic process contributing to the buildup of aggregated pathology. Recent studies confirm the notion that the Aβ42/Aβ40 ratio in CSF might better reflect brain levels of soluble Aβ protofibrils than insoluble Aβ fibrils in plaques in AD (5).

Similarly, fluid biomarkers reflecting elevated total tau (t-tau) levels also represent an indirect measure of a pathologic process. Interestingly, the measurement of elevated t-tau levels is believed to not primarily reflect increased tau aggregation but rather the release of tau protein from neurons as a result of ongoing neurodegeneration or neuronal injury. It is considered a nonspecific marker of axonal degeneration, thus rather representing the N category. Assessment of the N category may also be possible with fluid biomarkers by measurement of so-called neurofilament light chains (NfL), which appear nonspecifically increased in the respective fluids as a consequence of various forms of neuronal degeneration. Obviously, it does not allow detection of specific neurodegenerative patterns of disease, as the regional origin of the neuronal injury cannot be discerned. As a marker of tau deposition in CSF and plasma (T category), the measurement of hyperphosphorylated tau (p-tau) has been proposed. Importantly, p-tau markers such as p-tau217 or p-tau181 have demonstrated very promising performance to detect AD pathology even in plasma assays. These markers are therefore currently considered to be particularly promising candidates for biomarker-supported diagnostics of AD using simple plasma analysis. However, the specific reflection of a single neuropathology by these biomarkers is less obvious. It has been repeatedly demonstrated that these p-tau biomarkers seem to correlate even more closely with cerebral amyloid deposition in early stages of disease and stronger with tau neurofibril deposition or rather a mix of both pathologies in later stages of disease (6). Thus, p-tau CSF/plasma markers may not represent an unambiguous measure of tau pathology. In summary, it is important to be aware that fluid tau markers are not necessarily specific markers of the T category.

Importantly, the level of validation differs considerably among the various biomarkers. The PET biomarkers had to undergo complex in vivo versus post mortem validations for approval and are thus validated against the neuropathologic gold standard. Some of the fluid biomarkers have not yet completed this process or have been tested, for example, against PET imaging. As PET does not have 100% sensitivity and specificity, the reported results for the fluid markers need to be put into perspective. It is generally questionable which validation strategy is at all meaningful for the fluid biomarkers because they potentially depict completely different pathologic processes in the brain than what is measured by PET or histopathology. In any case, it seems inevitable that newly introduced biomarkers (and those that are yet to come) will be validated for their respective clinical purpose including real-world studies in populations with copathologies with possible autopsy validation and longitudinal performance metrics.

Taken together, imaging biomarkers basically allow a direct estimation of a status of the pathologic core pathologies of AD, which are also used for definite histopathologic diagnosis and on which the A/T/N classification has been based in its origin. Using these tools, the exact extent of neuropathologic changes in intensity, localization, and magnitude can be quantified for A, T, and N, respectively (disease extent). Also of note, obviously, only imaging but not fluid AD biomarkers are able to provide information on the pattern of the desired readout within the brain, with relevant clinical value for instance to differentiate different AD subtypes, such as posterior cortical atrophy, logopenic variant primary progressive aphasia, behavioral variant AD, and corticobasal syndrome AD. The fluid biomarkers, on the other hand, represent more indirect measures of ongoing pathologic processes associated with these pathologies. Although this does not directly quantify pathology that has actually occurred in the brain and it does not allow recognition of neurodegenerative patterns, the measurement of the underlying processes can potentially provide relevant complementary information to imaging with regard to current disease activity and may allow earlier detection of disease onset (Fig. 1).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Advantages and disadvantages of 3 AD biomarker categories. Advantages are in green, nondifferentiating features in yellow, and disadvantages in red.

DIFFERENCES IN THE AD BIOMARKER CLASSIFICATION SCHEMES (A/T/N VS. CORE)

The classification of available AD biomarkers according to the A/T/N scheme is not without problems. In particular, the categoric consideration may neglect stages and levels of pathology (i.e., 2 persons being A-positive/T-positive may be considerably different in disease stages, depending on levels of pathology in each category). Also, the A/T/N classification may falsely imply an uncritical interchangeability of the different biomarkers in the respective categories. This is as the underlying principles and the information obtained by the respective methods differ considerably. As such, the available biomarkers are not simply interchangeable; a categorization as A-positive with a CSF biomarker may differ significantly in meaning compared with the A definition with an imaging biomarker. For some fluid biomarkers, combinations are suggested for diagnostic purposes (e.g., t-tau/Aβ42), thus reducing specificity for the pathology. Also, it has been demonstrated that some fluid biomarkers are not reflecting a single pathology in a linear manner throughout the disease course. Finally, the level of validation and the sensitivity and specificity of the different biomarkers may vary within the different categories (see below).

More recently, a revised type of AD biomarker classification has been introduced, partly motivated to circumvent some of the limitations of the original A/T/N classification (3). This scheme divides between Core 1 and Core 2 biomarker categories, with Core 1 including biomarkers allowing early detection of AD (even in asymptomatic stages) and confirmation of AD pathology in symptomatic subjects and Core 2 used to increase confidence that AD is contributing to symptoms, to stage biologic disease severity and to inform on risk of short-term progression in asymptomatic subjects (in combination with Core 1). In this classification, the traditional A/T/N scheme has been abandoned to some extent. All traditional biomarkers of the A category are assigned to the Core 1 group (amyloid PET, CSF Aβ42/40). Also, several combinations of fluid biomarkers such as p-tau181/Aβ42 and t-tau/Aβ42 are grouped into to the Core 1 category. The T category has been divided into a T1 category (“secreted tau,” for example, p-tau217, p-tau181, and p-tau231) belonging to Core 1 and a T2 category (“AD-tau,” as measured by tau PET and more recently introduced fluid markers such as p-tau205, microtubule-binding region (MTBR)-243, non–p-tau fragments), belonging to Core 2. Reasons for this new classification attempt may be found in the need to divide between biomarkers turning positive early versus late in the disease process and to take into account that some of the fluid biomarkers (and their combinations) are not specific for one of the A/T/N categories. However, this classification has created some controversy and may also not be optimally suited to reflect the features particularly of the imaging biomarkers for several reasons. First, it dilutes specificity for diagnostic assessment of one particular neuropathology by mixing different biomarker types and their combinations. Core 1 biomarkers are suggested to represent AD neuropathology changes (ADNPC) more generally, that is, a mix of underlying pathologic features. This gives away the validated value particularly of the PET imaging biomarkers to specifically assess one particular pathology. For their approval, amyloid and tau PET imaging procedures were required to demonstrate specificity toward one particular neuropathology over other pathologies. One of the earliest PET biomarkers for amyloid plaques (¹⁸F-FDDNP) has not been universally accepted as a diagnostic tool because it binds to both amyloid and tau deposits and is therefore not pathology-specific (7). Now, it could be asked whether this tracer could be considered an ADNPC Core 1 biomarker according to the new classification. Particularly with respect to novel therapy approaches directed against specific AD neuropathologies, it may not be advisable to disregard the value of some biomarkers for the specific detection and quantification of clearly defined neuropathologies. The bundling of many different biomarkers in the Core 1 category also implies that these biomarkers are interchangeable or equivalent. However, this is not the case, neither in terms of specificity (specific imaging biomarkers vs. non–pathology-specific fluid biomarkers or their combinations) nor in terms of sensitivity. Again, a person being positive in one of the Core 1 biomarker tests could potentially be negative with another Core 1 test. Also, the level of validation for the different biomarkers is not necessarily comparable yet. In this context, it could also be discussed why the A category, unlike the T category, is not differentiated into secreted (i.e., fluid) and aggregated (i.e., PET), although there is no less reason to do so. The potential for staging disease, for example, by means of the Centiloid quantification of amyloid PET, is also neglected by the classification of any positive PET scan into 1 category. Finally, it has been demonstrated that amyloid pathology can often be present in other diseases such as DLB, and it may be debatable whether these patients should then be classified as having AD. In general, this novel AD biomarker concept, although having some obvious advantages with regard to simplification, seems to somewhat disregard the complementary value of the different biomarkers, especially between imaging and fluid markers.

DIFFERENCES IN AD BIOMARKER PRACTICABILITY

With regard to practical implementation, the 3 classes of available AD biomarkers also show some basic differences and advantages and disadvantages (Fig. 1). Regarding the available imaging biomarkers, some radiation exposure as well as the relatively high costs and limited availability are often cited as factors hampering PET diagnostics. Also, it is often stated that not all 3 categories (A/T/N) can be analyzed at once by means of a single imaging examination. However, for some of the imaging tests, including amyloid PET and tau PET, multimodal information such as assessment of perfusion (N) combined with proof a specific pathology may also be possible within a single examination procedure. In particular, tau PET may thus serve as a potential universal biomarker, allowing A, T, and N staging in a single imaging test. CSF diagnostics has the disadvantage of a higher invasiveness and a relevant frequency of side effects which have been described in a considerable proportion (>30%) of subjects examined. This includes back pain (17%), headache (19%), and sometimes but very rarely serious complications such as infection, spinal hematoma, or cerebral venous thrombosis. In contrast to imaging techniques, there are several contraindications such as anticoagulant medication, coagulopathy, bleeding diathesis and posterior fossa masses, Arnold-Chiari malformations, increased intracranial or CSF pressure, spinal pathologies, or skin infection at the puncture site (8). However, CSF analysis obviously allows recording of several parameters of pathology (A/T/N) simultaneously in 1 clinical examination.

The latter advantage also presumably applies to plasma analysis, which at the same time is less invasive and also could become more cost-effective to perform in the future, so that a very broad use (i.e., screening) could become possible. So far, however, still relatively limited experience is available for these biomarkers. In particular, the reliability and standardizability of the results need further validation, especially when used broadly, that is, outside dedicated memory clinic settings, for example, as a screening method for the general population. As discussed below in greater detail, other factors such as chronic kidney disease, obesity, and cardiovascular conditions and other disorders may affect blood biomarker assessment and lead to false-positive or false-negative results (9–11).

These relevant differences in level of validation, practical implementation, and informative value (Fig. 1) indicate a complementary role of the mentioned (groups of) AD biomarkers, which will be further explored below with reference to specific questions.

VALUE OF THE BIOMARKERS FOR ESTABLISHING DIAGNOSIS OF AD DEMENTIA

Regarding the categoric question of whether a patient with manifest dementia symptoms suffers from AD or not, a binary biomarker readout is required. Here, the 3 groups of biomarkers appear to show generally reasonable performance, particularly, as far as exclusion of AD is concerned. Following the “no AD without amyloid pathology” concept and as confirmed by various in vivo PET versus postmortem histopathology trials, a patient with dementia who has a negative amyloid PET scan is generally considered to not suffer from manifest AD. The same holds true for a negative tau PET scan, when testing patients with manifest symptomatic disease.

Regarding FDG PET, high sensitivity or specificity has likewise been reported (on the basis of characteristic topographic distribution patterns). With regard to the CSF biomarkers, reduced levels of Aβ42 (reduced Aβ42/Aβ40 ratios) as well as ratios of Aβ and tau levels have demonstrated good value for approximation of ongoing amyloid aggregation pathology in the brain. Thus, inconspicuous findings with these parameters are also not consistent with the diagnosis of AD. With the plasma biomarkers, reduced levels of Aβ42 have been described in patients with AD by means of sensitive novel assays. However, compared with the CSF Aβ42 values, the effect sizes of plasma Aβ42 levels are orders of magnitude smaller. These lower effect sizes make the blood test readouts more susceptible to variations in handling, the type of test kit or assay used, and to various biologic confounds (12–17).

Fluid markers of neuronal injury, such as the NfL, consistently show elevated mean levels in CSF and plasma analysis of AD dementia patients; however, the range of values is broadly overlapping with controls (12–15,18,19). Importantly, the NfL biomarkers are elevated not only in AD but also in other neurodegenerative disorders, thus they do not allow differential diagnosis (20).

With regard to plasma biomarkers, high effect sizes for AD dementia diagnosis were particularly reported for fluid p-tau (p-tau217 and p-tau181) (12–15,21,22). The specificity of these p-tau tests for the T category, however, has been challenged by the finding that plasma p-tau217 seems to correspond with cerebral Aβ pathology in earlier stages and with combined amyloid and tau pathology in later stages (6). Another more recently introduced plasma biomarker, MTBR-tau243, showed stronger associations with tau PET than with plasma p-tau217 and, thus, may potentially represent a more suitable T marker (23).

Importantly, as mentioned above, the value of plasma measurements for the diagnosis of AD dementia may be influenced by other factors such as the presence of chronic kidney disease, obesity, and cardiovascular conditions or medication which might lead to false-positive or false-negative results (9–11). Also, other factors including sex, race, or ethnicity have been discussed to potentially influence blood biomarker results. Consequently, a positive blood biomarker finding may indicate the presence of AD pathology but also could be an incidental finding (which may have relevant consequences, for example, regarding insurance coverage). Also, to date, many blood biomarker studies were performed in specialized centers or settings and may not yet be generalizable to typical patients with dementia. Therefore, further studies will be required to validate these novel plasma AD biomarkers at the population level in a true world environment and against a suitable gold standard.

Head-to-head imaging versus fluid biomarker comparison studies are most insightful to determine the potential interchangeability of these biomarkers. In this context, discordant cases are of particular interest. It has been demonstrated that 10%–20% of memory clinic patients have discordant Aβ PET and CSF Aβ42 results (24). With regard to interpreting these findings, there is some indication that CSF assessment may show advantages with regard to early preclinical detection of the disease onset (see section on early diagnosis), whereas PET imaging findings may have greater reliability with regard to detecting the actual presence of relevant AD-typical Aβ aggregation pathology. In a limited sample of patients with dementia in a memory clinic, discrepant in vivo amyloid PET and CSF Aβ42 findings were compared versus postmortem histopathology: 2 CSF-positive/PET-negative cases were A-negative on postmortem histopathology, and 1 CSF-negative/PET-positive case was A-positive according to the postmortem gold standard diagnostics (25). These findings are also consistent with the clinical impression of groups using both biomarker classes: It was shown that the main reason for ordering additional amyloid PET after CSF analysis was conflicting CSF results (either not supporting the clinical diagnosis or with discordant A/T status) (26). Similarly, another study showed that using PET imaging after CSF assessment increased diagnostic confidence and resulted in diagnostic changes, whereas adding CSF after PET imaging did not result in incremental diagnostic value. The authors concluded that—if available—amyloid PET should be prioritized over CSF biomarkers in the diagnostic work-up of patients investigated for suspected AD (27). A more recent study compared amyloid PET and CSF diagnostic assessment in more than 500 patients examined in a real-world memory clinic setting. In about 14%–16% of these patients, indeterminate findings were observed in CSF analysis, approximately 50% of which were found to be positive in subsequent PET examination. The authors concluded that if CSF analysis is performed as the first test for therapy planning, patients at borderline levels strongly benefit from additional amyloid PET imaging (24). Amyloid PET is also considered appropriate according to the current appropriate use criteria (AUCs) in cases with inconclusive CSF findings (28). In general, it is a relevant question for how to deal with discordant biomarker results and how these should be communicated to patients and their relatives. On the basis of current knowledge and in accordance with the guidelines discussed here, it could be concluded that, as of now, PET imaging can be considered the gold standard, which should be given greater weight in the case of discrepant results, and that additional PET diagnostics should be recommended whenever biomarker results are otherwise inconclusive. Appropriate diagnostic pathways should be systematically evaluated in the future.

Some of the observed differences between amyloid PET and CSF Aβ biomarkers may be explained by the different general properties of the biomarkers (disease activity vs. extent of pathology) but also by the different test sensitivities depending on the stage of disease (see below).

Interestingly, similar conclusions can be drawn with regard to T staging from head-to-head comparison studies, for example, analyzing flortaucipir PET and CSF p-tau181 data. Provost et al. demonstrated that the sensitivity and specificity of these markers depend on the disease stage within the AD continuum, with overall concordances between the 2 measures of only 68%–76% (29). In accordance with these data, concordance between the binary T classification as obtained by flortaucipir/RO948 tau PET, CSF p-tau181/217, and plasma p-tau181/217 was 66%–95% in the BioFINDER-2 and AD Neuroimaging Initiative cohorts (30).

VALUE OF THE BIOMARKERS FOR EARLY AD DIAGNOSIS and DISEASE COURSE PREDICTION

Early AD diagnosis or disease course prediction, specifically at the mild cognitive impairment (MCI) stage (about 50% of MCI subjects develop AD dementia within the following years), is increasingly gaining relevance for treatment decisions. Here, differences in the positivity ratio between the 3 AD pathology readouts are obtained: In general, in line with the hypothetical progression model of Jack et al. (31), it can be postulated that the A category markers are the first to become conspicuous, followed by the T category markers and the N category markers (Fig. 2). In general, however, this chronological classification cannot be taken as an unquestionable proof for a corresponding order of positivity of the respective biomarkers. This is as biomarker positivity is also bound to the detection limits and sensitivities of the respective methods. Also, the prevalence and quantities of the pathologies differ greatly, impacting effect sizes.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Hypothetical model of chronologic sequence of AD biomarker occurrence. ADL = activity of daily living; MTL = medial temporal atrophy; PHF = paired helical filament. (Reprinted with permission of (93).)

For amyloid PET imaging, several studies showed high sensitivity for predicting AD dementia in the stage of MCI: Roberts et al., for instance, showed a conversion rate of greater than 35% of amyloid-positive MCI patients to clinically manifest AD dementia within 5 y (32). The reported specificity values were variable in different trials (33), presumably because of different follow-up periods. Amyloid-positive subjects may not yet have shown conversion to manifest dementia within shorter follow-up periods, despite having early AD. In fact, it has been demonstrated that a relevant proportion of elderly subjects will be amyloid-positive even without cognitive symptoms (34). Less than 20% of these subjects may show development of mild cognitive decline within a 5-y follow-up period (32). Correspondingly, the current AUCs consider amyloid PET appropriate for establishing the prognosis of patients presenting with MCI but still uncertain in subjective cognitive decline at risk for AD and rarely appropriate in cognitively unimpaired subjects (28).

In this context, however, it is important to distinguish between the usual categoric definition of Aβ-positive or -negative and more sophisticated measures that are also available for continuous assessment of image data. These include (semi)quantitative evaluation and regional analysis of the imaging signal. Quantitative assessment of global amyloid burden using the Centiloid scaling is emerging as a valuable tool (see section on staging) and may also be relevant for early diagnosis and prediction (35). Recent studies have shown that subjects surpassing levels of approximately 16 within the gray area of the Centiloid values are likely to develop progressive amyloid deposition toward amyloid-positive stages. Thus, a 12–20 Centiloid window has been suggested for inclusion into early secondary prevention studies (36). In addition to quantification, regional assessment of amyloid burden may also offer great value in early diagnosis and prediction. Studies have shown that certain areas of the brain (e.g., the default mode network, temporal, cingulate, and occipital regions) are affected by amyloid pathology earlier than others (37,38). Interestingly, the longitudinal progression of deposits does not always appear to follow a uniform pattern. Collij et al. were able to show that subgroups with different trajectories of progression can be identified, which may be relevant for early prognostic assessment (39). Importantly, some papers have already demonstrated that specific regional patterns of amyloid distribution may allow improved prediction of future cognitive decline in cognitively healthy subjects (precuneus, subcortical, and parietal regions) and of future dementia in MCI subjects (cingulate, temporal, and frontal regions). This regional assessment added predictive value beyond global Aβ burden and CSF biomarkers (40). Furthermore, approaches combining quantitative with regional assessment such as the recently suggested “fill states” may have potential in this context (41). Obviously, it will be difficult to obtain comparable information from fluid biomarkers. In addition to quantification and regional assessment of amyloid aggregates, the sensitivity and value of PET imaging for reliable early diagnosis and prediction may be further improved with the advent of new high-resolution PET scanner technology and advanced computational and artificial intelligence–based approaches.

For the tau biomarkers, the first approved PET tracer, flortaucipir, was revealed to be able to reproduce advanced stages (Braak stages > IV) of pathology but to miss earlier stages (4). This may possibly be induced by a rather conservative interpretation guideline, suggesting that cases with isolated mesial tau tracer retention be read as tau-negative, presumably with the aim to increase specificity toward AD-typical manifest tau pathology and to reduce the false-positive impact of nonspecific tracer retention as well as to exclude cases with primarily age-related tauopathy (PART), that is, patients with tau aggregation (predominantly in the mesial temporal lobes) without isocortical Aβ aggregates. Currently, there is controversy surrounding PART, as the prognostic meaning of this finding with regard to developing AD dementia is not yet clear (42). According to the current AUCs, tau PET is considered rarely appropriate in clinically unimpaired and subjects with subjective cognitive decline (28).

However, with regard to short-term prediction of cognitive decline, the value of tau PET imaging might even be higher than that of amyloid PET imaging. Ossenkoppele et al. demonstrated that tau PET shows superior predictive value compared with amyloid PET or MRI (in amyloid-positive subjects) with regard to development of dementia in MCI (43). Consistently, the current AUCs for tau PET consider its use to inform on the prognosis in MCI as appropriate (28).

Even cognitively healthy amyloid-positive individuals who also were tau PET–positive showed a conversion rate to MCI of about 50% within 3.5 y, indicating a relevant prognostic value in this group (44). Importantly, this also included subjects with isolated mesial temporal tau aggregation, that is, cases which would not be read as “tau-positive” according to the current interpretation guidelines of the approved tracer. Thus, the ability to detect this pathology by means of imaging may be of prognostic relevance. It is not known if cases with this type of early tauopathy or PART can be detected or differentiated with fluid biomarkers. As mentioned above for amyloid PET, advanced quantitative and regional image assessment strategies and improved tracers may also further increase the value of tau PET with regard to early diagnosis and prediction of cognitive decline.

With regard to N category imaging markers, it has been demonstrated that FDG PET has added value for prediction, particularly of short-term decline from MCI to manifest AD dementia even in amyloid-positive subjects: Iaccarino et al. demonstrated that approximately 50% of patients with amyloid-positive MCI who had abnormal FDG PET scans converted to dementia within 2 y, whereas a high proportion of subjects who were amyloid-positive and FDG-negative remained stable for up to 10 y (45).

In general, for fluid biomarkers, findings somewhat similar to those of imaging biomarkers have been described. It was shown that CSF Aβ42 values may decrease long before clinical disease onset of cognitive decline. Stomrud et al. concluded in their work that a decrease in CSF Aβ42 occurs more than a decade before the onset of sporadic AD dementia and that CSF Aβ42 decline has already plateaued 9 y before AD dementia onset (46).

For early AD diagnosis and disease course prediction, a head-to-head comparison between fluid and imaging biomarkers provides the most relevant insights into their complementary value: Studies correlating CSF Aβ42 levels with amyloid PET changes implicate a nonlinear L-type relationship (Fig. 3) (47). These studies consistently suggest that CSF Aβ42 levels are reduced over a wide range before an increase in tracer accumulation can be verified by means of amyloid PET. On the other hand, a broad range of amyloid PET uptake values can be observed within a rather stable level of CSF values. This could be interpreted as follows: The beginning or early disease process can be detected by CSF analysis (with still unobvious amyloid PET findings), and the aggregation process, that is, the build-up of amyloid plaques, can only be detected by amyloid PET later in the disease process (with then already stable CSF reduction). This interpretation would be plausible both against the background of the different methodologic approaches and from a biologic point of view. Such an interpretation would imply that CSF Aβ42 measurement cannot be considered synonymous with amyloid PET imaging (although both are A category markers and are also considered Core 1 markers) but that there is a temporal offset in diagnostic value with very early changes in disease activity (CSF) reaching a plateau in initial disease stages and subsequent changes in manifest deposition of Aβ plaques reaching a plateau later (Fig. 3). Just recently, a study provided further evidence that the Aβ42/Aβ40 ratio in CSF might better reflect brain levels of soluble Aβ protofibrils than those of insoluble Aβ fibrils in plaques in AD (5). This would again indicate a complementary value of the 2 methods. Also, this interpretation would explain why studies have demonstrated a significantly lower progression rate in patients with AD dementia with MCI with abnormal CSF Aβ42 levels only (and normal amyloid PET) compared with subjects with both abnormal amyloid CSF and amyloid PET results (48). Also, Reimand et al. reported in a head-to-head comparison study of 768 memory clinic patients that the CSF Aβ42 readout was more sensitive in early AD stages, whereas the Aβ PET readout was more specific to AD pathology (49). This assumption is particularly relevant with regard to inclusion of patients in therapy trials and their treatment selection. Consequently, an appropriate use recommendation expert panel for AD therapy with aducanumab suggested that patients with abnormal CSF Aβ values and normal amyloid PET should not be selected for therapy (50).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3.

Relation between PET-based and fluid amyloid biomarkers of AD. L-shape association between CSF and PET readouts of amyloid biomarkers (left) and schematic presentation of complementary value of biomarkers dependent on time window within AD course (right). SUVR = SUV ratio. (Reprinted with permission of (94).)

In this context, it is important to note that a biomarker’s general capability to detect early forms of pathologies (e.g., soluble forms of Aβ) does not automatically define a high sensitivity of this biomarker to detect the disease early in its course. The latter depends on the reliability and methodologic sensitivity of the test. Importantly, a large population trial, using autopsy as a gold standard, was recently able to demonstrate data that amyloid PET was able to detect cortical Aβ deposition earlier than Aβ CSF biomarkers (51). Consistently, although fluid biomarkers may, in general, be capable of detecting early, soluble AD pathologies, intermediate findings and false-negative CSF results have been documented in patients with positive PET results, that is, showing already aggregated and later-stage pathology (24).

This may be even more relevant with regard to plasma biomarkers: It has been demonstrated that changes in Aβ42 levels can also be found in the plasma with sensitive assays. However, compared with the CSF Aβ42 values, the effect sizes of plasma Aβ42 levels are orders of magnitude smaller. These lower effect sizes make the blood test readouts more susceptible to variations in handling, the type of test kit or assay used, and to various biologic confounds. Consequently, their sensitivity with regard to early diagnosis may be limited (12–17). Thus, differences apply not only between imaging and fluid A or Core 1 biomarkers but also within fluid biomarkers.

Regarding plasma biomarkers, p-tau218 and p-tau181 may hold greater promise with regard to short-term prediction of cognitive decline compared with Aβ42, similarly as demonstrated for the imaging biomarkers. Again, particularly high rates of converters were observed in subjects positive for amyloid as well as for p-tau181, suggesting the added value of combined assessment (12,52). Importantly, as mentioned above, the p-tau assays, that is, CSF and plasma p-tau181/217, may not represent pure tau (T) biomarkers but have been demonstrated to reflect Aβ pathology in early stages and mixed Aβ and tau pathology in later stages (6,30). This may also explain why it was recently shown in a combined analysis of subjects with amnestic MCI and early amnestic dementia in the BioFINDER-2 and AD Neuroimaging Initiative cohorts that tau PET provided better prediction of cognitive decline over 2 y than p-tau217 (both in CSF and plasma) or CSF Aβ42/Aβ40 ratio (53).

VALUE OF AD BIOMARKERS FOR DETERMINATION OF DISEASE SEVERITY AND DISEASE STAGING

The ideal AD biomarker should, in its readout, correlate with disease severity, and for AD pathology biomarkers, should correlate with histopathologic staging of a specific type of neuropathology over all stages. Such biomarkers would, for instance, allow accurate assessment of disease progression and quantitative determination of therapy effects.

For the amyloid biomarkers, convincing evidence was provided in head-to-head in vivo PET imaging versus postmortem histopathology studies in which the PET readout was closely associated with the density of neocortical Aβ plaques (54,55). Also, Grothe et al. developed a PET-based hierarchical amyloid staging system that resembles neuropathologic staging (56). Importantly, quantification has been added as an adjunct to visual assessment in Europe (but is not yet currently included in the FDA labels). Particularly, the so-called Centiloid scale has been established as a measure for quantitative estimation of the level of amyloid pathology, with the great advantage of comparability even between different amyloid PET tracers. Values of more than 30 Centiloids are considered reliable for the detection of aggregated amyloid pathology, and values of less than 10 Centiloids are considered reliable for its exclusion (57,58). The Centiloid values were also used to detect and quantify the therapeutic success of the new antiamyloid therapies. No comparable measure has yet been established for fluid biomarkers. The reason why the AUCs currently still consider amyloid PET rarely suitable for staging and tracking disease progression may be that progression of symptomatic disease in later disease stages is no longer accompanied by progression of the amyloid burden, which often has already reached a plateau.

Many studies have also tested the ability of tau PET tracers to stage tau pathology (59–61). Recently, standardized quantification scales in analogy to the Centiloid classification have been introduced including the so-called Centaur scale (62), or the so-called fill states, quantifying the area of brain regions covered with aggregated tau pathology (41). It seems difficult to see how such staging approaches including regional distribution of pathology could be mirrored using fluid biomarkers. Accordingly, the recent Core criteria also suggest tau PET as the leading and currently only approved biomarker for staging disease severity (Core 2 biomarker). A recent therapy trial used tau PET–based staging for patient inclusion, aiming to exclude patients with early tau stages (who were expected to show only minor disease progression within the follow-up) (63). Interestingly, the recent Core criteria consider a main role of tau PET for staging biologic disease severity (3), whereas in the current AUCs for tau PET, its role to stage and track disease progress is considered to be uncertain (28).

For the main neurodegeneration N biomarkers, that is, structural MRI and FDG PET, it is common sense that they are suitable to provide in vivo information on the disease severity in symptomatic stages. In this context, it is surprising why, in the novel AD biomarker categorization proposal, they are not considered suitable for staging (i.e., to be Core 2 markers) in patients with established Core 1 positivity. In fact, it has been demonstrated that FDG PET and tau PET data are inversely correlated to substantial degree (64).

CSF biomarkers of amyloid (A or Core 1 biomarkers) may show potential to stage earlier or soluble pathology stages of disease. As mentioned above, studies correlating CSF Aβ42 levels with amyloid PET changes implicate a nonlinear L-type relationship between the two (Fig. 3) (47), indicating that a plateau is reached for CSF changes earlier than for amyloid PET, thus no longer reflecting progress of disease in later stages. In a trial using autopsy data, no association between CSF Aβ1–42 and AD neuropathological changes was observed, whereas CSF P-tau181 was associated with AD neuropathology in APOE ε4 noncarriers but also reached a plateau. The authors concluded that CSF biomarkers may not be well suited for staging or monitoring AD pathology (65).

With regard to CSF and plasma tau biomarkers, some evidence is available from postmortem cross-validation studies for a correlation between these readouts and the density of both Aβ plaques and tau tangles (19,65). In a head-to-head comparison study in 78 subjects with amyloid-positive cognitive impairment, the flortaucipir tau PET data correlated stronger with the severity of cognitive symptoms and that of brain atrophy than CSF p-tau181 (66).

Plasma p-tau values have been demonstrated to potentially reflect Aβ deposition in early stages and a mix of amyloid and tau pathology in later disease stages (6). This notion is confirmed by studies correlating p-tau plasma biomarkers with tau PET–based tangle load, demonstrating some correlations between the 2 (r > 0.6), which are apparently driven by later disease stages (high ADNPC) and show little correlation in patients with low ADNPC.

CSF or plasma markers of neurodegeneration may allow to some extent categoric disease staging with regard to the onset and presence of relevant neurodegeneration. However, regarding their origin, they presumably reflect ongoing neuronal injury than the stage of neuronal damage. This is confirmed by studies demonstrating elevated but not significantly different plasma NfL levels in amyloid-positive MCI and AD dementia (67). Again, the fluid N markers provide complementary and fundamentally different information than the imaging N markers.

For the fluid markers, compared with the imaging AD biomarkers, there is the general advantage of obtaining a multitude of different (A/T/N category) readout from 1 sample. As such, as a potential alternative to imaging-based AD staging, there are efforts ongoing to develop a data-driven staging scheme based on a multitude of different CSF tau readouts together with the CSF Aβ42/Aβ40 ratio (68). Also, the proposal was made to potentially allow future AD staging purely based on such a multitude of fluid markers, with a certain sequence in which this these markers become positive in the disease process: CSF Aβ42/Aβ40, p-tau181/217/231, p-tau205, MTBR-tau243, and non–p-tau fragments (3).

As reflected in common disease models, different imaging and fluid biomarkers may be suitable to provide complementary but different information on disease stages depending on the different phases of the disease. The fluid biomarkers may be suitable to stage earlier or soluble pathology phases of disease, whereas the imaging biomarkers may have advantages to stage later or aggregated pathology phases of disease. Generally, the A biomarkers may be suitable not only for earlier diagnosis but also for staging earlier phases of disease compared with the T and the N biomarkers. The recent Core 1 criteria do not currently take this possibility into account. It may be more difficult to interpret findings of the mixed biomarkers such as combinations and the p-tau plasma markers, which reflect varying ratios of neuropathologies in different phases of disease.

VALUE OF THE BIOMARKERS IN DIFFERENTIAL DIAGNOSIS OF DEMENTIA

With the emergence of antiamyloid drugs which will, as the downside to the much-desired positive clinical effect, bring not only relevant costs but also relevant side effects, an accurate biomarker-based differentiation between AD and non-AD dementias is becoming increasingly relevant. Also, there is increasing need to detect and differentiate atypical AD subtypes such as posterior cortical atrophy, logopenic variant primary progressive aphasia, behavioral variant AD, or corticobasal syndrome AD. Finally, differentiation and detection of other forms of dementia may have relevant prognostic and therapeutic consequences, such as avoiding neuroleptic medication in DLB (69).

Although, in principle, both imaging and fluid AD pathology biomarkers should be able to discriminate AD from non-AD dementias in the clinically manifest stages, AD subtype separation cannot be realized by amyloid PET, CSF, or plasma markers (19,20). This can currently only be accomplished by studying the distribution of the tau pathology by PET imaging (70) or by assessment of the patterns of neurodegeneration by FDG PET or MRI. Of note, tau PET using some of the second-generation tracers has the potential to also diagnose non-AD tauopathies such as progressive supranuclear palsy, corticobasal degeneration, or certain frontotemporal lobar degeneration subtypes via typical regional uptake patterns (71), a readout which, again, fluid biomarkers are currently not able to provide. The current AUCs consider amyloid PET to be appropriate for establishing diagnosis in MCI or dementia with typical as well as atypical features, whereas tau PET is considered appropriate in atypical or early onset but uncertain in typical cases (28).

With regard to differential diagnosis of non-AD forms of dementia, FDG PET has proven valuable for discriminating among various forms of neurodegenerative disorders, based on the detection of characteristic topographic phenotypes of hypometabolism (72). FDG PET may also support the diagnosis of more recently introduced types of disorders such as limbic predominant age-related TDP-43 encephalopathy (73). Importantly, fluid N biomarkers do not offer these options. Elevated NfL levels have been reported in various forms of neurodegenerative disorders, reflecting neuronal injury but not providing additional information on its regional distribution (19,20).

DLB may represent an important special case regarding differential diagnosis. It is well established that a high proportion of patients with DLB may have positive amyloid PET scans, thus not allowing a differential diagnosis between AD and DLB on the basis of amyloid PET (74). Interestingly, similar conclusions may be valid for fluid biomarkers. Similar levels of CSF Aβ42 in AD and DLB have been reported in several trials (19,20,75). Also, for plasma biomarkers, it has recently been demonstrated that plasma p-tau181 and p-tau231 were significantly elevated in DLB, particularly in a subgroup of patients with abnormal CSF Aβ42 levels (76). For a reliable differential diagnosis between DLB and AD, dopamine transporter imaging may therefore represent the current method of choice (77). In the future, α-synuclein seed amplification assays of CSF and potentially blood may add new options (78).

Another aspect in this regard refers to the fact that the probability of mixed pathologies leading to cognitive decline increases with age. Thus, respective biomarkers or biomarker combinations need to address the question of multiple pathologies. By doing so, a long-term vision might become realistic, that is, the testing of patients with dementia for all different pathologies presumably causally involved in neurodegeneration (Aβ, tau, α-synuclein, TDP-43), followed by the administration of a tailored therapy or combination, suitable to tackle the individual constellations of pathologies in the brain.

VALUE OF AD BIOMARKERS FOR TREATMENT QUALIFICATION AND THERAPY MONITORING

In the context of developing and applying novel AD pathology and targeting drugs, A/T/N biomarkers are of fundamental value. This refers to both treatment qualification and therapy monitoring (79). The concept of using A (and T) biomarkers to qualify patients with AD for treatment with antiamyloid drugs and to document therapy effects has been crucial for the successful approval of recent antiamyloid antibody therapies. After aducanumab, 2 promising antiamyloid antibodies, namely, lecanemab and donanemab, were approved or submitted for clinical approval in some countries, initially (in an accelerated approval pathway) on the basis of PET-based patient inclusion and treatment effect monitoring. The concept of combined amyloid PET and antiamyloid treatment may be considered as a promising extraoncologic example of theranostics (80). In general, biomarker-based validation of basic disease pathologies may also offer the possibility to apply novel drugs even to patients with presymptomatic AD. It is estimated that the use of AD biomarkers will steadily increase once new drug candidates, such as targeting tau aggregates, will emerge.

Within the clinical trials of the above-mentioned antiamyloid antibodies, amyloid PET imaging (rather than CSF Aβ assessment) was chosen as the method to qualify candidate patients for study inclusion. As it has already been discussed above that PET and CSF sampling are not interchangeable, it could be expected that PET imaging would also be required to establish the diagnosis for prescription of these drugs after their approval. This is particularly obvious because the above antibodies target Aβ aggregates or protofibrils and not earlier forms of Aβ pathology, which might turn CSF samples Aβ-positive.

Also, an interchangeable or parallel use of amyloid PET and CSF amyloid biomarkers to qualify patients for inclusion into antiamyloid drug testing trials (this also refers to therapy monitoring) bears the risks of including AD cases at different disease stages and, by doing so, of veiling positive drug effects. Obviously, these risks become more relevant the more such drug testing trials move into prodromal or even presymptomatic disease stages.

In this context, it is interesting to note that in the appropriate use recommendations for aducanumab, the following statements were made: If CSF results are ambiguous, amyloid PET imaging is recommended; individuals with abnormal CSF and normal amyloid PET lack evidence of amyloid plaques, which are the target of aducanumab; these patients should not be treated with aducanumab but rather reimaged with amyloid PET in 1–3 y (50). In contrast, the more recently introduced appropriate use recommendations for lecanemab state that patients qualify for the drug if they show “biomarker evidence (amyloid PET or CSF) for the AD pathophysiologic process.” However, it is also stated that “blood biomarkers for AD pathology…are not currently considered adequate…” for this purpose (81). With regard to donanemab, interestingly, the concept of PET imaging–based inclusion of study participants was even expanded from Aβ- to Aβ- plus tau imaging. In the phase III clinical trial (Trailblazer-Alz2), the study subjects were not only required to be positive on amyloid PET but also required to have a flortaucipir PET scan with an AD-typical uptake pattern (low/medium or high tau, indicated by a SUV ratio > 1.10 or positive visual read). Patients with very low or no tau were excluded (63). Again, one could have expected that for routine application of donanemab after approval similar PET-based inclusion criteria would have been required as in the drug trial. This type of patient stratification could not be similarly obtained from fluid tau biomarkers. Interestingly, according to the updated AUCs, not only amyloid PET but also tau PET is considered appropriate with regard to patient selection for amyloid targeting therapies, despite not being formally FDA-approved for this purpose (28). This further underlines the potential value of tau PET as a potential universal A/T/N biomarker.

It is important to realize that future therapy concepts will potentially aim to intervene at earlier stages of the disease. To this end, earlier diagnosis and standardized staging and stratification will be necessary to systematically investigate the effects of such therapeutic approaches. Once again, the specific advantages of imaging biomarkers could be particularly important for this purpose (see sections on early diagnosis and staging). Regarding fluid biomarkers, their potential to capture early (soluble pathology) disease processes may qualify them for therapy selection of future therapy and preventive concepts targeting early disease stages (in the preaggregated stages).

As far as blood biomarkers are concerned, their low invasiveness and potentially broad accessibility have put them in the spotlight as a potential screening tool for selecting patients for new therapies. An optimal screening test should be optimized for sensitivity, inexpensive and broadly available, and noninvasive or as little harmful as possible. Blood-based AD biomarker sampling holds the promise to possibly fulfill these criteria. Thus, we previously proposed using blood sampling as a screening method combined with imaging as a validation method for a staggered selection of patients for amyloid-targeting therapies (ATTs) (82). Similar considerations were taken up by other groups, for example, by the Global CEO Initiative on Alzheimer Disease BBM Workgroup (83). Based on our previous suggestions and with the goal to take into account the complementary strengths of imaging with CSF and blood biomarkers, we here propose a combined algorithm integrating CSF, blood, and imaging biomarker for patient selection (Fig. 4). This algorithm starts with blood biomarker–based screening of subjects considered to be at risk for AD. After blood biomarker assessment, subjects who are positive on the blood-based screening test and fulfill further inclusion criteria for therapy (clinical, MRI) may be forwarded to undergo PET imaging to verify the presence and extent and quantity of the drug target with high specificity, ultimately qualifying the subject for drug prescription in case of positivity. If PET imaging is not readily available and access/expertise with CSF biomarkers is locally provided, PET imaging could remain preserved to borderline or inconclusive CSF biomarker cases, as previously suggested (24). It needs to be mentioned that all current concepts (including those presented in this article) of how fluid and imaging of AD biomarkers might be used in an integrative manner suffer from the fact that diseases other than AD, which are likewise Aβ-positive, are not rigorously excluded and might as such be channeled to antiamyloid therapies. This point certainly needs more attention in the future. However, this algorithm currently represents a suggestion on how to reliably establish amyloid positivity, based on the current state of knowledge as compiled here. The implementation of such diagnostic pathways may face implementation barriers (costs, PET availability, insurance coverage), and thus, it would require systematic evaluation in the future. Nevertheless, we believe that this suggested concept provides a good summary of the current status of available biomarkers and emphasizes that no single category of biomarkers is systematically preferable to another with regard to treatment qualification of patients.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4.

Hypothetical concept for future use of imaging and fluid AD biomarkers to qualify patients for antiamyloid drug treatment.

In addition to patient selection, amyloid PET proved to have significant value in the ATT trials also for monitoring therapeutic efficacy. Although not yet formally FDA-approved for this purpose, lowering of amyloid PET tracer uptake was indirectly acknowledged by the FDA as a suitable surrogate biomarker of therapy response, “reasonably likely to predict a clinical benefit,” which served as a basis for accelerated approval of aducanumab and lecanemab (28,84). In the donanemab TRAILBLAZER-ALZ 2 phase III trial, patients were even switched to placebo when successfully meeting criteria for amyloid negativity (<11 Centiloids on any single PET scan or ≥11 to <25 Centiloids on 2 consecutive PET scans) (63). It is yet unclear if this will translate into clinical routine at some point. According to the updated AUCs, amyloid PET is considered appropriate for therapy monitoring of approved ATTs (28).

Concerning changes in tau PET as a consequence of antiamyloid therapy, results have been controversial, showing significant reduction in some trials (85) but not in others (63). However, tau PET is likewise able to monitor the biologic effect of anti-tau drugs (86).

CSF and plasma biomarkers have been used for therapy monitoring as secondary (in addition to PET imaging) biologic outcome parameters in some studies with some promising results. Interestingly, plasma biomarkers showed tendencies toward a normalization during therapy (i.e., increase of Aβ42/Aβ40 ratios and decrease of p-tau values). However, this trend would quickly reverse for the plasma biomarkers after discontinuation of the therapy, whereas the PET results would longer remain at a stable lower level (87). Consequently, no strong linear correlations between ATT-induced reductions in Centiloids and corresponding changes in plasma or CSF biomarkers have been reported so far to the best of our knowledge. Again, this underlines that fluid and imaging biomarkers are not interchangeable with regard to monitoring treatment response. It may be assumed that both AD biomarker classes measure different aspects and may hold complementary value to assess therapy response (i.e., imaging for removal of aggregated pathology vs. fluid biomarkers for mitigation of disease activity). More systematic work is necessary, however, to answer whether and to what extent blood-based Aβ and tau readouts might serve this purpose in the future (longitudinal/repeated CSF sampling is rather not suitable for patients). Regarding the plasma biomarkers, it should be considered that variable conditions such as kidney function might impact their levels (88). Thus, longitudinal (e.g., drug-induced) changes of cofactors such as kidney function might attenuate their potential to indirectly read out drug effects in the AD brain. As soon as therapies evolve that target earlier soluble disease pathologies, some of the fluid biomarkers could be beneficial in the therapy response evaluation (i.e., reduction of disease activity).

SUMMARY AND CONCLUSION

With the emergence of the ATN concept to define and diagnose AD on biologic grounds and the latest amyloid PET imaging–based accelerated FDA approval of lecanemab, AD biomarkers are indispensable to many researchers and clinicians to supplement clinical testing within the AD diagnostic algorithm. Here, the discussion of whether imaging and fluid AD biomarkers compete with each other or complement each other is culminating with high speed. The main take home messages of our perspectives in this regard are summarized.

As documented in this article, the different biomarkers are not equivalent: There are relevant biologic, technical, and practical differences among the different AD biomarkers available, even within the individual A/T/N or Core categories. This refers to all application scenarios, namely, AD dementia diagnosis and differential diagnosis, early AD diagnosis and AD dementia prediction, disease severity estimation and staging, antiamyloid antibody treatment qualification, and respective therapy monitoring. As such, imaging and fluid AD biomarkers are not simply interchangeable: they reflect different aspects and mechanisms of pathology. In general, fluid and imaging AD biomarkers complement each other. While imaging biomarkers directly report on disease extent and quantity of aggregates in the brain as well as regional distribution of pathology, fluid biomarkers indirectly mirror dynamic parameters of soluble disease activity.

Further, it is important to understand that the validation levels differ among different AD biomarkers. Many of the new emerging plasma biomarkers have only been tested in well-characterized specialty clinic populations and sometimes cross-validated against PET imaging. Such study designs are not suitable to generalize their results to the entire clinical field. Also, a cross-validation against the postmortem standard of truth, as done for the amyloid and tau PET techniques, is recommended. In general, more research in this regard is required, especially for the new emerging plasma biomarkers.

It is also obvious that the current AD biomarker classifications schemes such as the A/T/N or the more recent Core criteria have limitations. In particular, they somewhat neglect the complementary strengths and qualities of fluid and imaging biomarkers by bundling them in categories. Also, the categoric classification does not take into account the very important continuous changes and levels of abnormality within each category, which can be provided by the available biomarkers. Further, it is obvious that “positivity” of the fluid biomarkers (reflecting soluble pathology) is not identical to “positivity” of the imaging biomarkers (reflecting aggregated pathology). It may therefore be suggested to divide the A/T/N categories accordingly into soluble or aggregated pathology.

We currently do not see a one size fits all AD biomarker solution. From our perspective, there is no reason to systematically prefer one biomarker group over the other, such as considering CSF as primary biomarkers and imaging tools as secondary biomarkers. Instead, different diagnostic questions may require different AD biomarker applications or combinations. This is because the different biomarkers have different sensitivities and specificities for diagnosing AD and deciding on a specific treatment. Given their availability, we propose choosing the biomarker technique (imaging-based or fluid-based) depending on the actual question to be answered. This may include individual applications, combined applications, or applications in a meaningful sequence.

For some indications, several biomarkers may have relatively comparable value. This may include the categoric (yes/no) assessment of the probable presence of AD neuropathologic changes in advanced symptomatic disease.

Some other indications may benefit from the selection of the most suitable individual biomarker. For example, for early detection of incipient AD pathologic processes and early risk evaluation, fluid biomarkers might have advantages over PET imaging, whereas the opposite might be the case if estimation of the extent of manifest neuropathology is required for specific therapies (Fig. 4). Whenever information of the AD biomarker topology or pattern within the brain is relevant for differential diagnosis, PET imaging should be preferred over fluid biomarkers. The question of whether a PET imaging–based or a fluid biomarker should be primarily used for treatment qualification in clinical trials might depend on the actual therapeutic concept applied. As one example, Ossenkoppele et al. recently proposed preference for PET over CSF and plasma markers in the case of anti-tau immunotherapy and tau aggregation inhibitor testing, whereas the opposite was recommended in the case of tau production and posttranslational tau modification drug testing (89). Similar suggestions may apply for specific therapy control. In principle, imaging biomarkers are superior when quantification of a specific category of neuropathology is required and not the blanket detection of ADNPC. Also, for anti-AD drug testing in presymptomatic subjects, PET imaging may be predictive in terms of the time to symptom onset. Some fluid biomarkers, however, could be beneficial in therapy evaluation once therapies for earlier soluble disease pathologies are available.

Finally, some indications may require for meaningful combinations of different biomarkers. In the future, 1 application in which imaging and fluid AD biomarkers might complement each other might be a blood test for screening of subjects at risk, followed by specific PET imaging as a confirmatory test (Fig. 4). Excitingly, this combined fluid and imaging biomarker concept has the potential to be used to improve treatment qualification in not only AD but also other neurodegenerative disorders once respective fluid biomarkers, imaging biomarkers, and drugs are available for various proteinopathies (Aβ, tau , α-synuclein, TDP-43). Importantly, combined use of different biomarkers may not always be required in parallel, but potentially in a meaningful sequence, particularly regarding workflows of differential diagnosis. Here, the use of another biomarker (FDG PET) may only be required if the previous test was inconclusive (72). The same applies to patient selection for therapies in which PET may be useful in case of inconclusive CSF findings (24). In this context, the potentially universal value of tau PET providing information on A, T, and N in one test may also gain relevance in the future (90).

The assessment of the significance of individual biomarkers and their importance will continue to change in the future as a result of further developments. Significant developments are already foreseeable, particularly in the field of PET instrumentation. This concerns small brain-dedicated scanners, which will potentially improve the availability of the method. It also concerns a new generation of increasingly sensitive and high-resolution (neuro)scanners (91). Such advances in technology can improve sensitivity in the detection of lower concentrations of pathologic biomarkers, facilitating earlier diagnosis. The higher spatial resolution may allow small, sometimes early-affected brain regions, such as the hippocampus or even small brain nuclei, to be examined more accurately. Reduced noise and motion artifacts may also result, as well as more accurate quantification of tracer uptake, increasing the reliability of longitudinal studies tracking disease progression. Integrated PET/MRI scanner systems may additionally add value to multimodal integration and assessment of copathologies.

In addition to instrumentation, the introduction of deep learning and artificial intelligence techniques has great potential in image-based diagnostic assessment. Artificial intelligence may further improve the image quality and denoising data and decrease the dose. Machine learning models trained on large datasets may improve observer-independent assessment and allow standardized differential diagnostic classification and could potentially contribute to earlier diagnosis. Deep learning and artificial intelligence may also allow the integration of multimodal data (PET, MRI, fluid biomarkers, and cognitive scores) to predict disease progression and stratify patients by risk. They may even help by suggesting the most suitable diagnostic tools or therapies in the future (92).

Further advances can also be expected with regard to new tracers, potentially allowing the detection of specific pathologies other than amyloid and tau. New specific imaging and fluid biomarkers for non-AD neurodegenerative pathologies, such as α-synuclein, are eagerly awaited. Concerning the fluid AD biomarker techniques, groundbreaking developments are also expected, particularly in the field of plasma markers. It is estimated that plasma biomarkers may have the potential to gradually substitute the more invasive CSF assessment in the future and to provide novel insights into various neuropathologic processes.

For current and future portfolios of imaging and fluid AD biomarkers, a complementary rather than a competing use is recommended. Such an approach has optimal potential to improve AD dementia or differential diagnoses, early AD diagnosis or AD dementia prediction, disease severity estimation and staging, antiamyloid antibody treatment qualification, and respective therapy monitoring, to the good of our patients.

DISCLOSURE

Alexander Drzezga received research support from Siemens Healthineers, Life Molecular Imaging, GE HealthCare, AVID Radiopharmaceuticals, SOFIE, Eisai, Novartis/AAA, Ariceum Therapeutics; speaker/advisory board honoraria from Siemens Healthineers, Sanofi, GE HealthCare, Biogen, Novo Nordisk, Invicro, Novartis/AAA, Bayer Vital, Lilly, Peer View Institute for Medical Education, International Atomic Energy Agency; and stocks from Siemens Healthineers, Lantheus Holding, and Lilly. He holds the patent for ¹⁸F-JK-PSMA-7 (patent no. EP3765097A1). Henryk Barthel received research support from Life Molecular Imaging; consulting/speaker honoraria from Hermes Medical Solutions, IBA, Lilly, Eisai, Novartis/AAA, and GE HealthCare; reader honoraria from Life Molecular Imaging; dosing committee honoraria from Pharmtrace; and scientific advisory board honoraria from SOFIE and Positrigo. No other potential conflict of interest relevant to this article was reported.