Biomarkers in Alzheimer’s, Frontotemporal, Lewy Body, and Vascular Dementias

The most common neuroimaging finding in AD is cerebral atrophy, preferentially affecting the hippocampus and adjacent regions of the medial temporal lobes. In more advanced stages of the disease, atrophy extends into parietal and frontal association regions, with relative sparing of primary cortices such as motor and visual regions (6). Longitudinal studies have demonstrated that the rate and degree of gray matter atrophy are strongly associated with clinical indicators of AD severity and progression, with rates of cortical thinning in the inferior and medial temporal lobe and volume loss in the hippocampus serving as reliable indicators of conversion from MCI to AD upon repeat assessment (7–9). The rate of hippocampal volume loss is around 4.5% per year among patients with AD, 3.0% per year among those with MCI, and about 1.0% annual decline in healthy older adults (10).

Functional neuroimaging demonstrates patterns of reduced cellular metabolism in AD, often in the same regions in which atrophy occurs. One PET approach uses uptake of FDG as an indicator of glucose metabolism and closely associated neuronal activity. Patients with AD show significantly reduced cerebral glucose utilization in a characteristic pattern in the parietal and temporal lobes (11). Like cortical atrophy, the degree of hypometabolism correlates with disease severity, so that patients in the early stages of AD show reduced glucose uptake primarily circumscribed to temporal regions, which then extends into parietal and then also to frontal lobes and, ultimately, globally as the disease progresses (12–15).

PET ligands have also been developed that identify the two molecular neuropathological hallmarks of AD—amyloid plaques and neurofibrillary tangles. Pittsburgh Compound B (PiB) and now other, clinically available fluorinated ligands, including florbetapir, flutametamol, and florbetaben, have been shown to specifically bind cerebral tissue with high levels of amyloid deposition, as verified by neuropathological studies (16, 17). The presence of amyloid is the sine qua non of AD, with >96% of probable AD patients demonstrating significant amyloid accumulation, later verified at autopsy (18). However, amyloid is a poor indicator of disease progression, as the accumulation of plaques occurs early in the disease process, before observable cognitive decline (19). Cerebral amyloid burden has also been reported in the setting of dementia with Lewy bodies and Parkinson’s disease with dementia, with the former having a higher mean amyloid deposition (20, 21). Thus, although amyloid burden is necessary for the diagnosis of AD, it is not always specific to AD. More recently, novel PET ligands have been developed to specifically target tau (22). Interestingly, to date, tracers only appear to bind tau in AD and not tau in other tauopathies such as some FTLDs. As new and better tau ligands emerge that bind all forms of tau, regional distribution of tau tracer uptake may begin to serve as a diagnostic indicator of AD with perhaps greater specificity than amyloid deposition (23).

CSF is in intimate exchange with interstitial fluid of the brain and carries molecules secreted by neurons and glia as well as excreted substances, including metabolic waste and refuse. It is clinically accessible through minimally invasive lumbar puncture. Pathological proteins Aβ₄₂ and tau can both be measured reliably in the CSF using commercially available assays. Low levels of Aβ₄₂ protein and high levels of total and phosphorylated tau proteins are seen in AD (24, 25). Postmortem studies have found an inverse relationship between ventricular Aβ₄₂ CSF levels and cortical plaque load (26), leading to the assumption that “lower” Aβ₄₂ levels in the CSF are reflective of more extensive AD pathology. Decreased CSF Aβ₄₂ concentrations have also been found to relate inversely to high retention of amyloid radiotracer in the in vivo brain imaging of Aβ₄₂ pathology (16). The basis for this inverse relationship is presumed to be plaques that function as “sinks,” drawing soluble Aβ₄₂ from interstitial fluid that flows into or exchanges with CSF. Reduction in CSF Aβ₄₂ begins prior to symptom onset, and as such, normal CSF Aβ₄₂ levels in a person with dementia warrant further evaluation and reconsideration of an diagnosis of AD (27). On the other hand, low CSF Aβ₄₂ levels do not necessarily reflect brain amyloid deposition and can be seen in other conditions such as multiple sclerosis (27). Other forms of Aβ, including Aβ₄₀, can also be measured in CSF, and whereas some studies have suggested that the addition of CSF Aβ₄₀ to generate a 42/40 ratio may improve diagnostic accuracy in certain circumstances, this has not yet entered clinical practice (27). Microtubule-associated protein tau can also be measured in CSF, with levels of both total and phosphorylated tau found to be increased in AD (28). However, tau levels are also increased in stroke, neuroinflammatory conditions, and other disorders such as Creutzfeldt-Jakob disease. Despite ongoing research and the promise of new targets, the combination of low CSF Aβ₄₂ and high CSF tau predicts the presence of AD pathology with, perhaps, the greatest established accuracy of any biomarkers currently available (29, 30).

While CSF Aβ₄₂ and tau testing are available for clinical use, lumbar punctures are not routinely performed in dementia evaluations in many settings. Blood is considered a more accessible biofluid, but despite its practical advantages, no blood-based markers have yet found utility in clinical use. This is due, in part, to technical challenges, including that brain-derived proteins are diluted and often metabolized in blood and thus present in lower concentrations than in CSF, below the level of assay detection (31). The search for blood-based correlates of CSF Aβ₄₂ and tau levels has been unsuccessful to date—the relationship between plasma and CSF Aβ₄₂ levels is tenuous (32), as is that between plasma and CSF tau levels (33). New ultrasensitive assays are showing promise, however. Neurofilament light (NFL)—another neuron-specific cytoskeletal protein (21)—is one example of an emerging biomarker for AD and other neurodegenerative diseases. Results from previous studies suggest that patients with AD have increased plasma NFL levels, and plasma and CSF NFL levels are known to correlate (22). A recent study by Mattsson and colleagues (33), for example, demonstrated that plasma NFL correlated with CSF NFL, was increased in patients with MCI and AD dementia compared with controls, and had high diagnostic accuracy for patients with dementia, comparable with that of established CSF biomarkers. High plasma NFL also correlated with poor cognition, AD-related atrophy, and brain hypometabolism (18).

Approaches using techniques in which several biomarkers are analyzed simultaneously have identified potential “blood biomarker panels” (19), but findings from studies have unfortunately been hard to replicate (20). Such approaches attempt to leverage the burgeoning “omics” fields of proteomics (large-scale study of proteins), metabolomics (chemical fingerprints or metabolites generated as a result of cellular processes), and lipidomics (lipid pathways and networks). Casanova and colleagues, for example, recently reported using quantitative metabolomics to identify a panel of 10 plasma lipids that were highly predictive of conversion from normal cognition to AD but failed to replicate their findings in two larger datasets (34). Proitsi et al (2015) performed lipidomics on plasma samples from patients with late-onset AD, healthy controls, and individuals with MCI and identified a combination of 10 metabolites that classified patients with AD with 79% accuracy (35). Although one of the largest AD metabolomics studies to date, the sample size remained modest. Using a novel proteomics approach, Shah and colleagues (2016) used serum from patients with AD and matched controls analyzed by capillary liquid chromatography-electrospray ionization-tandem mass spectrometry, discovering 59 novel potential AD biomarkers, with 13 that recurred in more than one multimarker panel (36). These results, too, require replication in other larger, well-characterized cohorts.

Some AD biomarker experts note that a “combinatorial approach” will likely be required to identify a blood-based biomarker panel for AD (37). In the search for blood-based biomarkers, Lovestone notes that, “[S]amples and data matter. We need many samples, well curated and with as many data from other techniques measuring the same analytes as possible.” Moreover, “most ‘omics technologies remain in their infancy,” and there is no technical platform sufficiently superior to others to justify the enthusiasm sometimes expended on them by their supporters” (37).

As with AD, structural imaging such as high-resolution MRI is often undertaken as a first step in an FTLD workup. Structural MRI differentiates AD from FTLD by pattern of atrophy with high accuracy (41), with earliest volumetric differences appreciated in the insula and temporal lobes (42). Diffusion tenor imaging, or DTI, is also useful in distinguishing FTLD from AD, as changes in global white matter microstructure are more widespread in FTLD than in AD (43–45). bvFTD, characterized by marked behavioral changes and executive dysfunction, demonstrates cortical atrophy primarily involving frontal and lateral temporal lobes, with relative sparing of parietal regions (46). Meta-analyses have suggested that preferential tissue loss in the prefrontal, insular, anterior temporal, and anterior cingulate cortices, as well as volume loss in the thalamus and striatum, may be sensitive markers for distinguishing bvFTD from AD (47, 48). In the language variants of FTLD (semantic dementia and progressive nonfluent aphasia), cortical regions within the left hemisphere language network are particularly vulnerable to neurodegenerative processes, with distinct regional involvement mapping onto the language deficits unique to each subtype. In semantic dementia, atrophy is generally observed in the anterior and inferior temporal lobe, including the amygdala and hippocampus (49). However, because semantic dementia and AD both share pathology involving temporal lobe integrity, structural imaging is often used in conjunction with functional neuroimaging to facilitate diagnostic clarity. Hypometabolism of parietal regions differentiates AD from semantic dementia with greater specificity than reliance on cortical atrophy patterns alone (50). In progressive nonfluent aphasia, frontal areas involved in speech production are most vulnerable to neurodegeneration, with gray matter loss primarily occurring within the left inferior frontal gyrus and dorsolateral prefrontal cortex, as well as within the superior temporal gyrus and insula (49). Over time, cortical involvement extends to include more dorsal frontal regions, the parietal cortex, and superior temporal regions. Finally, because amyloid deposition is not typically observed in FTLD, amyloid PET scans can help differentiate between AD and FTLD (51).

Currently, the most helpful CSF biomarkers for FTLD are the same as for AD, because AD biomarkers can distinguish AD pathology from non-AD pathology (since reduction of Aβ₄₂ would not be expected in FTLD) (52). Several studies have reported that CSF total tau levels are lower in FTLD than in AD but higher than in controls (53, 54). In many FTLD cases, however, total tau levels can be normal. Elevated phosphorylated tau levels are typically seen in AD rather than other neurodegenerative diseases, with one study finding that reduced p-tau to t-tau ratio predicted TDP-43 pathology in FTLD (55). CSF Aβ₄₂ and tau are also valuable for differentiating between underlying AD and FTLD pathology in the differential diagnosis of primary progressive aphasia, in which an AD-like CSF profile is found in patients with logopenic variant primary progressive aphasia but not in patients with semantic or nonfluent aphasias (40).

NFL has been of increasing interest as a fluid biomarker for FTLD (40). Neurofilaments are a major constituent of the neuroaxonal cytoskeleton and play important roles in axonal transport, with increased NFL levels thought to reflect axonal damage (40). Two recent studies found elevated CSF NFL levels in FTLD, and these appeared to correlate with FTD disease severity (56, 57). Such data suggest that NFL may be a good biomarker of neurodegeneration and neural injury in general, rather than in any specific disease. As with AD, there is also considerable interest in inflammatory contributors to FTLD, such as proinflammatory cytokines tumor necrosis factor α (TNF-α) and transforming growth factor β (TGF-β), both of which have been found to be significantly increased in CSF in patients with FTLD compared with controls (58). While various changes in CSF and/or blood levels of cytokines have been found in FTLD, these changes, nevertheless, also seem to reflect nonspecific mechanisms, and many are also present in AD (40). Increased CSF TDP-43 has been reported in FTLD, though results of studies to-date have been mixed (40). Elevated TDP-43 is found in amyotrophic lateral sclerosis as well, a TDP-43 proteinopathy that often, though not always, is accompanied by FTLD (59).

There is considerable overlap in brain atrophic changes between AD and LBD; thus, the utility of MRI in this setting is limited (61). In a recent study, LBD, but not AD, pathology was associated with reduced amygdala volume in pathologically verified cases, but neither LBD nor AD pathology was associated with volume loss in hippocampus or entorhinal cortex, suggesting that other neuropathological factors accounted for atrophy in these structures (62). Atrophy in other cortical and subcortical structures has also been reported in LBD, including striatum, hypothalamus, and dorsal midbrain (63). Rates of atrophy in LBD (compared with AD) are mixed, with some studies suggesting higher rates in AD (64) and others suggesting higher rates in LBD (65).

SPECT and PET demonstrate abnormal glucose metabolism and perfusion deficits in parietal and occipital areas in DLB (66, 67). Radioligands such as ¹²³I-FP-CIT have also been developed to track dopamine transporter (DAT) loss with SPECT imaging (68). Reduced binding in the striatum reflects DAT loss in the substantia nigra, a disease process highly specific to LBD (69, 70). Preliminary tau PET studies suggest a gradient of increasing tau binding from cognitively normal Parkinson’s disease (absent to lowest) to DLB (intermediate) to AD (highest). However, given the presence of overlapping pathologies, future α-synuclein PET ligands are expected to have the best potential for distinguishing LBD from AD (71).

Currently, there are no blood or CSF markers available for diagnosis of LBD (61). α-synuclein has been studied as a potential CSF biomarker, but results from studies have been mixed. Quantification of CSF α-synuclein may help to differentiate the synucleinopathies (DLB, PD, and multiple system atrophy) from AD, but CSF α-synuclein levels do not appear to differ significantly across the synucleinopathies themselves (72, 73). Others have reported low Aβ₄₂ levels in DLB compared with those in controls (74) and cases of PDD (75), though such levels are also seen in AD. Studies have also suggested that CSF Aβ₃₈ as well as Aβ₄₂/Aβ₄₀ and Aβ₄₂/Aβ₃₈ ratios may discriminate AD from DLB, though the evidence base for this remains limited (76, 77).

Neuroimaging that demonstrates tissue damage of a presumed vascular origin defines the core diagnostic feature of VCID. While functional neuroimaging techniques (such as ASL-MRI) are capable of measuring in vivo cerebral blood perfusion, more commonly used clinical indicators of VCID are structural MRI sequences. In MRI, T2-weighted sequences such as fluid-attenuated inversion recovery (FLAIR) scans are most sensitive to vascular-related lesions—often described upon radiological review as deep white matter hyperintensities, periventricular hyperintensities (leukoaraiosis), and lacunar infarcts, as well as susceptibility-weighted imaging (SWI) protocols sensitive to old hemorrhagic lesions common in cerebral amyloid angiopathy. Although the presence of overt stroke is easily captured by structural MRI, it is the more insidious and diffuse ischemic tissue damage resulting from prolonged exposure to hypertension and other vacular risk factors that underlies the bulk of VCID pathophysiology in the absence of larger strokes (80). Among patients presenting with VCID, the quantitative burden of leukoaraiosis is generally related to the severity of cognitive deficits, with the extent of tissue damage correlating with worse cognitive performance and dementia severity. However, the presence of white matter signal abnormalities in older adults is not always an indicator of VCID, as patients can exhibit extensive leukoaraiosis while maintaining generally intact cognition. Indeed, over 96% of community-dwelling older adults over the age of 65 demonstrate white matter signal abnormalities on MRI, though only 20% have a lesion burden that is associated with decreased cognition and/or gait changes (85).

As is the case for FTLD, CSF markers of the greatest interest in VCID include Aβ₄₂, total tau, and phospho-tau, primarily to rule out (or rule in) the presence of AD pathology. It is important to note, however, that dementia is often of mixed etiology, particularly so for AD and VCID. Aβ₄₂ levels in AD are significantly lower than in “pure” VCID (86), though decreased Aβ₄₂ levels within the range of those in AD have also been reported (87). VCID cases with a low incidence of AD-related pathology (e.g., CADASIL) demonstrate Aβ₄₂ values falling between those of AD and control cases (87). CSF Aβ₄₀ levels have been reported to be lower in VCID when compared with AD and controls, and it has been suggested that the Aβ₄₂/Aβ₄₀ ratio may help to discriminate between AD and VCID cases (88).

Cerebrovascular damage seen in VCID may occur through pathways mediated by oxidative stress, inflammation, and the accumulation of advanced glycation end products, among others (89), with BBB alterations likely compromising the cerebral microenvironment (90). As such, markers that capture key drivers of these processes may ultimately have utility as biomarkers for VCID. The ratio of CSF to serum albumin, a marker of BBB structural and functional integrity, is increased in VCID (91), though this is a nonspecific finding that may not distinguish VCID from AD (92). Increased levels of CSF myelin lipid sulfatide (93) and NFL (94) have also been described in VCID, suggesting mechanisms of demyelination and axonal damage that occur in the setting of cerebrovascular injury.

Apart from CSF, serum C-reactive protein (CRP) and homocysteine levels are nonspecific vascular risk factor biomarkers, and increased levels have been reported in the presence of vascular lesions and VCID (95, 96). Elevated levels of serum homocysteine, however, can also be seen in AD (97), although this may be reflective of comorbid vascular pathology (89).