Part 1 of a two-part story.
As researchers analyze data in longitudinal biomarker studies, they find that some previously estimated biomarker trajectories do not hold up. At the Alzheimer’s Association International Conference 2017, held July 16–20 in London, one tenet to tumble was the idea that biomarkers change along the neatly stacked sigmoidal curves drawn in progression models of AD that have become so familiar to the field. In reality, in-person serial data now suggest that biomarkers change in a range of different patterns, and vary by brain region and disease stage. Rates of change accelerate and decelerate at different points in time; some markers, for example cerebrospinal fluid p-tau181, even reverse course from rising to falling as disease worsens.
These revelations not only add nuance to disease diagnosis, progression, and prognosis, they also complicate interpretation of biomarker outcome measures in clinical trials, speakers said. “In order to plan interventions for a chronic disease, researchers need good estimates for when a biomarker starts changing and how it behaves,” noted Eric McDade of Washington University in St. Louis.
Much of the longitudinal data presented at AAIC came from the Dominantly Inherited Alzheimer Network (DIAN), which enrolls adults from families with autosomal-dominant AD. Because they are younger than people with sporadic AD, early pathological signs in their brains are not confounded by age-related changes, allowing researchers a glimpse into a pure AD process. In this population, some changes, such as hippocampal atrophy and tau tangles, occurred later than researchers had expected based on prior cross-sectional and sporadic AD data. The findings discussed at AAIC are preliminary as, in some cases, they are based on but two or three time points. More data will refine estimates of biomarker change, researchers say.
This story sums up several DIAN presentations. Part 2 of this story covers other AAIC talks presenting longitudinal data from sporadic AD that added insight into age-related changes in ApoE4 carriers, and from people who maintain excellent memory into old age.
Changing Trajectories. Rates of amyloid accumulation (left column), neuronal hypometabolism (center), and tissue atrophy (right) vary by time and brain region. Top, inferior temporal (orange), precuneus (blue), and insula (green); bottom, caudate (purple), putamen (red), and hippocampus (turquoise). [Courtesy of Brian Gordon, AAIC 2017.]
Longitudinal Data Spring Surprises
The DIAN observational study started enrolling in January 2009 (Nov 2008 news series). Longitudinal data remain limited, in part because DIAN participants who were many years away from their expected age at onset initially came in for comprehensive testing only once every three years. Moreover, many participants left the observational study to join DIAN-TU treatment trials, and others left because they became too ill to travel.
At AAIC, McDade reported findings based on data collected through July 2016. This covered 251 carriers of pathological mutations in APP, PS1, or PS2, and 160 noncarriers. The cohort’s average age was 39 and a little more than half had made two or more clinic visits, which take three to four days each and encompass the full suite of imaging and fluid biomarkers and cognitive and clinical tests. From this initial dataset, the researchers were surprised to learn that some biomarkers changed more rapidly than they had expected based on DIAN cross-sectional results.
To be sure, some biomarkers did behave as anticipated. Rates of change for amyloid PET and cerebrospinal fluid Aβ42 in mutation carriers diverged from those in noncarriers about 25 years before the expected onset of symptoms (Jul 2012 news). Whereas CSF Aβ42 levels fell most rapidly early in the disease, with the rate of change slowing as people approached symptom onset, brain amyloid deposition as per PET continued accelerating even after symptoms appeared. Other researchers at AAIC noted that CSF and PET measures followed their own trajectories in sporadic disease as well (Aug 2017 conference news).
Likewise, glucose metabolism in the precuneus started to wane at 15 years before expected symptom onset, just as researchers anticipated from the cross-sectional data. Decline was gradual until symptoms began, at which point it accelerated sharply.
That said, other biomarkers tracked differently than expected. Hippocampal atrophy and cognitive decline both appeared later than in the cross-sectional study (Bateman et al., 2012). There, hippocampi in mutation carriers appeared shrunken as early as 10 to 15 years before symptom onset, while cognition dipped five years before. Longitudinal data showed, however, that hippocampal atrophy accelerated only a year before symptoms appeared. Scores on a cognitive composite only began to worsen two years prior to symptoms, while the CDR-SOB did not budge until symptoms emerged. “The longitudinal changes [in hippocampal atrophy] correlate closely with cognition, which is what we might expect,” McDade noted. McDade emphasized that this young population has no age-related hippocampal atrophy, which might explain why shrinkage shows up only as symptoms begin. A separate analysis of DIAN data recently pegged hippocampal atrophy to symptom onset as well (Kinnunen et al., 2017).
The biggest surprise in the data was the behavior of CSF total tau and phosphorylated tau. In cross-sectional data, these markers are first elevated about 15 years before symptom onset and steadily climb as people approach and even pass diagnosis. In longitudinal data, however, total tau levels were already high in mutation carriers when they entered the study, and the researchers saw little difference in a person over time, with the rate of change close to zero. Probably the rise occurs so early in the disease that it did not show up in these data, McDade said. Phospho-tau followed a different course. Its level stayed high in mutation carriers until symptom onset, at which point it dropped dramatically. Previous longitudinal data from DIAN had hinted at a drop in CSF total tau and p-tau once symptoms emerged (Mar 2014 news). The findings emphasize that CSF total tau and p-tau measure different things, McDade said. He noted, however, that these proteins were measured using older ELISA-based assays, rather than newer automated systems that are more accurate (Aug 2015 conference news; Apr 2017 conference news). He plans to repeat the CSF analysis using high-performance systems.
Tau PET Distinct in Familial Disease
Could tau PET clarify how tau pathology evolves in familial AD? This marker was recently added to DIAN, and longitudinal data are not yet available. However, initial cross-sectional results already have surprised researchers. At AAIC, Tammie Benzinger of WashU presented tau PET data from 50 DIAN participants. Fourteen were symptomatic carriers with an average age of 50, 20 were asymptomatic carriers averaging 39 years old, and 16 were non-carriers with an average age of 38. The asymptomatic carriers were about 11 years shy of their estimated age of onset on average, while the symptomatic carriers averaged two years past it.
To Benzinger’s surprise, tau PET was negative in asymptomatic carriers up until the age of onset. This contrasts with late-onset AD, where researchers have found positive tau PET signals in amyloid-positive people who have no clinical symptoms (Mar 2016 news; Aug 2016 conference news). Moreover, DIAN mutation carriers at onset age or beyond had intense uptake of tau tracer, two to three times higher than that seen in symptomatic late-onset AD. Benzinger had reported similar preliminary findings from growing numbers of DIAN participants last year, noting that in familial disease, tau tangles seem to storm onto the scene late (Feb 2016 conference news; Aug 2016 conference news). Those hints are reproduced in this larger dataset.
In addition, the pattern of deposition in the brain varied from that seen in LOAD. In DIAN, tau deposition appears concentrated in posterior and precuneus regions, which also become hypometabolic in DIAN participants, Benzinger noted. Furthermore, the tau PET signal correlated with atrophy in these regions.
What do these findings of a late tau PET signal mean? Intriguingly, the signal shows up around the same time as CSF p-tau drops in longitudinal samples. McDade noted that existing tau tracers bind to hyperphosphorylated paired helical filaments of tau. Thus, tau tracers may be detecting the same form of tau as CSF p-tau assays. Possibly, as tau tangles spread across the brain, they absorb soluble p-tau, leading to its drop in CSF, much as amyloid plaques are believed to soak up soluble Aβ42, McDade speculated.
CSF total tau, by contrast, increases long before PET detects tangles in the brain. Total tau likely reflects neuronal injury and the lysing of cells, and bears little relationship to tangle deposition, McDade said.
Some researchers at AAIC speculated that differences in tau pathology in young and old brains may explain the late but rapid rise of the tau PET signal in young mutation carriers. Older brains accumulate some age-related tangles without developing notable cognitive decline. In the young DIAN brains, however, tau tangles seem tightly linked to clinical symptoms.
At AAIC, Keith Johnson of Massachusetts General Hospital, Boston, reported that in the Harvard Aging Brain Study, the older the person, the weaker the link between tangles and cognitive decline. HABS tracks cognitive change in older adults who start out cognitively healthy. William Jagust of the University of California, Berkeley, said he has seen a related phenomenon in ADNI data: Older people with AD take up less tau tracer than younger AD patients do. Overall, tau seems less informative about disease in older brains. The reason is unclear, although some researchers have previously suggested that tau may no longer track decline at the end stage of the disease because too many neurons have already died. “These findings may be telling us something fundamental about the biology of tau,” Benzinger agreed. However, she also noted that the AV1451 tau tracer used in DIAN binds off-target to the glial inflammatory marker monoamine oxidase. Thus, some of the intense signal in young mutation carriers could be a result of neuroinflammation, Benzinger suggested.
Changing Biomarker Trajectories
The longitudinal findings highlight a dilemma for researchers planning to use biomarkers in clinical trials: They cannot yet predict how a biomarker might change with treatment. For example, if an intervention raised CSF p-tau, would that indicate improvement or worsening of disease? It might depend on exactly where in the disease trajectory the person was when treatment started, researchers agreed. The same applies to CSF Aβ, where researchers have wondered for years what a treatment benefit would look like. They also do not know how a therapy that shifts one marker might affect others. Would a treatment that lowers brain amyloid below the threshold for PET positivity stop the spread of tau or improve cognition? If a runaway process has already taken hold in the brain, perhaps not, McDade noted. For these reasons, cognitive impairment will remain the key outcome measure for trials, he predicted.
Other data from DIAN may help researchers pick the best biomarker for a particular disease stage by showing when and where each marker changes most rapidly. Brian Gordon at WashU presented longitudinal imaging findings from 88 symptomatic carriers, 141 presymptomatic carriers, and 148 non-carriers in DIAN. Most participants had made two clinic visits. Gordon compared amyloid PET, FDG PET, and structural MRI scans for 34 cortical and seven subcortical regions. He found tremendous variety in rates of change by brain region and disease stage (see image above).
For example, amyloid plaque load climbed fastest in the precuneus up until about 10 years before symptom onset. At that point, precuneus plaque accumulation slowed down, and after symptoms appeared, began to fall. From –10 years onward, plaques grew the fastest in the inferior temporal lobe. In the hippocampus, on the other hand, virtually no plaque deposited at any point in disease, belying data from LOAD. Thus, a trial in a very early preclinical familial population might best track amyloid PET in the precuneus, while the inferior temporal lobe signal would be more telling for later disease stages. “These would be the best regions to look at to judge the potential effectiveness of an anti-amyloid drug,” Gordon wrote to Alzforum.
Intriguingly, the precuneus also showed the most dramatic changes in FDG PET signal and brain volume throughout most of the course of disease. This region seems particularly susceptible to AD pathology, Gordon noted. The hippocampus appeared to shrink steadily throughout the whole disease course, from –30 to +10 years. “Rather than considering aggregate, summary measures of pathology, clinical trials should take into account these regional patterns and look at areas of the brain that have the most optimal signal properties,” Gordon wrote.
Another complication for DIAN studies regards how particular mutations may affect the results. At AAIC, Jasmeer Chhatwal at Massachusetts General Hospital, Boston, presented new data suggesting that specific mutations cause unique patterns of amyloid deposition. Early studies had reported that plaques form first in the striatum in familial disease, although later research noted tremendous variability in DIAN (Jun 2007 news; Mar 2012 conference news).
From 129 DIAN participants, who between them had made 181 clinic visits, Chhatwal stratified amyloid PET data by mutation. He found that only those people who carried a presenilin 1 mutation in either transmembrane domain 2 or 8 developed early striatal plaque. People with presenilin 2 transmembrane domain 2 mutations, by contrast, accumulated cortical plaque first. For all other mutations, accumulation occurred at about the same time in the two regions. Chhatwal did not analyze other regions such as the precuneus. It is still unclear why mutations have these distinct effects, he noted, suggesting that animal studies may be able to parse out the mechanisms. Intriguingly, a previous study found that presenilin 2 mutations churn out more intracellular Aβ42 than presenilin 1 mutations do (Jun 2016 news).—Madolyn Bowman Rogers
- Longitudinal Data Say: Nope, CSF Markers Do Not Track Progression
- Paper Alert: DIAN Biomarker Data Show Changes Decades Before AD
- CSF and Brain Markers Highlight Different Facets of Dementia
- DIAN Longitudinal Data Surprises With Late Drop in Tau
- CSF Aβ Assays Remain Fickle: Robots to the Rescue?
- Are CSF Assays Finally Ready for Prime Time?
- Tau PET Aligns Spread of Pathology with Alzheimer’s Staging
- Tau PET Studies Agree—Tangles Follow Amyloid, Precede Atrophy
- Tau Tracers Track First Emergence of Tangles in Familial Alzheimer’s
- Brain Imaging in DIAN: Atrophy Rates—Check. Tau PET—Not Yet.
- eFAD Research Surprise: In Mutation Carriers, Amyloid Starts in Striatum
- Miami: When Does Amyloid Deposition Start in Familial Alzheimer’s?
- Lodged in Late Endosomes, Presenilin 2 Churns Out Intraneuronal Aβ
- Bateman RJ, Xiong C, Benzinger TL, Fagan AM, Goate A, Fox NC, Marcus DS, Cairns NJ, Xie X, Blazey TM, Holtzman DM, Santacruz A, Buckles V, Oliver A, Moulder K, Aisen PS, Ghetti B, Klunk WE, McDade E, Martins RN, Masters CL, Mayeux R, Ringman JM, Rossor MN, Schofield PR, Sperling RA, Salloway S, Morris JC. Clinical and biomarker changes in dominantly inherited Alzheimer's disease. N Engl J Med. 2012 Aug 30;367(9):795-804. PubMed.
- Monogenetic and Sporadic Alzheimer’s Disease: Same or Different?
- Brain Imaging in DIAN: Atrophy Rates—Check. Tau PET—Not Yet.
- DIAN Longitudinal Data Say Cognition Goes Earlier Than Previously Thought
- Midlife CSF Crisis? Markers May Herald Future AD
- API Biomarker Data Mirror DIAN’s, Support Progression Models