This concludes our two-part series. See also Part 1.
23 May 2009. Chugging along full-steam ahead with analysis of data from live brain imaging, fluid biomarkers, and genetic screens, the Alzheimer’s Disease Neuroimaging Initiative (ADNI) was originally conceived as an exploratory project to identify the best measures for tracking disease progression. However, the focus of the $64 million initiative has widened, perhaps shifted, of late. This was the sense people took away from a 26 April internal data presentations meeting and from ADNI presentations at the Annual Meeting of the American Academy of Neurology (AAN) held during the same week, both in Seattle. “When ADNI started, it was about using these methods as outcome measures for AD and mild cognitive impairment (MCI) patients,” said Bill Jagust of the University of California, Berkeley, who heads ADNI’s fluorodeoxyglucose (FDG)-positron emission tomography (PET) core. “Now I think it’s about using these measures as biomarkers to predict what’s going to happen to people early in the disease course.” This shift comes in part from technological advances that enable detection of brain Aβ pathology years in advance of clinical symptoms, and from a growing body of research hinting that the early amyloid may predict impending dementia in otherwise healthy seniors (see ARF related news story). Mining the ADNI data, which is freely downloadable, scientists are getting a better handle on the pathological cascade that leads to AD—how these events proceed in time, how they are linked, and which biomarkers can best capture them.
In Seattle, Jagust reviewed a set of studies his group has done to address relationships between biomarkers—brain amyloid as measured by Pittsburgh Compound-B (PIB), glucose metabolism as determined by FDG-PET, and cerebrospinal fluid (CSF) levels of Aβ and tau. “What actually motivated me to do this [first] study was the high number of PIB-positive controls,” he said. (About half of ADNI’s normal elderly volunteers have high amyloid load on PIB-PET.) When Jagust and colleagues looked at CSF measures in these PIB-positive folks, most were low for CSF Aβ1-42. Overall, the agreement was about 90 percent for PIB and CSF Aβ, and weaker (55-75 percent) for PIB and tau. However, while the amyloid markers correlated tightly with each other, neither seemed to track well with cognitive tests. Specifically, cross-sectional associations between amyloid markers and MMSE were weak. Glucose metabolism measures held greater promise in this regard: baseline FDG-PET not only correlated with baseline MMSE and ADAS-Cog scores, but also predicted and paralleled ADAS-Cog change. The researchers calculated that using FDG-PET in lieu of ADAS-Cog could cut sample size by nearly a third in a 12-month AD trial aiming to detect a 25 percent treatment effect. FDG-PET also came out on top—relative to CSF biomarkers, MRI hippocampal grey matter density, and ApoE genotype—in a test of the relative efficacy of these measures at predicting conversion and tracking cognitive decline in MCI patients over two years, Jagust reported.
In a separate biomarker analysis using ADNI data, Cliff Jack of the Mayo Clinic in Rochester, Minnesota, compared the relative merits of CSF measures (total tau, phospho-tau, and Aβ1-42) with MRI—that is, a computer program that gauges disease severity from atrophy patterns on MRI scans. His group developed this MRI algorithm—the STructural Abnormality iNDex (STAND) score—and presented data at the 2008 International Conference on Alzheimer’s Disease (ICAD) showing that it can distinguish between MRI scans of healthy people and AD patients with 90 percent accuracy. In Seattle last month, Jack reported that the MRI algorithm seemed to work a tad better than the CSF measures at predicting which clinical group a given person belonged to, and at tracking with cognitive and functional measures. “MRI slightly outperforms CSF biomarkers in predicting future Clinical Dementia Rating (CDR) change, and time to MCI-AD conversation,” claimed Jack, who heads ADNI’s MRI core. To Jack, it makes biological sense that the MRI biomarker in this study tracked most closely with cognitive status and did the best at predicting short-term future cognitive change, because MRI measures neurodegeneration. MRI marks cumulative injury, whereas CSF tau reflects active neuronal injury, he noted.
In collaboration with Jack and other ADNI investigators, Liana Apostolova of the University of California, Los Angeles, reported at AAN that a three-dimensional method of studying hippocampal atrophy shows good agreement with CSF biomarkers. Aβ, and ratios of tau/Aβ and phospho-tau/Aβ, correlated best with brain loss revealed by MRI; tau and phospho-tau showed weaker correlations. Several years ago, Paul Thompson of UCLA came up with the computer algorithm that assembles MRI slice images into 3D maps to measure brain volume loss more precisely. “The new thing is our automated technique,” Apostolova told ARF. “You can get volumes of the whole dataset in a minute. It’s that beautiful.”
In other biomarker news, researchers led by Satoshi Minoshima and first author Eric Petrie of the University of Washington, Seattle, report that high CSF levels of tau or phospho-tau in healthy people were linked to reduced glucose metabolism in several brain areas affected in early AD. The results appear in this month’s issue of the Archives of Neurology (Petrie et al., 2009). And in an independent analysis by Martin Ingelsson and first author Elin Blom of Uppsala University, Sweden, and colleagues, CSF tau biomarkers (total tau and phospho-tau) and ApoE4 genotype (Blom et al., 2009) did well at predicting which MCI patients would decline to AD. In addition, people with high CSF T-tau or P-tau and two ApoE4 alleles progressed faster from MCI to AD. That study was published online 7 May in the journal Dementia and Geriatric Cognitive Disorders.
“Each biomarker is probably different at different stages of disease. That’s what we’re getting a handle on now,” Jagust said. “The later you go, the more you want biomarkers that are tied to degeneration. The best markers for that are changes in MR volumes, changes in glucose metabolism, maybe changes in CSF phospho-tau or tau. The earlier you go, the more you’re going to want markers related to β amyloid deposition.”
At AAN, Charles DeCarli of the University of California, Davis, presented ADNI data that underscore the importance of a multi-modal approach for measuring clinical outcomes with biomarkers. He stressed that the best markers for neurodegeneration may not have the final word when it comes to gauging disease progression. “When we think about what affects cognitive function, we have to consider not only the loss of the brain cells but also the direct toxic effects that may occur in addition to cell loss,” DeCarli said in an ARF interview in advance of his AAN talk. For example, Aβ oligomers wreak havoc on neuronal transmission (Shankar et al., 2008 and ARF related news story) without necessarily killing cells. Likewise, tau is released into the brain as neurons die but is not just a marker for cell death, DeCarli said. “It is a sign of injury as much as it is a sign of death.”
To dissect Aβ’s direct and indirect effects on cognition, DeCarli’s team analyzed 312 ADNI participants (91 controls, 151 MCI, 70 AD) with baseline CSF data and repeated MRI and cognitive measures. The researchers used multiple regression analysis to systematically test the effects of CSF Aβ, CSF tau, and brain volume loss on one-year cognitive change. DeCarli reported that CSF Aβ levels correlated strongly with ADAS-Cog change but that Aβ’s direct influence on cognition was quite small. Much of its effect on cognition was indirect, mediated by cell injury (tau) and cell death (progressive atrophy). However, when the researchers used a more specific measure of episodic memory instead of the ADAS-Cog, Aβ had a stronger direct effect on that, DeCarli said. The upshot of this is that Aβ and tau have both direct and indirect effects on cognition. Hence, “in a clinical trial, you could move Aβ or tau measures and have a cognitive outcome without affecting brain structure,” he told ARF.
Where is all this research on AD biomarkers leading? One of ADNI’s long-term goals is to develop surrogate markers that the U.S. Food and Drug Administration will ultimately sanction to stand in for slower cognitive measures (e.g., the ADAS-Cog, CDR sum of boxes) in AD clinical trials, said principal investigator Michael Weiner, of the University of California, San Francisco, in an Alzforum interview. However, ADNI is not “settling down on one or two methods,” he noted. “Some methods may be good at predicting change but not as good for tracking or monitoring change.” In a similar vein, just because one measure gives the most power to detect change in one regard does not mean it is also the best way to detect a treatment effect, he told ARF. (See Weiner-Potter exchange below).
Weiner stressed that the ADNI findings thus far are preliminary. The Seattle meetings featured analysis of the one-year data, with two-year data collection set to conclude this fall. “One year in AD is an eye blink. Most clinical trials last 18 months to two years,” Weiner said. “At meetings I’ve stood up and said, ‘This is all very exciting, but don’t go away thinking you’ve heard the last word. These are not the publications. These are talks. Most of these data were calculated in the last three weeks.’”
That said, ADNI is charging ahead with plans for its second phase, ADNI 2. In November, ADNI investigators will apply to renew their ADNI 1 grant—to the tune of $40 million—with the U.S. National Institutes of Health, Weiner said. Later this month, they will submit an application for a $24 million “Grand Opportunities” (GO) grant, offered as part of the American Recovery & Reinvestment Act. David Lee of the Foundation for the National Institutes of Health confirmed in an e-mail to ARF that industry support for ADNI 2 is anticipated at around the same level as the current ADNI—at least $20 million.
ADNI 2 will focus on milder patients, so-called eMCI (early MCI) folks. Exact criteria for this population have not yet been finalized, but Weiner shared a few guidelines. These study volunteers cannot be considered normal, nor can they have stroke, cancer, or other exclusion criteria, he said. In addition, they cannot meet the current criteria for amnestic MCI; they would be somewhere between normal and aMCI.
Studies in early MCI people would feed into another long-term goal of ADNI, i.e., AD prevention trials. “We don’t want to wait for people to get impaired and demented,” Weiner said. “We want to prevent disease.” He says the field already has a pretty good idea who is prone to develop AD—i.e., those with small hippocampi, low CSF Aβ, positive PIB-PET scans, ApoE4 genotype. Prevention trials would involve offering a potential treatment to healthy people with such characteristics and following them for a long time, Weiner said.
In the meantime, he hopes ADNI can provide impetus for future data-sharing research initiatives. “One thing that makes ADNI unique and a model for all science in the world is that all of our data are put up on the Web and available to everybody without any embargo at all,” Weiner said. “In my opinion, all scientists should at some point, perhaps when submitting papers, put their raw data on the Web, especially if it’s taxpayer-supported. Scientists don’t own their data.”—Esther Landhuis.
This concludes our two-part series. See also Part 1.