Little by Little—Standardizing, Validating Those Biomarkers
The devil lurks in the details. Biomarkers stand poised to transform Alzheimer’s research, but scientists first must iron out pesky technical issues and ensure that markers give reliable, repeatable results. Two new papers in Archives of Neurology describe some of this essential, albeit less glamorous, research—one on the fluid and one on the imaging fronts. In the first, published online May 9, researchers led by Anne Fagan at Washington University in St. Louis, Missouri, report the results of a head-to-head comparison of two commonly used assays for measuring cerebrospinal fluid (CSF) levels of the AD markers tau, phospho-tau, and Aβ42. They found that, although the assays gave widely varying absolute values, they were equally effective at identifying underlying amyloid pathology. The other paper, in the May print edition, describes work led by Alireza Atri at Massachusetts General Hospital and Reisa Sperling at Brigham and Women’s Hospital, Boston. Their groups examined the test-retest reliability of task-related functional MRI over time in a group of patients with mild AD. The scientists found the scan results were stable, suggesting that fMRI maps would be sensitive to small functional changes and could potentially be used to measure drug effects in small clinical trials.
New diagnostic criteria for AD call for more use of biomarkers in research and clinical trials (see ARF related news story). CSF biomarkers look particularly promising for widespread use, but the field has been dogged with quality control problems and wide variability among assays. Several groups are leading initiatives to identify sources of assay variation and to standardize methods. The largest is the Quality Control (QC) program sponsored by the Alzheimer’s Association and led by Kaj Blennow at Sahlgrenska University Hospital, Mölndal, Sweden, now ongoing in over 60 centers worldwide (see ARF related news series), including Fagan’s. Separately, Fagan is also interested in the problem of CSF biomarker reliability, and chose to home in on the effects of assay type.
One factor that explains inconsistent results is the difference among commercial assay kits. The INNOTEST enzyme-linked immunosorbent assay (ELISA), a plate-based, single analyte immunoassay made by Innogenetics NV, Ghent, Belgium, has been widely used, but is now seeing competition from newer bead-based immunoassays that employ Luminex’s xMAP technology and allow researchers to measure multiple biomarkers in the same well. Innogenetics makes one such xMAP-based test, the INNO-BIA AlzBio3, which measures Aβ42, tau, and phospho-tau. Blennow’s group found that both tests give diagnostically useful results, but many researchers wondered why one test consistently gave numerically different results than the other when tested on the identical CSF sample (see Olsson et al., 2005). Importantly, several groups, including Fagan’s, have shown that low CSF Aβ42, as measured by INNOTEST, correlates with amyloid deposits in the brain as seen by positron emission tomography with Pittsburgh Compound B (PIB-PET) (see, e.g., ARF related news story on Fagan et al., 2006). This has not been shown for the xMAP platform, however.
To rectify this, Fagan used data from her ongoing longitudinal study on healthy aging and dementia at the Knight Alzheimer’s Disease Research Center at WashU. She evaluated about 100 participants who received a PIB-PET scan and had CSF drawn within a two-year span. Although most participants were cognitively normal, the cohort included a handful of people with mild or moderate dementia, and also spanned a wide range of PIB-PET values. The WashU group used the INNOTEST kit, then shipped CSF samples to Leslie Shaw and John Trojanowski at the Alzheimer’s Disease Neuroimaging Initiative (ADNI) Biomarker Core at the University of Pennsylvania, Philadelphia. Shaw and colleagues analyzed the same samples with the xMAP platform INNO-BIA test. As suspected, the absolute values obtained by each method varied by two- to sixfold. Even so, test results correlated with each other, and with amyloid plaques in the brain equally well, the authors found. Fagan said that they tested the Innogenetics kits because they had experience with these assays, emphasizing that her group is not promoting any particular product.
One intriguing finding was that the ratio of either total tau to Aβ 42, or phospho-tau to Aβ42, proved highly accurate for determining whether a person’s brain contained amyloid plaques. The WashU group had previously shown that these ratios predict cognitive decline (see ARF related news story on Fagan et al., 2007). That the ratios correlated better with brain amyloid than did any single biomarker strengthens the idea that they might be the most valuable measure to use in clinical research, Fagan suggested. A key question yet to be answered is what a given CSF biomarker value means for an individual’s risk of dementia. Definitively answering that requires longitudinal studies, Fagan said, in which researchers look at the change in biomarkers and cognitive performance within individuals over time. Her group is collecting those data now.
The study results also mean that researchers can use either ELISA or xMAP-based tests with confidence, Fagan told ARF. Coauthor Hugo Vanderstichele at Innogenetics agreed: “If you have good tools for detection, then the clinical validity will be independent of the platform. However, well-designed studies like this are needed to convince the community,” he wrote to ARF.
Most labs have more experience in using ELISA-based assays, and they are simpler to use, Fagan said. The newer xMAP platform is quicker, however, and requires less CSF, since all three biomarkers can be measured in one well. Nonetheless, comparing results among the tests remains problematic, due to the large difference in absolute values and the fact that the numbers are not easy to interconvert. For example, Aβ42 in one CSF sample measured about 500 pg/mL by INNOTEST, but around 275 pg/mL by INNO-BIA, while another sample read 750 pg/mL by INNOTEST and around 325 pg/mL by INNO-BIA. Although it might be possible to find an equation to convert among them, it likely will not be that simple, Fagan said. Eventually, vendors will need an international reference standard to allow them to bridge results among technology platforms, Vanderstichele suggested.
In the short term, the answer may be to use assay-specific reference and cutoff values, said Henrik Zetterberg at Sahlgrenska University Hospital. Zetterberg leads, along with Blennow, the international Quality Control program. Scientists interviewed for this story agreed that in the long term, the field will probably adopt one uniform test, although which one remains up in the air. One problem with commercial kits, Zetterberg said, is that reagents vary from batch to batch. Because of this, labs that want to ensure longitudinal stability in the values they report need to maintain many aliquots of internal control CSF samples with clinically relevant tau and amyloid concentrations, and test each incoming kit to make sure it gives the expected values. “I think the winner on this market will be the kit vendor that produces a high-quality kit with very good longitudinal stability,” Zetterberg said.
The second paper looks at imaging biomarkers. In the imaging field, “ADNI has done a great job of standardizing acquisition for volumetric MRI,” Sperling wrote to ARF. The continuation program, ADNI2, is gathering data to help standardize resting functional MRI, said Clifford Jack at the Mayo Clinic in Rochester, Minnesota. Jack heads the ADNI MRI core. Resting fMRI is the most widely used form of fMRI in AD research, because it is simple to implement, Jack said. On the other hand, task-related fMRI, in which people must perform a cognitive task while in the machine, is more complicated and requires special equipment. Cognitive neuroscientists have made the greatest use of task fMRI, usually on young, healthy volunteers such as psychology graduate students. In the AD field, most studies of task fMRI have been cross-sectional, with little data available on how this measure performs over time in a cognitively impaired population.
Nonetheless, many scientists are interested in using task fMRI in early-phase clinical trials, said first author Alireza Atri. The question has been, Does task fMRI have sufficient sensitivity and reliability to detect changes in early-phase AD? To answer this, Atri and colleagues used data from the placebo arm of a small drug trial. The 12-week trial consisted of 24 patients with mild AD, all of whom were taking donepezil. The 12 patients in the treatment arm received memantine as well. Participants underwent task fMRI at baseline, six weeks, and 12 weeks. In the task, participants saw paired names and faces. They had to assess whether the face and name went well together, and then try to remember the combination. Atri noted that this is a challenging job that activates multiple brain areas. The brain usually processes names in the left hemisphere, faces in the right, and integrates the information with the medial temporal lobe and hippocampus, he said.
The scientists compared the scans of the 12 people in the placebo arm and found that overall brain activation stayed stable and repeatable over the 12 weeks of the study. When they homed in on the hippocampus, the researchers saw a slight downward trend in activation, particularly in the right hippocampus, which may have been more challenged by this primarily visual task. Intriguingly, participants who had worse clinical dementia ratings at baseline showed a stronger downward trend in the right hippocampus over 12 weeks. The good reproducibility over time in this placebo trial implies that in a drug trial, researchers would be able to pick up small changes in brain activation due to a drug effect. With power analysis, Atri calculated that for trials, task fMRI could potentially detect a 25 percent change in hippocampal activity due to a treatment effect using cohorts as small as 50 people, and a 50 percent change with only 15 people.
The results demonstrate that “it’s feasible to repeatedly perform task-related fMRI in AD patients and get high-quality, usable, consistent data. We did four of them in each patient over a 24-week span—96 in total,” Atri said. Task fMRI would be most appropriate for small, early, proof-of-concept clinical studies that seek a signal of effect, Atri said, though he stressed fMRI would be an exploratory biomarker in such a case and should be combined with other measurements. He noted that it remains to be seen if this technique would work for hundreds of patients across multiple centers. Jack agreed that the technique shows promise, noting “[The researchers] demonstrated good reproducibility, which is the key element for a biomarker for a clinical trial.” Sperling wrote to ARF, “I personally do not think that task fMRI is likely to be that useful in clinical diagnosis or in large-scale pivotal studies, but I do hope it may prove useful in finding an early signal of efficacy that will predict long-term response.” Atri pointed out that, because task fMRI places demands on areas of the brain most vulnerable to AD, it may provide a more sensitive measure of dysfunction than resting fMRI does, potentially providing more information from fewer people. Atri said the next step is to assess how and where in the brain the fMRI signature correlates over time with changes in neuropsychological measures. He is also looking at the treatment arm of the study to see if the addition of memantine to stable donepezil therapy leads to any change in task fMRI signals.—Madolyn Bowman Rogers
- Worldwide Quality Control Set to Tame Biomarker Variation
- Brain Imaging Speaks Volumes about AD and the Aβ Sink
- Biomarker Roundup: Collecting Clues from MRIs to RNAs
- Olsson A, Vanderstichele H, Andreasen N, De Meyer G, Wallin A, Holmberg B, Rosengren L, Vanmechelen E, Blennow K. Simultaneous measurement of beta-amyloid(1-42), total tau, and phosphorylated tau (Thr181) in cerebrospinal fluid by the xMAP technology. Clin Chem. 2005 Feb;51(2):336-45. Epub 2004 Nov 24 PubMed.
- Fagan AM, Mintun MA, Mach RH, Lee SY, Dence CS, Shah AR, Larossa GN, Spinner ML, Klunk WE, Mathis CA, Dekosky ST, Morris JC, Holtzman DM. Inverse relation between in vivo amyloid imaging load and cerebrospinal fluid Abeta42 in humans. Ann Neurol. 2006 Mar;59(3):512-9. PubMed.
- Fagan AM, Roe CM, Xiong C, Mintun MA, Morris JC, Holtzman DM. Cerebrospinal fluid tau/beta-amyloid(42) ratio as a prediction of cognitive decline in nondemented older adults. Arch Neurol. 2007 Mar;64(3):343-9. Epub 2007 Jan 8 PubMed.
- Fagan AM, Shaw LM, Xiong C, Vanderstichele H, Mintun MA, Trojanowski JQ, Coart E, Morris JC, Holtzman DM. Comparison of analytical platforms for cerebrospinal fluid measures of β-amyloid 1-42, total tau, and p-tau181 for identifying Alzheimer disease amyloid plaque pathology. Arch Neurol. 2011 Sep;68(9):1137-44. PubMed.
- Atri A, O'Brien JL, Sreenivasan A, Rastegar S, Salisbury S, Deluca AN, O'Keefe KM, Laviolette PS, Rentz DM, Locascio JJ, Sperling RA. Test-retest reliability of memory task functional magnetic resonance imaging in Alzheimer disease clinical trials. Arch Neurol. 2011 May;68(5):599-606. PubMed.
To make an annotation you must Login or Register.
University of California, San Diego
This study is timely in comparing two widely used platforms (ELISA and bead-based multiplex assays [X-MAP]) for measuring cerebrospinal fluid (CSF) analytes related to Alzheimer’s disease. The results support assays for CSF Aβ42 and total tau as being reliable across platforms, although at present, precise cutoff levels need to be established by users. Measurement of CSF levels of phospho-tau181 showed greater variability between the platforms, for unclear reasons. A particular strength of the study was amyloid imaging on all participants, allowing comparison of cutoffs of CSF biomarkers with amyloid positive or negative scans. The study does not discuss why each platform reads out different concentrations of Aβ42, tau and p-tau181 in CSF, although levels of each of these analytes were correlated using different assays; the strength of correlation does not allow for a facile conversion of levels between platforms.
A different standardization problem—how to obtain consistent readouts (and low variation) within a laboratory when assays are run intermittently on different days (rather than in large batches)—is the subject of ongoing research and harmonization efforts spearheaded by Kaj Blennow and others.View all comments by Douglas Galasko
Make a Comment
To make a comment you must login or register.