Blood Biomarkers Plus Gene Status May Predict Cognitive Decline
by Richard Robinson for Neurology Today:
Using a machine-learning approach, investigators identified four plasma proteins that correlate with cerebrospinal fluid amyloid-beta-1-42 status; these proteins plus apolipoprotein E4 genotype predicted amyloid positivity.
Levels of four blood proteins, when combined with apolipoprotein E4 (APOE4) gene status, can predict the level of amyloid-beta (Abeta) peptide in the cerebrospinal fluid (CSF) and the likelihood of conversion from mild cognitive impairment (MCI) to Alzheimer's disease, according to a new study published in the March 11 issue of Scientific Reports.
The finding may provide an inexpensive and less invasive method to screen large numbers of people for those most at risk of cognitive decline, enhancing recruitment for clinical trials seeking to treat Alzheimer's disease at the earliest possible stage.
A growing body of work has shown that a decline in CSF Abeta is a sign that it is accumulating in the brain, according to lead author Noel Faux, PhD, of IBM Australia in Carlton. Recent research has shown that that decline may occur up to a decade before PET imaging can detect amyloid in the brain, making it a valuable biomarker of impending brain pathology.
But lumbar puncture is invasive and not ideal for large-scale screening, which led Dr. Faux and colleagues to search for some combination of blood-borne surrogate markers that were correlated with lower levels of CSF Abeta.
“Blood is the information highway of the body,” Dr. Faux said. “If things go wrong, you are likely to get a signal in the blood.” He noted that brain amyloid PET imaging has previously been shown to correlate with both blood and CSF biomarkers, strengthening the case that blood is likely to carry some trace of Alzheimer's-related changes in the CSF. The new study was conducted as part of the Alzheimer's Disease Neuroimaging Initiative (ADNI), a long-term search for biomarkers in Alzheimer's disease.
Study Methods, Findings
To determine which of the hundreds of substances in blood might be relevant to CSF Abeta, Dr. Faux examined 566 individuals in the ADNI cohort whose ages and APOE4 status were known, and for whom blood samples were available, and divided the group into training and validation cohorts. The training set included 58 individuals who were cognitively normal, 198 with mild cognitive impairment, and 102 with Alzheimer's disease. Most individuals were in their 70s, and each had had a lumbar puncture to determine their level of CSF Abeta.
The team used the levels of 149 proteins and 138 metabolites as inputs into a machine learning algorithm known as a “random forest.” In this approach, thousands of decision trees are constructed, each using different subsets of individuals and different subsets of available measures (that is, levels of some but not all of the proteins and metabolites). To generate a new prediction, every decision tree in the random forest is evaluated and a majority vote is taken across all trees to determine whether that individual is predicted to have normal or abnormal Abeta levels. The prediction is then compared to the individual's actual Abeta level.
“The random forest method has been used for almost 20 years to build predictive models from many types of data, including noisy biological signals,” Dr. Faux said. “By taking all the information you have, constructing many simple classifiers and taking a majority vote, you can generate robust predictions, in this case predicting those who have low CSF Abeta, without needing to make many assumptions about the nature of the data.”
The results of the analysis indicated that, when combined with age and APOE4 status, levels of four proteins—chromogranin-A, eotaxin-3, apolipoprotein E, and the Abeta 1-42 peptide—significantly improved the prediction of low CSF Abeta over the baseline model of age and positive APOE4 status alone. Sensitivity was increased from 0.751 for the baseline model to 0.807 for the best model incorporating the four blood proteins.
“Rather than screening thousands of individuals with a lumbar puncture, we envision using a blood test, combined with a gene test, to find those who are positive for a decline in Abeta,” who could then be definitively tested with either lumbar puncture or PET scan, said Dr. Noel Faux.
Dr. Faux then applied the model to the test cohort, 198 individuals with mild cognitive impairment whose Abeta CSF levels were unknown. He used the blood analytes plus age and gene status to predict the CSF Abeta level, and therefore the likelihood of conversion from MCI to Alzheimer's disease over 120 months.
He found that those individuals whose CSF Abeta was predicted to be low converted to Alzheimer's disease significantly faster than those with a predicted higher level of Abeta in the CSF. Further, he found that predicted CSF Abeta correlated well with brain amyloid as measured by PET imaging in those individuals for whom imaging was available.
“This method will still need to be refined, and is not yet robust enough to make lumbar puncture unnecessary,” Dr. Faux said, “but for clinical trials, we need to screen a large number of people to identify those who are most suited for enrollment,” based on their amyloid status. “Rather than screening thousands of individuals with a lumbar puncture, we envision using a blood test, combined with a gene test, to find those who are positive for a decline in Abeta,” who could then be definitively tested with either lumbar puncture or PET scan.
Regarding the need for an APOE gene test, he said, “we don't see that as a limitation, since if you are doing a blood draw, including a single-gene test is relatively cheap.”
Beyond the implications for trial enrollment, Dr. Faux added, this study, in finding further objective correlates of clinical disease, “should help advance the movement toward defining Alzheimer's disease in terms of biomarkers.”
Expert Commentary
“The ability to replace expensive and invasive PET or CSF amyloid evaluation with a low-cost blood test would be a major development for the field,” said Samantha Burnham, PhD, senior research scientist at Commonwealth Scientific and Industrial Research Organisation in Parkville, Victoria, who studies Alzheimer's disease biomarkers.
“A blood test for amyloid would most likely be used to pre-screen individuals for confirmatory PET or CSF analysis where amyloid positivity is a criterion for the trial. This would allow a much wider screening protocol for amyloid to be undertaken, in turn reducing the cost of recruitment for clinical trials. In the advent of a disease-modifying therapy, this would also be an ideal population level screening medium.”
Commenting on the study, Gil Rabinovici, MD, professor of neurology at University of California, San Francisco, said: “Identifying blood-based biomarkers is incredibly important, since clinical diagnosis is insufficient to predict amyloid status. There has been a lot of progress in the past few years,” including with mass spectrometry approaches, “but the technology has not been clearly scalable,” since the assays require advanced laboratory technology, and results are very dependent on how the sample is handled immediately after the blood draw. The approach in the current study appears to be more robust, he said.
“The ability to replace expensive and invasive PET or CSF amyloid evaluation with a low-cost blood test would be a major development for the field,” said Dr. Samantha Burnham.
“Of special importance is the high sensitivity the authors achieved,” he said. “That is really important, because it allows one to rule out those who don't have amyloid, and then in those who test positive, to confirm it with a more specific test.”
One limitation of the study, he noted, is that it did not address the predictive ability of the biomarker panel in APOE4 non-carriers. “It would be reassuring to know that it was predictive in those individuals as well, and that the effect was not mainly driven by APOE status,” Dr. Rabinovici said. “And it would be useful to see how well this model does in predicting brain amyloidosis in cognitively normal individuals,” who may turn out to be the most treatment-sensitive group for therapies designed to prevent amyloid build-up.”