16 Jan 2015

A computer model does a pretty good job of predicting where in the brain Alzheimer’s pathology will strike next, according to a paper in the January 20 Cell Reports. Ashish Raj and colleagues at Weill Cornell Medical College in New York, who published their model in 2012, now have compared how its performance stacks up against four years’ worth of longitudinal imaging data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). They found tight correlations between predicted and actual areas of atrophy and hypometabolism. However, without a longer time course, it is difficult to say for sure how accurate the model is, noted experts who spoke with Alzforum.

Glass Brains. Weill's "glass brains” depict where atrophy will occur. The bigger the marble-like sphere, the worse the pathology. Colors indicate frontal lobe (blue), parietal (purple), occipital (green), temporal (red), and subcortical (yellow). The brain on the left comes from a baseline MRI, the others are computer predictions for five (middle) and 10 (right) years later. [Image courtesy of Cell Reports, Raj, et al.]

Many Alzheimer’s researchers have come to suspect that plaques and tangles spread along the brain’s natural fiber tracts, tracing a standard route from the hippocampus to the temporal, parietal, and prefrontal cortices. The pathological proteins leave in their wake a trail of atrophy, visible on magnetic resonance images, and hypometabolism, seen with positron emission tomography. Raj based his model on the connectome of healthy people, and then assumed that the atrophy and hypometabolism would diffuse their next destination along the route of least resistance—that is, axon fiber tracts (see Alzforum live discussion).

The atrophy patterns the computer came up with looked much like the real MRIs of people with AD at a single time point (Raj et al., 2012). Now, Raj and colleagues have tested the model against longitudinal data from 99 people who had Alzheimer’s when they were first scanned and 151 who had mild cognitive impairment, now known as preclinical AD. They were all scanned multiple times over the course of four years.

The computer program divides the brain into 90 regions and, based on voxel intensity, assigns a level of atrophy or hypometabolism, depending on the scan performed. Raj and colleagues use a system developed at Cornell to visualize those data in what they call “glass brains,” because they represent pathology with glass marble-like spheres (see image above). The program uses the standard brain connectivity to predict where atrophy or hypometabolism will appear next. To determine the rate of progression, the authors input two different scans, taken at baseline and then again six to 12 months later.

Then, the authors compared their projections to the final scans performed two to four years after baseline. Most of the time, the true spread of atrophy and hypometabolism was close to that predicted by the model (see image below). This match supports the hypothesis on which the model was based, namely, that pathology travels along axon fibers, noted Konstantinos Arfanakis of the Illinois Institute of Technology in Chicago, who was not involved with the study. The model also predicted which people with MCI would progress to AD—those whose subcortical and temporal lobes were in the projected path of the pathology. These areas are most affected in AD, Raj said.

How good is the model? “It does look like it is going to turn out to be accurate,” commented Lary Walker of Emory University in Atlanta, who was not involved in the study. However, he added, two to four years represents a short time in the progression of dementia. Raj and colleagues made projections out to five and 10 years, but they may find it difficult to prove the accuracy of those predictions, pointed out Clifford Jack of the Mayo Clinic in Rochester, Minnesota. “Unfortunately, neither [Raj et al.] nor anyone else has the data to test this hypothesis,” Jack said. People with severe dementia typically drop out of studies as their condition worsens, he explained, so those five- to 10-year scans are hard to come by.

The model could be useful in the clinic and in trials, but not quite yet, said Raj. He plans to incorporate other information, such as blood and cerebrospinal fluid markers, and cognitive assessments, to fine-tune the model’s predictions. In the current study, the authors found that including fluid biomarkers did not make a difference to the predictions of where pathology would spread, probably because any change in biomarkers was already reflected in the baseline images. However, he wants to include any quantifiable clinical data to make the best possible predictions about a person’s risk for dementia and what brain functions might be affected in the near future. Walker commented, “That is entirely theoretical at this point, but these studies indicate that it is at least a possibility down the road.” Before the method can be used in the clinic, the researchers may want to verify their model’s performance in other cohorts, suggested Arfanakis. The ADNI cohort consists of people who meet strict clinical criteria, he noted, and said he would also like to see evidence of the model’s efficacy in elderly people “off the street.”

The computer predictions could also help scientists select the most appropriate cohorts for clinical trials. For example, researchers testing a drug that slows progression from MCI to AD might want to eliminate subjects whose scans indicate they will not progress to dementia. Scientists could also use the predictions to determine whether a medication has the desired effect, Raj said. They could use a person’s baseline scan to obtain their untreated trajectory, then compare that to the real progression during a trial, and thus determine if the treatment diminished the spread of atrophy or hypometabolism. Each patient would then act as his or her own control.

Raj’s network diffusion model may apply to other neurodegenerative diseases caused by pathological proteins traveling axon tracts. The study authors are already testing whether they can predict pathological changes in Parkinson’s, too.—Amber Dance

Comments

No Available Comments

Make a Comment

To make a comment you must login or register.

References

Webinar Citations

1. “Network Epicenters” in Healthy Connectome Predict Dementia 16 Apr 2012

Paper Citations

1. Raj A, Kuceyeski A, Weiner M. A network diffusion model of disease progression in dementia. Neuron. 2012 Mar 22;73(6):1204-15. PubMed.

Further Reading

Papers

1. Zhou J, Gennatas ED, Kramer JH, Miller BL, Seeley WW. Predicting regional neurodegeneration from the healthy brain functional connectome. Neuron. 2012 Mar 22;73(6):1216-27. PubMed.
2. Dopper EG, Rombouts SA, Jiskoot LC, den Heijer T, de Graaf JR, de Koning I, Hammerschlag AR, Seelaar H, Seeley WW, Veer IM, van Buchem MA, Rizzu P, van Swieten JC. Structural and functional brain connectivity in presymptomatic familial frontotemporal dementia. Neurology. 2014 Jul 8;83(2):e19-26. PubMed.
3. Frost B, Diamond MI. Prion-like mechanisms in neurodegenerative diseases. Nat Rev Neurosci. 2010 Mar;11(3):155-9. PubMed.
4. Jucker M, Walker LC. Pathogenic protein seeding in Alzheimer disease and other neurodegenerative disorders. Ann Neurol. 2011 Oct;70(4):532-40. PubMed.
5. Kane MD, Lipinski WJ, Callahan MJ, Bian F, Durham RA, Schwarz RD, Roher AE, Walker LC. Evidence for seeding of beta -amyloid by intracerebral infusion of Alzheimer brain extracts in beta -amyloid precursor protein-transgenic mice. J Neurosci. 2000 May 15;20(10):3606-11. PubMed.

News

1. Cortical Hubs Found Capped With Amyloid 13 Feb 2009
2. Functional Connectivity Predicts Aβ Deposition in Mice 29 Mar 2012
3. Insidious Spread of Aβ: More Support for Synaptic Transmission 5 Nov 2010
4. Mice Tell Tale of Tau Transmission, Alzheimer’s Progression 3 Feb 2012
5. The Epidemic in Your Head? New Model Casts Amyloid as Intra-Brain Contagion 20 Nov 2014
6. A High-Resolution Wiring Diagram of the Mammalian Brain 4 Apr 2014
7. Synthetic Synuclein Corrupts Native Along Mouse Brain Networks 26 Apr 2012