Unique Brain 'Fingerprint' May Predict Therapeutic Response

Batya Swift Yasgur, MA, LSW

July 19, 2018

A unique "brain fingerprint" may help identify the most beneficial therapeutic intervention for individual patients with neurologic disorders such as Alzheimer's disease (AD), potentially sparing millions from undergoing ineffective treatment, new research suggests.

Investigators used computational brain modeling and artificial intelligence techniques to analyze positron emission tomography (PET) and MRI from over 300 patients with AD and healthy controls.

They found that the personalized therapeutic intervention fingerprint (pTIF) predicted the effectiveness of targeting biological factors, such as brain amyloid/tau deposition, inflammation, and neuronal functional dysregulation, to control the patient's disease course.

Moreover, patients who shared a given pTIF subtype had similar gene expression, which supported the relevance of the pTIF framework in biomarker-driven assisted therapeutic interventions.

"Although needing further validation, the introduced concept is a promising tool for patient stratification," lead author Yasser Iturria-Medina, PhD, assistant professor, Department of Neurology and Neurosurgery, Faculty of Medicine, McGill University, and primary investigator, Ludmer Centre for Neuroinformatics & Mental Health, Montreal, Quebec, Canada, told Medscape Medical News.

"It has broad implications for both the future identification of effective individualized treatments and the selective enrollment of patients in clinical trials and could provide a more effective medical care assisted by individual sophisticated models, undesired secondary effects, and the substantial reduction of pharmaceutical/clinical costs," he said.

The study was published online June 14 in NeuroImage.

Pinpointing Optimal Interventions

Unlike "generalized medicine," personalized medicine (PM) is "based on the optimization of treatment plans for individual patients through consideration of particular characteristics," such as molecular, macroscopic, and medical information, the authors write.

Previous studies of PM have been limited by misuse of association analyses, incorrect extrapolation of group-based statistical inferences for identifying individual disease biomarkers, and the "paradoxical confidence in broad clinical/cognitive categories for validating new patient subtypes," they add.

Recent advances in brain modeling using network-based approaches have focused on the spreading of normal and pathological functional signals or local interactions among different biological factors, and dynamic network modeling has contributed to understanding dissimilar brain mechanisms.

Additional work has used control theory to predict functional and cognitive response of the brain under the influence of external experimental interventions.

Incorporating "multiple relevant biological factors, rather than only functional neuronal signals" has led to proposing an "integrative multifactorial causal model of brain organization and control," the researchers note.

This approach "allows accurate characterization of the intra-brain factor-factor causal interactions, the spreading of multifactorial pathological signals through different brain networks…and assessment of the effectiveness of either single-target or combinatorial therapeutic interventions."

The pTIF is based on the Multifactorial Causal Model (MCM) of Brain (dys)Organization, Iturria-Medina explained.

The MCM framework and control theory and constitute "a set of multivariate metrics constructed according to the needed energy required to either stop the patient's pathologic progression or revert its condition to a healthy state," the authors explain.

The pTIF is "inferred from individual multimodal longitudinal imaging data…characterizing each patient's multifactorial causal interactions and dynamic brain changes in response to potential external (therapeutic) inputs."

Patterns of pTIF, when applied to aging or patients with neurodegenerative disorders, "significantly predict the individual variability in plasma gene expression (GE) profiles and represent a significantly more accurate GE predictor than the traditional clinical/cognitive categories."

"We were motivated by the fact that there are millions of patients under therapeutic treatment that will not benefit them," said Iturria-Medina

"I believe that it is possible to identify the most beneficial intervention for each patient, in correspondence with the tenders of PM," he added.

Predicting Individual Therapeutic Needs

To investigate the effectiveness of pTIF, the researchers used data from a total of 1006 patients with multimodal brain imaging (n = 944) and/or blood GE expression data (n = 744), taken from the Alzheimer's Disease Neuroimaging Initiative (ADNI).  

The pTIFs were estimated and analyzed for 331 participants, with at least four different imaging modalities and four longitudinal data acquisitions.

Of those, a subset of participants (n = 256) presented GE from blood samples that were used in the differential GE analysis.

The researchers also used GE data from a reference group consisting of 74 additional patients who did not have symptoms of cognitive/clinical deterioration.

To assess cognition, participants received the Mini-Mental State Examination and the Alzheimer's Disease Assessment Scale-Cognitive Subscales 11 and 13.

The blood RNA of participants was analyzed, and PET/MRI images were used to quantify and map seven different biological properties.

Participants were longitudinally scanned by using six different neuroimaging modalities: structural MRI, fluorodeoxyglucose PET, florbetapir PET, arterial spin labeling, resting functional MRI, 18F-AV-1451 PET, and diffusion-weighted MRI.

The researchers also estimated multimodal connectivity, including vascular and anatomic networks.

The pTIF was defined as "the set of cost-energy/deformation values obtained for each patient, which are inversely proportional to the effectiveness of each intervention."

The researchers evaluated cognitive/clinical properties, GE (from blood samples) and/or six different biological factors in the brain (intra-brain amyloid β (Aβ) proteins, tau proteins, glucose metabolism, cerebral blood flow, resting-state functional activity, and/or structural tissue patterns) in healthy and diseased participants (n = 331).

They found that the pTIF "vastly outperforms cognitive and clinical evaluations on predicting individual variability in…GE profiles."

The researchers compared three patients by using factor-specific cost-energy deformation levels, considering only single-target interventions.

The factor-specific deformations reflected diverse individual therapeutic requirements, dependent on each patient's level of multifactorial biological alterations and intra-brain causal and spreading mechanisms.

For example, one patient might benefit more from a vascular or metabolic-based intervention, while the second might benefit more from an agent targeting intra-brain Ab levels, and the third might need combination rather than single-factor interventions.

After regrouping the patients according to their predicted primary single-target interventions, the researchers found that these pTIF-based subgroups "present distinctively altered molecular pathway signatures, supporting the across-population identification of dissimilar pathological stages, in active correspondence with different therapeutic needs."

Approximately 30% of all the patients were calculated to gain primary benefit from promoting increased grey matter density (ie, increasing neurogenesis), followed by glucose metabolism-, anti-Ab-, anti-tau -, functional-, and vascular- focused treatments (22%, 22%, 11%, 9%, and 6%, respectively).

Across all intervention-specific subgroups, the researchers found 1101 significantly differentially expressed genes associated with 103 functional pathways, with each subgroup characterized by a distinctive set of altered genes and functional pathways.

"We found that [by] using multimodal imaging and computation models, it is possible to predict each patient's therapeutic needs," Iturria-Medina commented.

"The predictions are in correspondence with the individual molecular properties, which validate our findings and the computational techniques we used," he continued.

Preparing for the Future

Commenting on the report for Medscape Medical News, Keith Fargo, PhD, director of scientific programs & outreach at the Alzheimer's Association, who was not involved with the research, called the pTIF an "interesting concept that enables us to look at multiple domains in the brain with imaging."

However, he cautioned, it's not yet clear "how this will translate into informing what kinds of medications or therapeutics we should use for a particular patient, since our current medications are symptomatic and don't actually slow, halt, or even reverse the brain pathologies we see."

Current approaches, including healthy eating, exercise, and good blood pressure and cholesterol control, may be beneficial for reducing risk for cognitive decline and even dementia as people age, but "we don't yet have major clinical trials showing the effectiveness of those things in people with brain disease — they might work with different kinds of patients, but large clinical trials are still underway."

The pTIF model "could be of great benefit, so that when these future treatments come on board, we'll know which is best for which particular patient rather than for every person who has dementia," he suggested.

"It's good that they are doing this research now," he concluded.

Iturria-Medina added that the study's results "highlight the imprecision of the traditional clinical evaluations and categories for understanding the individual therapeutic needs, evidencing the positive impact of using multimodal data and data-driven techniques in the clinic, in addition to the medical doctor's criteria and evaluations."

The research was funded by the Government of Canada's Banting postdoctoral fellowship and Brain Canada through the Canada Brain Research Fund with the financial support of Health Canada. Data collection was funded by the ADNI and its associated institutions. Iturria-Medina and coauthors and Fargo have disclosed no relevant financial relationships.

Neuroimage. Published online June 14, 2018. Abstract

For more Medscape Neurology news, join us on Facebook and Twitter


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.