Because neurofibrillary tangles correlate with cognitive decline in Alzheimer’s disease, a drop in the tau PET signal could show a drug is working. In the November 27 Science Advances, researchers led by Michael Ewers at Ludwig-Maximilians University, Munich, suggest a way to improve the sensitivity of this signal. The scientists used the fact that tau tangles spread through connected brain regions. By identifying the initial epicenter of accumulation in each person, they could predict where the pathology would appear on follow-up scans. This allowed them to detect small local changes in the tau PET signal that might not have been obvious on a global scan. Notably, the method was more sensitive than quantifying tau spread by Braak staging. The researchers calculated that this individualized approach could lower the number of participants needed in a trial arm by as much as two-thirds. “This method acknowledges and addresses the heterogeneity of disease progression,” Ewers told Alzforum.

  • Tangle accumulation can start in distinct brain regions in different people.
  • This results in individual patterns of tangle spread that vary from Braak staging.
  • By taking this into account, researchers sharpened the sensitivity of tau PET.

Justin Sanchez at Massachusetts General Hospital, Boston, agreed the approach has potential. “That this method is fully automated is very encouraging and ought to make it broadly applicable across tau PET studies … this is exciting work and a big step forward in the direction of precision medicine,” he wrote to Alzforum (full comment below).

Ewers and colleagues previously reported that combining a baseline tau PET scan with a functional connectivity map of the brain could predict areas of future tangle accumulation in 94 AD patients (Jan 2020 news). This is needed because patterns of tangle spread are highly heterogenous. For example, researchers led by Oskar Hansson at Lund University, Sweden, recently reported four distinct patterns of tangle progression in AD patients, each with similar prevalence, suggesting there is no one typical trajectory (Vogel et al., 2020). This makes generalized models of spread poor predictors of progression in any given patient.

Predictable Path. Tangles spread from their initial epicenter in the brain to the most closely connected quartile of regions (Q1) first, then distally connected regions (Q4) last. [Courtesy of Franzmeier et al., Science Advances.]

In the new study, Ewers and colleagues set out to test the performance of their individualized model. First author Nicolai Franzmeier expanded on his previous dataset, analyzing 254 amyloid-positive people and 247 amyloid-negative controls from two longitudinal cohorts, ADNI and BioFINDER. The amyloid-positive participants ran the gamut from cognitively healthy to AD dementia. All had flortaucipir tau PET scans. In general, across the whole group, patterns of tangle accumulation reflected the expected Braak staging, with unimpaired participants at Braak stage 1/2 and those with dementia at advanced stages.

Connectivity Tells the Tale. Over time, tangles accumulate most in those brain regions (Q1) closest to the initial epicenter. [Courtesy of Franzmeier et al., Science Advances.]

However, at the individual level, there were discrepancies. In classic Braak staging, tangles start in inferior temporal regions. Many participants instead had tangle hotspots only in the medial parietal lobe, the lateral temporoparietal lobe, or the occipitotemporal lobe. Altogether, the authors identified nine distinct epicenters in these cohorts at baseline. These epicenters were fairly evenly represented across the population. For example, in the ADNI cohort of 213 amyloid-positive participants, each group contained an average of 20 people, with the most common tau epicenter occurring in 43 people, and the least common in 12.

Next, Franzmeier and colleagues developed a general connectivity map of the human brain using resting-state fMRI data from 1,000 cognitively healthy participants in the Human Connectome Project. To do this, they divided the neocortex into 200 regions and determined the connections between them. Using this map, they identified the 50 regions most closely connected to each tangle epicenter. The authors hypothesized that these regions, dubbed Q1 for first quartile, were the ones most likely to show future tangle accumulation. Indeed, in cross-sectional baseline data for each tau epicenter subtype, Q1 regions took up more tau tracer than did more distantly connected ones.

Individualized Measure Does Best. The connectivity-based Q1 regions better predicted future tangle accumulation than did regions based on Braak staging. [Courtesy of Franzmeier et al., Science Advances.]

What about over time? A subset of 57 ADNI and 33 BioFINDER tau-positive participants had a second scan occurring an average of 1.5 to two years after the first. As expected, in follow-up scans the tau signal rose the most in Q1 brain regions, and the least in the most distant Q4 regions (see image above). Importantly, Q1 regions outperformed Braak staging regions or a temporal-meta region of interest for detecting small changes (see image below). The authors calculated that their method would lower the number of trial participants needed to see a 20 percent decline in tangle accumulation by one-quarter to two-thirds of that needed when using Braak staging.Because it uses general connectivity maps, this method does not require researchers to assess connectivity in individual patients, and so any PET center could adopt it. Franzmeier noted that brain connectivity is highly conserved, varying little between cohorts. Nonetheless, there could be some individual differences, and Alzheimer’s pathology itself is known to alter connectivity (Jul 2012 news; Aug 2013 news; Aug 2016 conference news). The authors are examining whether individualized fMRI connectivity maps would improve the predictive power of their approach.

Franzmeier believes their method will help individualize patient care. “This gives us a patient-tailored understanding of disease progression, rather than treating everyone as equal,” he told Alzforum.—Madolyn Bowman Rogers


  1. This is really excellent work, highlighting the heterogeneity in tau accumulation patterns. Especially interesting is the subject-specific modeling of tau spread, using each subject’s individual tau PET to identify the regions that were most likely to accumulate tau. This individualized sampling of target regions of interest (ROIs) increased sensitivity to detect tau changes and reduced the sample size needed in simulated trials. The fact that this method is fully automated is very encouraging and ought to make it broadly applicable across tau PET studies.

    For widespread use, some questions remain and details would need to be worked out. For example, how does each subject’s target ROI, being unique, affect interpretation of a significant result (or lack thereof)? How does this individualized sampling perform at different points along the AD continuum (i.e., in cognitively normal versus mild cognitive impairment; in those with little tau versus those with a lot)? How does tau positivity probability (TPP) compare to SUVr as an outcome measure? Since TPP is sample-dependent, which sample should be used to define this for a clinical trial?

    Overall, this is exciting work and a big step forward in the direction of precision medicine.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Connectivity, Not Proximity, Predicts Tau Spread
  2. Communication Breakdown: Multiple Networks Decline in AD Brains
  3. Brain Connectivity Reveals Preclinical Alzheimer’s Disease
  4. Homing in on Early Alzheimer’s Biomarkers: Does Connectivity Hold the Key?

Paper Citations

  1. . Characterizing the spatiotemporal variability of Alzheimer's disease pathology. medRxiv August 24, 2020.

Further Reading

Primary Papers

  1. . Patient-centered connectivity-based prediction of tau pathology spread in Alzheimer's disease. Sci Adv. 2020 Nov;6(48) Print 2020 Nov PubMed.