The idea that tau pathology creeps through the brain via neural circuitry may have gained its strongest support yet from a study published January 17 in Nature Communications. Drawing on longitudinal tau PET imaging data from ADNI and BIOFINDER cohorts, researchers led by Michael Ewers of Ludwig-Maximilians-University in Munich reported that in people with amyloid plaque deposition, tau tangles crop up sequentially in functionally connected regions, as opposed to regions that are merely nearby. A combination of baseline tau-PET and functional connectivity predicted where, and to what extent, tau would accumulate next. The data support the idea that tau pathology spreads via neuronal circuitry, as opposed to simply drifting into neighboring areas.

  • Tau tangles accumulate at similar rates in connected regions of the brain.
  • Spatial proximity alone does not predict pattern of tau spread.
  • Baseline tau and connectivity together do.

That tau tangles overtake the brain in a stereotypical pattern in AD was shown by neuropathology nearly 30 years ago (Braak and Braak, 1991). The advent of tau tracers has allowed scientists to track tangles in living people, and shown that Aβ deposition instigates the spread of neurofibrillary tangles from medial temporal lobe out into the neocortex (Feb 2018 news; May 2019 newsJun 2019 news). Cross-sectional studies, including a recent one from Ewers and colleagues, have found that the timing and extent of tau accumulation appears highly correlated among regions that are functionally connected, i.e., that fire in sync with each other (Hoenig et al., 2018; Franzmeier et al., 2019; Vogel et al., 2019). 

However, cross-sectional data cannot completely decipher longitudinal processes. First author Nicolai Franzmeier and colleagues analyzed serial data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and Swedish BIOFINDER cohorts. ADNI participants included 28 people without evidence of Aβ deposition, and 53 people with amyloid who were either cognitively normal or had mild cognitive impairment. The BIOFINDER set included 16 people without amyloid and 41 with amyloid, who spanned the AD spectrum from unimpaired to dementia. For both cohorts, participants had a baseline tau-PET scan and at least one additional scan about a year later.

Divided Brain. Researchers defined 400 brain regions. For each one, they measured its tau accumulation and functional connectivity to every other region. Regions that form part of larger networks are color coded. [Courtesy of Franzmeier et al., Nature Communications, 2020.]

The researchers first measured by how much tangles accumulated in each of 400 brain regions. As seen in previous studies, people without amyloid deposition had little to no evidence of tau tangles at baseline or follow-up, while those with amyloid not only had more tangles at baseline, but also had still more tau pathology at follow-up. Those with MCI or AD had more deposition than those who were cognitively normal.

How did patterns of tau accumulation in people with amyloid relate to functional connectivity? ADNI participants had resting-state functional MRI scans around baseline, which the researchers used to chart connectivity between 400 brain regions. BIOFINDER did not use fMRI, so the researchers used fMRI data compiled from 500 different people in the Human Connectome Project to create a connectivity template. Across 400 brain regions in both cohorts, the researchers found a strong positive association between functional connectivity and increase in tangles. In other words, the more closely two regions fired together, the more likely they were to have accumulated similar amounts of tau between baseline and follow-up.

Aβ Sparks Tau. In ADNI (left) and BIOFINDER (right) cohorts, tau PET shows tangle growth between baseline and follow-up (colored areas) only in people who have amyloid deposition. [Courtesy of Franzmeier et al., Nature Communications, 2020]

This link between functional connectivity and tangle burden held true regardless of the absolute amount of tau in a given region, though the link was strongest where tau-PET signals were highest. The association was consistent across all seven canonical brain networks, for example the default mode network, dorsal attention network, and frontoparietal control network. It held up even when the scientists controlled for how close together two regions were to each other, suggesting that synaptic connections, rather than proximity in the same neighborhood, dictated tau accumulation.

Franzmeier and colleagues next asked to what extent functional connectivity and/or spatial proximity predict patterns of tau spreading from one area to the next. Weaving baseline tau and functional connectivity into computational models, they were able to foretell changes in tau-PET seen one year later. Spatial proximity was not a strong predictor of tangles, but it did improve models of tau spread based on functional connectivity. Essentially, the findings suggested that tau spreads fastest between functionally connected regions that are also close to each other.

“No matter where tau pathology occurs, the functional connectivity of that region to another one seems to be the best predictor of where tau occurs next,” said Ewers.

The data are consistent with the long-held hypothesis that hyperphosphorylated, misfolded tau spreads between neurons via synaptic connections, corrupting healthy tau along the way (Jun 2009 newsJun 2016 news). However, both Franzmeier and Ewers emphasized that the data are also consistent with the idea that interconnected regions become sequentially vulnerable to tau pathology.

David Jones of the Mayo Clinic in Rochester, Minnesota, agreed. “This and related studies are unable to differentiate purely reductionist mechanisms, such as transneuronal spreading, from complex systems models that include cascading network failures,” he wrote. “These two model classes would produce similar patterns in neuroimaging data, but would have vastly different mechanism and therapeutic implications across the disease spectrum.”

Tau propagation is likely a multifactorial process that depends on both transneuronal propagation and regional vulnerability, said Sylvia Villeneuve of McGill University in Montreal. Future studies that consider gene-expression differences between brain regions might illuminate the role of regional vulnerability, she said. For example, one study reported that regions prone to neurodegenerative disease tend to express genes that promote protein aggregation, while another implicated a combination of functional connectivity and gene expression in the spread of α-synuclein pathology in PD (Aug 2016 news; Zheng et al., 2019). 

Villeneuve added that incorporating structural measures of connectivity—i.e., diffusion tensor imaging that traces white-matter connections—could also clarify what drives tau’s travel patterns. Ewers and Franzmeier agreed, noting that the two types of connectivity measures complement each other. While DTI exposes direct physical connections, resting-state functional MRI reveals both direct and indirect connections between regions. For example, two regions that fire together might be connected via one or two intermediate regions. Whether tau tangles accumulate more synchronously between directly versus indirectly connected regions could be explored by incorporating both imaging measures, they said.

Alexander Drzezga of the University Hospital of Cologne in Germany wondered how the known erosion of neural networks in AD might factor into the findings. Patterns of functional connectivity are known to change with disease progression. In particular, as tau accumulates, circuitry tends to crumble. “It is unclear whether the pattern of propagation in patients with affected network function is still determined by patterns of ‘previous (healthy)’ or by ‘current (diseased)’ pathways of high connectivity,” Drzezga wrote to Alzforum. In short, the causal relationship between network dysfunction and tau tangles needs more investigation (see full comment below).

The researchers believe longitudinal tau PET imaging is suitable to gauge treatment effects in clinical trials for therapies targeting tau. Using a functional connectivity template similar to the one devised for the BIOFINDER cohort, along with a patient’s baseline tau-PET scan, clinicians could predict where tau is most likely to accumulate next. In a follow-up tau-PET scan, they could assess how well a given treatment stopped tau in its tracks.—Jessica Shugart


  1. This work by Franzmeier and colleagues is an elegant study that further illuminates the relationship between tau deposition and the functional network architecture of the brain. A particularly interesting outcome of the work lies in the discovery of a longitudinal relationship between the increase in tau deposition and the functional connectivity of certain brain regions. More strongly connected brain regions showed a stronger covariance in the longitudinal tau deposition. In addition, the authors demonstrate the intriguing possibility to predict future tau deposition of a brain region based on its connectivity strength to other tau-seeding brain regions, their respective tau load, and the approximate length of the connections.

    The longitudinal aspect of this paper is of great relevance, because the possibility of predicting the course of the disease could play a major role, e.g., in the interpretation of follow-up data collected in therapy trials. The paper provides further convincing arguments for the relationship between functional connectivity and tau deposition, i.e., the network degeneration hypothesis. Of particular interest is the finding that regions with low-tau accumulation rates showed high connectivity to regions with similarly low-tau accumulation rates. This implies that connectivity strength per se is not the driver but rather the pathway of tau propagation.

    These results are in good agreement with the hypothesis of "neuronal spreading" of tau deposits depending on synaptic activity. It should not be forgotten, however, that a direct proof of this hypothesis cannot yet be deduced from these and other similar data. Other theories of the causal neuropathology of Alzheimer's disease are in principle also consistent with these results. Regions of strong functional connectivity may also be similar in their molecular/cell-biological nature, e.g., with regard to genetic background (as mentioned by the authors themselves), but also as far as other factors such as metabolic, vascular/perfusion-related conditions, etc., are concerned. The activity-induced tau release described by the authors themselves could lead to increased tau formation (or "missorting") at both ends of a functional connection in regions of high neuronal activity, without necessarily requiring a "spreading" of the tau pathology. Lifelong increased neuronal or metabolic activity within certain highly active networks could lead to a "wear and tear" effect, which could occur in the affected regions simultaneously, in correlation with their connectivity. It is also conceivable that a so-far-unidentified noxious factor agent that is causally upstream of the tau accumulation could spread along the affected networks, entailing tau aggregation. In this context, inflammatory effects are also to be considered. It would also be important to further investigate the connection with structural connectivity already discussed by the authors

    Further factors requiring additional research would be time-course and direction of tau distribution across the brain. Of note, the authors demonstrate a link between connectivity and tau accumulation in parallel, i.e., simultaneously at “both ends” of connected brain regions. However, in case of a neuronal spread, we would expect a certain delay between the starting point and destination of the spreading pathology. This is, of course, methodologically difficult to ascertain. Also, the analysis of functional connectivity does not allow a direct statement about the directionality of the connection. That is, in the models examined by the authors, it would have to be assumed that the tau pathology spreads in both directions, i.e. upstream from soma to dendrites or downstream from soma along the axon. Further molecular biological research would be important to address this question.

    As far as can be discerned, the authors seem to have used different fMRI connectivity data sets for the two populations studied, one (BioFINDER) from an independent healthy control collective (healthy connectivity) and one (ADNI3) from the patients themselves (possibly already affected connectivity). This may explain some of the differences between the results in the two groups and raises some questions: It is assumed that functional connectivity itself is impaired already early in the course of Alzheimer's disease (possibly by the protein pathology that is deposited). This means that the pattern of functional connectivity changes with disease progression, and thus, eventually also the spread of pathology. It seems to be legitimate to use connectivity data from healthy controls in order to obtain information on whether the pathology spreads along the "usual" connectivity pathways, i.e., those typical for healthy controls. This seems to be the case, but it does not allow direct conclusions on actual distribution pathways in patients. It is unclear whether the pattern of propagation in patients with affected network function is still determined by patterns of "previous (healthy)" or "current (diseased)" pathways of high connectivity.

    It seems paradoxical, at least at first glance, that high tau accumulation should still correlate positively with high connectivity in advanced disease stages. On the contrary, it is assumed that tau contributes to a breakdown of the functional connection (e.g., through reduced microtubular function or synaptic dysfunction). Thus, if high tau accumulation in a network contributes to network breakdown, then tau should eventually be high where network connectivity is low (i.e., correlation of high tau burden with low connectivity would be expected). This also has been demonstrated topographically, e.g., by the high tau deposition in the default-mode network and the known loss of connectivity of this network.

    In principle, it is also conceivable that the tau deposition could be a reaction to a disturbance of functional connectivity (e.g., as an attempt to improve or to reorient the connection of the affected neuron). In this case, tau would form in "previously" highly interconnected brain regions, as soon as connectivity between those regions declined. From this point of view, the attempt to block the spread of tau by targeted therapeutic reduction of neural activity or connectivity could be counterproductive and harmful.

    Finally, the question of the interplay between tau and amyloid remains open. The authors report, in line with previous work, that tau accumulation was accelerated in amyloid-positive individuals in their study. This seems plausible, but the pathophysiological interaction with tau is still unclear, especially with regard to network function. It is often discussed that amyloid, especially in soluble form, may exert a synaptotoxic effect. If this is the case, the result would be reduced functional connectivity, which would then contribute to a reduced rather than increased spread of the tau pathology. This illustrates that research will still have to focus on coherently merging different disease hypotheses.

    In summary, the current study provides compelling evidence for an interrelation between functional connectivity and longitudinal tau accumulation in the brain, supporting the network degeneration hypothesis. The study nicely adds to current disease hypotheses and also stimulates the need for additional research in this interesting field. Particularly the question about the relationship between tau deposition and functional connectivity before and after disease onset (with resulting changes in network function) requires further attention.

  2. This paper carries forward an emerging theme in the literature, which they reference, that Alzheimer’s disease is a network disease of the brain. By that I mean that the anatomy, or the space the disease occupies, conforms to what has been dubbed the brain’s cortical-hippocampus memory system. This can be viewed also as the brain’s default mode network, with which the hippocampus has an intimate relationship. Of interest here is that such networks have been identified largely by functional brain imaging, first with PET and now routinely with fMRI. They can be seen in resting-state fMRI data as very specific regional correlations in the fMRI signal (a very surprising and important discovery). Along with the default mode network, others go by names such as dorsal and ventral attention, salience, control, and the more commonly known ones such as motor and visual.

    The idea that AD would reside in one of these networks is certainly of interest and that it is a network concerned with memory makes sense. But what is emerging is that AD not only resides within a network but, as the disease progresses, it “moves” from one part of the network, in this case from the hippocampus, to the back end of the default mode network where AD pathology (i.e., plaques, then tangles) begins in the back of the brain and moves forward. What this paper addresses is how that might occur with regard to tau accumulation. The authors are suggesting that within a functional network tau is passed from one part of the network to another. To put it simply, relationships matter in the progression of the disease. This is an interesting but not an entirely new idea. It is nice to see it so well-articulated here but exactly how this is occurring remains to be fully understood.

  3. This work elegantly addresses an important question: Does tau spread through synaptically connected brain regions as Alzheimer's disease progresses? By using detailed functional connectivity maps and, importantly, two independent cohorts of individuals, the authors provide compelling evidence that neuronal connections, rather than anatomical proximity, are key for propagation of tau pathology.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Imaging Clinches Causal Connections between Aβ, Tau, Circuitry, and Cognition
  2. Longitudinal Tau-PET Links Aβ to Subsequent Rise in Cortical Tau
  3. Serial PET Nails It: Preclinical AD Means Amyloid, Tau, then Cognitive Decline
  4. Traveling Tau—A New Paradigm for Tau- and Other Proteinopathies?
  5. Excited Neurons Release More Aberrant Tau
  6. Aggregation-Prone Gene Expression Signature Mapped in Brain

Paper Citations

  1. . Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991;82(4):239-59. PubMed.
  2. . Networks of tau distribution in Alzheimer's disease. Brain. 2018 Feb 1;141(2):568-581. PubMed.
  3. . Functional connectivity associated with tau levels in ageing, Alzheimer's, and small vessel disease. Brain. 2019 Apr 1;142(4):1093-1107. PubMed.
  4. . Spread of pathological tau proteins through communicating neurons in human Alzheimer’s disease. bioRxiv. May 30, 2019
  5. . Local vulnerability and global connectivity jointly shape neurodegenerative disease propagation. PLoS Biol. 2019 Nov;17(11):e3000495. Epub 2019 Nov 21 PubMed.

Further Reading

Primary Papers

  1. . Functional brain architecture is associated with the rate of tau accumulation in Alzheimer's disease. Nat Commun. 2020 Jan 17;11(1):347. PubMed.