In neurodegenerative diseases, specific brain regions take the brunt of pathology and atrophy. However, differences from one patient to another make it hard to predict the precise path of progression for any one person. Now, scientists led by Jesse Brown and Bill Seeley of the University of California, San Francisco, report that they used a single structural MRI from 72 patients with frontotemporal dementia to predict which areas of their brains would next succumb to the disease. They determined where the disease likely originated—its epicenter—and used common functional connectivity maps to extrapolate regions likely to atrophy. In practice, the predictions correlated with brain volume loss over the next several years. The method provides an individualized biomarker for early clinical trials, the authors suggest.

  • Baseline MRI scans pinpoint disease epicenter for individual patients.
  • Model predicts future atrophy among functionally connected regions.
  • Prediction model could be biomarker for early stage clinical trials.

“Patient-tailored biomarkers represent a critical step forward for the field,” wrote Alexandre Bejanin, Research Institute of the Hospital de la Santa Creu i Sant Pau, Barcelona, Spain, to Alzforum. “They can provide core information on the clinical evolution of patients with neurodegenerative disorders, play a crucial role in monitoring disease progression in clinical trials, and guide future prevention programs and treatments.” Bejanin was not involved in the study.

“I find this work to be important and interesting in that the authors provide a novel framework for single-subject outcome measures based on network physiology,” wrote David Jones, Mayo Clinic, Rochester, Minnesota, to Alzforum. “This provides more evidence that the key biology at play in neurodegenerative diseases of the brain involves the functional physiology of large-scale brain networks that support mental functioning.” Jones was also uninvolved in the study.

Mapping Predicted Change. For this person with bvFTD, a baseline scan (top) determined areas of atrophy (dark red) and the estimated epicenter of disease (bright red with yellow outline). Then, researchers used their connectivity-based model (bottom) to predict areas of worsening (pink) and new (orange) volume loss. Those areas largely matched the patient’s actual atrophy (middle). [Courtesy of Brown et al., 2019.]

A prominent theory for the progression of brain diseases argues that misfolded, pathological proteins pass from neuron to neuron via trans-synaptic spread (de Calignon et al., 2012; Frost and Diamond, 2010). Based on that idea, Seeley and colleagues previously proposed that disease starts in one region of the brain, the epicenter, and spreads to new regions that are functionally connected (Seeley, 2017). 

Most of the evidence for this came from cross-sectional data. It remained to be proven if such a model predicted longitudinal atrophy in individual patients. In the current study, the authors sought to do just that in patients with either behavioral variant frontotemporal dementia (bvFTD) or the semantic variant of primary progressive aphasia (svPPA), two forms of frontotemporal dementia with distinct patterns of atrophy. They guessed that within a particular disorder, individual patients would have distinct epicenters that would dictate differences in overall disease progression.

Brown and colleagues examined structural MRI scans taken from 70 patients and 265 healthy controls. From the controls, the researchers derived a typical path of normal age-related volume loss. Then for each patient, the researchers calculated the difference between a region’s expected volume for that age and sex, and the actual measurement. That gave a map of a patient’s own disease-related atrophy pattern. To find its epicenter, they overlaid these patterns onto intrinsic connectivity maps, which show the functional connections between different regions in the normal human brain. The region’s intrinsic connectivity map that most closely resembled the constellation of atrophied regions in the patient’s brain was deemed the epicenter of disease.

In general, epicenters overlapped among patients with the same syndrome. In patients with bvFTD, they most often resided in the anterior cingulate cortex and frontoinsular cortex. In people with svPPA, the epicenter tended to be near the anterior temporal lobe. Even so, exact epicenter locations still varied considerably from person to person. Interestingly, while these epicenters always exhibited signs of atrophy, they were not necessarily the areas of greatest volume loss. Within each disease, atrophy patterns differed substantially between people.

To identify which regions would atrophy as disease progressed, the researchers built a prediction model that took into account three factors: how functionally connected a region was to the epicenter, how much its nearest neighbors had already been shrinking, and its baseline volume loss. From this, the researchers determined likely atrophy patterns for 42 bvFTD and 30 svPPA patient volunteers. Those participants then had an average of three more scans separated by six to 14 months, and the researchers compared the expected with the actual loss.

For most, those atrophy patterns matched up fairly well, with a correlation coefficient of 0.65. For 16 patients, 13 with bvFTD, the model failed to predict volume loss, with an average correlation coefficient of -0.04. Most of these patients had limited baseline atrophy and an unclear epicenter.

The researchers found that the rate of volume loss differed among brain regions. The most shrinkage occurred not in the epicenters themselves, but among their first-degree network neighbors. Possibly, the epicenter had already degenerated as much as it would, while first-degree neighbors were just getting started, the authors reasoned. “This has fundamental implications for clinical trials that would use imaging-derived indexes as an outcome,” wrote Bejanin. “The best regions to assess the effect of disease-modifying drugs should not be those primarily targeted by the disease, nor the most atrophied, but those most connected to these areas.”

The model isn’t refined enough to be used in clinical trials just yet, said Seeley. He noted that in this study, pooled functional connectome data of a group of healthy individuals were used to predict connectivity, but using a patient’s own connectome would likely improve predictions. But with better accuracy, this model could provide a proof-of-concept indicator for early to mid-stage clinical trials. For example, if a therapy leads to less atrophy than expected, it could encourage stakeholders to proceed with a confirmatory trial. Rik Ossenkoppele, Amsterdam University Medical Center, the Netherlands, added that the model could allow placebo and treatment groups to be better matched for expected atrophy rates. Perhaps a group of “fast progressors” could be identified and included to keep trials shorter and smaller, he suggested.

Seeley said that a more distant goal is to predict atrophy well enough to advise patients in the clinic about what symptoms to expect in the future. He expects that the model will be useful in other neurodegenerative diseases, including Alzheimer’s, although the double proteinopathy of Aβ and tau will likely render the picture more complicated.—Gwyneth Dickey Zakaib

Comments

  1. This is very elegant work from Jesse Brown and colleagues. I found two aspects of the study particularly noteworthy. The first is the idea of a personalized prediction model of longitudinal brain atrophy based on the trans-neuronal spreading hypothesis. The second is the introduction of a concept called “nodal hazard,” which is a regional risk measure of future atrophy based on the degree of baseline atrophy in regions that are highly functionally connected. Compared with previous group-level approaches, an individualized metric of rate and directionality of imminent brain atrophy has important potential ramifications for clinical practice and clinical trials. For example, since brain atrophy is intimately linked to clinical disease progression, this connectivity-based method may prove useful for the prognosis of various neurodegenerative disorders. Also, placebo and treatment groups could be carefully matched for expected atrophy rates in clinical trials. Moreover, identifying a group of “fast-progressors” may allow more-efficient screening of potential drug candidates (i.e., shorter duration and fewer persons needed).

    A major advantage of this method is that it only requires a baseline MRI scan, coupled with graph-theory-derived information on intrinsic functional connectivity properties from fMRI scans obtained in healthy subjects. This study thus represents an important first step toward prediction of biological disease progression at the individual level. Yet, further refinements are needed.

    First, the method seems to work better for svPPA (a focal and neuropathologically homogeneous disorder [mostly TDP-43 Type C]) than for bvFTD (affecting more widespread neocortical areas and caused by a myriad of brain pathologies). Second, the predictive power decreased when baseline atrophy levels were mild. This might indicate that the method needs substantial information on the emerging neurodegenerative pattern, which may hamper application in early disease stages. Third, although highly correlated, atrophy does not equal the underlying pathology. Thus, some of the direct effects of pathology on cognition and/or behavior (i.e., not mediated by atrophy) may not be captured. Unfortunately, there are currently no selective PET tracers available that bind TDP-43 or FTLD-tau aggregates, but it would be very interesting to test this method using Aβ and tau PET data in individuals with Alzheimer’s disease.

    Finally, some caution about the accuracy of the disease epicenter location is warranted based on the retrospective nature of its definition in this study. Overall, akin to previous publications from this group (Seeley et al., 2009; Zhou et al., 2012), the current work will most likely be guiding many future scientific studies, and will hopefully result in an individualized prediction model that directly benefits patients.

    References:

    . Neurodegenerative diseases target large-scale human brain networks. Neuron. 2009 Apr 16;62(1):42-52. PubMed.

    . Predicting regional neurodegeneration from the healthy brain functional connectome. Neuron. 2012 Mar 22;73(6):1216-27. PubMed.

  2. I find the work by Brown et al. to be important and interesting in that they provide a novel framework for single-subject outcome measures based on network physiology. This provides more evidence that the key biology at play in neurodegenerative diseases of the brain involves the functional physiology of large-scale brain networks that support mental functioning.

    As in previous studies, the methods employed in this study cannot differentiate competing models of network-based neurodegeneration. Reductionist models involving protein spreading along large-scale networks are not distinguishable from complex systems-based models that predict network collapse based in part on functional properties of the large-scale networks themselves.

  3. Jesse Brown and colleagues combined healthy functional connectome with patient baseline MRI to predict individual patterns of longitudinal gray-matter atrophy in patients with behavioral variant frontotemporal dementia (bvFTD) and semantic variant primary progressive aphasia (svPPA). Patient-tailored biomarkers represent a critical step forward for the field. Indeed, they can provide core information on the clinical evolution of patients with neurodegenerative disorders, play a crucial role in monitoring disease progression in clinical trials, and guide future prevention programs and treatments. Among many analyses presented in this excellent paper, I found three of them particularly interesting.

    First, the authors extensively explored the topographical heterogeneity in gray-matter atrophy in bvFTD and svPPA and used functional connectivity maps (derived from healthy controls) to define patient-tailored epicenters (i.e., site of the onset of the disease for each patient). In line with previous studies (Whitwell et al., 2009; Ranasinghe et al., 2016), their results support a certain within-group heterogeneity in the pattern of atrophy, which may result from distinct epicenters and disease severity. Most patient-tailored epicenters either overlapped across subjects or appeared to involve regions belonging to similar functional brain networks. This suggests that bvFTD and svPPA may arise from neurodegeneration that initially started in different brain regions belonging to the same syndrome-specific network.

    Second, the authors identified three region-wise metrics that, together with baseline volume, independently predicted future atrophy in patients: 1) the shortest path length to the patient-tailored epicenter (SPE), and the 2) nodal and 3) spatial hazard, which reflect the regional risk of subsequent atrophy based on the atrophy of most connected areas (nodal) or most spatially neighboring areas (spatial). Importantly, high nodal hazard scores led to volume loss over and above those associated with SPE or baseline atrophy. This means that a high degree of atrophy in most connected regions is a crucial factor to predict regional neuronal loss. Altogether, these results represent a major contribution to the network-based neurodegeneration framework by showing that 1) each atrophied brain region plays an active role in neurodegeneration spreading even if not primarily targeted by the disease, and 2) SPE remains a significant contributor to regional loss even when neurodegeneration has progressed. Results also imply that 3) neurodegeneration propagates not only into connected brain areas but also into spatially close regions. Further work is needed to determine if these two types of propagation are underlined by similar biological mechanisms.

    Finally, the use of a nonlinear generalized additive model very nicely evidenced that longitudinal changes in gray-matter volume are nonlinear in FTDs, as previously shown in AD (Sabuncu et al., 2011). Brain regions with high degrees of atrophy are less prone to show longitudinal changes than regions with intermediate degrees of atrophy. This has a fundamental implication for clinical trials that would use imaging-derived indexes as an outcome: The best regions to assess the effect of disease-modifying drugs in demented patients should not be those primarily targeted by the disease, nor the most atrophied, but those most connected to these areas.

    Together with previous investigations (Iturria-Medina et al., 2014; Raj et al., 2015), this study paves the way for patient-tailored approaches. Yet, there are still several challenges to translate these results into clinical practice, as individual predictions are still moderate (27/152 scans were “inaccurately” predicted [r2<0.06] and median r2=0.42 for the remaining scans) and the remaining unexplained variance might be even more challenging to predict. As mentioned by the authors, future models likely will be refined by including the patient’s functional and structural connectome. Additional model improvement may also imply to adjust the nodal and spatial hazard scores by the time interval between scans, to account for the presence of other pathological lesions (e.g., white-matter lesions), and incorporate other baseline neuroimaging, CSF, and genetic data.

    Overall, this study provides important insights into the mechanisms underlying the progression of neurodegeneration and offers promising perspectives to translate imaging-based method into clinical applications and individualized approaches.

    References:

    . Epidemic spreading model to characterize misfolded proteins propagation in aging and associated neurodegenerative disorders. PLoS Comput Biol. 2014 Nov;10(11):e1003956. Epub 2014 Nov 20 PubMed.

    . Network Diffusion Model of Progression Predicts Longitudinal Patterns of Atrophy and Metabolism in Alzheimer's Disease. Cell Rep. 2015 Jan 14; PubMed.

    . Distinct Subtypes of Behavioral Variant Frontotemporal Dementia Based on Patterns of Network Degeneration. JAMA Neurol. 2016 Sep 1;73(9):1078-88. PubMed.

    . The dynamics of cortical and hippocampal atrophy in Alzheimer disease. Arch Neurol. 2011 Aug;68(8):1040-8. PubMed.

    . Distinct anatomical subtypes of the behavioural variant of frontotemporal dementia: a cluster analysis study. Brain. 2009 Nov;132(Pt 11):2932-46. PubMed.

  4. This paper builds nicely on the realization that the spread of neurodegenerative diseases within the human brain is confined to specific, functionally defined networks. In the case of the behavioral variant of frontotemporal dementia, this involves the brain’s salience network, which explains the social-emotional features of the illness. In the case of Alzheimer’s disease, the featured network is the brain’s default mode network or, more generally, the brain’s hippocampal-cortical memory network. Again, this explains nicely the severe memory deficits encountered in persons with Alzheimer’s disease.

    The reason why these diseases originate and progress within specific brain networks is a critical unanswered question, but that does not preclude using this information to predict disease progression. However, one of the challenges in using this information is our ability to relate it to individual patients. Most brain-imaging studies with fMRI are based on large group averages. While this provides valuable data, those data are not easily extended to individual patients, where individual differences become an important consideration.

    This paper presents and convincingly defends a novel approach to monitoring changes within specific brain networks over time in individual subjects. The value of this approach is that it provides a means of monitoring the effect of interventions in individual subjects using an approach that is generally available. With such a tool in hand, our challenge is to identify effective interventions!

  5. The paper by Brown et al. tackles one of the biggest challenges in neurodegenerative disease: resolving the incredible heterogeneity of disease trajectories in dementia, both across and within syndromes. No two patients will show the same disease trajectory, making accurate diagnosis difficult and prognosis anyone’s guess. Jesse Brown and colleagues elegantly combined well-established neuroimaging approaches to develop a predictive model of atrophy spread.

    Importantly, their patient-tailored approach is a significant step forward compared with previous group-level analyses, and adds to recent investigations of a similar scope (Schmidt et al., 2016; Raj et al., 2012; Weickenmeier et al., 2018). Informed by the network degeneration theory and the epicenter model, the authors test the hypothesis that disease pathology in FTD spreads through functionally connected brain regions, derived from the healthy rs-fMRI connectome. The model performed well overall but failed to predict trajectories of atrophy in a moderate proportion of cases (27/152). Most of these cases were not severe or did not conform to prototypical disease presentations (i.e., C9ORF72 mutations carriers). As noted by the authors, further refinements to this model are needed to reduce the heterogeneity and explain the variance associated with brain trajectories in atypical cases.

    An alternative approach to the healthy rs-fMRI connectome is to use patient-derived anatomical connectivity measures. Indeed, recent developments in DWI analysis, such as track-weighted imaging methods (Calamante, 2017; Raffelt et al., 2017), provide a powerful means to study structural and functional connectivity simultaneously. Structural MRI methods require less abstraction, have clearer biological correlate and can be measured against the gold standard of molecular pathology in postmortem tissue. The epicenter model as a starting point of the disease does not fully address the cascade of molecular, metabolic, vascular, and functional changes that begins years before emergence of clinical syndromes. Once atrophy reaches a significant magnitude, the disease may have progressed too far for disease-modifying treatments to be effective.

    This work has the major advantage of only requiring a baseline MRI scan, which is routinely acquired as part of a clinical assessment. Further, since most neurodegenerative conditions show progressive brain atrophy, the method presented here has promising applications beyond FTD. Although atrophy provides a good proxy of underlying pathology, the precise nature of their relationship is yet to be uncovered. In particular, FTD is associated with incredibly heterogeneous pathologies, often coexisting. Therefore, mapping atrophy alone (even at the individual level), though important for disease monitoring, cannot effectively inform treatment development. As such, while this paper provides important avenues for development of clinical trials, further research, including other biomarkers of pathology, is needed.  

    In my opinion, where this paper makes a significant contribution is in informing disease staging and prognosis for individual patients. The emergence of particular clinical syndromes follows closely the location and temporal properties of underlying brain atrophy. While we are still far from adapting these neuroimaging approaches into clinical practice, this paper represents a significant step forward and a proof of concept of patient-tailored prediction of disease progression.

    References:

    . Simulating disease propagation across white matter connectome reveals anatomical substrate for neuropathology staging in amyotrophic lateral sclerosis. Neuroimage. 2015 Apr 11;124(Pt A):762-769. PubMed.

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

    . Multiphysics of Prionlike Diseases: Progression and Atrophy. Phys Rev Lett. 2018 Oct 12;121(15):158101. PubMed.

    . Track-weighted imaging methods: extracting information from a streamlines tractogram. MAGMA. 2017 Aug;30(4):317-335. Epub 2017 Feb 8 PubMed.

    . Investigating white matter fibre density and morphology using fixel-based analysis. Neuroimage. 2017 Jan 1;144(Pt A):58-73. Epub 2016 Sep 14 PubMed.

Make a Comment

To make a comment you must login or register.

References

Paper Citations

  1. . Propagation of tau pathology in a model of early Alzheimer's disease. Neuron. 2012 Feb 23;73(4):685-97. PubMed.
  2. . Prion-like mechanisms in neurodegenerative diseases. Nat Rev Neurosci. 2010 Mar;11(3):155-9. PubMed.
  3. . Mapping Neurodegenerative Disease Onset and Progression. Cold Spring Harb Perspect Biol. 2017 Aug 1;9(8) PubMed.

Further Reading

Primary Papers

  1. . Patient-Tailored, Connectivity-Based Forecasts of Spreading Brain Atrophy. Neuron. 2019 Dec 4;104(5):856-868.e5. Epub 2019 Oct 14 PubMed.