Resting state MRI: a new marker of prodromal neurodegeneration?

This scientific commentary refers to ‘Basal ganglia dysfunction in idiopathic REM sleep behaviour disorder parallels that in early Parkinson’s disease’, by Rolinski et al. (doi:10.1093/brain/aww124).

Like most neurodegenerative diseases, Parkinson’s disease has a long prodromal period; that is, a time when symptoms or signs have emerged, but are not sufficiently advanced to diagnose clinical parkinsonism. This condition can currently be diagnosed by detecting combinations of motor and non-motor signs. Diagnostic criteria for prodromal Parkinson’s disease have recently been published (Berg et al., 2015). Unlike other diseases, prodromal Parkinson’s disease diagnosis relies mostly upon clinical examination. Established biomarkers of prodromal Parkinson’s disease are relatively limited; only PET/SPECT scanning and substantia nigra (SNpc) ultrasound have some prospective evidence of predictive value (Berg et al., 2015). This is a major limitation for the field; we need biomarkers to both diagnose prodromal Parkinson’s disease (to identify candidates for neuroprotective therapy) and to assess progression (to give a quantifiable biological outcome for neuroprotective trials). In this issue of Brain, Rolinski and co-workers suggest a potential new biomarker of prodromal Parkinson’s disease: resting state functional MRI (Rolinski et al., 2016).

If MRI could be used as a biomarker, it would offer a number of key advantages: it has a well-established standardized acquisition and infrastructure, and is globally available. However, a typical clinical MRI in Parkinson’s disease is normal. If a test is normal in established disease, it is hard to imagine using it to diagnose a prodromal state. However, as MRI techniques advance, this may change. For example, with increased magnet strength and sequences sensitive to brain iron (susceptibility weighted imaging, neuromelanin sequences), the architecture of the substantia nigra can be delineated with greater precision. Studies have suggested that a loss of the normal hyperintense signal corresponding to nigrosome 1 can reliably distinguish patients with SNpc degeneration from controls (Bae et al., 2016). Although not useful in differential diagnosis of Parkinson’s disease (scans are abnormal in progressive supranuclear palsy and multiple system atrophy), this may have promise for prodromal Parkinson’s disease. Similarly, detailed analyses of atrophy and diffusion tensor imaging show abnormalities in Parkinson’s disease, although the reproducibility and reliability of these techniques for diagnosis at the individual level has yet to be established.

Why might resting state functional MRI work? Landmark studies by Seeley and co-workers (2009) showed that neurodegeneration can selectively attack regions that have highly correlated patterns of spontaneous blood oxygen level-dependent fluctuations on functional MRI, measured when subjects are ‘at rest’ (i.e. lying inactive in the scanner). These correlations indicate regions that are functionally connected into networks (note that this does not require anatomical connections, although regions are often anatomically connected as well). This network structure may be a fundamental organizing principle of brain function, with over 30 independent networks proposed. If networks are targeted in disease, this may mean that cells that connect together, die together. The most common example is the targeting of the default mode network in Alzheimer’s disease, but this may be a principle that applies to many neurological disorders. These findings accord with animal models showing that neurodegeneration can propagate via prion-like spread of pathological proteins along synaptic connections (Luk et al., 2012). Moreover, recent studies using MRI volumetry have found evidence of network-based degeneration in Parkinson’s disease that may potentially track along pathways connected to the substantia nigra (Zeighami et al., 2015).

If Parkinson’s disease is also a network disease, what does this mean for resting state functional MRI? Rolinski and co-workers have previously shown that patients with established Parkinson’s disease have reduced functional connectivity within a basal ganglia network (Szewczyk-Krolikowski et al., 2014). This abnormal network connectivity was rescued with levodopa therapy. Smaller reductions in connectivity were found in the cortex, perhaps reflecting executive dysfunction. These changes were present in early Parkinson’s disease and did not clearly worsen with longer disease duration, suggesting that they might be found in prodromal Parkinson’s disease. This is the hypothesis Rolinski et al. have tested here.

The model of prodromal Parkinson’s disease they chose to use is idiopathic REM sleep behaviour disorder (RBD). The large majority of patients with ‘idiopathic’ RBD are in fact currently in stages of prodromal Parkinson’s disease; >70% develop neurodegenerative synucleinopathy over 10–15 years, including Parkinson’s disease, multiple system atrophy, and dementia with Lewy bodies (which is also usually associated with Parkinson’s disease) (Berg et al., 2014; Postuma et al., 2015). Given this proportion, if a synucleinopathy marker is abnormal in idiopathic RBD, it is likely to be a prodromal Parkinson’s disease marker. Rolinski et al. used a data-driven whole-brain approach to examining networks in RBD (in contrast to seed-based approaches or region of interest approaches, which test specific areas based on a priori hypotheses). They found that network functional connectivity is indeed abnormal in RBD. Results look generally similar to those in Parkinson’s disease, with most abnormalities occurring within the basal ganglia network. There were also non-significant trends towards cortical abnormalities, consistent with the substantial cognitive deficits in RBD (and the fact that RBD is also a prodromal dementia marker). This is not surprising, but is already an important advance.

What is especially notable, and surprising, about these results is the strength of the effect. Generally, because of interindividual variability, functional MRI measures are not sufficiently robust to be used at the individual level. Resting state functional MRI in particular is highly variable, even within individuals. However, here the summary estimate could distinguish patients with RBD from controls with 96% sensitivity and 78% specificity. This is approaching the accuracy seen in diagnostic testing. This sensitivity strikingly contrasts with more conventional measures of basal ganglia function; for example, only 40% of patients with RBD demonstrate dopaminergic denervation on PET/SPECT (Iranzo et al., 2010). Moreover, the analysis finds no asymmetry, suggesting that a ‘floor’ effect is reached early in the course of RBD. Therefore, it appears that network analysis reveals abnormalities that start earlier than currently measurable assays of cell death or dopamine denervation.

What are the next steps? The obvious first step is to follow patients with RBD over time. This is not only to confirm predictive value, but also to see how changes in the network evolve in prodromal synucleinopathy (or if they evolve at all). Given the potential floor effect, this analysis should probably not simply be between RBD convertors and non-convertors; if nearly all patients with RBD have prodromal synucleinopathy, and functional MRI is abnormal early in the disease course, the comparison will be negative. Rather, the analysis should assess RBD converters at various prodromal time points versus controls. This would also allow direct assessment of how far in advance functional MRI abnormalities precede defined parkinsonism or dementia. Given the potential floor effect in these data, one might suspect that functional MRI would be more useful as a diagnostic rather than stage marker. That is, it could be used to test for early prodromal Parkinson’s disease, but would not be used as a biomarker of progression (for example, in neuroprotective trials). Second, these findings need to be reproduced; in particular, this promising sensitivity and specificity must be confirmed. Confirmation may be possible soon; for example, the ‘Parkinson’s Progression Markers Initiative’ (PPMI) cohort will include a prodromal cohort with at least structural MRI scanning (this will be open-access data) and there are numerous groups assessing MRI in RBD. Third, it would be of interest to see if seed-based approaches generate similar findings, perhaps testing seeds that examine propagation along synaptic connections (Rolinski et al. did not find abnormalities with an SNpc seed-based approach, but this may be because of technical difficulties and the limits of anatomical resolution). Finally, it should be noted that RBD might be a strong marker of a subtype of Parkinson’s disease (Fereshtehnejad et al., 2015). It will be important to see if similar changes are observed in patients with prodromal Parkinson’s disease without idiopathic RBD (although finding a large population with sufficiently high risk will be a considerable challenge).

And as for clinical practice, is it time to consider functional MRI in idiopathic RBD? Perhaps it is too soon to say, at least for conventional clinical MRI. Analysis of the variables used in the current study is not trivial, requiring centres with research expertise. But then again, MRI data can be stored. Obtaining a 3 T structural MRI that includes a sequence sensitive to SNpc changes (e.g. susceptibility-weighted or neuromelanin imaging) plus a resting state acquisition may eventually be a worthwhile investment.

References