In this longitudinal study of preHD, we used the completely automated SIENA tool to evaluate whole-brain change. We found significantly larger PBVC (more severe atrophy) in preHD individuals compared to controls over the course of one year. The increased PBVC was particularly striking in a close-to-onset preHD subgroup, supporting a previous finding of accelerating preHD atrophy in the striatum using manual segmentation13
. The longitudinal decrease in brain volume was also evident at the voxel-level in periventricular regions – consistent with the well-established basal ganglia atrophy in preHD.
The ideal MRI biomarker for assessing neurodegeneration in preHD would be objectively measured, consistent with known disease pathology, and serve as a predictor for clinical outcomes (such as conversion to manifest HD)8–11
. Our results were achieved with a fully-automated analysis. We used the SIENA software tool, which has been shown to be very reliable with an estimated error in brain volume change as low as 0.15%and to be robust to varying image quality, slice thickness, and different pulse sequences22, 23
. This error rate is smaller than the group differences we identify in our study. In comparison with other semi-automated techniques to assess longitudinal change, SIENA has demonstrated lower error rates and higher sensitivity in the detection of subtle differences in atrophy38
. Moreover, SIENA is freely available and quick to run. Clearly, this fully-automated analysis method meets the objectivity criterion of a preHD biomarker.
A useful longitudinal MRI biomarker should also reveal atrophy that is consistent with known preHD pathology. Here, voxel-wise analysis of longitudinal brain edge displacement revealed changes to periventricular regions, consistent with the well-established profile of basal ganglia atrophy in preHD13–15, 17, 18, 39
. However, preHD pathology also includes non-striatal changes such as decreases in white matter volume19, 20
and cortical changes16, 18, 40, 41
. Although our voxel-wise analysis points to basal ganglia atrophy, a striking (and probably the most useful) finding here is the overall brain atrophy measure (i.e. PBVC). This measure reflects the total amount of brain edge displacement across time. Thus it is likely sensitive to pathology at multiple levels, including basal ganglia, cortical gray, and white matter, even if not all these changes are reflected in voxel-level differences.
Biomarkers in preHD should also serve as predictors for known future clinical outcomes8–11
. We have shown that individuals closest to estimated disease onset drove group differences in atrophy, suggesting that SIENA-derived PBVC may be a good predictor of manifest HD conversion. Furthermore, of the 37 preHD participants who entered our study at the beginning of the year, the two who later converted to manifest HD also had the largest PBVC, consistent with the possibility that PBVC predicts imminent clinical onset. A survival analysis after future follow-up of our remaining 35 preHD participants will further clarify the utility of PBVC as a predictor of clinical onset.
PBVC correlated with the Aylward YTO estimate but not the Langbehn YTO estimate. Differences in methodology may account for this discrepancy. The Langbehn YTO estimate is constrained to always be positive27, 42
, whereas the Aylward YTO estimate may be negative when the current age exceeds the estimated age-of-onset26
. Thus the Aylward method may allow for greater variability among those closest to onset that is lost when using the Langbehn method.
A limitations of SIENA is the inability to provide full regional specificity as to the locus of atrophy. Our findings of periventricular change, though suggestive regional atrophy, remain unspecific as to the actual locus of change. Indeed, future regionally specific automated analysis tools hold great promise as preHD biomarkers. To date, the sole automated study of caudate atrophy found a high effect size of 0.9 when comparing preHD and controls over a two year period14
. Though we show a medium effect size when comparing preHD and control groups, we note that the SIENA whole-brain methodology detected group differences in a shorter time period, i.e. one year. A longer study period would likely accentuate group differences. Moreover, SIENA has the advantage of being freely available, quick, and easy to run.
Furthermore, we observed a high effect size (0.92), as large as that for the abovementioned caudate specific measure14
, when comparing PBVC in controls with close-to-onset preHD participants (Aylward YTO ≤ 6 years). This provides useful information about statistical power needed to plan neuroprotective treatment studies in preHD individuals over a one-year period. For instance, a study designed using SIENA to test a treatment that can slow the yearly rate of atrophy by 50% may only require about 74close-to-onset preHD individuals in each experimental arm (treatment vs. placebo). Although the assumption of a 50% benefit may be overly optimistic, the feasibility of such a study strongly argues for the utility of SIENA as a preHD biomarker.
In summary, we have demonstrated the potential of SIENA, a fully-automated and robust method, to detectatrophy in preHD over a mere one year period. We have also shown that year-end PBVC values may predict known disease outcomes such as conversion to manifest disease. These findings provide proof-of-concept regarding biomarker development for disease detection in preHD, as well as quantitative insights into how to power upcoming trials of neuroprotection.