Findings suggest converging evidence from varying markers for detectable abnormalities beginning between one and two decades before diagnosis of HD. The relationships between estimated years to diagnosis and motor scores, striatal volumes, odour recognition and cognitive measures were strikingly consistent. On this estimated time scale, all domains suggest a curvilinear pattern of disease commencement. This period of commencement is followed by more rapid and constant change in the last years prior to diagnosis. Our findings are consistent with previous reports suggesting an association of striatal volumes with estimated diagnosis based on CAG repeat.17 18 22
The current sample size is nearly 10 times larger than previous reports and thus serves both a confirmatory and refining role. Additionally, demonstration of similar estimated time lines for cognitive, sensory and motor changes is novel and significantly extends these findings. We have illustrated a pervasive pattern of change, occurring on the same time scale, which extends from biological measures such as striatal volumes to several different clinical aspects of the HD phenotype.
Our finding that associations between years to diagnosis and other variables remained, even after controlling for motor signs, strongly suggests that apparent cognitive and sensory dysfunction cannot be explained solely on the basis of emerging motor signs interfering with task performance. It is noteworthy that these associations remained, regardless of whether the assessment was laden with motor demands (speeded finger tapping, self-timed finger tapping) or not (verbal learning, odour identification). The consistency of associations between the estimated genetic time scale and these diverse cognitive measures suggests that multiple corticostriatal circuits are simultaneously involved in early disease, although further research is needed to more explicitly track the brain behaviour associations suggested here. Striatal volumes also remained curvilinearly related to estimated years to diagnosis after controlling for motor signs. The analyses therefore indicate that neither cognitive nor striatal measurements are wholly redundant with the clinical motor examination. It follows that all of these can likely be combined and leveraged to improve the accuracy of individualised prognosis. Many of these markers may be candidate surrogate endpoints (bearing in mind that promotion from candidate to true surrogate is exceedingly difficult).32
However, markers that cannot meet criteria for this elusive role can still be useful for risk stratification—which is still quite valuable for increasing clinical trial power in a rare disease.
Quantitative estimation of the impact of these markers in either role (longitudinal outcome or baseline stratification) is a substantial topic in its own right. In the near future, we will address the details of using baseline Predict-HD findings for increased statistical power and trial design efficiency in a separate manuscript.
There are some important caveats to consider when interpreting these findings. Firstly, the cross sectional associations require longitudinal validation. With planned follow-up periods of up to 7 years, the Predict-HD study will eventually yield appropriate data to attempt this validation. Furthermore, the study was designed to eventually yield a sufficient sample of individuals who become diagnosed, allowing CAG based estimates of average diagnosis age to be validated or refined. Despite excellent goodness of fit in the original model derivation,3
the time scale discussed in the current article remains provisional until longitudinal study is complete.
Even if the time scale is accurate, estimated time to diagnosis is clearly an imperfect proxy for actual time to diagnosis, which is of course unknown in these subjects. Although relationships based on estimated time to diagnosis cannot perfectly represent the relationship that would be seen if true years to diagnosis were known, it can be shown that an accurate expected time proxy leads to approximations providing useful bounds on the true average relationships. For example, the (essentially) uniformly concave or convex nature of each curve in allows invocation of Jensen’s inequality33
to suggest that we are estimating an upper limit to the true time course of mean change. Further mathematical arguments beyond the realm of this paper also show that it is unlikely that this upper bound dramatically overestimates the true mean time of onset of these deficits. Hence while resisting the temptation to over interpret these curves, we feel comfortable in our claim that they are strong evidence that detectable deficits begin between 10 and 20 years before HD diagnosis in the average CAG expanded subject. A more detailed and mathematical exposition regarding the limits of interpretability for this class of prognostic model will also be forthcoming in a separate report.
Which of the outcomes described here is the strongest marker of developing HD? Again, we must remember that the models we have presented actually reflect the relationship between these markers and another known marker—CAG repeat length (corrected for age). While the relative strengths of these relationships provide a tentative basis for judgment, they are not guaranteed to reflect the relative prognostic importance that we will eventually observe with actual clinical diagnosis times. Similarly, we cannot yet know to what extent each measure independently contributes prognostic information. For example, striatal volume shows the strongest relationship to estimated prognosis in our analyses. Given the well known central role of basal ganglia deterioration in HD, we will not be surprised if this measure remains the strongest individual predictor when we can analyse observed onset. Nonetheless, it is possible that its relative strength in the current analysis could be due to an especially tight CAG length association that does not necessarily translate into the best incremental independent prediction of true prognosis.
Our findings of cognitive, sensory, motor and striatal volumetric changes well before disease diagnosis are unequivocal, but it remains unknown whether these changes are of functional significance until our longitudinal data are acquired. Also, it is not known whether there are very subtle abnormalities, perhaps imperfectly reflected in these markers, prior to the evident upper limits that we have estimated here. Although not included in the current report, the Predict-HD study includes a comparison group of participants who are from HD families but are known not to have the CAG expansion. Once sufficient data from this sample become available, it may be possible to detect subtle abnormalities even earlier than the current report suggests.
Current hopes for reducing the burden of neurodegenerative disease rely on the idea of preventing disease onset and slowing progression so that people at risk may live out a longer portion of their lifespan as healthy, fully functioning individuals. For this to be possible, promising therapeutic agents must be tested for their ability to slow early disease progression. These cross sectional findings from Predict-HD indicate the approximate time scale of measurable disease development, and suggest disease markers which may be candidate surrogates for use in preventive trials. Furthermore, even markers that fail as treatment response surrogates will have great value for risk stratification in such trials. With validation and refinement through longitudinal study, it will be possible in just a few more years to confirm such markers, making it feasible to initiate the first preventive trials in individuals with the HD CAG expansion prior to functional decline and diagnosis.