Skip to main content
NCBI home page
UKPMC Funders Author Manuscripts logoLink to UKPMC Funders Author Manuscripts
. Author manuscript; available in PMC: 2021 Oct 25.
Published in final edited form as: Sci Transl Med. 2020 Dec 16;12(574):eabc2888. doi: 10.1126/scitranslmed.abc2888

Mutant huntingtin and neurofilament light have distinct longitudinal dynamics in Huntington’s disease

Filipe B Rodrigues 1,#, Lauren M Byrne 1,#, Rosanna Tortelli 1, Eileanoir B Johnson 1, Peter A Wijeratne 2, Marzena Arridge 3,4, Enrico De Vita 3,4,5, Naghmeh Ghazaleh 6, Richard Houghton 6, Hannah Furby 6, Daniel C Alexander 2, Sarah J Tabrizi 1,7, Scott Schobel 8, Rachael I Scahill 1, Amanda Heslegrave 7,9, Henrik Zetterberg 7,9,10,11, Edward J Wild 1,†
PMCID: PMC7611886  EMSID: EMS137096  PMID: 33328328

Abstract

The longitudinal dynamics of the most promising biofluid biomarker candidates for Huntington’s disease (HD)— mutant huntingtin (mHTT) and neurofilament light (NfL)—are incompletely defined. Characterizing changes in these candidates during disease progression could increase our understanding of disease pathophysiology and help the identification of effective therapies. In an 80-participant cohort over 24 months, mHTT in cerebrospinal fluid (CSF), as well as NfL in CSF and blood, had distinct longitudinal trajectories in HD mutation carriers compared with controls. Baseline analyte values predicted clinical disease status, subsequent clinical progression, and brain atrophy, better than did the rate of change in analytes. Overall, NfL was a stronger monitoring and prognostic biomarker for HD than mHTT. Nonetheless, mHTT has prognostic value and might be a valuable pharmacodynamic marker for huntingtin-lowering trials.

Introduction

Despite knowledge of its monogenetic cause, no treatments have been shown to slow neurodegeneration in Huntington’s disease (HD) (1, 2). However, multiple approaches aimed at lowering the production of the causative mutant huntingtin (mHTT) protein are in human clinical trials (3–5). The ultimate goal, treating mutation carriers early to prevent disease onset, will require prevention trials in premanifest HD mutation carriers (preHD).

Successful target engagement by the first targeted huntingtin-lowering therapeutic tested in patients with HD—the antisense oligo-nucleotide (ASO) tominersen (formerly IONIS-HTTRx/RG6042)—was demonstrated by dose-dependent mHTT reduction in cerebrospinal fluid (CSF) in a phase 1/2 trial (3), quantified by ultrasensitive immunoassay (6). A reliable CSF-to-brain mHTT relationship has been established in animal studies (7, 8). This notable success led to the first phase 3 trial of such a drug, whose primary outcomes are the total functional capacity (TFC) score of the Unified Huntington’s Disease Rating Scale (UHDRS) in the United States and a composite UHDRS (cUHDRS) measure combining motor, functional, and cognitive scores in Europe (9). Such clinical rating scales quantify overt clinical manifestations but are less sensitive to detect deterioration or its therapeutic benefit in preHD (10–13), making their use as outcomes in prevention trials problematic. Although clinically relevant, they are also far removed from the core disease mechanism: neuronal injury by the huntingtin gene product. Quantifying biochemical manifestations of neurodegeneration can inform our understanding of pathobiology and the development and testing of novel therapies.

To this end, we recently showed that CSF concentrations of mHTT, the toxic pathogenic protein, and neurofilament light (NfL), an axonal protein indicative of neuronal injury, are among the earliest detectable changes in HD and are strongly associated cross-sectionally with baseline measures of clinical severity and brain volume (14). In the longitudinal TRACK-HD cohort, we showed that blood NfL concentration independently predicts subsequent onset, clinical progression, and brain atrophy in HD over 3 years (15).

One unexpected finding from the phase 1/2 ASO trial was a transient elevation of NfL in CSF around 5 months after first dose (3). In the subsequent open-label extension of this trial (16), NfL rose again and then fell in the context of ongoing huntingtin suppression, almost to baseline by month 8. Although it remains unexplained mechanistically, this observation attests to the combined value of mHTT and NfL—as potential pharmacodynamic/response, safety, and monitoring biomarkers—to highlight changes of note in the “undiscovered country” of huntingtin lowering. The relationship between early biomarker changes and subsequent clinical outcomes will, of course, be critical to follow. It also calls for a more detailed understanding of how these markers change over time throughout the natural history of HD to help the interpretation of their behavior in clinical trials.

Here, we present the mHTT and NfL findings from the 2-year prospective longitudinal HD-CSF study, in which an 80-participant cohort of HD mutation carriers and controls underwent clinical assessments, sampling of CSF and plasma, and magnetic resonance imaging (MRI), under strictly standardized conditions. We report here the longitudinal dynamics of CSF mHTT and NfL, studied and compared head-to-head in the natural history of HD. We assessed and compared the ability of the biomarkers at baseline, as well as their rates of change, to predict longitudinal progression in clinical and neuroimaging measures. We used advanced methods including random forest methodology and event-based modeling (EBM), permitting the exploration of nonlinear relationships and multiple variables at once. Last, we performed computational clinical trial simulations to provide insight into how they could be used and combined in the therapeutic context.

Results

The HD-CSF cohort

Seventy-four (92.5%) of the 80 baseline participants returned for the 24-month follow-up assessments. Three (4%) of the 74 opted out of doing the follow-up lumbar puncture but agreed to blood and phenotypic data collection (Fig. 1). No preHD participant progressed to manifest HD during the follow-up period. A more detailed version of the study flow is provided in fig. S1.

Fig. 1. HD-CSF study participant disposition.

Fig. 1

Open in a new tab

Baseline visit (n = 80) was performed 24 months (± 3 months) before the follow-up visit (n = 74). Optional repeat sampling visits occurred 6 to 8 weeks after baseline. A more detailed version including all study assessments is provided in fig. S1.

Full cohort characteristics are shown in table S1. Disease groups were well matched for gender and differed, as expected, in HD clinical, cognitive, and imaging measures. Age differed between groups due to the control group (50.68 years ± 11.0) being matched to all HD mutation carriers and manifest HD (56.02 years ± 9.36) being more advanced in their disease course than preHD (42.38 years ± 11.04), as previously reported (14).

Longitudinal dynamics of mHTT and NfL

Longitudinal trajectories of each analyte within individuals are shown in Fig. 2 (A to C). For NfL in CSF and plasma, there was little overlap in the trajectories of HD mutation carriers and healthy controls. Mixed-effects models depict distinct patterns of longitudinal analyte dynamics (Fig. 2, D to F). CSF and plasma NfL showed a sigmoidal pattern over time in HD mutation carriers, initially accelerating and then later slowing, compared to a slow linear rise with aging in controls. CSF mHTT, on the other hand, rose linearly with age, albeit with more variability.

Fig. 2. Longitudinal dynamics of mHTT and NfL over 24 months.

Fig. 2

Open in a new tab

(A to C) Individual participant trajectories. Connected dots are measurements within the same participant. Disease groups are color-coded as per (J). (D to F) Modeled biomarker trajectories. Model (solid lines), 95% (colored areas), and 99% (dashed lines) bias-corrected accelerated confidence intervals were generated from generalized mixed-effects models. For CSF mHTT, age was used as first-order fixed effect, whereas for HD mutation carriers for NfL, there was a first- and second-order fixed effect for age. All models included CAG repeat count and had a random intercept for participant and corresponding random slopes for age. Dots represent the observed values. For ease of visual interpretation, two individual data points (>6000 pg/ml) were included in the model but excluded from the figure. (G to I) Modeling genetic dose-response relationships to show associations between biomarkers, age, and CAG repeat count. Solid lines were produced from our observations using the models above; dashed lines are predictions outside the range of our observations. Separate figures with individual data points for each individual CAG repeat count are provided in fig. S2, and the distribution of age and CAG within the cohort are supplied in fig. S3. Gray diamonds show the age of predicted onset for each CAG length [as per (35)]. Colored diamonds show the age at which HD mutation carrier trajectories are most likely to depart from healthy control trajectories for each CAG repeat count, generated by change-point analysis. (J) Annualized rates of change and 95% confidence intervals. For each biomarker, estimates were computed as the average of the rate of change in 1000 simulations per group of study participants (healthy controls, preHD, and manifest HD). Note that (A), (D), and (G) referring to mHTT do not depict values for controls. For CSF mHTT, shaded areas mark the limits of detection (LoD, 8 fM) and quantification (LoQ, 25 fM) of the assay. CSF, cerebrospinal fluid; mHTT, mutant huntingtin; N/A, not applicable; NfL, neurofilament light.

To explore the “genetic dose-response relationship” between the causative gene mutation and each biofluid measure, we queried our models for the interaction between CAG and age in HD mutation carriers (linear effect for CSF mHTT, P < 0.0001; and nonlinear effect for CSF and plasma NfL, P = 0.006 and P < 0.001, respectively; Fig. 2, G to I, and fig. S2). Using a change-point analysis, we estimated the disease burden score (DBS; a combination of age and CAG) at which each analyte in HD mutation carriers starts to deviate from controls. The DBS change points for each analyte (CSF mHTT, 188.9; CSF NfL, 248.6; plasma NfL, 236.1; fig. S4) were used to annotate Fig. 2 (G to I) with the age of expected departure from controls, for each CAG length.

On the basis of simulations of the models in Fig. 2 (D to F), CSF NfL rose in manifest HD by 98.85 pg/ml per year, in preHD by 79.16 pg/ml per year, and in controls by 20.05 pg/ml per year (Fig. 2J). Similar relative findings were seen for plasma NfL rates of change (controls, 0.28 pg/ml per year; preHD, 0.84 pg/ml per year; manifest HD, 1.04 pg/ml per year; Fig. 2J).

Prognostic value for overall HD progression of baseline analyte versus its rate of change

We assessed the clinical associations of each analyte using the cUHDRS, a composite score derived from large natural history cohorts and combining motor, functional, and cognitive symptoms to reflect overall HD clinical severity across clinically important domains (17). All three analytes were associated with cUHDRS cross-sectionally at both baseline and follow-up (fig. S5). To assess the prognostic value of each analyte for HD progression, we first examined whether their baseline values predicted subsequent change in cUHDRS. All models included age as an independent variable to account for its known confounding effects. Second-tier models also included CAG, a known effect modifier and fundamental driver of HD pathology, as an independent variable. This two-step approach allows the exploration of whether associations, accounting for acquired factors, are driven by the core genetic driver of HD or have predictive and/or prognostic value independent from it. Significant associations with subsequent cUHDRS change were found for all three analytes [CSF mHTT: r = −0.31, 95% confidence interval (CI) = −0.57 to −0.03, P = 0.026; CSF NfL: r = −0.38, 95% CI = −0.52 to −0.18, P < 0.0001; plasma NfL: r = −0.47, 95% CI = −0.63 to −0.25, P < 0.0001; Fig. 3, A to C]. When CAG was added to the model, only the association with baseline plasma NfL remained significant (CSF mHTT: r = −0.11, 95% CI = −0.48 to 0.18, P = 0.513; CSF NfL: r = −0.21, 95% CI = −0.48 to 0.00, P = 0.098; plasma NfL: r = −0.33, 95% CI = −0.58 to −0.08, P = 0.011; Fig. 3, A to C, and table S2).

Fig. 3. Longitudinal associations of mHTT and NfL with disease progression quantified by cUHDRS.

Fig. 3

Open in a new tab

Associations between (A to C) baseline values or (D to F) annualized rate of change in each analyte and the annualized rate of change in the cUHDRS. Pearson’s partial correlation coefficient adjusted for age and P values are presented. Gray figures were estimated including CAG in the model. Dashed horizontal lines mark no change in cUHDRS, with negative values representing deterioration. Dashed vertical lines mark no change in the biomarker. The baseline values (G to I) and annualized rate of change (J to L) for each biomarker compared between fast (n = 24) and slow progressors (n = 30), defined as participants with an absolute decrease in cUHDRS greater than or equal to 1.2 over the follow-up period. (M) Random forest plot for prediction of annualized rate of change in cUHDRS including baseline and rate of change of biofluid biomarkers as predictors. Higher relative importance score indicates a greater relative importance of the variable in predicting worsening cUHDRS score. Relative importance scores were based on the mean decrease in Gini score. Distributions were generated from rerunning the model 100 times. The boxes show the median and 25 and 75% percentiles, whereas whiskers are the lower and upper adjacent values (1.5 times the interquartile range minus the 25% percentile or plus the 75% percentile). Dots are values under and above the adjacent values. Violin plots represent ranking distributions. Black diamonds represent the median. cUHDRS, composite Unified Huntington’s Disease Rating Scale; DBS, disease burden score; Δ, annualized rate of change.

Next, we assessed whether the rate of change of each analyte gave any additional prognostic information beyond that given by a single baseline measurement. The rate of change in plasma NfL was associated with the rate of change in cUHDRS, whereas CSF NfL showed weaker association and CSF mHTT showed no association (CSF mHTT: r = −0.17, 95% CI = −0.49 to 0.08, P = 0.253; CSF NfL: r = −0.20, 95% CI = −0.43 to −0.01, P = 0.049; plasma NfL: r = −0.34, 95% CI = −0.65 to −0.12, P = 0.013; Fig. 3, D to F). These associations for rate of change were weaker than those of the baseline values and did not survive adjustment for age when also including CAG in the models (CSF mHTT: r = −0.09, 95% CI = −0.46 to 0.15, P = 0.539; CSF NfL: r = −0.09, 95% CI = −0.33 to 0.10, P = 0.401; plasma NfL: r = −0.22, 95% CI = −0.54 to −0.00, P = 0.111; Fig. 3, D to F, and table S2).

In our analysis of fast and slow progressors (≥1.2 or <1.2 decline in cUHDRS, respectively), baseline CSF mHTT, CSF NfL, and plasma NfL were all significantly higher in faster progressors (CSF mHTT mean difference (MD) = 19.27 fM, 95% CI = 10.74 to 27.80, P < 0.0001; CSF NfL MD = 1066.05 pg/ml, 95% CI = 532.62 to 1599.48, P < 0.0001; plasma NfL MD = 11.44 pg/ml, 95% CI = 6.45 to 16.43, P < 0.0001; Fig. 3, G to I). Only plasma NfL remained associated after adjustment for age and including CAG in the model (CSF mHTT MD = 7.38 fM, 95% CI = −0.92 to 15.68, P = 0.081; CSF NfL MD = 449.75 pg/ml, 95% CI = −61.52 to 961.02, P = 0.087; plasma NfL MD = 5.59 pg/ml, 95% CI = 1.24 to 9.93, P = 0.032; Fig. 3, G to I).

We repeated this analysis using the rate of change in each analyte. Only the rate of change of plasma NfL was significantly higher in faster progressors (CSF mHTT MD = 0.67 fM/year, 95% CI = −1.75 to 3.09, P = 0.584; CSF NfL MD = 230.23 pg/ml per year, 95% CI = −29.32 to 489.78, P = 0.082; plasma NfL MD = 2.83 pg/ml per year, 95% CI = 1.03 to 4.62, P = 0.002; Fig. 3, J to L). This did not survive adjustment for age and including CAG in the models (CSF mHTT MD = −1.04 fM/year, 95% CI = −3.69 to 1.61, P = 0.440; CSF NfL MD = 27.07 pg/ml per year, 95% CI = −253.56 to 307.70, P = 0.849; plasma NfL MD = 1.22 pg/ml per year, 95% CI = −0.71 to 3.15, P = 0.210; Fig. 3, J to L).

We next used random forest analysis to compare head-to-head and illustrate the relative importance ranking of the three biofluid analytes, including their baseline value and annualized rate of change, in prediction of HD clinical progression, alongside other established predictors (age, CAG, and DBS). Using change in cUHDRS as a continuous outcome, baseline values ranked as stronger predictors than annualized rates of change (Fig. 3M). Similar results were obtained for prediction of fast versus slow progression (fig. S6A).

Prognostic value of biofluid biomarkers for clinical, imaging, and cognitive measures

Next, we compared the prognostic power of each analyte, in terms of both its baseline value and its rate of change, to predict progression in individual clinical and MRI measures (Fig. 4, figs. S7 and S8, and table S2, A and B). Baseline measurements of all analytes had associations with subsequent decline in several clinical and imaging measures (Fig. 4 and fig. S7). Baseline CSF mHTT was associated with worsening in TFC (r = 0.40, P = 0.001) and atrophy of whole brain (r = 0.38, P = 0.008), white matter (r = 0.64, P < 0.0001), gray matter (r = 0.59, P < 0.0001), and caudate (r = 0.45, P = 0.002) but had no association with UHDRS total motor score (TMS) and cognitive measures (r < 0.20, P > 0.242). Baseline NfL, in both CSF and plasma, was associated with progression in all measures except TMS [CSF NfL: TFC, r = 0.39, P < 0.0001; symbol digit modalities test (SDMT), r = 0.20, P = 0.039; Stroop word reading (SWR), r = 0.22; Stroop color naming (SCN), r = 0.26, P = 0.016; verbal fluency—categorical (VFC), r = 0.23; whole brain, r = 0.44, P < 0.0001; white matter, r = 0.58, P < 0.0001; gray matter, r = 0.37, P = 0.002; caudate, r = 0.47, P < 0.0001; plasma NfL: TFC, r = 0.46, P < 0.0001; SDMT, r = 0.32, P = 0.009; SCN, r = 0.30, P = 0.014; whole brain, r = 0.56, P < 0.0001; white matter, r = 0.63, P < 0.0001; gray matter, r = 0.47, P < 0.0001; caudate, r = 0.42, P < 0.0001; Fig. 4, fig. S7, and table S2A]. After including age and CAG in the models (fig. S8 and table S2B), associations remained significant only between baseline CSF mHTT and subsequent change in white matter (r = 0.45, P < 0.0001) and gray matter (r = 0.46, P = 0.0001); between baseline CSF NfL and subsequent change in TFC (r = 0.25, P = 0.032), white matter (r = 0.37, P = 0.009), and caudate (r = 0.32, P = 0.007); and between baseline plasma NfL and SDMT (r = 0.29, P = 0.023), whole brain (r = 0.45, P < 0.0001), white matter (r = 0.46, P < 0.0001), and gray matter (r = 0.30, P = 0.023).

Fig. 4. Comparison of prognostic abilities of mHTT and NfL for clinical and imaging measures.

Fig. 4

Open in a new tab

Matrices show the Pearson’s partial correlation coefficients adjusted for age only for associations between baseline values or annualized rate of change (Δ) of each analyte and the annualized rate of change in clinical and imaging measures, each expressed such that higher positive values denote clinical worsening. Color coding displays the magnitude and the direction of the association. Coefficients with corresponding confidence intervals and P values for each combination are provided in table S2. Scatter plots with individual data points for each association are provided in fig. S7. r, Pearson’s partial correlation coefficient; SDMT, symbol digit modalities test; SCN, Stroop color naming; SWR, Stroop word reading; TFC, UHDRS total functional capacity; TMS, UHDRS total motor score; VFC, verbal fluency—categorical.

In contrast, the rate of change in each analyte had no associations with progression in any measure apart from change in TMS (mHTT: r = 0.46, P < 0.0001; CSF NfL: r = 0.32, P = 0.012; Fig. 4 and table S2A). These associations remained after adjustment for age and including CAG (r = 0.43, P = 0.001 and r = 0.18, P = 0.032, respectively; fig. S8 and table S2B).

To compare the relative prognostic ability of the biofluid and imaging biomarkers within a single model, we repeated the random forest analysis, including only the baseline values for biofluid analytes and imaging biomarkers. Using change in cUHDRS as a continuous outcome, biofluid analytes ranked as stronger predictors than imaging biomarkers (Fig. 5). Similar results were obtained for prediction of fast and slow progression (fig. S6B).

Fig. 5. Associations of baseline markers with clinical progression.

Fig. 5

Open in a new tab

Random forest plot for prediction of annualized rate of change in cUHDRS including baseline biofluid and imaging biomarkers as predictors. Higher relative importance score indicates a greater relative importance of the variable in predicting worsening cUHDRS score. Note that this figure uses the same technique as Fig. 3M but this model examines a different set of variables. Relative importance scores were based on the mean decrease in Gini score. Distributions were generated from rerunning the model 100 times. Violin plots represent ranking distributions.

Using a receiver operating characteristics (ROC) curve analysis, we compared the discriminatory ability of each analyte’s baseline value and rate of change, to distinguish between different clinical states: controls versus HD mutation carriers and between preHD versus manifest HD. For all three analytes, rate of change had poor ability to distinguish in either comparison; areas under the curves (AUCs) were about 0.5 (no better than chance). Baseline concentrations had excellent discriminatory ability with AUCs greater than 0.8 (Fig. 6). Under each condition and for each analyte, the AUC for the baseline measurement was significantly greater than that for its rate of change [controls versus HD mutation carriers: P < 0.0001 (CSF mHTT), P < 0.0001 (CSF NfL), and P < 0.0001 (plasma NfL); preHD versus manifest HD: P = 0.001 (CSF mHTT), P < 0.0001 (CSF NfL), and P < 0.0001 (plasma NfL); Fig. 6].

Fig. 6. Discriminatory ability of mHTT and NfL for disease state.

Fig. 6

Open in a new tab

ROC curves comparing the discriminatory ability of baseline values for each analyte and its annualized rate of change to distinguish (A) between healthy controls and HD mutation carriers and (B) between preHD and manifest HD. P values are for comparison between the baseline AUC and rate of change AUC. AUC, area under the curve; HD, manifest Huntington’s disease; preHD, premanifest Huntington’s disease.

Simulating clinical trials with biofluid biomarker surrogate end points

The data thus far suggest that these analytes indicate current clinical state and might have prognostic value for clinical decline. We used longitudinal data from the HD-CSF cohort to run computationally simulated clinical trials using CSF mHTT, CSF NfL, and plasma NfL as possible surrogates for clinical progression. These simulations assume that the intervention-induced change in the analyte emulates the change expected in clinical state by an intervention. Figure 7 depicts the relationships between statistical power, sample size, and trial duration for such trials, using a nominal 20% drug effect per year on the biomarker trajectory. For longitudinal change in NfL, CSF, or plasma, fewer than 100 participants per arm are needed to show an effect over 9 months. More than 10,000 participants per arm would be required to achieve 80% power for a similar trial using the lowering of CSF mHTT as a surrogate outcome over 24 months. Note that this calculation is based on mHTT release and does not apply to any trial in which the intervention reduces its production directly and lowers the protein below baseline values. Comparable simulations for TFC and cUHDRS, commonly used as clinical trial end points, are shown in fig. S9.

Fig. 7. Statistical power, sample size, and trial duration.

Fig. 7

Open in a new tab

Monte Carlo simulations predicting the statistical power of using (A) CSF mHTT, (B) CSF NfL, and (C) plasma NfL in the clinical trial context, contingent on sample size per arm, and trial duration in months. Two-arm parallel design clinical trials with no attrition or placebo effect with a constant effect size of 20% reduction in each analyte per year were simulated. Each pixel represents 1000 simulated clinical trials, generated using generalized mixed-effects models shaped to estimate the longitudinal trajectories of each biomarker (as in Fig. 2). The main effect in each simulation repetition was calculated as the interarm mean difference in the mean change from baseline, using generalized linear models adjusted for CAG. Statistical power was calculated as the proportion of trial simulations with a P < 0.05 for the main effect.

Technical validation of cross-sectional baseline results across assays

CSF mHTT, CSF NfL, and plasma NfL were remeasured in baseline samples using the same methods used for the follow-up samples to perform the longitudinal analysis. Comparing the remeasured values with those previously published at baseline, batch, assay, or storage effects did not affect our samples (Fig. 8 and table S3). We used the remeasured data to replicate our previously published cross-sectional findings (14). This included each analytes’ intergroup differences (fig. S10), associations with clinical and imaging measures (figs. S11 to S13 and table S4), ROC curves and AUC analysis (fig. S13, A and B), analyte correlations (fig. S13, C and D), and EBM (fig. S13, E and F). All results for NfL were similar to those previously published (14). For CSF mHTT, the remeasured values had stronger associations with clinical measures, and, where it previously lacked statistically significant association with brain volumes, the remeasured mHTT was now associated with gray matter and caudate volume (fig. S12 and table S4).

Fig. 8. Time in freezer and comparison of remeasurement of baseline samples.

Fig. 8

Open in a new tab

(A to C) Time in freezer and analyte of interest. (D to F) Associations between repeated baseline measurements for CSF mHTT (purple; D and G), CSF NfL (blue; B and E), and plasma NfL (red; C and F) within controls (n = 20) and HD mutation carriers (n = 60; table S3). (G to I) Bland-Altman plots for remeasurement of baseline samples. The difference between the two measurements is plotted against the mean of the two measurements. Mean difference is represented by the black solid horizontal lines (table S3). The upper and lower 95% confidence intervals are designated with dashed horizontal lines. Each point represents one participant.

Replication of cross-sectional results in follow-up data

We used the samples and data from the 24-month follow-up to examine whether our cross-sectional findings held true in the same cohort 2 years later (figs. S14 to S17). All results for NfL were similar to those previously published and to the remeasured baseline data. For CSF mHTT, we replicated the stronger associations with clinical measures and found further stronger associations with all brain volumes, similar to those for NfL (figs. S15 and S16 and table S5). Our data-driven event-based model for staging participants based on the totality of their baseline data was validated longitudinally, showing that at follow-up, nearly all participants had an EBM stage greater than or equal to the baseline EBM (indicating progression; fig. S18).

Discussion

Here, we present the 24-month results of HD-CSF, a longitudinal study of biofluid biomarkers in HD mutation carriers and matched controls, with longitudinal clinical and MRI data. We have characterized and compared the longitudinal dynamics of mHTT and NfL in CSF in HD, defining the trajectories of these biofluid biomarkers and the inflection points at which they depart from healthy controls in the natural history of HD. Although rates of change in the analytes had some prognostic value, a single measurement at baseline of each analyte exhibited stronger ability to predict subsequent clinical decline, brain atrophy, and disease state. How our prognostic findings add to what was known about these biomarkers is summarized in context in table S6. Using clinical trial simulations, we showed that NfL might be used as an outcome measure of neuronal protection and disease progression, to run trials of feasible duration.

mHTT in CSF, as well as NfL in CSF and plasma, all rose detectably within participants over 2 years. Over the whole course of the disease, mHTT increases rose linearly with age, whereas NfL rose in a more sigmoidal pattern. The disease-associated rise in NfL was more consistent, whereas mHTT was more variable within individuals. The NfL trajectory was distinct from that in healthy controls, with little overlap. This suggests that monitoring these biomarkers against an age-relevant reference range derived from the healthy population might be clinically meaningful. The dynamics of both CSF mHTT and NfL were also CAG dependent, revealing longitudinally the genetic dose-response relationships that we demonstrated previously for plasma NfL in the TRACK-HD cohort (15). Changepoint analysis identified the approximate age and analyte concentration at which HD mutation carriers became detectably different from controls, for a given CAG repeat length. Defining these points of deflection from the trajectories in healthy controls may help us move toward models based on CAG repeat length that could be used to enrich or stratify clinical trial participants and could eventually be used to personalize treatment approaches (18).

Here, we assessed the clinical prognostic potential of biofluid biomarkers against the cUHDRS, a composite measure derived from large cohort datasets to have high signal-to-noise ratio as a longitudinal measure of disease progression (17). mHTT and NfL concentrations predicted change in cUHDRS, affirming their potential as biomarkers of HD progression. We show that both CSF mHTT and NfL each have prognostic value for subsequent clinical decline and brain atrophy, as seen previously for plasma NfL in the TRACK-HD cohort.

For the rate of change in analytes, there was an association with change in TMS but no other measures. That the rate of change in each analyte had lesser prognostic and discriminatory power than their baseline values may appear unexpected. Longitudinal studies of NfL in other genetic neurodegenerative diseases, including dominantly inherited Alzheimer’s disease (AD) (19, 20) and frontotemporal dementia (FTD) (21), have revealed that the rate of change in NfL was a stronger predictor of disease progression, with accelerated rate of change in those who converted from presymptomatic to symptomatic. The HD-CSF study was not designed to assess predictors of conversion from preHD to manifest HD, but the rate of change in plasma NfL did show prognostic value in our comparison of fast versus slow progressors. NfL is an axonal protein but not specific to neuronal subpopulations or to a given disease pathology. It is likely that each disease exhibiting neuronal dysfunction will have a distinct longitudinal NfL profile. It is notable that HD has some of the highest elevated concentrations of CSF NfL compared to other neurological diseases studied to date, greater than concentrations in AD and FTD that have more rapid clinical progression (22). A baseline measure encompasses the totality of a disease’s effects up to the point in a person’s life; by comparison, in a slowly progressive disease, even a 2-year change value captures a relatively small further difference.

The relative importance rankings generated from unbiased random forest analyses suggest that a single measurement of plasma NfL has equivalent, if not superior, prognostic value to that of DBS, one of the strongest predictors of HD progression. Further, the baseline biofluid biomarkers were stronger predictors of clinical progression than in brain volumes. This is supportive of a single measurement of biofluid biomarkers being closely indicative of the dynamic pathological processes driving disease progression.

Our computational clinical trial simulations offer a means to plan future clinical trials that would use the lowering of these biomarkers as surrogate end points. They suggest that a trial using plasma NfL with as few as 100 participants per arm run over 6 months would have over 90% power to show 20% slowing of the expected longitudinal trajectory of the analyte and, under our assumptions, the slowing of clinical decline. Using CSF NfL, the same trial would need to be run over 9 months to achieve the same power. This is promising when compared to the several hundred participants per arm required to achieve the same effect size over 24 months for both cUHDRS and TFC, current clinical end points of the ongoing phase 3 huntingtin-lowering trial (9). Currently, clinical trials tend to focus on specific subsets of patients, whereas our simulations are based on a broad range of the disease spectrum including from preHD and manifest HD. In earlier disease stages, such as for prevention trials in preHD, the sample sizes are likely to be higher because longitudinal change is slower. Dedicated study in a large preHD cohort would be needed to attain the statistical power to model this group independently. Another caveat here is that we assume that the effect size lowering in NfL would be equivalent to the same effect size for improvement in clinical outcome. Until there is a clinically efficacious intervention for HD, this hypothesis cannot be tested. Because of a slower rate of increase, and, more importantly, the intra- and intersubject variability in the change over time, in our simulations, a much larger participant number would be needed to show a deflection in the trajectory of mHTT as a surrogate outcome (not a pharmacodynamic outcome) by slowing disease progression alone. However, given the existing associations of CSF mHTT as a prognostic and pharmacodynamics response marker, it is likely that this measure, too, will turn out to have some predictive utility in the drug trial context. NfL is less variable and appears to be a better marker of HD progression and prognosis than mHTT. However, it is important to reiterate that mHTT retains its intrinsic value as a direct measure of the causative neurotoxin and as a means of assessing the on-target effects of huntingtin-lowering agents. To achieve this purpose, very small participant numbers are required, as shown by our previous cross-sectional power calculations and findings from the first human trial of such an agent (3, 14). Our finding that lower mHTT concentrations predict slower progression is potentially important in this respect. We also show that long-term freezer storage of samples does not adversely affect quantification of mHTT and NfL.

In our previously published baseline analysis (14), CSF mHTT concentration was not cross-sectionally associated with any brain volume measure. However, in our replication of cross-sectional analyses, we found associations between CSF mHTT and caudate and gray matter volume at baseline and strong associations with all brain volumes with the samples from the 24-month collection. This was likely because the performance of the mHTT assay, which has been enhanced through its use in clinical trial programs, has improved so as to reduce variability in the latest analysis. Further, baseline CSF mHTT values also predicted subsequent brain atrophy, confirming that mHTT concentration in CSF per se has prognostic potential in HD, beyond its intrinsic appeal as a therapeutic target. It remains important to note that several participants’ mHTT values were below the lower limit of quantification (LLoQ) of the assay, an indication that more sensitive huntingtin assays will be needed, especially in the realm of preHD and prevention trials.

Our study is not without limitations. Despite being a large longitudinal natural history cohort CSF collection with matched MRI data in HD, the sample number remains modest. We lack the granularity to make predictions of clinical outcome at the individual level. Larger sample numbers and greater follow-up durations may bring us closer to developing models that could inform clinical prognosis. We are also limited to two longitudinal time points. The lack of associations between rate of change in analyte and the rate of clinical decline was likely driven by both sets of rates being derived from the same time period. Having additional time points would permit the comparison of rates of change in analyte with rates of clinical decline in the subsequent time period. For instance, change in NfL in year 1 may predict brain atrophy in year 2. The range of HD mutation carriers in this study does not cover the whole spectrum of HD. This study was not designed to detect the very earliest disease-related alterations nor what happens in the later stages of disease. Nevertheless, our range is broader than what is currently recruited for clinical trials. Last, the sample sizes for TFC and cUHDRS generated from the clinical trial simulations were larger compared to ongoing trials. Given the range in HD clinical severity, the variability is larger in our cohort, and we used a conservative effect size, all of which drive larger sample size.

Efforts are well underway to address these issues: HDClarity, a multinational CSF collection initiative for HD, has amassed more than 600 CSF and plasma samples across the disease spectrum and is now accumulating longitudinal samples over repeated annual intervals (23). There are also ongoing efforts to build a larger information set with a broader range of age and CAG distributions encompassing studies such as HD-CSF, HDClarity, TRACK-HD (24), PREDICT-HD (25), and HD-YAS (26).

These insights into the longitudinal dynamics of mHTT and NfL shed light on the biology of HD in human mutation carriers and will be of immediate value in the design and conduct of disease-modifying clinical trials, especially as we enter the era of prevention trials where qualified surrogate end points will be fundamental (27). The case is far from made that either of the markers is a true surrogate of clinical progression; we will only know this once we have disease-modifying intervention. Looking ahead, some centers are already incorporating blood NfL measurement into shared clinical decision-making in neurological disease (18). Continued study may reveal a role for mHTT and NfL in guiding decision-making for individuals living with HD.

Materials And Methods

Study design

HD-CSF was a prospective single-site study with standardized longitudinal collection of CSF, blood, and phenotypic data [online protocol: digital object identifier (DOI): 10.5522/04/11828448.v1) (14, 28). Eighty participants were recruited. All phenotypic assessment measures were predefined for HD-CSF based on metrics shown to have the largest effect sizes for predicting HD progression (24).

This study was performed in accordance with the principles of the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice standards. Ethical approval was obtained from the London Camberwell St. Giles Research Ethics Committee (15/LO/1917). Before undertaking study procedures, all participants gave informed consent that was obtained by clinical staff. Power calculations, inclusion criteria, and clinical and imaging characterization are as previously described (see Supplementary Methods) (14).

Sample collection and processing

Sample collection and processing were as previously described (14). All collections were standardized for time of day after overnight fasting and processed within 30 min of collection using standardized equipment. Blood was collected within 10 min of CSF and processed to plasma.

Analyte quantification

CSF and plasma NfL were quantified in duplicate using the Neurology 4-Plex B assay on the Simoa HD-1 Analyzer (Quanterix), per the manufacturer’s instructions. A 4x dilution for blood samples was performed automatically by the HD-1 Analyzer, and CSF samples were manually diluted 100x in the sample diluent provided before loading onto the machine. The limit of detection (LoD) was 0.105 pg/ml, and LLoQ was 0.500 pg/ml. NfL was over the LLoQ in all samples. The intra-assay coefficient of variance (CV) (calculated as the mean of the CVs for each sample’s duplicate measurements) for CSF NfL and plasma NfL was 5.0 and 3.7%, respectively. The interassay CVs (calculated as the mean of the CVs for analogous spiked positive controls provided by the manufacturer and used in each well plate) for CSF NfL and plasma NfL were 2.7 and 8.4%, respectively. We previously quantified NfL in the same baseline samples using an enzyme-linked immunosorbent assay (ELISA) (NF-Light, UmanDiagnostics) in CSF and 1-plex Simoa kit (NF-Light, Quanterix) in plasma (14). In both biofluids, agreement between assays was good (Fig. 8).

CSF mHTT was quantified in triplicate using the same 2B7-MW1 immunoassay as at baseline (SMC Erenna platform, Merck) (6). The LoD was 8 fM, and LLoQ was 25 fM. All control samples were below the LoD of the assay except one subject’s baseline remeasured sample. These were imputed as 0 fM for analysis purposes. Twenty-seven (21%) samples were below the LLoQ and were included in subsequent analyses. One preHD had CSF mHTT below the LoD in their remeasured baseline sample and was not included in the analyses. The intra-assay CV for CSF mHTT was 14.1%. Hemoglobin contamination was quantified using a commercial ELISA (E88-134, Bethyl Laboratories) by Evotec. Only one sample (2.186 μg/ml) had hemoglobin over the recommended threshold (2 μg/ml) (29). Assays were run using same-batch reagents, blinded to clinical data.

Statistical analysis

As previously observed (14, 15), NfL distributions were right skewed; therefore, log-transformed values were used for analytical purposes. Because of their known effects on HD, models included age and CAG repeat count as covariates.

Cross-sectional analyses

To validate previous findings and compare assays, we replicated the cross-sectional analyses from the study baseline (14), using remeasured baseline data and newly collected 24-month follow-up data. To investigate intergroup differences, we applied generalized linear regression models estimated via ordinary least squares, with analyte concentration as the dependent variable, with group membership and age as independent variables, and then with group membership, age, and CAG as independent variables. To study associations in HD mutation carriers between the analytes and clinical or imaging measures, we used Pearson’s partial correlations adjusted for age and for age and including CAG in the models. Bias-corrected and -accelerated bootstrapped 95% CIs were calculated for mean differences and correlation coefficients. To understand the discriminatory power of the studied analytes, we produced ROC curves for each analyte to differentiate healthy controls from HD mutation carriers and preHD from manifest HD and compared AUCs, formally using the method suggested by DeLong and colleagues (30).

Longitudinal modeling

For modeling analyte trajectories over time, generalized mixed-effect models were performed, estimated via restricted maximum likelihood, with analyte concentration as the dependent variable. Independent models were developed for healthy controls and HD mutation carriers. Only HD mutation carriers were modeled for mHTT. For CSF mHTT in HD mutation carriers, the model had fixed effects for age and CAG, a random intercept per participant and a random slope for age. A similar model was used for healthy controls for NfL in CSF and plasma. HD mutation carriers were modeled with fixed effects for age (first and second order) and CAG, and random slopes for age were included for both CSF and plasma NfL.

Change-point analysis

We used an offline Bayesian change-point algorithm to estimate the most likely disease time at which a given biomarker changes from a normal to abnormal state. The algorithm was adapted from Zhou et al. (31) and estimates the marginal likelihood that the data over a segment of time are generated by a given underlying model. We explicitly model changes in both the mean and covariance and use a minimally informative prior. The change point is then inferred by a change in likelihood of the underlying model (fig. S4). Because we want to estimate the point of change from normality to abnormality, we use data from all groups (control, preHD, and HD) to fit the model over each time segment.

Rates of change simulations

The longitudinal models above were used to estimate rate of change from simulated data. Model parameters, age and CAG distributions, and sample sizes were mimicked from the HD-CSF cohort. Each simulation was repeated 1000 times and run independently for each analyte for each participant subgroup (healthy controls, preHD, and manifest HD).

Correlation and ROC analyses

Associations of the analytes’ baseline values and of their rates of change, with clinical and imaging changes, were assessed using Pearson’s partial correlations adjusted for age and for age and including CAG in the models. Rates of change were computed as the 24-month follow-up value minus the baseline value divided by the follow-up time in years. To further explore clinical prognostic value, we divided mutation carriers into nominally “fast” and “slow” progressors at the previously described cUHDRS minimal clinically important difference for decline (absolute 1.2-point reduction) (32). Intergroup differences were investigated with generalized linear regression estimated via ordinary least squares, with analyte concentration or rate of change as dependent variable, group membership and age, and then group, age, and CAG as independent variables. Bias-corrected and -accelerated bootstrapped 95% CIs were calculated for correlation coefficients and mean differences. ROC curves were produced, and AUCs were compared for the ability of analytes’ baseline values and rates of change to differentiate healthy controls from mutation carriers and preHD from manifest HD using the method of DeLong et al. (30).

Clinical trial simulation

We used 1000 repetitions, a parallel design without attrition or placebo effect, a pseudo-control arm emulating the observed longitudinal trajectories, and an intervention arm with constant 20% annualized reduction in the analyte of interest. Synthetic datasets were generated with Monte Carlo simulations using mixed-effect models matching the longitudinal models above, including all gene expansion carrier participants. Main effects were estimated as interarm mean difference in the mean change from baseline, adjusted for CAG using generalized linear models estimated as above. Data were analyzed using Stata/MP 16 (StataCorp).

Random forest analyses

To complement our analyses, we used random forest methodology, a supervised ensemble-learning approach based on building multiple independent decision trees from bootstrapped samples (33, 34). This allows the detection of nonlinear relationships and simultaneous ranking of predictors of clinical change in multivariable analyses. We implemented random forests on the following outcomes: annualized rate of change in cUHDRS and cUHDRS as a binary outcome—fast and slow progressors defined as above. Models were first run with biofluid biomarkers including their baseline value and annualized rate of change and then with biofluid and imaging biomarkers baseline value as predictor variables. Age, CAG repeat count, and DBS were included in all models to serve as comparator variables. Each random forest had 1000 trees and was based on bootstrapped samples with replacement, and three randomly sampled predictor variables were considered for splitting each node. Relative importance rankings were based on the mean decrease Gini score across all trees, where a higher mean decrease in Gini indicates greater predictor variable relative importance at predicting outcomes. To explicitly test the stability of our results and generate ranking distributions, we reran the model 100 times, each containing 80% of the possible observations (randomly selected). Random forests were implemented using R randomForest package (34).

Supplementary Material

Supplementary Material

Acknowledgments

We would like to thank all the participants from the HD community who donated samples and gave time to take part in this study. This paper is dedicated in memory of G. Smith (1957 to 2020), a tireless warrior in the fight against HD.

Funding

This work was supported by the Medical Research Council UK (to E.J.W. and S.J.T.), the CHDI Foundation (to E.J.W.), the Huntington’s Disease Society of America (to L.M.B.), the Hereditary Disease Foundation (to L.M.B.), the Wellcome Trust [to S.J.T., Wellcome Collaborative Award In Science (200181/Z/15/Z); and to E.D.V., Wellcome/EPSRC Centre for Medical Engineering (WT 203148/Z/16/Z)], the Department of Health’s NIHR Biomedical Research Centres funding scheme (to D.C.A.), the U.K. Dementia Research Institute that receives its funding from DRI Ltd. (to S.J.T. and H.Z.), Alzheimer’s Society (to S.J.T.), Alzheimer’s Research UK (to S.J.T.), F. Hoffmann–La Roche Ltd. (to E.J.W.), Horizon 2020 Framework Programme (to D.C.A.), Engineering and Physical Sciences Research Council (to D.C.A.), the Swedish Research Council (to H.Z.), the European Research Council (to H.Z.), Swedish State Support for Clinical Research (to H.Z.), and the Innovative Medicines Initiative Joint Undertaking under EMIF grant (to H.Z.).

Footnotes

Author contributions:

E.J.W. designed the study with the input of H.Z., R.I.S., S.S., and S.J.T. F.B.R. and L.M.B. were involved in participant recruitment. Eligibility, clinical examinations, and sample collection were performed by F.B.R., L.M.B., and R.T. Imaging assessments were conceived by R.I.S., E.B.J., and E.D.V. Data were acquired by E.D.V., M.A., and E.B.J. and processed by R.I.S. and E.B.J. A.H. and L.M.B. processed and analyzed the patient samples. F.B.R. developed and performed the statistical analysis. P.A.W. and D.C.A. developed and performed the EBM and change-point analyses. N.G., H.F., R.H., and S.S. developed and performed the random forest analysis. F.B.R., L.M.B., and E.J.W. interpreted the data and wrote the manuscript, and all authors contributed to reviewing the manuscript.

Competing interests:

F.B.R., L.M.B., R.T., E.B.J., P.A.W., D.C.A., S.J.T., R.I.S., A.H., H.Z., and E.J.W. are University College London employees. M.A. is a University College London Hospitals NHS Foundation Thrust employee. E.D.V. is a King’s College London employee. N.G., R.H., H.F., and S.S. are full-time employees of F. Hoffmann–La Roche Ltd. F.B.R. has provided consultancy services to GLG and F. Hoffmann–La Roche Ltd. L.M.R. has provided consultancy services to GLG, F. Hoffmann–La Roche Ltd., Genentech, and Annexon Inc. R.I.S. has undertaken consultancy services for Ixitech Ltd. S.J.T. receives grant funding for her research from the Medical Research Council UK, the Wellcome Trust, the Rosetrees Trust, Takeda Pharmaceuticals Ltd., Vertex Pharmaceuticals, Cantervale Limited, NIHR North Thames Local Clinical Research Network, U.K. Dementia Research Institute, and the CHDI Foundation. In the past 2 years, S.J.T. has undertaken consultancy services, including advisory boards, with Alnylam Pharmaceuticals Inc., Annexon Inc., DDF Discovery Ltd., F. Hoffmann–La Roche Ltd., Genentech, PTC Bio, Novartis Pharma, Takeda Pharmaceuticals Ltd., Triplet Theraputics, UCB Pharma S.A., University College Irvine, and Vertex Pharmaceuticals. All honoraria for these consultancies were paid through the offices of UCL Consultants Ltd., a wholly owned subsidiary of University College London. H.Z. has served at scientific advisory boards for Roche Diagnostics, Wave, Samumed, and CogRx; has given lectures in symposia sponsored by Biogen and Alzecure; and is a cofounder of Brain Biomarker Solutions in Gothenburg AB, a GU Ventures-based platform company at the University of Gothenburg. E.J.W. reports grants from Medical Research Council UK, CHDI Foundation, and F. Hoffmann–La Roche Ltd. during the conduct of the study and personal fees from F. Hoffman-La Roche Ltd., Triplet Therapeutics, PTC Therapeutics, Shire Therapeutics, Wave Life Sciences, Mitoconix, Takeda Pharmaceuticals Ltd., and Loqus23. All honoraria for these consultancies were paid through the offices of UCL Consultants Ltd., a wholly owned subsidiary of University College London. University College London Hospitals NHS Foundation Trust has received funds as compensation for conducting clinical trials for Ionis Pharmaceuticals, Pfizer, and Teva Pharmaceuticals.

Data and materials availability

The data that support the findings of this study are available at DOI: 10.5522/04/12709391 (https://figshare.com/s/c4b7f3aa731aebf9a9d7). All data associated with this study are present in the paper or the Supplementary Materials.

References

  • 1.MacDonald ME, Ambrose CM, Duyao MP, Myers RH, Lin C, Srinidhi L, Barnes G, Taylor SA, James M, Groot N, Farlane HM, et al. A novel gene containing a trinucleotide repeat that is expanded and unstable on Huntington’s disease chromosomes. Cell. 1993;72:971–983. doi: 10.1016/0092-8674(93)90585-e. [DOI] [PubMed] [Google Scholar]
  • 2.Travessa AM, Rodrigues FB, Mestre TA, Ferreira JJ. Fifteen years of clinical trials in Huntington’s disease: A very low clinical drug development success rate. J Huntingtons Dis. 2017;6:157–163. doi: 10.3233/JHD-170245. [DOI] [PubMed] [Google Scholar]
  • 3.Phase 1-2a IONIS-HTTRx Study Site Teams. Tabrizi SJ, Leavitt BR, Landwehrmeyer GB, Wild EJ, Saft C, Barker RA, Blair NF, Craufurd D, Priller J, Rickards H, Rosser A, et al. Targeting Huntingtin expression in patients with Huntington’s disease. N Engl J Med. 2019;380:2307–2316. doi: 10.1056/NEJMoa1900907. [DOI] [PubMed] [Google Scholar]
  • 4.Rodrigues FB, Wild EJ. Huntington’s disease clinical trials corner: February 2018. J Huntingtons Dis. 2018;7:89–98. doi: 10.3233/JHD-189001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Rodrigues FB, Quinn L, Wild EJ. Huntington’s disease clinical trials corner: January 2019. J Huntingtons Dis. 2019;8:115–125. doi: 10.3233/JHD-190001. [DOI] [PubMed] [Google Scholar]
  • 6.Wild EJ, Boggio R, Langbehn D, Robertson N, Haider S, Miller JRC, Zetterberg H, Leavitt BR, Kuhn R, Tabrizi SJ, Macdonald D, et al. Quantification of mutant huntingtin protein in cerebrospinal fluid from Huntington’s disease patients. J Clin Invest. 2015;125:1979–1986. doi: 10.1172/JCI80743. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Southwell AL, Smith SEP, Davis TR, Caron NS, Villanueva EB, Xie Y, Collins JA, Ye ML, Sturrock A, Leavitt BR, Schrum AG, et al. Ultrasensitive measurement of huntingtin protein in cerebrospinal fluid demonstrates increase with Huntington disease stage and decrease following brain huntingtin suppression. Sci Rep. 2015;5:12166. doi: 10.1038/srep12166. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Ducray PS, Frances N, Smart K, Norris D, Kordasiewicz H, Guenther A, Wild E, Schobel S. Translational pharmacokinetic/pharmacodynamic (PK/PD) modeling strategy to support RG6042 dose selection in Huntington’s disease (HD) (S16.005) Neurology. 2019;92:S16.005 [Google Scholar]
  • 9.Hoffmann-La Roche. “A study to evaluate the efficacy and safety of intrathecally administered RO7234292 (RG6042) in patients with manifest Huntington’s disease”. 2019. NCT03761849.
  • 10.Members of the MDS Committee on Rating Scales Development. Mestre TA, van Duijn E, Davis AM, Bachoud-Lévi A-C, Busse M, Anderson KE, Ferreira JJ, Mahlknecht P, Tumas V, Sampaio C, Goetz CG, et al. Rating scales for behavioral symptoms in Huntington’s disease: Critique and recommendations. Mov Disord. 2016;31:1466–1478. doi: 10.1002/mds.26675. [DOI] [PubMed] [Google Scholar]
  • 11.Members of the MDS Committee on Rating Scales Development. Mestre TA, Busse M, Davis AM, Quinn L, Rodrigues FB, Burgunder J-M, Carlozzi NE, Walker F, Ho AK, Sampaio C, Goetz CG, et al. Rating scales and performance-based measures for assessment of functional ability in Huntington’s disease: Critique and recommendations. Mov Disord Clin Pract. 2018;5:361–372. doi: 10.1002/mdc3.12617. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Members of the Movement Disorder Society Committee on Rating Scales Development. Mestre TA, Forjaz MJ, Mahlknecht P, Cardoso F, Ferreira JJ, Reilmann R, Sampaio C, Goetz CG, Cubo E, Martinez-Martin P, Stebbins GT. Rating scales for motor symptoms and signs in Huntington’s disease: Critique and recommendations. Mov Disord Clin Pract. 2018;5:111–117. doi: 10.1002/mdc3.12571. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Members of the MDS Committee on Rating Scales Development. Mestre TA, Bachoud-Levi A-C, Marinus J, Stout JC, Paulsen JS, Como P, Duff K, Sampaio C, Goetz CG, Cubo E, Stebbins GT, et al. Rating scales for cognition in Huntington’s disease: Critique and recommendations. Mov Disord. 2018;33:187–195. doi: 10.1002/mds.27227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Byrne LM, Rodrigues FB, Johnson EB, Wijeratne PA, De Vita E, Alexander DC, Palermo G, Czech C, Schobel S, Scahill RI, Heslegrave A, et al. Evaluation of mutant huntingtin and neurofilament proteins as potential markers in Huntington’s disease. Sci Transl Med. 2018;10:eaat7108. doi: 10.1126/scitranslmed.aat7108. [DOI] [PubMed] [Google Scholar]
  • 15.Byrne LM, Rodrigues FB, Blennow K, Durr A, Leavitt BR, Roos RAC, Scahill RI, Tabrizi SJ, Zetterberg H, Langbehn D, Wild EJ. Neurofilament light protein in blood as a potential biomarker of neurodegeneration in Huntington’s disease: A retrospective cohort analysis. Lancet Neurol. 2017;16:601–609. doi: 10.1016/S1474-4422(17)30124-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Hoffmann-La Roche. “An open-label extension study to evaluate the safety, tolerability, pharmacokinetics, and pharmacodynamics of RO7234292 (ISIS 443139) in Huntington’s disease patients who participated in prior investigational studies of RO7234292 (ISIS 443139)” ( NCT03342053, 2017)
  • 17.TRACK-HD, COHORT, CARE-HD, and 2CARE Huntington Study Group Investigators. Schobel SA, Palermo G, Auinger P, Long JD, Ma S, Khwaja OS, Trundell D, Cudkowicz M, Hersch S, Sampaio C, Dorsey ER, et al. Motor, cognitive, and functional declines contribute to a single progressive factor in early HD. Neurology. 2017;89:2495–2502. doi: 10.1212/WNL.0000000000004743. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Leppert D, Kuhle J. Blood neurofilament light chain at the doorstep of clinical application. Neurol Neuroimmunol Neuroinflamm. 2019;6:e599. doi: 10.1212/NXI.0000000000000599. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Weston PSJ, Poole T, O’Connor A, Heslegrave A, Ryan NS, Liang Y, Druyeh R, Mead S, Blennow K, Schott JM, Frost C, et al. Longitudinal measurement of serum neurofilament light in presymptomatic familial Alzheimer’s disease. Alzheimers Res Ther. 2019;11:19. doi: 10.1186/s13195-019-0472-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Dominantly Inherited Alzheimer Network. Preische O, Schultz SA, Apel A, Kuhle J, Kaeser SA, Barro C, Gräber S, Kuder-Buletta E, LaFougere C, Laske C, Voglein J, et al. Serum neurofilament dynamics predicts neurodegeneration and clinical progression in presymptomatic Alzheimer’s disease. Nat Med. 2019;25:277–283. doi: 10.1038/s41591-018-0304-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Genetic Frontotemporal dementia Initiative (GENFI) van der Ende EL, Meeter LH, Poos JM, Panman JL, Jiskoot LC, Dopper EGP, Papma JM, de Jong FJ, Verberk IMW, Teunissen C, Rizopoulos D, et al. Serum neurofilament light chain in genetic frontotemporal dementia: A longitudinal, multicentre cohort study. Lancet Neurol. 2019;18:1103–1111. doi: 10.1016/S1474-4422(19)30354-0. [DOI] [PubMed] [Google Scholar]
  • 22.NFL Group. Bridel C, van Wieringen WN, Zetterberg H, Tijms BM, Teunissen CE. Diagnostic value of cerebrospinal fluid neurofilament light protein in neurology: A systematic review and meta-analysis. JAMA Neurol. 2019;76:1035–1048. doi: 10.1001/jamaneurol.2019.1534. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.University College London, CHDI Foundation Inc. “HDClarity: A multi-site cerebrospinal fluid collection initiative to facilitate therapeutic development for Huntington’s disease”. 2016. NCT02855476.
  • 24.TRACK-HD Investigators. Tabrizi SJ, Scahill RI, Owen G, Durr A, Leavitt BR, Roos RA, Borowsky B, Landwehrmeyer B, Frost C, Johnson H, Craufurd D, et al. Predictors of phenotypic progression and disease onset in premanifest and early-stage Huntington’s disease in the TRACK-HD study: Analysis of 36-month observational data. Lancet Neurol. 2013;12:637–649. doi: 10.1016/S1474-4422(13)70088-7. [DOI] [PubMed] [Google Scholar]
  • 25.PREDICT-HD Investigators and Coordinators of the Huntington Study Group. Paulsen JS, Long JD, Ross CA, Harrington DL, Erwin CJ, Williams JK, Westervelt HJ, Johnson HJ, Aylward EH, Zhang Y, Bockholt HJ, et al. Prediction of manifest Huntington’s disease with clinical and imaging measures: A prospective observational study. Lancet Neurol. 2014;13:1193–1201. doi: 10.1016/S1474-4422(14)70238-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Scahill RI, Zeun P, Osborne-Crowley K, Johnson EB, Gregory S, Parker C, Lowe J, Nair A, O’Callaghan C, Langley C, Papoutsi M, et al. Biological and clinical characteristics of gene carriers far from predicted onset in the Huntington’s disease young adult study (HD-YAS): A cross-sectional analysis. Lancet Neurol. 2020;19:502–512. doi: 10.1016/S1474-4422(20)30143-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.FDA. Biomarker Qualification: Evidentiary Framework Guidance for Industry and FDA Staff. 2018.
  • 28.Rodrigues FB, Byrne LM, Tortelli R, Johnson EB, Wijeratne PA, Arridge M, De Vita E, Ghazaleh N, Houghton R, Furby H, Alexander DC, et al. Longitudinal dynamics of mutant huntingtin and neurofilament light in Huntington’s disease: The prospective HD-CSF study. medRxiv. 2020:2020.03.31.20045260. doi: 10.1126/scitranslmed.abc2888. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Fodale V, Boggio R, Daldin M, Cariulo C, Spiezia MC, Byrne LM, Leavitt BR, Wild EJ, Macdonald D, Weiss A, Bresciani A. Validation of ultrasensitive mutant huntingtin detection in human cerebrospinal fluid by single molecule counting immunoassay. J Huntingtons Dis. 2017;6:349–361. doi: 10.3233/JHD-170269. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach. Biometrics. 1988;44:837–845. [PubMed] [Google Scholar]
  • 31.Alzheimer’s Disease Neuroimaging Initiative. Zhou W, Zhang J, Ye F, Xu G, Su H, Su Y, Zhang X. Plasma neurofilament light chain levels in Alzheimer’s disease. Neurosci Lett. 2017;650:60–64. doi: 10.1016/j.neulet.2017.04.027. [DOI] [PubMed] [Google Scholar]
  • 32.Trundell D, Palermo G, Schobel S, Long JD, Leavitt BR, Tabrizi SJ. Validity, reliability, ability to detect change and meaningful within-patient change of the cUHDRS. J Neurol Neurosurg Psychiatry. 2018;89:A48 [Google Scholar]
  • 33.Breiman L. Random forests. Mach Learn. 2001;45:5–32. [Google Scholar]
  • 34.Liaw A, Wiener M. Classification and regression by randomForest. R News. 2002;2:18–22. [Google Scholar]
  • 35.International Huntington’s Disease Collaborative Group. Langbehn DR, Brinkman RR, Falush D, Paulsen JS, Hayden MR. A new model for prediction of the age of onset and penetrance for Huntington’s disease based on CAG length. Clin Genet. 2004;65:267–277. doi: 10.1111/j.1399-0004.2004.00241.x. [DOI] [PubMed] [Google Scholar]
  • 36.Sled JG, Zijdenbos AP, Evans AC. A nonparametric method for automatic correction of intensity nonuniformity in MRI data. IEEE Trans Med Imaging. 1998;17:87–97. doi: 10.1109/42.668698. [DOI] [PubMed] [Google Scholar]
  • 37.Freeborough PA, Fox NC, Kitney RI. Interactive algorithms for the segmentation and quantitation of 3-D MRI brain scans. Comput Methods Prog Biomed. 1997;53:15–25. doi: 10.1016/s0169-2607(97)01803-8. [DOI] [PubMed] [Google Scholar]
  • 38.Freeborough PA, Fox NC. The boundary shift integral: An accurate and robust measure of cerebral volume changes from registered repeat MRI. IEEE Trans Med Imaging. 1997;16:623–629. doi: 10.1109/42.640753. [DOI] [PubMed] [Google Scholar]
  • 39.Hobbs NZ, Henley SMD, Wild EJ, Leung KK, Frost C, Barker RA, Scahill RI, Barnes J, Tabrizi SJ, Fox NC. Automated quantification of caudate atrophy by local registration of serial MRI: Evaluation and application in Huntington’s disease. NeuroImage. 2009;47:1659–1665. doi: 10.1016/j.neuroimage.2009.06.003. [DOI] [PubMed] [Google Scholar]
  • 40.Fox NC, Freeborough PA. Brain atrophy progression measured from registered serial MRI: Validation and application to Alzheimer’s disease. J Magn Reson Imaging. 1997;7:1069–1075. doi: 10.1002/jmri.1880070620. [DOI] [PubMed] [Google Scholar]
  • 41.Ledig C, Heckemann RA, Hammers A, Lopez JC, Newcombe VFJ, Makropoulos A, Lotjonen J, Menon DK, Rueckert D. Robust whole-brain segmentation: Application to traumatic brain injury. Med Image Anal. 2015;21:40–58. doi: 10.1016/j.media.2014.12.003. [DOI] [PubMed] [Google Scholar]
  • 42.Ashburner J, Friston KJ. Voxel-based morphometry—The methods. NeuroImage. 2000;11:805–821. doi: 10.1006/nimg.2000.0582. [DOI] [PubMed] [Google Scholar]
  • 43.Christensen GE, Rabbitt RD, Miller MI. Deformable templates using large deformation kinematics. IEEE Trans Image Process. 1996;5:1435–1447. doi: 10.1109/83.536892. [DOI] [PubMed] [Google Scholar]
  • 44.Hobbs NZ, Henley SMD, Ridgway GR, Wild EJ, Barker RA, Scahill RI, Barnes J, Fox NC, Tabrizi SJ. The progression of regional atrophy in premanifest and early Huntington’s disease: A longitudinal voxel-based morphometry study. J Neurol Neurosurg Psychiatry. 2010;81:756–763. doi: 10.1136/jnnp.2009.190702. [DOI] [PubMed] [Google Scholar]
  • 45.TRACK-HD investigators. Tabrizi SJ, Scahill RI, Durr A, Roos RAC, Leavitt BR, Jones R, Landwehrmeyer GB, Fox NC, Johnson H, Hicks SL, Kennard C, et al. Biological and clinical changes in premanifest and early stage Huntington’s disease in the TRACK-HD study: The 12-month longitudinal analysis. Lancet Neurol. 2011;10:31–42. doi: 10.1016/S1474-4422(10)70276-3. [DOI] [PubMed] [Google Scholar]
  • 46.Freeborough PA, Fox NC. Modeling brain deformations in Alzheimer disease by fluid registration of serial 3D MR images. J Comput Assist Tomogr. 1998;22:838–843. doi: 10.1097/00004728-199809000-00031. [DOI] [PubMed] [Google Scholar]
  • 47.Fonteijn HM, Modat M, Clarkson MJ, Barnes J, Lehmann M, Hobbs NZ, Scahill RI, Tabrizi SJ, Ourselin S, Fox NC, Alexander DC. An event-based model for disease progression and its application in familial Alzheimer’s disease and Huntington’s disease. NeuroImage. 2012;60:1880–1889. doi: 10.1016/j.neuroimage.2012.01.062. [DOI] [PubMed] [Google Scholar]
  • 48.Young AL, Oxtoby NP, Daga P, Cash DM, Fox N, Ourselin S, Schott JM, Alexander DC. O2-05-01: A data-driven model of biomarker changes in sporadic Alzheimer’s disease. Alzheimers Dement. 2014;10:P172. doi: 10.1093/brain/awu176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Eshaghi A, Marinescu RV, Young AL, Firth NC, Prados F, Jorge Cardoso M, Tur C, De Angelis F, Cawley N, Brownlee WJ, De Stefano N, et al. Progression of regional grey matter atrophy in multiple sclerosis. Brain. 2018;141:1665–1677. doi: 10.1093/brain/awy088. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Wijeratne PA, Young AL, Oxtoby NP, Marinescu RV, Firth NC, Johnson EB, Mohan A, Sampaio C, Scahill RI, Tabrizi SJ, Alexander DC. An image-based model of brain volume biomarker changes in Huntington’s disease. Ann Clin Transl Neurol. 2018;5:570–582. doi: 10.1002/acn3.558. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Constantinescu R, Romer M, Oakes D, Rosengren L, Kieburtz K. Levels of the light subunit of neurofilament triplet protein in cerebrospinal fluid in Huntington’s disease. Parkinsonism Relat Disord. 2009;15:245–248. doi: 10.1016/j.parkreldis.2008.05.012. [DOI] [PubMed] [Google Scholar]
  • 52.Vinther-Jensen T, Börnsen L, Budtz-Jørgensen E, Ammitzbøll C, Larsen IU, Hjermind LE, Sellebjerg F, Nielsen JE. Selected CSF biomarkers indicate no evidence of early neuroinflammation in Huntington disease. Neurol Neuroimmunol Neuroinflamm. 2016;3:e287. doi: 10.1212/NXI.0000000000000287. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material
EMS137096-supplement-Supplementary_Material.pdf (4.1MB, pdf)

Data Availability Statement

The data that support the findings of this study are available at DOI: 10.5522/04/12709391 (https://figshare.com/s/c4b7f3aa731aebf9a9d7). All data associated with this study are present in the paper or the Supplementary Materials.

RESOURCES