Resting-state fMRI fails to detect disease progression in a multicenter randomized clinical trial of Alzheimer's disease
Coimbra Alexandre1, Farshid Faraji1, Alexander de Crespigny1, Lee Honigberg1, Robert Paul1, and David Clayton1

1Research and Early Development, Genentech, South San Francisco, CA, United States

Synopsis

RS-fMRI was implemented in two multicenter clinical trials of a novel therapeutic for AD. Although data of good quality were acquired, none of three functional connectivity metrics (FCMs) showed significant progression associated with disease in placebo-treated patients: changes in connectivity in this mild-to-moderate AD population were less than the measurement precision. Significant cognitive decline and brain atrophy were observed. Test-retest precision was similar to other single-center studies. Operational and acquisition improvements could increase data quality (though difficult in multicenter trials), but more sensitive analysis will be needed for RS-fMRI to be a useful tool for the development of AD therapeutics.

Background

Single-center observational (1,2) and therapeutic (3) studies have used resting-state fMRI (RS-fMRI) to suggest that functional connectivity is altered in patients with Alzheimer’s disease (AD). To test robustness in a multi-center setting, RS-fMRI was included in two Phase II trials of a novel AD drug. We examined: (1) data quality and adherence to acquisition protocol, (2) test-retest variability and minimal detectable difference of three commonly used functional connectivity metrics (FCMs), and (3) sensitivity of FCMs to detect disease progression in the placebo group.

Methods

A total of 572 scans were acquired from 111 patients with mild-to-moderate AD (MMSE 18-26) across 19 sites in North America and Europe participating in two randomized clinical trials (NCT01343966, NCT00997919). Scanners with different field strengths (7 1.5T and 12 3.0T) and manufacturers (12 Siemens, 5 GE, 2 Philips) were used. Patients were scanned prior to initiating therapy and then at weeks 7, 15, 23, 35, 47, 59 and 73. Consecutive pairs of scans acquired at 8-week intervals were treated as test–retest data and pooled for analysis of FCM variability. The scanning protocol consisted of a 2D single-shot GRE-EPI and a 3D high-resolution T1. Sites were allowed to use locally established RS-fMRI scan protocols with the exception of the following requirements: 3000-ms TR, 140 time points, 64x64 matrix, 224x224-mm FOV, 3.5-mm slices, ~160-mm SI coverage, 50-ms (1.5T) or 30-ms (3.0T) TE, 90˚ (1.5T) or 80˚ (3.0T) flip angle, local fat saturation method, and high-order shimming (if available). Subjects were asked to rest, remaining awake with eyes closed. EPI data were brain-extracted, motion-corrected, spatially and temporally filtered and signal-intensity normalized, registered to brain-extracted T1 scans, and subsequently to MNI152 standard space using non-linear registration algorithms. Mean CSF, white matter, global brain signal, and 6 motion-correction parameters were regressed out of the EPI data by general linear model fitting. Coefficient of variation (CoV) in gray matter (GM) was defined as the voxelwise average of the relative standard deviations of the temporal signals. Head motion was assessed with FSL/MCFLIRT. Three FCMs were computed: 1) goodness-of-fit (GOF) for the default mode network, following FSL/MELODIC independent component analysis with automatic dimensionality estimation; 2) average GM z-score of GLM regression using the mean posterior cingulate cortex (PCC) as the seed region (SEED); and 3) correlation coefficients (CORR) between mean PCC and precuneus signals. Intra-class correlation (ICC) for test–retest pairs were calculated. Sample size curves were derived to infer minimal detectable FCM changes. Groupwise annualized rates of change (ARC) were computed.

Results

Deviations from prescribed imaging protocol were observed at 10 sites. Most were minor (e.g. deviations in TE/TR) and occurred consistently per site. Significant variation in raw GM SNR was observed across sites, but was highly consistent between test–retest pairs of scans from any specific site. GM CoV showed less inter-site variation than SNR. Median CoV was 1.0, decreasing to 0.75 after regressing out nuisance variables, decreasing to 0.25 after temporal/spatial filtering. The ICC=0.55 calculated for test–retest pairs was moderate. Sample size calculations for detecting specific percent changes with 80% power at 0.05 significance level for GOF and CORR show that 40–80 patients are needed to detect 20–30% change. For SEED, 20–40 patients are sufficient to detect 7–10% change, since standard deviation of SEED test–retest difference is small relative to the mean. The ARCs in the pooled placebo cohorts were: GOF=0.03±0.56/year, CORR=0.05±0.27/year, SEED=-0.14±0.64/year. No ARC was significantly different from zero for the FCMs, however significant ARC for ADAS-Cog12 score (-7.2±8.7 points/year, p<0.001) and whole brain volume (-2.0±1.0%/year, p<0.001) was observed for this cohort.

Conclusion

In two global Phase II multicenter clinical trials, RS-fMRI data-acquisition protocol was established, data of good quality were acquired and variability of FCMs was assessed. Of three FCMs tested, none showed significant progression associated with disease in placebo-treated patients, indicating that actual changes in functional connectivity in this narrowly defined AD population were less than the measurement precision of the investigated metrics. Meanwhile, significant cognitive decline and atrophy were observed. Test–retest precision of FCMs after pre-processing steps was similar to reports from other single-center test–retest studies of RS-fMRI (1,2). Aside from implementing operational and acquisition improvements to potentially increase SNR across sites, which may be difficult to achieve under the constraints of multicenter trial, a significantly more sensitive analysis approach will be needed for RS-fMRI to prove useful as a tool for the development of novel drugs in AD.

Acknowledgements

No acknowledgement found.

References

1. Greicius M.D., et al. Proc Natl Acad Sci 2004; 101:4637-42.

2. Sheline Y.I., et al. Biol Psychiatry 2010; 67:584-7

3. Li W., et al. NeuroImage 2012, 60:1083-91.

4. Coimbra A., et al. ISMRM 2010.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3756