1
Appendix e-1
Functional data pre-processing and independent component analysis (ICA) of RSFC (for all participants)
The following analyses were performed on all three groups of participants.
1.1.Pre-processing.The first four volumes of the EPI images were discarded to allow for stabilization of the magnetic field. Slice-time correction, image registration, motion correction, and spatial smoothing (full width at half maximum = 6 mm) were then performed using Analysis of Functional NeuroImages (AFNI) software, available at e1, e2. After the global linear trend was removed, the imaging data were band-pass filtered using a Fourier transform procedure (0.01-0.08 Hz) to extract the low frequency fluctuations (LFFs). These LFFs were then scaled and converted into percent signal change.
1.2.Independent component analysis (ICA). The pre-processed data were then subjected to independent components analysis (ICA) using the FMRIB Software Library’s (FSL) Melodic software ( ICA separates a set of signals into independent maximally non-Gaussian components (ICs) e3. ICA has been used recently to identify functional connectivity in resting-state functional MRI scans e4, e5. During the decomposition of ICs, the optimal number of ICs for each participant was determined using the Laplace approximation to the Bayesian evidence for a probabilistic principal components analysis model e4,e 6. Each IC has a time course pattern. The voxels whose time series were correlated significantly with the IC’s time course comprise a spatial pattern. The voxel’s value is the loadings of the voxel’s time series on the IC. Sample maps associated with each IC were then transformed to z-scores. The z-scoresreflected the degree to which a given voxel’s time series was correlated with the specific IC’s time course.All data sets were then normalized into the standard Montreal Neurological Institute (MNI) space.
Next, following previous research e5, a modified quantitative procedure was used to select the IC in each participant that matched most closely the spatial maps of interest. This was done as follows: A speech-language network spatial maptemplate was built and validated with a two-step procedure.First, a multi-session temporal concatenation procedure in the Melodic software was performed to identify a common spatial map across all participants. This procedure was similar to the analysis described above, except that here the procedure constrainedthe spatial maps to be the same across all participants. For each resultingz-map, a combined Gaussian/γ model of the probability attenuationof background noise was then applied to the distribution ofvoxel intensities to determine which voxels were significantlymodulated by the associated time course, with a posteriorprobability threshold (of activation) of p0.5 e7. Clusters were considered significant when they includedmore than 500 volumes. Next, group-level spatial maps were selected that included clusters within the bilateral IFC and insula, middle and superior temporal gyrus, inferior parietal lobe, primary and pre-motor areas, basal ganglia, and cerebellum (see Figure e-1A)e8. After the identified spatial map was validated by a group-level ICA map (see Figure e-1B) and a block-designed word reading task-induced activation map (see Figure e-1C), the map was used as a template for individual participants’ ICselection.
The spatial correlation coefficients between the z-scores of the speech-language spatial map template and those of the ICs for each participant were computed. The IC with the largest correlation coefficient was selected as the IC of interest.
The speech-language task was a word reading task. Fifty three-syllable low frequency (< 50 per million) words were selected from a standard word database. The scanning session was composed of seven task-baseline alternative blocks. Within each block, the task lasted for 27.5 s, 33.5 s, or 38.5 s, randomly, whereas the baseline lasted 16.5 s. Within each task trial, the word was first presented on the screen and stayed on the display for 3500 ms, and then the screen was blank for 500 ms. Finally, a row of *s was presented on the screen for 1500 ms. When the word appeared on the screen, participants were required to read it aloud as fast and accurately as possible. When the *s disappeared from the screen, the participants were required to stop reading no matter whether they had finished speaking or not. The voice responses were digitally recorded via an MRI-compatible microphone. All participants were able to read and produce the words within 3500 ms. During baseline, the participants were required to attend to the fixation point “+” on the screen and do nothing else. A sparse sampling technique was used to acquire the images after the 1500 ms of *s whilst participants were not speaking so as to avoid articulation-induced movement artifacts.
Cortical thickness measurements
Cortical surface reconstruction and cortical thickness measurements were performed using the FreeSurfer toolkit ( the structural image data of each participant were motion-corrected and the volume data were obtained. Cerebral grey/white matterwas then segmented and the grey/white matter boundaries were estimated e9. Topological defects in the grey/white boundaries were corrected e10. The boundaries were then used for a deformable surface algorithm designed to find the pial surface with submillimeter precisione11. Cortical thickness measurements were obtained by calculating the distance between the grey/white boundary and pial surface at each of approximately 160,000 points (per hemisphere) across the cortical mantlee11.
The surface representing the grey/white border was “inflated” e12, differences among individuals in the depth of gyri and sulci were normalized, and each participant’s reconstructed brain was then morphed and registered to an average spherical surface representation that optimally aligned sulcal and gyral features across participants e12, e13. Thickness measures were then mapped to the inflated surface of each participant’s reconstructed brain e12. The data were smoothed on the surface using an iterative nearest-neighbor averaging procedure. One hundred iterations were made, which is equivalent to applying a two-dimensional Gaussian smoothing kernel along the cortical surface with a full width at half maximum of 10 mm. The data were then resampled into a common spherical coordinate system e13. Further details of the method are availableelsewheree9, e11, e12, e14, e15.
The RSFC differences between PDS+ patients and fluent controls after the intervention
After the intervention there was only one brain area, the left pars-opercularis (PO, BA44, x, y, z = -54, 14, 16, z = -3.677)(see blue areas in Figure e-2) that showed significantly lower RSFC strength in PDS+ patients than in fluent controls, and no brain areas showed greater RSFC strength in PDS+ patients than in fluent controls.
1
e-References
e1.Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162-173.
e2.Cox RW, Jesmanowicz A. Real-time 3D image registration for functional MRI. Magn Reson Med 1999;42:1014-1018.
e3.Beckmann CF, Smith SM. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans Med Imaging 2004;23:137-152.
e4.Beckmann C, DeLuca M, Devlin J, Smith S. Investigations into resting-state connectivity using independent component analysis. Philosophical Transactions of the Royal Society B: Biological Sciences 2005;360:1001-1013.
e5.Greicius MD, Srivastava G, Reiss AL, Menon V. Default-mode network activity distinguishes Alzheimer's disease from healthy aging: evidence from functional MRI. Proc Natl Acad Sci U S A 2004;101:4637-4642.
e6.Beckmann C, Smith S. Probabilistic independent component analysis for functional magnetic resonance imaging. Medical Imaging, IEEE Transactions on 2004;23:137-152.
e7.Beckmann CF, Jenkinson M, Smith SM. General multilevel linear modeling for group analysis in FMRI. NeuroImage 2003;20:1052-1063.
e8.Indefrey P, Levelt WJ. The spatial and temporal signatures of word production components. Cognition 2004;92:101-144.
e9.Dale AM, Fischl B, Sereno MI. Cortical surface-based analysis. I. Segmentation and surface reconstruction. NeuroImage 1999;9:179-194.
e10.Fischl B, Liu A, Dale AM. Automated manifold surgery: constructing geometrically accurate and topologically correct models of the human cerebral cortex. IEEE Trans Med Imaging 2001;20:70-80.
e11.Fischl B, Dale AM. Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proc Natl Acad Sci U S A 2000;97:11050-11055.
e12.Fischl B, Sereno MI, Dale AM. Cortical surface-based analysis. II: Inflation, flattening, and a surface-based coordinate system. NeuroImage 1999;9:195-207.
e13.Fischl B, Sereno MI, Tootell RB, Dale AM. High-resolution intersubject averaging and a coordinate system for the cortical surface. Hum Brain Mapp 1999;8:272-284.
e14.Kuperberg GR, Broome MR, McGuire PK, et al. Regionally localized thinning of the cerebral cortex in schizophrenia. Arch Gen Psychiatry 2003;60:878-888.
e15.Salat DH, Buckner RL, Snyder AZ, et al. Thinning of the cerebral cortex in aging. Cereb Cortex 2004;14:721-730.
Figuree-1 The speech-language network template. (A)The speech-language network template obtained from themulti-session temporal concatenation procedure of Melodic software. (B) Group-level speech-language network obtained from one-sample t-test after IC selection. (C) An activation map based on an additional word-reading task performed by the same participants. Note the visually detectable similaritiesamongthe three maps. For (A), p0.5 7; for (B) and (C), p < 0.005, uncorrected.
1
Figure e-2 Regions with reductions (blue) or increases (red) of RSFC in PDS+ patients relative to controls based on post-intervention data. The pre-intervention data are drawn as enclosed curves that are superimposed onto the post-intervention data. The green and pink curves represent intervention-induced reductions or increases of RSFC, respectively. The white arrow highlights the left PO which showed an exact overlap between pre- and post-intervention data. p < 0.05, corrected.