R. M. Sanchez-Panchuelo, S. Francis, R. Bowtell and D. Schluppeck J Neurophysiol 103:2544-2556, 2010

R. M. Sanchez-Panchuelo, S. Francis, R. Bowtell and D. Schluppeck J Neurophysiol 103:2544-2556, 2010. First published Feb 17, 2010; doi:10.1152/jn.01017.2009

You might find this additional information useful...

This article cites 60 articles, 12 of which you can access free at: http://jn.physiology.org/cgi/content/full/103/5/2544#BIBL Updated information and services including high-resolution figures, can be found at: http://jn.physiology.org/cgi/content/full/103/5/2544 Additional material and information about Journal of Neurophysiology can be found at: http://www.the-aps.org/publications/jn

This information is current as of June 1, 2010 . Downloaded from jn.physiology.org on June 1, 2010

Journal of Neurophysiology publishes original articles on the function of the nervous system. It is published 12 times a year (monthly) by the American Physiological Society, 9650 Rockville Pike, Bethesda MD 20814-3991. Copyright © 2005 by the American Physiological Society. ISSN: 0022-3077, ESSN: 1522-1598. Visit our website at http://www.the-aps.org/. J Neurophysiol 103: 2544–2556, 2010. First published February 17, 2010; doi:10.1152/jn.01017.2009.

Mapping Human Somatosensory Cortex in Individual Subjects With 7T Functional MRI

R. M. Sanchez-Panchuelo,1 S. Francis,1 R. Bowtell,1 and D. Schluppeck2 1Sir Peter Mansﬁeld Magnetic Resonance Centre, School of Physics and Astronomy and 2Visual Neuroscience Group, School of Psychology, University of Nottingham, Nottingham, United Kingdom

Submitted 19 November 2009; accepted in ﬁnal form 15 February 2010

Sanchez-Panchuelo RM, Francis S, Bowtell R, Schluppeck D. are being revisited. Several fMRI studies have investigated the Mapping human somatosensory cortex in individual subjects with 7T cortical representation of the hand using pneumatic or piezo- functional MRI. J Neurophysiol 103: 2544–2556, 2010. First pub- electric stimuli applied to the glabrous skin of the fingertips lished February 17, 2010; doi:10.1152/jn.01017.2009. Functional magnetic resonance imaging (fMRI) is now routinely used to map the (Duncan and Boynton 2007; Francis et al. 2000; Huang and topographic organization of human visual cortex. Mapping the de- Sereno 2007; Overduin and Servos 2008; Schweizer et al. tailed topography of somatosensory cortex, however, has proven to be 2008; Weibull et al. 2008). However, the spatial resolution of more difficult. Here we used the increased blood-oxygen-level-depen- the cortical maps produced in most of these studies is limited dent contrast-to-noise ratio at ultra-high field (7 Tesla) to measure the by the use of large voxel sizes or normalization and averaging Downloaded from topographic representation of the digits in human somatosensory of results across subjects (Francis et al. 2000; Gelnar et al. cortex at 1 mm isotropic resolution in individual subjects. A “traveling 1998; Kurth et al. 2000; Maldjian et al. 1999; Nelson and Chen wave” paradigm was used to locate regions of cortex responding to periodic tactile stimulation of each distal phalangeal digit. Tactile 2008; Overduin and Servos 2008; Weibull et al. 2008). stimulation was applied sequentially to each digit of the left hand from Ultra-high magnetic field (7T) provides access to fMRI data thumb to little finger (and in the reverse order). In all subjects, we with high-spatial resolution and high signal-to-noise ratio found an orderly map of the digits on the posterior bank of the central (SNR). Resolution is particularly important when studying jn.physiology.org sulcus (postcentral gyrus). Additionally, we measured event-related somatosensory cortex as maps of the body surface in S1 (even responses to brief stimuli for comparison with the topographic map- the relatively over-represented maps of the finger tips) are ping data and related the fMRI responses to anatomical images small, and the postcentral gyrus (PCG) is prone to significant obtained with an inversion-recovery sequence. Our results have important implications for the study of human somatosensory cortex and partial volume effects (PVE) due to its large surface-to-volume underscore the practical utility of ultra-high field functional imaging ratio (Scouten et al. 2006) and narrow width (Fischl and Dale with 1 mm isotropic resolution for neuroscience experiments. First, 2000) as well as the interdigitation of the pre- and postcentral on June 1, 2010 topographic mapping of somatosensory cortex can be achieved in 20 gyri (White et al. 1997). Achieving high-spatial resolution min, allowing time for further experiments in the same session. generally means compromising blood-oxygenation-level-de- Second, the maps are of sufficiently high resolution to resolve the pendent (BOLD) contrast-to-noise ratio (CNR), volume cov- representations of all five digits and third, the measurements are erage, or temporal resolution. Statistical power can thus be- robust and can be made in an individual subject. These combined come limiting at lower field, particularly for sensory tasks advantages will allow somatotopic fMRI to be used to measure the where the BOLD signal is low in comparison with visual and representation of digits in patients undergoing rehabilitation or plastic motor tasks, even with a large number of repeated measure- changes after peripheral nerve damage as well as tracking changes in normal subjects undergoing perceptual learning. ments (Schweizer et al. 2008). 7T provides increased BOLD sensitivity (Gati et al. 1997; Van der Zwaag et al. 2009; Yacoub et al. 2001), which can be INTRODUCTION exploited to improve the spatial resolution and/or reduce the number of trials. Higher spatial specificity can also be achieved High resolution measurements of the somatotopic mapping in gradient echo (GE) BOLD data at 7T due to the attenuation of the hand in human cerebral cortex are important for under- of intravascular signal from veins (Gati et al. 1997; Ogawa standing disturbed sensory representations in neurological dis- et al. 1998; van der Zwaag et al. 2009; Yacoub et al. 2001). orders (Butterworth et al. 2003) and for tracking cortical However, the extravascular signal from venules and large veins re-organization in rehabilitation following peripheral nerve that immediately drain the capillary bed may still cause some damage or stroke. Precise knowledge of the spatial layout of spread of the BOLD response. primary somatosensory cortex (S1) is also crucial for under- Here we used fMRI at ultra-high field (7 T) to generate standing how top-down influences, such as attention, modulate complete high-resolution maps of the digit representations in sensory inputs to cortex. the primary somatosensory cortex (S1) in individual subjects. Penfield was the first to map human somatosensory cortex We aimed to maximize efficiency in mapping all five digits and perioperatively (Penfield and Boldrey 1937). With the advent to assess the use of a “traveling wave” paradigm as has been of noninvasive neuroimaging techniques, such as functional widely applied to retinotopic mapping (Engel et al. 1994, 1997; magnetic resonance imaging (fMRI), these early measurements reviewed in Wandell et al. 2007); a similar approach was first Address for reprint requests and other correspondence: D. Schluppeck, used in the somatosensory system to map the ventral surface of Visual Neuroscience Group, School of Psychology, University of Nottingham, the left arm (Servos et al. 1998). In addition, we validated the Nottingham, NG7 2RD UK (E-mail: [email protected]). traveling wave activity against a simple event-related protocol,

2544 0022-3077/10 Copyright © 2010 The American Physiological Society www.jn.org MAPPING HUMAN SOMATOSENSORY CORTEX 2545 assessed the contribution of different tissues to the measured scan parameters: SENSE factor 2 in the right-left direction, TE ϭ 25 responses, and characterized the temporal properties of the ms, flip angle FA ϭ 90°, TR ϭ 2.4 s. A total of 22 contiguous axial event-related fMRI response in S1. slices spanning the right primary somatosensory cortex were acquired The results reveal a robust, consistent map of digit repre- with 1 mm isotropic resolution and a field of view (FOV) of 192 mm sentations in individuals along the posterior aspect of the in the anterior-posterior (AP) direction and 72 mm in the right-left (RL) direction. The reduced FOV in the phase encoding direction central sulcus. In contrast to several previous studies that (RL) required the use of outer-volume suppression (OVS) (Pfeuffer inferred cytoarchitectonic subregions by the existence of iso- et al. 2002) to prevent signal fold-over. The locations of the imaging lated peaks in thresholded statistical maps, our data show no stack and OVS slab are shown in Fig. 3A for a representative subject. such clear segregation (see e.g., Nelson and Chen 2008). Magnetic field inhomogeneity was minimized by using a local, image- based shimming approach (Poole and Bowtell 2008; Wilson et al.

2002). This involved generating a B0-field map from the difference in METHODS ϭ phase of two gradient echo images with echo times of TE1 6msand ϭ Subjects TE2 6.5 ms; skull-stripping the B0-field map (BET, FSL) (Smith 2002) so that field optimization could be focused on brain regions; and Five subjects experienced in fMRI experiments participated in this determining the shim coil currents up to second order needed to study with written consent. Procedures were conducted with approval minimize the field inhomogeneity inside a cuboidal region containing from the University of Nottingham ethics committee. Each subject the central sulcus (highlighted in Fig. 3A). The shim calculation took participated in three scan sessions: two sessions at 7T, one session to Ͻ30 s (Matlab R2007, 2.0 GHz Pentium 4 processor). The efficacy of measure the topographic representation of digits in the somatosensory the image-based shimming procedure was compared with that of the cortex with a traveling wave paradigm, and one session to characterize FASTMAP approach (Gruetter 1993) as implemented on the Philips the spatial and temporal properties of responses to vibrotactile digit

scanner, by acquiring B0-field maps and EP images after shimming on Downloaded from stimulation in an event-related task design. In addition, one session was the same cuboidal target region with each procedure and converting to performed at 3T to obtain high-resolution T1-weighted (MPRAGE, 1 a geometric distortion or spatial offset map in pixel units (deviations mm isotropic) anatomical images of the whole brain. in Hz converted to pixels using a 22.8 Hz bandwidth per pixel of the EPI sequence). Stimuli and paradigm To classify voxels in the functional, T2*-weighted data into tissue type, we also acquired three inversion recovery EP data sets (with the Tactile stimuli were delivered to the digit tips of the left hand using same bandwidth, matrix size, and shim settings to ensure that any five independent, custom-built, MR-compatible, piezoelectric devices. residual geometric distortions were matched to the functional data). jn.physiology.org Each stimulator delivered a somatosensory stimulus at a frequency of The inversion delays (TI) of 450, 1,200, and 2,300 ms were chosen to 30 Hz with ϳ1 mm displacement applied over a ϳ1mm2 area of null white matter, gray matter, and cerebrospinal fluid (CSF), respec- contact (Dancer Design, UK; Fig. 1A). tively. T1-weighted anatomical images with the same slice prescrip- Two experimental paradigms were used: a traveling wave paradigm tion, coverage, and resolution as the functional data (MPRAGE in which the sensory stimulation created a wave of activity across sequence with linear phase encoding order, TE ϭ 2.14 ms, TR ϭ 14 cortical regions containing a somatotopic map of the hand and an ms, FA ϭ 10°, TI ϭ 960 ms, 2 averages) were also acquired during on June 1, 2010 event-related paradigm in which stimulation was simultaneously ap- each 7T scanning session. T1-weighted anatomical images of the plied to all five digits for brief periods. The traveling wave paradigm whole brain (3D MPRAGE, 1 mm isotropic resolution, linear phase involved applying stimuli sequentially to different digits. Each digit encoding order, TE ϭ 3.7 ms, TR ϭ 8.13 ms, FA ϭ 8°, TI ϭ 960 ms) was stimulated for 3 s, with an OFF period of 1.8 s between stimulation were also acquired from each subject at 3T. These images, which of adjacent digits. A whole cycle of stimulation of all five digits thus displayed less B1-inhomogeneity-related signal intensity variation took 24 s. Digits were stimulated in a “forward” sequence from thumb than images acquired at 7T, were used for gray matter segmentation to little finger (digit 1 to digit 5) and in a “backward” sequence and cortical flattening. (starting at the little finger and moving to the thumb; Fig. 1, B and C). In both cases, this resulted in each digit experiencinga3sstimulus Data processing followed by a 21 s OFF period. Similar paradigm designs have been extensively used to form retinotopic maps of the visual cortex (as The effective local geometric distortion due to residual magnetic reviewed in Wandell et al. 2007). In our study, a single traveling wave field inhomogeneity was assessed for the two different shimming fMRI experiment (scan) consisted of 10 cycles, resulting in a scan- techniques by converting field offsets to spatial shifts based on the duration of 240 s. A total of six repeats of the traveling wave fMRI 22.8 Hz bandwidth per pixel of the EPI data. The EPI data were also experiment were performed in a single fMRI session, alternating the visually compared against the MPRAGE data. The image-based repeats between forward and backward stimulation sequences. For the shimming technique significantly reduced geometric distortions (see event-related paradigm, all five digits were simultaneously stimulated RESULTS) and so was used for all subsequent fMRI data acquisitions. for an ON period of 3 s (matching the individual digit stimulation Functional image analysis was performed with a combination of period used in the traveling wave paradigm) with random interstimu- custom-written software (mrTools, NYU, http://www.cns.nyu.edu/ lus intervals of 18, 19, or 20 s. Each functional experiment consisted heegerlab/wiki/; VISTA, Stanford) running in Matlab 7.4 (Math- of 12 trials (228 s duration), and three experiments were run on each works, Natick, MA) and FSL (FMRIB, Oxford, UK) (Smith et al. subject in a single fMRI session. 2004). fMRI data were motion-corrected within and between scans in a given session (for traveling wave and event-related data) using a Functional MRI data acquisition robust motion correction algorithm (Nestares and Heeger 2000), and the fMRI time series was then high-pass filtered to eliminate slow fMRI data were collected on a 7T Philips Achieva system (Philips signal drift. Finally, each voxel’s time series was divided by its mean Medical Systems) using a volume transmit head coil and 16-channel intensity to convert the data from image intensity to units of percent- receive coil. To minimize head motion, subjects were stabilized using age signal change. No spatial filtering was applied to the data. The a customized MR-compatible vacuum pillow (B.U.W. Schmidt) and mean of the T2*-weighted fMRI time-series was calculated and foam padding. Data were collected using T2*-weighted, gradient subsequently calculated statistical maps overlaid on this image. For echo, single shot, echo planar imaging (GE-EPI) with the following additional visualization, the statistical maps from traveling wave and

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org 2546 R. M. SANCHEZ-PANCHUELO, S. FRANCIS, R. BOWTELL, AND D. SCHLUPPECK

A B

C i Downloaded from

ii jn.physiology.org on June 1, 2010 iii

FIG.1. A: single piezo-electric stimulator device. The gray arrow indicates the protruding tip which is applied to skin (ϳ1 mm thickness); direction of movement is in and out of the plane. B: illustration of the traveling wave paradigm. Vibrotactile stimuli were applied to the fingertips in “forward” ordering from thumb to little finger or “backward” ordering from little finger to thumb. The backward scans can be time reversed and time-shifted to cancel any residual hemodynamic lag. C: illustration of expected temporal delays in functional magnetic resonance imaging (fMRI) responses in traveling wave paradigm due to the effect of hemodynamics and experimental design. I: timing diagram for the forward sequence. For voxels responding to stimulation of digit 1, the only delay (w.r.t. the onset of cycle) is due to hemodynamics (ϩph). The additional arrow indicates this delay. Ii: timing diagram for the backward sequence, again with additional arrow indicating the delay (ϩph). Iii: time-reversal of the backward sequence leading to reversal of the inherent hemodynamic delay. The asymmetry in the stimulation paradigm (3 s stimulation is followed by an 1.8 s off period) results in the sequence being delayed (or biased) by 1.8 s. iv: summary of the effects of hemodynamic delay and bias due to the stimulation paradigm. If the time series are simply averaged, the resulting time course will retain a delay (of magnitude ϩbias/2). To remedy this, we advance the phase values of the time-reversed backward scan by 1.8 s giving the difference between the 2 phase values for a given region of interest (ROI) equal to twice the inherent hemodynamic delay.

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org MAPPING HUMAN SOMATOSENSORY CORTEX 2547 event-related data were transformed to the space of the anatomical computed by finding the ordinary least-squares solution to the equa- MPRAGE data acquired at 3 T (see Schluppeck et al. 2005). tion S⅐HT ϩ noise ϭ BT, where B is the measured (mean-subtracted) BOLD time course, S the stimulus convolution matrix, H the to be T Traveling wave data analysis estimated event-related fMRI response, and [] the transpose opera- tion. This yields the best least-squares estimate for the event-related Standard Fourier-based analysis methods (Engel et al. 1994) were fMRI response. Statistical activation maps were computed from the applied voxel-by-voxel to generate somatotopic maps. We computed amount of variance in the original fMRI time course accounted for by the coherence between the time series and the best-fitting sinusoid at events modeled (using the estimated event-related response) to be the 0.04167 Hz stimulus repetition frequency, the phase of the time-locked to stimulus presentations, r2 ϭ 1 Ϫ variance(residual)/ response at that frequency, and the amplitude of the best-fitting variance(original). The residual was calculated as the difference sinusoid. The coherence measures the contrast to noise (Engel et al. between the estimated time course (based on the least-squares solu- 1994, 1997) of the response, taking a value near 1 when the fMRI tion) and the original time course. signal modulation at the stimulus frequency is large relative to the To estimate the hemodynamic lag from the event-related data, the noise (in other frequency components) and a value near 0 when there measured fMRI responses in ROIs defined for each digit from the is no signal modulation or when the signal is small compared with the phase measured in the traveling wave experiment were fitted using noise. The phase measures the temporal delay of the fMRI signal with the model of the hemodynamic response function (HRF) as described respect to the onset of the stimulus cycle and therefore for this in Eq. 1 (Friston et al. 1998; Glover 1999). The best-fit parameters paradigm, reflects the spatial location of the stimulus (which digit) on were estimated by nonlinear least squares fitting and the time-to-peak the hand. A somatotopic map should therefore be visible on the of the fitted response was then determined numerically to provide an cortical surface as a smooth progression of early to late phase values, estimate of the hemodynamic delay. corresponding to the different digits of the hand.

The time series data for the “forward” (from digits 1 to 5) and Cortical segmentation and flattening Downloaded from “backward” (from digits 5 to 1) stimulus sequences were combined following the standard approach used in retinotopic mapping (Engel Inversion recovery (white, gray, and CSF nulled) images were used et al. 1997; Sereno et al. 2001) to remove the effect of the hemody- to visualize the location of the activity on the EP images (data not namic delay in visualizing the somatotopic maps. This process is shown). To display activation on a cortically flattened surface, a illustrated in Fig. 1C. Both the “forward” and “backward” scans were registration algorithm (Nestares and Heeger 2000) was first used to time-shifted (advanced by 2 TRs), to approximately cancel any align the MPRAGE anatomical images acquired at the end of each residual hemodynamic delay at each voxel, the backward scans were functional session to the co-registered fMRI data. The high-resolution then time reversed, and finally the forward and transformed backward MPRAGE images acquired at 7T were then registered to the whole- jn.physiology.org scans were averaged. The resulting mean time series at each voxel in head 3T MPRAGE data. The 3T MPRAGE images were then used in the average data (and its associated phase value) therefore approxi- a subset of subjects to create flattened visualizations of the cortical mately reflected the timing of the forward ordering scans, in which activity restricted to gray matter voxels. Cortical segmentations were digits were stimulated sequentially from thumb (digit 1) to little finger obtained using a combination of tools [SurfRelax (Larsson 2001); (digit 5). We then subdivided the range of phase values in steps of FSL distribution, FMRIB Software Library (Smith et al. 2004)] using ␲ methods previously described (Schluppeck et al. 2006). It is important

2 /5 corresponding to the somatotopic representations of digits 1–5. on June 1, 2010 To incorporate information about the local spatial support of statisti- to note that data were projected onto surface representations as a final cally significant responses, threshold-free cluster enhancement visualization step only. For the flat maps shown in the Results (Fig. 6), (TFCE) (Smith and Nichols 2009) was applied to the coherence map it was ascertained that the gray/white matter segmentations of the (H ϭ 2, E ϭ 0.5, neighborhood connectivity ϭ 6). Regions of interest MPRAGE images for each particular subject were in good agreement (ROIs) comprising voxels in the top percentile of the resulting image with the EPI data (see also Fig. 2). were then interrogated to produce fMRI time series, response (Fou- rier) amplitude, and related statistics for the different scans. RESULTS To estimate the hemodynamic lag from the traveling wave data, the time courses of the average forward and time-reversed backward The image-based shimming method significantly reduced scans (to which no time shift had been applied) were compared in geometric distortions in the EPI data. Figure 2 shows example detail. If we assume that after time reversal of the backward scans the images obtained using the manufacturer’s implementation of time delay of the response is equal, but opposite in sign to that in the the FASTMAP shimming method (B and C) compared with the forward scans, then the difference in phase of the best-fitting sinusoids image-based shimming method (F and G). The image-based to the two time courses corresponds to twice the hemodynamic delay shimming method produces a better correspondence of EPI (this is defined as the uncorrected hemodynamic delay – see RESULTS). data and (undistorted) anatomical MPRAGE images. The reduction of residual geometric distortion is illustrated by the Event-related data analysis field-maps (D and H) and histograms of spatial offset (E and I), A model free deconvolution analysis procedure (Boynton et al. which show that distortion can be restricted to less than one 1996; Buckner et al. 1998; Burock et al. 1998) was used to identify voxel (1 mm) over the shimming region. consistently active areas in the event-related data. Specifically, we Analysis of the traveling wave data showed that the signal used the data-driven analysis approach developed by Gardner et al. from a region of gray matter extending along the posterior bank (2005) to reconstruct event-related fMRI responses to digit stimula- of the central sulcus and the crown of the postcentral gyrus was tion without making any prior assumptions about the shape of the significantly modulated at a frequency corresponding to the response (or that of the underlying hemodynamic response function). 24 s period of stimulation in all subjects. Figure 3 shows data First, the time series of each voxel in the three individual scans was high-pass filtered with a cutoff frequency of 0.01 Hz and converted to from a representative subject (subject 2) with statistical maps percent signal change. Data from the three scans were then concate- superimposed on the mean T2*-weighted images from the nated to allow estimation of the response using all the acquired functional scan. One scan (10 stimulus cycles, 100 volumes) event-related data. The event-related fMRI response for a single was sufficient to form statistical maps, but to improve the SNR, event-type (all 5 fingers stimulated for 3 s) at a given voxel was then responses were averaged across up to six scans before perform-

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org 2548 R. M. SANCHEZ-PANCHUELO, S. FRANCIS, R. BOWTELL, AND D. SCHLUPPECK

A B C D E

F G H I Downloaded from

FIG. 2. Effect of shimming on geometric distortion in echo planar imaging (EPI) images with pixel bandwidth of 22.8 Hz and phase-encoding (and hence distortion) in the right-left direction. A: skull-stripped anatomical image (MPRAGE, 1 mm isotropic resolution). Highlighted square indicates shimming volume. Orientation of stack as shown in Fig. 3A. B: EPI acquired with FASTMAP shimming method. C: EPI from B (hot colors) superimposed on undistorted MPRAGE image (greyscale) from A. Note the significant mismatch between echo-planar and MPRAGE images. D: field map obtained with the FASTMAP shimming jn.physiology.org method. Colors indicate the magnitude of geometric distortion in the phase-encode direction (right-left) in mm. The black square represents the border of the shim box. E: histogram of the geometric distortion, proportion of voxels in shim box (y axis) with given spatial offset (in mm, x axis). F–I as in B–E for data acquired using the image-based shimming method. Note the reduced distortions in the field map (H) with the histogram (I) centered close to 0 and narrowed for the image based shimming method. ing the standard Fourier analysis. The resulting coherence map and robust, with CNRs of Յ40, but response strengths varied (unthresholded) for one subject (average of 6 scans) is shown across voxels as well as subjects. on June 1, 2010 as a semi-transparent overlay in Fig. 3D (left) and following The time courses measured in the traveling wave experi- TFCE correction (right). The coherence map indicates that ments showed high consistency across repeats and locations on some voxels show nearly perfectly sinusoidal signal modula- the somatotopic map. Figure 5 shows the fMRI responses from tion (coherence values close to 1.0). The temporal delay, or the traveling wave paradigm (for subject 2) for two small phase, of the measured periodic response at a given point in the regions of interest (ROI details in figure) sited along the brain was then used to classify voxels corresponding to each posterior bank of the central sulcus in different slices for two stimulated digit. The resulting phase/”digit” map (Fig. 3E) sets of three scan averages [3 “forward” (f) scans; 3 “back- showed consistency in the inferior-superior direction (across ward” (b) scans] that were acquired in interleaved order (f, b, slices in the acquired stack) and an orderly mapping of phase f, b,.). The presence of the large peak at 10 cycle/scan in the values along the postcentral gyrus. Smaller phase delays, Fourier spectrum clearly indicates that the responses are dom- corresponding to digits 1 and 2 were present more laterally and inated by the effects of the somatosensory stimulation. Contri- inferiorly, while larger phase delays, corresponding to digits butions to the response from frequency components other than 3–5, more medially and superiorly. Figure 4 shows correspond- the first harmonic are negligible (Fig. 5, C and F). ing phase maps for all five subjects scanned; for each subject, Figure 6 compares the r2 map calculated from the event- a somatotopic map is clearly visible. Table 1 lists the average related data (A) with a coherence map from the traveling wave spatial extent of each individual digit representation and their analysis (C) on a flattened surface (for subject 5). Similar SD, highlighting the variability in sulcal size and position activation patterns are seen in the primary somatosensory across subjects. The areas of activation were localized to cortex for both the traveling wave and event-related paradigms. cortical gray matter as is evident from the images. For com- The mean time courses of event related data extracted from parison, Table 2 shows the distribution of coherence, mean ROIs, which were independently defined by the somatotopic fMRI response amplitude and CNR across the functional ROIs mapping are shown in Fig. 6B. Note that the estimated hemo- obtained in each subject. The CNR was computed as the dynamic lag from the event-related data does not strongly magnitude of the peak at 10 cycle/scan divided by the average depend on a particular choice of r2 threshold (Table 3). The solid of the magnitude of the high-frequency (33–50 cycle/scan) lines represent the corresponding fits to the hemodynamic re- components in the Fourier spectrum (Fig. 5, C and F). The sponse function (Friston et al. 1998). Figure 6D indicates that fMRI responses in individual voxels were large (with maxi- voxels with the highest r2 values in the event-related data also tend mum values of the order of 10–24% peak-to-peak modulation) to give most significant responses in the traveling wave experi-

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org MAPPING HUMAN SOMATOSENSORY CORTEX 2549

A B CD

central sulcus

E Downloaded from jn.physiology.org

FIG.3. A: location of the imaging stack for zoomed fMRI (red) superimposed on MPRAGE scan for a representative subject. The gray shaded area indicates the on June 1, 2010 location of the outer volume suppression slab and the blue lines show the location of the shim box. B: single slice of the MPRAGE image (1 mm isotropic MPRAGE) collected in the same location as the T2*-weighted EPI data. The red line indicates the location of central sulcus, the green box the cropping used in E. C: mean T2* weighted EP image obtained by averaging 100 repeats of 1 functional MRI scan; because fMRI data contain slight residual geometric distortions with respect to the MPRAGE images, statistical images are superimposed on the average EPI data. Red line, as in B. D: statistical map superimposed on mean EPI image. Transparent colors, coherence with best-ﬁtting sinusoid at frequency of stimulus paradigm. Note values close to 1, indicating high statistical signiﬁcance. Left: no thresholding applied.

Right: statistical image obtained with threshold-free cluster enhancement, TFCE. E: statistical map (from the average of 6 scans) superimposed on mean T2*-weighted EPI data for all slices in stack. Axial slices are numbered from most superior (1) to most inferior (22). Colors indicate the phase values from traveling wave paradigm. Note that the statistical map is thresholded based on application of TFCE to the coherence map (see D). For this subject, the ROI contained 1,998 voxels. The mean coherence value was 0.53 Ϯ 0.13 (minimum-to-maximum range: 0.35–0.94), the mean fMRI response amplitude across all voxels 3.28 Ϯ 3.2% (range: 0.72–21.38), and the mean raw image intensity of the EP image was 22,434 Ϯ 6,957 (range: 4,014–50,071). ment. The uncorrected hemodynamic lag estimated from the trav- response are tightly coupled. Here we perform a simulation to eling wave data is shown in Table 3. The event-related study assess any potential bias in the estimation of the hemodynamic provided an independent estimate of the hemodynamic delay in lag from traveling wave data. each voxel, and so for two subjects, we compared the average Traveling wave signals were simulated using the timing of hemodynamic delay estimated from the event-related data to that stimulus delivery and a model of the HRF, H(t), formed from obtained from the traveling wave analysis (Table 3) using the the sum of two gamma functions (Friston et al. 1998) methods and correction described in Fig. 7. The uncorrected hemodynamic lag estimated from the traveling wave data is t a1 ͑t Ϫ d ͒ t a2 ͑t Ϫ d ͒ H͑t͒ ϭ ͩ ͪ expͫϪ 1 ͬ Ϫ cͩ ͪ expͫϪ 2 ͬ (1) significantly shorter than that measured from the event-related d b d b paradigm. 1 1 2 2 ϭ ϭ where d1 a1b1 and d2 a2b2. Data were simulated for a Assessing systematic errors arising using Fourier analysis to range of HRF time-to-peak (TTP) values by varying parameter ϭ estimate hemodynamic lag a1 from 3 to 9 while other parameters were fixed at a2 2; a1; ϭ ϭ ϭ b1 0.9 s; b2 0.9 s; c 0.35 (Glover 1999) (Fig. 7D). The The method of estimating the hemodynamic lag from the simulated data were then analyzed using the Fourier-method traveling wave data relies on the assumption that the fMRI time and the corresponding TTP of the hemodynamic delay was series can be well approximated by a sinusoidal function the estimated. Based on this analysis, we calculated a correction period of which equals the length of the cycle (24 s) and that factor for the traveling wave data. Figure 7 shows the steps of the hemodynamic delay and phase of the fitted sinusoidal the simulation for digit 1. The fMRI time series is simulated

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org 2550 R. M. SANCHEZ-PANCHUELO, S. FRANCIS, R. BOWTELL, AND D. SCHLUPPECK

4 3 5 2 1

subject 1 6 scans B

4 3 2 5 1

subject 2 6 scans

C Downloaded from C 4 3 5 2 1

jn.physiology.org subject 3 6 scans

3 4 2 on June 1, 2010 5 1 subject 4 2 scans E

3 4 2 5 1

subject 5 6 scans

FIG. 4. Statistical maps (average of 6 scans except for subject 4 with 2 scans) for each of 5 subjects scanned, superimposed on mean T2*-weighted EPI data for all slices in the stack, equivalent to Fig. 3E. Axial slices are numbered from most superior (1) to most inferior (22). Colors indicate the phase values from traveling wave paradigm. Statistical maps are thresholded using TFCE. from the asymmetric HRF (Eq. 1) convolved with the stimulus (Fig. 1C) that was included in the estimate of the uncorrected input function. The simulated fMRI response for the backward hemodynamic delay. scan is time reversed and the responses for each of the forward and To estimate the inaccuracy in this uncorrected measure of time-reversed backward scans are then fitted to a sinusoid of hemodynamic lag, the uncorrected hemodynamic lag from the period 24 s. Assuming the hemodynamic lag is equal and traveling wave data (as estimated from the phase difference in opposite for the forward and time-reversed backward scans, the the sinusoidal fits) was plotted against the lag of the simulated hemodynamic lag can then be estimated as half the difference HRF (as measured by the time to peak). For our simulated between the delays found for the two conditions (light gray parameters, the hemodynamic lag from the traveling wave lines in Figs. 7A and 7C). However, we must also take into analysis (y) is related to the lag of the simulated HRF (x)bya account an additional delay of 1.8 s for the time reversed data parabolic relationship (y ϭ 0.03x2 ϩ 0.70x Ϫ 0.51), reflecting

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org MAPPING HUMAN SOMATOSENSORY CORTEX 2551

TABLE 1. Spatial extent of digit representations in traveling wave maps averaged over subjects

D1 D2 D3 D4 D5

M-L, mm 26.2 Ϯ 5.9 23.2 Ϯ 4.0 20.4 Ϯ 5.0 20.6 Ϯ 4.4 15.2 Ϯ 4.4 A-P, mm 22.6 Ϯ 2.1 21.2 Ϯ 4.6 16.0 Ϯ 1.9 15.6 Ϯ 2.9 12.5 Ϯ 4.3 F-H, mm 18.4 Ϯ 0.9 18.0 Ϯ 0.9 16.6 Ϯ 1.9 16.6 Ϯ 2.6 15.2 Ϯ 2.9 Volume, mm3 313 Ϯ 140 373 Ϯ 226 141 Ϯ 47 225 Ϯ 76 120 Ϯ 60

Spatial extent of digit representations in travelling wave maps averaged over subjects for the medio-lateral (M-L), anterior-posterior (A-P), and foot-head (F-H) directions. Mean Ϯ SD for each digit averaged across subjects. an underestimation of the hemodynamic lag in the traveling strengths due to the supralinear increase in CNR in the microvas- wave data. This relationship was used to correct the traveling culature (Gati et al. 1997; Yacoub et al. 2003) and suppression of wave hemodynamic delay, Table 3. Both methods yield esti- intravascular signal in draining veins distant from the site of ϳ mates of hemodynamic delay that are comparable, but the neuronal activity due to shortening of venous blood T2 ( 7 ms) delay from the event related paradigm is still longer. (Duong et al. 2003; Gati et al. 1997; Ogawa et al. 1998; Thulborn et al. 1982; Yacoub et al. 2003). DISCUSSION Another factor that may have improved spatial speciﬁcity in this experiment compared with previous studies is the para- These results demonstrate that robust maps of the represen- digm design. First, tactile stimuli were delivered with a piezo-

tation of the fingers in human somatosensory cortex can be Downloaded from obtained with 1 mm isotropic resolution at 7T using a traveling electric device at a stimulation frequency chosen to maximize the wave paradigm and focal piezoelectric stimulation applied at response in rapidly adapting somatosensory receptors the recep- 30 Hz to the fingertips. Areas of significant signal modulation tive fields of which are known to densely tile the glabrous skin on were confined to the posterior aspect of the central sulcus and the digits (Darian-Smith 1982). Second, the traveling wave para- the crown of the postcentral gyrus and showed an orderly digm (Engel et al. 1997) is a form of differential paradigm and is progression of phase of modulation reflecting the ordered likely to suppress nonspecific activation common to all digits, somatotopic representation of digits 1–5 in primary somato- such as the BOLD signal in large veins draining from tissue jn.physiology.org sensory cortex (S1). The representation of digit 1 (thumb) was spanning multiple digit representations in the postcentral gyrus, most inferior and lateral, whereas digits 2–5 were represented thus improving digit specific mapping. We compared the extent of at increasingly superior and medial locations (Figs. 3 and 4). activation maps for the traveling wave and event-related para- Unthresholded statistical maps confirm that the pattern of digms and found these to be similar, but the estimates of the activation is almost exclusively on the postcentral side of the hemodynamic lag were significantly shorter for the traveling wave central sulcus and the crown of the postcentral gyrus. This is in paradigm (Table 3). The estimated hemodynamic delay is only on June 1, 2010 contrast to some previous reports that have found that somato- slightly affected by the choice of thresholds. We performed a sensory representations extend further anteriorly to the precen- simulation to correct for the systematic errors introduced by the tral aspect of the sulcus (Moore et al. 2000). Fourier analysis in the estimating the hemodynamic lag. This The fMRI responses in individual voxels were large (from 10 to reduced the discrepancy, but a difference of ϳ1 s still remains. 24% peak-to-peak modulation, Table 2) and robust, with CNRs of This discrepancy may in part be due to no slice-time correction Յ40. This can be compared with signal modulations of 3–4% for being applied in our data analysis (to avoid interpolation effects), most sensory fMRI studies at 3T (Francis et al. 2000; Stippich and this could give rise to an average 1.2 s error in the measured et al. 1999) using 3 mm isotropic voxels. This large signal hemodynamic lag. The greater lag observed in the event-related modulation at 7T is expected if a linear increase in BOLD signal estimates may also indicate an increased venous contribution (de with field strength can be realized (Gati et al. 1997; Yacoub et al. Zwart et al. 2005; Hulvershorn et al. 2005) that is reduced by 2001). Here the increased CNR at ultra-high field has been using a traveling wave paradigm. exploited to achieve high-spatial resolution, reducing partial vol- The traveling wave paradigm has previously been applied to uming of activated tissue with nonactivated gray matter and white the study of somatosensory cortex only at coarse resolution matter/CSF. Potentially this can lead to larger and more consistent using long cycle lengths (Huang and Sereno 2007; Overduin hemodynamic response modulations arising from cortical gray and Servos 2004) or to form an incomplete map of all digits of matter. In addition, 7T offers intrinsic improvements in spatial the hand. Overduin and Servos (2008) used a similar design to specificity for GE BOLD data compared with lower field stimulate single digits from tip to base but did not use the Fourier-based analysis methods described in this study. TABLE 2. Details of fMRI response for traveling wave paradigm The spatial resolution of earlier functional MRI measurements (typically with voxel volumes of 10–30 mm3) (Kurth Coherence Value Response Amplitude, % et al. 2000; Maldjian et al. 1999; Nelson and Chen 2008) is Subject (minimum Ϫ maximum) (minimum Ϫ maximum) CNR unfavorable with respect to the size of the somatotopic map. In 1 0.51 Ϯ 0.12 (0.35–0.89) 2.8 Ϯ 1.4 (0.6–11.9) 36.7 Ϯ 7.0 our high resolution (1 mm isotropic) functional measurements, 2 0.53 Ϯ 0.13 (0.35–0.94) 3.3 Ϯ 3.2 (0.7–21.4) 39.5 Ϯ 11.6 the extent of the sensory maps spanning D1–D5 was greater in 3 0.45 Ϯ 0.11 (0.39–0.89) 3.5 Ϯ 1.5 (1.2–15.6) 32.5 Ϯ 6.3 the medial-lateral than AP or foot-head directions (Table 1). 4 0.39 Ϯ 0.08 (0.30–0.67) 4.2 Ϯ 2.0 (1.6–24.6) 19.4 Ϯ 3.4 5 0.57 Ϯ 0.13 (0.27–0.89) 2.6 Ϯ 1.3 (0.7–12.5) 41.7 Ϯ 5.4 Digits 1 and 2 occupied the largest extent within the somatotopic map, whereas digit 5 has the smallest representation fMRI, functional magnetic resonance imaging; CNR, contrast-to-noise ratio. (Table 1). Note that the activation lies along the tortuous

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org 2552 R. M. SANCHEZ-PANCHUELO, S. FRANCIS, R. BOWTELL, AND D. SCHLUPPECK

A D

B E

C F Downloaded from jn.physiology.org

ROI 1: subject 2, slice 11, volume 51 mm ROI 1: subject 2, slice 9, volume 36 mm

phase: 2.8 rad (digit 1) phase: 2.8 rad (digit 4) on June 1, 2010

coherence: 0.69 coherence: 0.57

response amplitude: 6.68 % response amplitude: 3.98 %

FIG. 5. Time course plots from 2 small ROIs (columns) for 2 different stimulus conditions (rows). Details of the ROI are provided at the bottom of the figure. A: each trace, mean fMRI response in ROI1 for each of 3 scans with a forward stimulus (advancing from digits 1–5). B: fMRI responses in ROI1 for backward sequence. C: amplitude spectrum of average time course across all scans in ROI1. Black circle, magnitude at stimulus alternation frequency. Thick black line, high-frequency components used to calculate contrast-to-noise ratio (CNR). D–F: corresponding data for ROI2. postcentral gyrus (Fig. 5) and with 1 mm isotropic resolution Chen 2008; Overduin and Servos 2008), known to exist from this leads to a band which is ϳ29 mm in length in the anatomy, histology, and human/non-human single-unit physi- postcentral gyrus for D1–D5 mapping. The failure to identify ology. These areas cannot currently be accurately identified clear somatotopic maps in previous studies may have been due using MRI in vivo but can roughly be assigned based on to the limits imposed by sampling a relatively small spatial anatomical MR images: area 1 is located at the crown of the map with comparatively coarse resolution. For example use of postcentral gyrus, area 2 in the postcentral sulcus, area 3a in the 3 mm isotropic resolution would yield somatotopic maps fundus of the central sulcus, and area 3b at the rostral bank of containing at best a third of the number of voxels measured in the postcentral gyrus. Microelectrode mapping studies in pri- our study (assuming the best 1-dimensional tiling); at worst mates (e.g., Pons et al. 1987; Sur et al. 1984) have suggested 1/33 taking into account volume. Even in those previous that multiple representations of the body arise within primary studies, where high-spatial resolution was used, the reduced somatosensory cortex with two complete body surface repre- contrast-to-noise necessarily reduced statistical power sentations occupying cortical fields 3b and 1. Additionally it (Schweizer et al. 2008). In addition in several previous studies, has been suggested that area 2 contains an orderly representa- results were averaged across-subjects, often registered into tion of predominantly deep body tissues, whereas area 3a may Talairach/MNI space. This causes blurring but also doesn’t constitute a fourth representation. Several studies have com- ensure exact alignment of anatomical structures across sub- pared functional data to these areas using probabilistic cytoar- jects, especially when their spatial extent is relatively small chitectonic maps derived from postmortem brains (see e.g., (Fischl and Dale 2000). Geyer et al. 1999; see also Geyer et al. 2000; Grefkes et al. Previous studies, some based on relatively coarse resolution 2001; Schleicher et al. 2000). However, thresholding of statis- measurements, have emphasized the division of significant tical maps from fMRI experiments at a single significance clusters of fMRI responses in S1 into four anatomical subdi- value can lead to erroneous conclusions about multiple foci of visions (areas 1, 2, 3a, and 3b) (Moore et al. 2000; Nelson and activation corresponding to distinct anatomical subregions (the

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org MAPPING HUMAN SOMATOSENSORY CORTEX 2553

A C

3 4 2 0 5 1

4.3

0 2 Phase (rad) B D

r2>0.16, ttp=6.15 r2>0.09, ttp=6.00 r2>0.06, ttp=5.85 r2>0.01, ttp=5.85 Downloaded from jn.physiology.org FIG.6. A: coherence map based on the r2 values from the event-related data (subject 5). B: mean event-related fMRI responses in S1 for subject 5. Symbols, mean event-related time course for voxels in the functional ROI defined in the traveling wave experiment (error bars, SE, are smaller than the plot symbols). Colors, data were thresholded according to the distribution of r2 value obtained for each voxel in the event-related analysis [ROI contain voxels with r2 over 0, 25, 50, and 75% of the r2 distribution (as for Table 3 for all subjects)], results plotted separately for the lowest (dark) to highest (bright) threshold. Solid lines, corresponding fits to the hemodynamic response function. ttp indicates the hemodynamic delay of the curve. C: coherence map from the traveling wave analysis (subject 5). Similar activation patterns can be seen for the primary somatosensory cortex for the event-related paradigm (A). D: direct comparison of the traveling wave and event-related data for subject 5 indicates that voxels with the highest r2 values (obtained in the event-related experiment) also tend to give most significant responses in the traveling wave experiment (linear regression: y ϭ 0.29x Ϫ 0.04, Pearson correlation:0.39). Color, image intensity of underlying on June 1, 2010

T2*-weighted EPI data. iceberg effect). Further, the spatial information in the proba- ical information (referring to the layout of the cyto-architec- bilistic atlases currently does not capture the anatomical pattern tonic subregions in postmortem samples in humans or mon- of sulci and gyri, which makes accurate assignment to Brod- keys), or using statistical maps that were computed and then mann regions difficult. thresholded to reveal several peaks that in turn were assigned, Given the size of the somatotopic maps of digit representa- post hoc, to the different subregions. tions that we find, the spatial resolution and signal-to-noise Eickhoff et al. (2007) show that with careful consideration limits imposed by fMRI data acquisition, and analysis methods of the layout of the maps in somatosensory cortex (S2) and of previous studies, we believe that the assignment of cyto- with reference to the monkey literature, such anatomical- architectonic subregions based on the currently available sim- functional assignments can be attempted, but in principle these ple mapping data alone is extremely problematic. Additionally, measures require an even higher spatial resolution than for S1. the anatomical variability in somatosensory cortex is known to In the study by Overduin and Servos (2004) where an attempt be quite large (see e.g., White et al. 1997). In all previous was made to map out the cytoarchitectonic subregions, the imaging studies where this assignment was attempted, areas somatotopic maps were found to cover a distance of 50 mm in were classified either purely on the basis of geometric/anatom- the rostral-caudal direction (10 coronal slices of 5 mm slice

TABLE 3. Hemodynamic delay of traveling wave and event-related data

Hemodynamic Delay, s

Event-Related Estimates, s Uncorrected Traveling Corrected Traveling Subject Wave, s Wave, s ROI A ROI B ROI C

3 3.92 Ϯ 0.04 5.09 Ϯ 0.04 6.85 Ϯ 0.50 6.82 Ϯ 0.54 6.74 Ϯ 0.44 5 3.12 Ϯ 0.32 4.33 Ϯ 0.32 6.22 Ϯ 0.23 6.15 Ϯ 0.23 5.95 Ϯ 0.23

Values are means Ϯ SD. Hemodynamic delay measured using traveling wave (both uncorrected and corrected) and event-related data. ROIs were deﬁned for each digit from the traveling wave statistical map. For the event-related data, the hemodynamic delay is estimated for each digit ROI at different r2 thresholds (ROI A–C threshold levels chosen to include voxels with r2 values Ͼ 25, 50, and 75% of the r2 distribution).

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org 2554 R. M. SANCHEZ-PANCHUELO, S. FRANCIS, R. BOWTELL, AND D. SCHLUPPECK

C Downloaded from

DEHRF with different a1 Phase vs TTP 0.05 10 a1 3

0.04 4 8 jn.physiology.org 5 6 7 0.03 8 6 9

0.02 4 fMRI response 0.01 2 on June 1, 2010 Phase difference (s)

0 0

−0.01 −2 0 5 10 15 20 25 0 2 4 6 8 10 Time (s) TTP (s)

FIG.7. A: the stimulus input function of the forward scan of digit 1 (black line) convolved with the double gamma variate hemodynamic response function ϭ ϭ ϭ ϭ ϭ [HRF; 6 s time-to-peak (TTP); Eq. 1, a1 6, a2 12, b1 0.9 s, b2 0.9 s, c 0.35] to simulate the forward fMRI response (thick black line). The simulated waveform (blue line) is then fitted to a sinusoidal wave of period 24 s (light gray line). The sinusoidal waveform shows a reduced time-to-peak compared with the model response. B: the stimulus input function of the backward scan for digit 1 (dark gray line), and the convolution with the double gamma variate HRF with 6 s TTP (as shown in A) to simulate the backward fMRI time course (thick black line). C: time reversed backward scan for digit 1 (thick black line) and fitted sinusoidal waveform (light gray line). The sinusoidal waveform shows a delayed time to peak compared with the double gamma variate HRF. The hemodynamic delay from the traveling wave analysis is estimated by first calculating the time between peaks of the 2 fitted sinusoid waveforms (light gray lines) in A and C. A delay of 1.8 s is then added to account for the time between stimulation of each digit (as shown in Fig. 1C). The resulting time difference is then ϭ divided by 2 to give an estimate of the hemodynamic delay. D: simulated double gamma variate HRF with a1 3–9tosimulate the range of hemodynamic delays found in the brain. E: simulated TTP of hemodynamic delay as determined from the traveling wave analysis [from the phase difference in the sinusoidal fits (A) and (C)] vs. the true hemodynamic TTP from the simulated HRF (D). Data show a parabolic relationship of y ϭ 0.03x2 ϩ 0.70x Ϫ 0.51. thickness), over double that found in our study and in previous the fingers. Given the relatively low indentation amplitude of work (e.g., Moore et al. 2000; Nelson and Chen 2008). our stimuli to the surface of the fingers, we did not expect to In this study, we found statistical maps with spatially con- see responses dominated by neurons thought to be restricted to tiguous representations of all digits. Stimulation of the finger- areas 3a and 2. tips mainly activated the rostral bank of the postcentral gyrus Although the pattern of the somatotopic maps is clearly (defined operationally as area 3b), in agreement with a previ- visible in the axial planes in which our EPI data were acquired, ous study using piezoelectric stimulation where “apparent” with the direction of changing phase values in the map running area 3b has shown some evidence of somatotopic organization along the central sulcus, surface rendering can provide a (Schweizer et al. 2008). However, the spatial extent of the clearer sense of the spatial layout of topographic maps. There- maps reported by Schweizer et al. (2008) was much smaller fore to ease visualization of the two-dimensional digit topog- than our data, suggesting that previous functionally defined raphy, the statistical maps were rendered on inflated and maps have greatly underestimated the cortical representation of flattened representations of the cortical surface (Fig. 6). For

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org MAPPING HUMAN SOMATOSENSORY CORTEX 2555 such visualization, accurate alignment between functional data be a useful method to measure spatial changes in the cortical and anatomical images is essential. The image-based shimming representations on the millimeter scale in, e.g., patients under- method has been shown to reduce distortions to less than one going rehabilitation or plastic changes after peripheral nerve voxel (see METHODS and Fig. 2). Despite the relatively small damage as well as tracking changes in normal subjects under- residual geometric distortion between EP and anatomical im- going perceptual learning. ages, problems in assigning responses to different sides of the postcentral sulcus may still result, although these can to some ACKNOWLEDGMENTS extent be overcome by image processing and careful segmen- We thank our three anonymous reviewers for helpful comments on the tation and cortical unfolding (e.g., any voxel the activation of manuscript. which is partial volumed over voxels, or voxels where there is simultaneous activation on both sides of the sulcus, could be GRANTS excluded from further analysis). To enable more detailed This work was funded by the Medical Research Council with grants to R. surface-based analysis, segmentations have to be based on Bowtell, S. Francis, R. Sanchez-Panchuelo and supported by Biotechnology anatomical images with higher spatial resolution (e.g., 0.5 mm and Biological Sciences Research Council Grant BB/G008906/1 to D. Schluppeck and S. Francis. R. Sanchez-Panchuelo is supported by a Marie isotropic voxels), and future studies will assess the feasibility Curie Early Stage Research Training Fellowship, D. Schluppeck by a Research of using whole head anatomical MPRAGE data acquired at 7T Councils UK Academic Fellowship. in conjunction with normalization methods that can correct the nonuniform signal intensity (Van de Moortele et al. 2009). REFERENCES Additionally, small amounts of blurring due to spatial spread- Boynton GM, Engel SA, Glover GH, Heeger DJ. Linear systems analysis of

ing of the hemodynamic response or caused by resampling in functional magnetic resonance imaging in human V1. J Neurosci 16: Downloaded from the motion correction step can aggravate such effects. Signal 4207–4221, 1996. may appear in the precentral gyrus due to pial veins that drain Buckner RL, Goodman J, Burock M, Rotte M, Koutstaal W, Schacter D, Rosen B, Dale AM. Functional-anatomic correlates of object priming in the postcentral gyrus that share their extravascular signal with humans revealed by rapid presentation event-related fMRI. Neuron 20: the precentral gray matter. However, note that the signal in the 285–296, 1998. precentral aspect of the sulcus in the traveling wave data shown Burock MA, Buckner RL, Woldorff MG, Rosen BR, Dale AM. Random- in Fig. 6C follows a progression of phase values consistent ized event-related experimental designs allow for extremely rapid presen- with the digit representations, indicating that such signal is tation rates using functional MRI. Neuroreport 9: 3735–3739, 1998. jn.physiology.org Butterworth S, Francis S, Kelly E, McGlone F, Bowtell R, Sawle GV. likely caused by blurring due to venules specific to each digit, Abnormal cortical sensory activation in dystonia: an fMRI study. Mov and not larger veins (which should lead to larger swaths of Disord 18: 673–682, 2003. identical values in the phase maps). Also the spatial extent of Darian-Smith I. Touch in primates. Annu Rev Psychol 33: 155–194, 1982. activity on the precentral sulcus is larger for the event-related de Zwart JA, Silva AC, van Gelderen P, Kellman P, Fukunaga M, Chu R, Koretsky AP, Frank JA, Duyn JH. Temporal dynamics of the BOLD data, suggesting an increasing venous contribution to this data. fMRI impulse response. Neuroimage 24: 667–677, 2005. It remains to be seen whether a further reduction in image Duncan RO, Boynton GM. Tactile hyperacuity thresholds correlate with on June 1, 2010 voxel size below 1 mm can lead to significant improvements in finger maps in primary somatosensory cortex (S1). Cereb Cortex 12: the achievable resolution in cortical maps. Artifacts due to 2878–2891, 2007. subject motion, respiration, and the cardiac cycle may domi- Duong TQ, Yacoub E, Adriany G, Hu XP, Ugurbil K, Kim SG. Micro- vascular BOLD contribution at 4 and7Tinthehuman brain: gradient-echo nate at such small voxel sizes. In this study, subjects’ heads and spin-echo fMRI with suppression of blood effects. Magn Resonance were stabilized using customized MR compatible vacuum Med 49: 1019–1027, 2003. pillows and foam padding, and realignment results showed that Eickhoff SB, Grefkes C, Zilles K, Fink GR. The somatotopic organization of subjects could maintain a stable head position within a frac- cytoarchitectonic areas on the human parietal operculum. Cereb Cortex 17: 1800–1811, 2007. tions of mm for several minutes. Second, the width of the Engel SA, Glover GH, Wandell BA. Retinotopic organization in human cortical point spread function (PSF) of the gradient-echo signal visual cortex and the spatial precision of functional MRI. Cereb Cortex 7: may be ultimately limiting; at 7T, the PSF has been shown to 181–192, 1997. be of the order of 2.3 mm for gray matter with vessels masked Engel SA, Rumelhart DE, Wandell BA, Lee AT, Glover GH, Chichilnisky out; this increased to 3.2 mm if veins were not excluded EJ, Shadlen MN. fMRI of human visual cortex. Nature 369: 525, 1994. Fischl B, Dale AM. Measuring the thickness of the human cerebral cortex (Shmuel et al. 2007). However, recent studies of layer-specific from magnetic resonance images. Proc Natl Acad Sci USA 97: 11050– activation using gradient echo BOLD (Harel et al. 2006; 11055, 2000. Koopmans et al. 2009; Ress et al. 2007) suggest that the Francis ST, Kelly EF, Bowtell R, Dunseath WJ, Folger SE, McGlone F. intrinsic spatial resolution is of the order of the submillimeter fMRI of the responses to vibratory stimulation of digit tips. Neuroimage 11: 188–202, 2000. level if contributions from pial veins are excluded. Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R. The results of this study show that fMRI at 7T provides an Event-related fMRI: characterizing differential responses. Neuroimage 7: excellent tool for studying the functional organization of cortex 30–40, 1998. at a spatial resolution not easily achievable at lower field Gardner JL, Sun P, Waggoner RA, Ueno K, Tanaka K, Cheng K. Contrast strengths. Because a somatotopic map can be obtained rela- adaptation and representation in human early visual cortex. Neuron 47: Ͻ 607–620, 2005. tively quickly for an individual subject ( 20 min scanning Gati JS, Menon RS, Ugurbil K, Rutt BK. Experimental determination of the time), at 7T, these methods will allow researchers to study BOLD field strength dependence in vessels and tissue. Magn Reson Med 38: more complex aspects of somatosensory processing in fol- 296–302, 1997. low-up experiments in the same session; making use of exact Gelnar PA, Krauss BR, Szeverenyi NM, Apkarian AV. Fingertip representation in the human somatosensory cortex: an fMRI study. Neuroimage 7: functional co-localization with an independent subject-specific 261–283, 1998. somatotopic map. The maps are of sufficiently high resolution Geyer S, Schleicher A, Zilles K. Areas 3a, 3b, and 1 of human primary to resolve the representations of all five digits, and this could somatosensory cortex. Neuroimage 10: 63–83, 1999.

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org 2556 R. M. SANCHEZ-PANCHUELO, S. FRANCIS, R. BOWTELL, AND D. SCHLUPPECK

Geyer S, Schormann T, Mohlberg H, Zilles K. Areas 3a, 3b, and 1 of human Schluppeck D, Curtis CE, Glimcher PW, Heeger DJ. Sustained activity in primary somatosensory cortex. II. Spatial normalization to standard anatom- topographic areas of human posterior parietal cortex during memory-guided ical space. Neuroimage 11: 684–696, 2000. saccades. J Neurosci 26: 5098–5108, 2006. Glover GH. Deconvolution of impulse response in event-related BOLD fMRI. Schluppeck D, Glimcher P, Heeger DJ. Topographic organization for de- Neuroimage 9: 416–429, 1999. layed saccades in human posterior parietal cortex. J Neurophysiol 94: Grefkes C, Geyer S, Schormann T, Roland P, Zilles K. Human somatosen- 1372–1384, 2005. sory area. II. Observer-independent cytoarchitectonic mapping, interindi- Schweizer R, Voit D, Frahm J. Finger representations in human primary vidual variability, and population map. Neuroimage 14: 617–631, 2001. somatosensory cortex as revealed by high-resolution functional MRI of Neuroimage Gruetter R. Automatic, localized in vivo adjustment of all first- and second- tactile stimulation. 42: 28–35, 2008. Scouten A, Papademetris X, Constable RT. Spatial resolution, signal-to- order shim coils. Magn Reson Med 29: 804–811, 1993. noise ratio, and smoothing in multi-subject functional MRI studies. Neuro- Harel N, Lin J, Moeller S, Ugurbil K, Yacoub E. Combined imaging- image 30: 787–793, 2006. histological study of cortical laminar specificity of fMRI signals. Neuroim- Sereno MI, Pitzalis S, Martinez A. Mapping of contralateral space in age 29: 879–887, 2006. retinotopic coordinates by a parietal cortical area in humans. Science 294: Huang RS, Sereno MI. Dodecapus: an MR-compatible system for somato- 1350–1354, 2001. sensory stimulation. Neuroimage 34: 1060–1073, 2007. Servos P, Zacks J, Rumelhart DE, Glover GH. Somatotopy of the human Hulvershorn J, Bloy L, Gualtieri EE, Leigh JS, Elliott MA. Spatial arm using fMRI. Neuroreport 9: 605–609, 1998. sensitivity and temporal response of spin echo and gradient echo bold Shmuel A, Yacoub E, Chaimow D, Logothetis NK, Ugurbil K. Spatio- contrast at 3 T using peak hemodynamic activation time. Neuroimage 24: temporal point-spread function of fMRI signal in human gray matter at 7 216–223, 2005. Tesla. Neuroimage 35: 539–552, 2007. Koopmans PJ, Barth M, Norris DG. Layer-specific BOLD activation in Smith SM. Fast robust automated brain extraction. Human Brain Mapp 17: human V1. Hum Brain Mapp In press. 143–155, 2002. Kurth R, Villringer K, Curio G, Wolf KJ, Krause T, Repenthin J, Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, Schwiemann J, Deuchert M, Villringer A. fMRI shows multiple somato- Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE, Downloaded from topic digit representations in human primary somatosensory cortex. Neuro- Niazy RK, Saunders J, Vickers J, Zhang Y, De Stefano N, Brady JM, report 11: 1487–1491, 2000. Matthews PM. Advances in functional and structural MR image analysis Larsson J. Imaging Vision: Functional Mapping of Intermediate Visual and implementation as FSL. Neuroimage 23, Suppl 1: S208–219, 2004. Smith SM, Nichols TE. Processes in Man (PhD thesis). Karolinska Institutet, 2001. Threshold-free cluster enhancement: addressing problems of smoothing, threshold dependence and localisation in cluster Maldjian JA, Gottschalk A, Patel RS, Detre JA, Alsop DC. The sensory inference. Neuroimage 44: 83–98, 2009. somatotopic map of the human hand demonstrated at 4 Tesla. Neuroimage Stippich C, Hofmann R, Kapfer D, Hempel E, Heiland S, Jansen O, Sartor 10: 55–62, 1999. K. Somatotopic mapping of the human primary somatosensory cortex by Moore CI, Stern CE, Corkin S, Fischl B, Gray AC, Rosen BR, Dale AM. fully automated tactile stimulation using functional magnetic resonance jn.physiology.org Segregation of somatosensory activation in the human rolandic cortex using imaging. Neurosci Lett 277: 25–28, 1999. fMRI. J Neurophysiol 84: 558–569, 2000. Sur M, Wall JT, Kaas JH. Modular distribution of neurons with slowly Nelson AJ, Chen R. Digit somatotopy within cortical areas of the postcentral adapting and rapidly adapting responses in area 3b of somatosensory cortex gyrus in humans. Cereb Cortex 18: 2341–2351, 2008. in monkeys. J Neurophysiol 51: 724–744, 1984. Nestares O, Heeger DJ. Robust multiresolution alignment of MRI brain Thulborn KR, Waterton JC, Matthews PM, Radda GK. Oxygenation volumes. Magn Reson Med 43: 705–715, 2000. dependence of the transverse relaxation time of water protons in whole

Ogawa S, Menon RS, Kim SG, Ugurbil K. On the characteristics of blood at high field. Biochim Biophys Acta 714: 265–270, 1982. on June 1, 2010 functional magnetic resonance imaging of the brain. Annu Rev Biophys van de Moortele PF, Auerbach EJ, Olman C, Yacoub E, Ugurbil K, Biomol Struct 27: 447–474, 1998. Moeller S. T(1) weighted brain images at 7 Tesla unbiased for proton Overduin SA, Servos P. Distributed digit somatotopy in primary somatosen- density, T(2) contrast and RF coil receive B(1) sensitivity with simultaneous sory cortex. Neuroimage 23: 462–472, 2004. vessel visualization. Neuroimage 46: 432–446, 2009. Overduin SA, Servos P. Symmetric sensorimotor somatotopy. PLoS ONE 3: van der Zwaag W, Francis S, Head K, Peters A, Gowland P, Morris P, e1505, 2008. Bowtell R. fMRI at 1.5, 3 and 7T: characterising BOLD signal changes. Penfield W, Boldrey E. Somatic motor and sensory representation in the Neuroimage 47: 1425–1434, 2010. cerebral cortex of man as studied by electrical stimulation. Brain 60: Wandell BA, Dumoulin SO, Brewer AA. Visual field maps in human cortex. 389–443, 1937. Neuron 56: 366–383, 2007. Pfeuffer J, van de Moortele PF, Yacoub E, Shmuel A, Adriany G, Weibull A, Gustavsson H, Mattsson S, Svensson J. Investigation of spatial Andersen P, Merkle H, Garwood M, Ugurbil K, Hu XP. Zoomed resolution, partial volume effects and smoothing in functional MRI using functional imaging in the human brain at 7 Tesla with simultaneous high artificial 3D time series. Neuroimage 41: 346–353, 2008. spatial and high temporal resolution. Neuroimage 17: 272–286, 2002. White LE, Andrews TJ, Hulette C, Richards A, Groelle M, Paydarfar J, Pons TP, Wall JT, Garraghty PE, Cusick CG, Kaas JH. Consistent features Purves D. Structure of the human sensorimotor system. I. Morphology and of the representation of the hand in area 3b of macaque monkeys. Somato- cytoarchitecture of the central sulcus. Cereb Cortex 7: 18–30, 1997. sens Res 4: 309–331, 1987. Wilson JL, Jenkinson M, de Araujo I, Kringelbach ML, Rolls ET, Jezzard Poole M, Bowtell R. Volume parcellation for improved dynamic shimming. P. Fast, fully automated global and local magnetic field optimization for Magma 21: 31–40, 2008. fMRI of the human brain. Neuroimage 17: 967–976, 2002. Ress D, Glover GH, Liu J, Wandell B. Laminar profiles of functional activity Yacoub E, Duong, TQ, Van De Moortele PF, Lindquist M, Adriany G, in the human brain. Neuroimage 34: 74–84, 2007. Kim SG, Ugurbil K, Hu X. Spin-echo fMRI in humans using high spatial Schleicher A, Amunts K, Geyer S, Kowalski T, Schormann T, Palomero- resolutions and high magnetic fields. Magn Reson Med 49: 655–664, 2003. Gallagher N, Zilles K. A stereological approach to human cortical archi- Yacoub E, Shmuel A, Pfeuffer J, Van de Moortele PF, Adriany G, tecture: identification and delineation of cortical areas. J Chem Neuroanat Andersen P, Vaughan JT, Merkle H, Ugurbil K, Hu X. Imaging brain 20: 31–47, 2000. function in humans at 7 Tesla. Magn Reson Med 45: 588–594, 2001.

J Neurophysiol • VOL 103 • MAY 2010 • www.jn.org