www.elsevier.com/locate/ynimg NeuroImage 40 (2008) 1552 – 1560
Spatiotemporal frequency tuning of BOLD and gamma band MEG responses compared in primary visual cortex Suresh D. Muthukumaraswamy⁎ and Krish D. Singh CUBRIC (Cardiff University Brain Research Imaging Centre), School of Psychology, Cardiff University, Park Place, Cardiff CF103AT, UK Received 3 October 2007; revised 24 January 2008; accepted 27 January 2008 Available online 14 February 2008
In this study, the spatial and temporal frequency tuning characteristics of the MEG gamma (40–60 Hz) rhythm and the BOLD response in primary visual cortex were measured and compared. In an identical MEG/fMRI paradigm, 10 participants viewed reversing square wave gratings at 2 spatial frequencies [0.5 and 3 cycles per degree (cpd)] reversing at 5 temporal frequencies (0, 1 6, 10, 15 Hz). Three-dimensional images of MEG source power were generated with synthetic aperture magnetometry (SAM) and showed a high degree of spatial correspondence with BOLD responses in primary visual cortex with a mean spatial separation of 6.5 mm, but the two modalities showed different tuning characteristics. The gamma rhythm showed a clear increase in induced power for the high spatial frequency stimulus while BOLD showed no difference in activity for the two spatial frequencies used. Both imaging modalities showed a general increase of activity with temporal frequency, however, BOLD plateaued around 6–10 Hz while the MEG generally increased with a dip exhibited at 6 Hz. These results demonstrate that the two modalities may show activation in similar spatial locations but that the functional pattern of these activations may differ in a complex manner, suggesting that they may be tuned to different aspects of neuronal activity. © 2008 Elsevier Inc. All rights reserved.
Introduction The high spatial resolution and non-invasive nature of functional magnetic resonance imaging (fMRI) have led to it becoming one of the most popular tools for measuring brain function in human neuroscience. fMRI, however, provides only an indirect measure of neural activity by measuring the changes in cerebral metabolism that are coupled in a complex manner to changes in neural activity. An improved understanding of the relationship between the generation of the BOLD signal and actual neuronal activity has emerged in recent years with the development of simultaneous recording techniques of either BOLD (Logothetis et al., 2001) or optical imaging (Arieli and Grinvald, 2002; Niessing et al., 2005) with the recording ⁎ Corresponding author. Fax: +44 29 2087 0339. E-mail address:
[email protected] (S.D. Muthukumaraswamy). Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2008.01.052
of local field potentials. These studies demonstrate that local field potentials are a better predictor of the BOLD response than multiunit spiking activity (Logothetis et al., 2001); although each is significantly correlated with the other. The invasive nature of these techniques limits their use to animal populations. Because of this limitation, a need exists to understand the relationship between fMRI and neuronal activity as recorded in humans with commonly used non-invasive techniques such as M/EEG. In recent years, the oscillatory activity of the M/EEG has been increasingly studied in a number of frequency bands and modulation of these oscillations corresponds with a number of functions including memory, perception and cognition. Oscillatory activity is of particular interest when making comparisons to BOLD because oscillatory activity appears to be well spatially correlated with the BOLD response (Brookes et al., 2005; Singh et al., 2002). The analysis technique synthetic aperture magnetometry (SAM) (Robinson and Vrba, 1999) provides an excellent technique to localise in space, time and frequency, changes in cortical oscillatory power associated with stimulus events. Because sustained oscillatory changes are analysed with similar temporal paradigms to fMRI (lasting several seconds and beyond) can be used. Several previous studies have adopted this approach and demonstrated that various oscillatory components of the oscillatory MEG signal and BOLD are spatially correlated. It has been found in more cognitive paradigms that desynchronisation in the alpha and beta band is inversely correlated with BOLD increases (Singh et al., 2003a,b, 2002). Conversely, in primary visual cortices, synchronisation effects are seen in higher frequency bands (Adjamian et al., 2004; Brookes et al., 2005; Fawcett et al., 2004). A recent comparison of local field potential recordings and BOLD in humans has demonstrated a similar relationship (Mukamel et al., 2007). Brookes et al. (2005) used a static checkerboard stimulus with 5 participants and demonstrated that gamma ERS spatially co-localised with the BOLD response in primary visual cortex. While this demonstrates spatial coincidence between the two measures because only one stimulus type (condition) was used, the functional relationship between the two is still unexamined. In this experiment, we sought to examine the functional relationship for spatial and temporal frequency tuning between the MEG and fMRI. The cortical responses of single neurons in the macaque visual
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
cortex have been well described. In terms of temporal frequency tuning cells in V1 exhibit increased firing rates up to 8 Hz and then a decrease (Foster et al., 1985). With respect to spatial frequency, cells in macaque V1 are optimally responsive between 2 and 4 cpd (Foster et al., 1985). This is similar to gamma band MEG responses that exhibit a spatial frequency tuning effect, with a peak response at 3 cpd (Adjamian et al., 2004). The cortical responses of cells in V1 for both macaque and cat do not show interaction effects between different spatial and temporal frequencies suggesting they can be used as orthogonal stimulus parameters (Tolhurst and Movshon, 1975; Foster et al., 1985). Spatial and temporal frequency tuning have been separately reported with fMRI (Singh et al., 2000) and MEG (Adjamian et al., 2004; Fawcett et al., 2004), however, due to a number of experimental variations used across these studies the data are difficult to compare directly. Specifically, the spatiotemporal frequency tuning fMRI experiment of Singh et al. (2000) used centrally fixated drifting sinusoids, the spatial frequency MEG experiment of Adjamian et al. (2004) used centrally fixated square wave gratings and the temporal frequency tuning MEG experiment of Fawcett et al. (2004) used single visual quadrant square wave checkerboard stimuli. Therefore, in this experiment, we manipulated spatial and temporal frequency tuning with square wave gratings using identical subjects and stimulus parameters in both MEG and fMRI with the aim of directly comparing their spatiotemporal frequency tuning characteristics. Methods Participants, stimuli and paradigm Ten healthy right-handed volunteers including six females and four males (mean age 27.5 years; range, 23–36 years) participated in the experiment after giving informed consent. All procedures were
1553
approved by the local Ethics Committee. One other participant completed the fMRI component but was not asked to complete the MEG due to poor fMRI data quality and similarly a further participant completed the MEG component but due to excessive artefact (head movement) was not asked to complete the fMRI component. Identical participants, stimuli and paradigm-design were used for both the MEG and fMRI sessions. This consisted of reversing square-wave vertical gratings presented at different spatial and temporal frequencies displayed using Presentation software (Neurobehavioral Systems Inc.). Grating stimuli were presented at two spatial frequencies (3 and 0.5 cpd) each at five temporal frequencies (0, 1, 6, 10, and 15 Hz). These frequencies were chosen such that each image reversal would be an integer multiple of the frame rate of the video-card driving the projection systems at 60 Hz at 1024 × 768 pixels. For the MEG, a Sanyo XP41 LCD back-projection system was used and for the MRI a Canon Xeed SX60 projector. The ability of both our sets of stimulus projection equipment to accurately display the temporal frequencies required was verified using a photodiode and examining the Fourier spectrum of its digitally converted output. Stimuli were presented in the lower left visual field with the upper right corner of the stimulus located 1° horizontally and vertically from a small fixation cross. Stimuli subtended 8° both horizontally and vertically. Participants were instructed to maintain fixation for the entire experiment and to maintain attention to press a response key at the termination of each stimulation period. 500 ms offset jitter was added to each stimulus duration. On separate days, each participant underwent fMRI or MEG scanning consisting of four identical runs each lasting 10 min back to back. Each run consisted of forty trials in a 15 s boxcar design with a 10 s rest period followed by a 5 s active stimulation period. Our temporal design was chosen to satisfy several competing design considerations. Firstly, we wanted the same timing parameters for both the MEG and fMRI. Secondly, we wanted relatively short but
Fig. 1. Example MEG gamma 40–60 Hz SAM and BOLD images from the high spatial frequency, 6 Hz temporal frequency stimulus for participant I. The functional images displayed are unthresholded amplitude maps for both modalities. MEG values are pseudo-T values and BOLD measures are arbitrary units.
1554
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
more trials so that we could examine the temporal characteristics of the MEG signal change, and thirdly, we wanted long enough block lengths to allow the BOLD response to occur and recover. The 10 different stimuli were presented in randomised fashion in each of the blocks (each stimulus was presented four times in a run). MRI/fMRI acquisition and analysis MRI data were acquired on a 3-T General Electric HDx scanner with an eight channel receive only head RF coil (Medical Devices). fMRI data were acquired using a gradient echo EPI sequence taking 37 axial slices of the whole brain at 3 mm isotropic voxel resolution using a 64 × 64 matrix size, echo time of 35 ms, 90° flip angle and a TR of 2.5 s. For each participant, a 3D FSPGR scan with 1 mm isotropic voxel resolution was also obtained. Both MEG and fMRI data were coregistered to this high resolution structural scan. Analysis of fMRI data was analysed using the FSL software library (www.fmrib.ox.ac.uk/fsl). The following pre-processing was applied; motion correction using MCFLIRT (Jenkinson et al., 2002); non-brain removal using BET (Smith, 2002); spatial smoothing using a Gaussian kernel of FWHM 5 mm; mean-based intensity normalisation of all volumes by the same factor and high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with σ = 50 s). For each 10-min run, the GLM was used to model each of the 10 conditions using a 5 s on/10 s off boxcar to describe each stimulus. This boxcar function was convolved with a standard HRF to account for hemodynamic effects. To combine the four runs for each individual a second level analysis was performed using a fixed effects model, by forcing the random effects variance to zero in FLAME [FMRIB's Local Analysis of Mixed Effects (Beckmann et al., 2003; Woolrich et al., 2004)].
MEG acquisition and analysis Whole-head MEG recordings were made using a 275-channel radial gradiometer system sampled at 600 Hz. An additional 29 reference channels were recorded for noise cancellation purposes and the primary sensors were analysed as synthetic third order gradiometers (Vrba and Robinson, 2001). 3 of the 275 channels were turned off due to excessive sensor noise. At the commencement of each active period of photic stimulation, a TTL pulse was sent to the MEG system. The location of three fiduciary markers (nasion, left and right preauricular) was monitored continuously through the MEG acquisition. In the event that participants moved more than 5 mm from their initial position, the experiment was paused between trials and participants were asked by the experimenter to adjust their head position until it was less than 5 mm from the original position as detected by the MEG system and then the experiment continued. Offline, each data set was band-pass-filtered using a fourth-order bi-directional IIR Butterworth filter into four frequency bands 0–20, 20–40, 40–60 and 60–80 Hz. Evenly spaced frequency bands were used so that the accuracy of covariance matrix estimation would be equal for each frequency band (Brookes et al., in press). The SAM beamformer algorithm as implemented in the CTF software (Robinson and Vrba, 1999) was used to create differential images of source power (pseudo-T statistics) for 5 s of baseline (−5 to 0 s) compared to 5 s of visual stimulation (0 to 5 s). Only 5 s of the baseline period was used for baseline estimation to achieve balanced covariance estimation between stimulated and unstimulated states. Details of the calculation of SAM pseudo-T source image statistics are described in detail in a number of sources (Cheyne et al., 2003; Hillebrand et al., 2005; Robinson and Vrba, 1999; Singh et al., 2003a,b; Vrba and Robinson,
Fig. 2. Group temporal frequency tuning curves for peak responses for the fMRI and MEG Gamma (40–60 Hz) band in primary visual cortex. In order to plot both modalities on the same scale ordinate, magnitudes have been expressed as a percentage of the maximum response hence each modality has one anchor point at 100%. The high spatial frequency stimulus was a 3 cpd square wave grating and the low spatial frequency 0.5 cpd.
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
2001). To achieve MRI/MEG co-registration, prior to the MEG acquisition, fiduciary markers were placed at fixed distances from anatomical landmarks identifiable in participants' anatomical MRIs (tragus, eye centre). Fiduciary locations were verified afterwards using high-resolution digital photographs. For source localisation, a multiple, local-spheres-forward model was derived by fitting spheres to the brain surface extracted by BET (Huang et al., 1999). Estimates of the three-dimensional distribution of source power were derived for the whole head at 3 mm isotropic voxel resolution for each subject, frequency-band and condition. Virtual sensors were generated by using SAM beamformer coefficients obtained using the individual condition covariance matrices band pass filtered between 0 and 100 Hz and returning time-series from peak locations in primary visual cortex for each trial (Robinson and Vrba, 1999). Time–frequency analysis of virtual sensors was conducted using slepian multitapers (Jarvis and Mitra, 2001; Mitra and Pesaran, 1999; Percival and Walden, 1993; Thomson 1982) as implemented in the FieldTrip toolbox (www.ru. nl/fcdonders/fieldtrip). Time–frequency spectrograms are presented as a percentage change from the baseline energy for each frequency band.
1555
Group-level analysis (both MEG and fMRI) For group analysis of both MEG and fMRI data sets, images were normalised using FLIRT into MNI template space using an affine transform. Non-parametric permutation tests were conducted using the full permutation set (1024) for each condition with 5 mm variance smoothing and thresholded using the omnibus test statistic value at p b .05 (Nichols and Holmes, 2002; Singh et al., 2003a,b). Results Fig. 1 illustrates example MEG and fMRI activations from a single participant using the unthresholded amplitude maps for the BOLD response and the MEG gamma response. It can be seen that there is a high spatial overlap between the MEG response and the medial activation in the fMRI response. While the MEG gamma image is dominated by a single activation, the fMRI response has several foci that extend into more lateral visual areas. Statistical group images for the SAM images showed significant activation for the MEG in the 40–60 Hz (gamma) frequency for all 10 of the conditions, located in medial visual cortex (BA 17/18). No
Fig. 3. Grand-averaged time–frequency spectrograms from MEG virtual sensors constructed in peak locations in primary visual cortex for each condition. Energy values are represented as a percentage change from the baseline energy (calculated for each frequency band). These spectrograms represent a combination of both evoked and induced power.
1556
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
consistent oscillatory changes were seen in the lower frequency bands (10 conditions × 2 frequency bands = 20 analyses). Significant effects were seen in all 10 conditions for the 60- Hz frequency band in locations similar to the 40- Hz frequency band. Inspection of individual data showed that these responses were weaker and more spatially variable than the 40- Hz band, suggesting that this frequency band was reflecting the upper frequency part of the activity seen in the 40- Hz band. Because the images were dominated by 40- Hz gamma band activity, we quantified the peak 40- Hz band activity (in primary visual cortex) and compared it to the BOLD amplitude measures in primary visual cortex for each individual for each of the 10 conditions. Fig. 2 illustrates the results of this analysis and plots the spatial and temporal frequency tuning curves of the MEG gamma source power and the BOLD response. In order to plot the two modalities on the same ordinate, amplitudes have been expressed as a percentage of the maximum amplitude measure for each modality. This figure demonstrates that a clear difference exists in the spatial frequency tuning of the MEG gamma band at each temporal frequency. The high spatial frequency stimulus of 3 Hz elicited approximately three times more source power in this band than the 0.5 Hz spatial frequency
stimulus. However, this pattern is not reflected in the spatial frequency tuning for BOLD for which the high and low spatial frequency stimuli show an almost identical pattern. With regards to temporal frequency tuning the two modalities also show differences. BOLD showed an increasing response up to 6 Hz and then plateaued whereas MEG shows a generally increasing amount of source power with temporal frequency but a decrease at 6 Hz for both spatial frequencies. For the MEG data in Fig. 2, a repeated-measures two-way ANOVA found significant main effects of spatial [F(1,9) = 40.7, p b .001] and temporal frequency [F(4,36) = 14.78, p b .001] but no interaction effect [F(4,36) = 1.66, p = .181]. For the corresponding BOLD data in Fig. 2, a significant effect of temporal frequency [F(4,36) = 24.26, p b .001] was found but no main effect of spatial frequency [F(1,9) = 0.086, p = .77] or interaction effect [F(4,36) = 1.86, p = .139]. To further investigate the pattern and time course of activity in the MEG, virtual sensors were constructed at peak locations in primary visual cortex. The results of a time–frequency analysis of these virtual sensors are represented in Fig. 3. Alpha desynchronisation can be seen following stimulus onset for both the high and low spatial frequency stimuli but a Wilcoxon rank sum test found no significant differences between the integrated active (0–5 s) power for low versus high spatial
Fig. 4. Grand-averaged time–frequency spectrograms from the same virtual sensors in Fig. 3. Prior to computation waveforms were averaged in the time-domain hence these spectrograms represent only evoked power changes in the virtual sensor. The 2f response corresponding to the steady state evoked field can be clearly seen for the 6, 10 and 15 Hz conditions.
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
1557
Fig. 5. Comparison of evoked (n.s.) and induced + evoked gamma (40–60 Hz) (p b .05) power for low and high spatial frequencies for the 0 Hz temporal frequency stimulus. Induced + evoked power was obtained by averaging the spectrograms in Fig. 3 and evoked power by averaging the spectrograms in Fig. 4.
frequency. Integrating the power of these spectrograms from 40 to 100 Hz demonstrates almost identical tuning curves to those displayed in Fig. 2. Fig. 4 displays the time–frequency analysis of the evoked virtual sensor waveforms. In comparison to Fig. 3, the alpha desynchronisation is absent from these plots. In this figure, a strong response at 2 times the temporal frequency of the stimulus can be seen for the 6, 10 and 15 Hz conditions for both spatial frequencies. This 2f response corresponds to the steady state evoked field. For the 1 Hz temporal stimuli, bursts of power at each stimulus reversal can be more clearly seen than in Fig. 3. To compare the effect of evoked versus induced activity on spatial frequency, we further analysed the 0 Hz condition where only one burst of evoked activity occurs at stimulus onset. Power was integrated between 1 and 5 s to obtain measures of induced + evoked activity from Fig. 3 and between 0.2 and 0.5 from the spectrograms in Fig. 4 for evoked activity. The obtained levels of gamma power are plotted in Fig. 5. A Wilcoxon rank sum test on these data found a significant effect for the induced + evoked gamma activity ( p b .05) but not for any of the evoked activity time windows ( p N .05).
Discussion In this experiment, we performed an identical fMRI/MEG experiment in 10 participants and measured spatiotemporal frequency curves of the BOLD response and the gamma rhythm (40–60 Hz) in primary visual cortex. The most striking finding of the current experiment is the clear difference in the amplitude of the gamma rhythm found for stimuli of different spatial frequencies whereas the BOLD response exhibited no difference. This finding is consistent with previous fMRI and MEG studies in the area. In the MEG experiment of Adjamian et al. (2004), a peak gamma response was found at 3 cpd and it was greatly attenuated at 0.5 cpd per degree. Using fMRI (Singh et al., 2000), it has been demonstrated that spatial frequency tuning curves for V1 are relatively flat for the range of spatial frequencies used in the present experiment (and up to around 9 cpd where a falloff is exhibited). In the Adjamian et al. (2004) experiment, stationary square waving grating stimuli were used, whereas in the Singh et al. (2000) paper, the stimuli were sinusoidally moving gratings. The present experiment eliminates these interexperiment variables and held as many experimental factors constant
between modalities as possible, demonstrating that in primary visual cortex, the gamma MEG rhythm is sensitive to spatial frequency whereas BOLD is not. Our data are also consistent with single-cell recording studies, which demonstrate no interactions between spatial and temporal frequency tuning for the range of stimuli that we used (Foster et al., 1985). A number of studies examining fMRI temporal frequency tuning have found a similar pattern of results to the present study. Usually, this consists of a peak in the temporal frequency function at about 8 Hz followed by either a plateau (Ozus et al., 2001; Parkes et al., 2004) as seen in our data or a decrease in the function (Singh et al., 2000, 2003a,b; Thomas and Menon, 1998). These response functions are similar to those found in single unit studies in primary visual cortex (Foster et al., 1985). One possibility is that the plateau of the BOLD response is caused by the maximal stimulus contrast used in this experiment. It has been found in single-cell recording studies that there is an interaction between contrast level and temporal frequency (Albrecht, 1995; Holub and Morton-Gibson, 1981). The responsiveness of cells in V1 generally increases linearly with contrast up to a saturation level. This threshold level is dependent on temporal frequency characteristics of the stimulus, demonstrating an interaction between contrast and temporal frequency exists in single-cell responses in primary visual cortex. To our knowledge, there has been no systematic fMRI study that has parametrically varied both temporal frequency and contrast level to examine contrast saturation effects and most previous studies have been conducted with maximum contrast in order to elevate signal to noise levels. However, one BOLD (Thomas and Menon, 1998) and one PET (Mentis et al., 1997) study used red flickering LEDs as stimuli and found similar temporal frequency tuning curves to other high contrast stimulus studies suggesting that hemodynamic temporal frequency tuning curves are not contrast dependent. It has been found in several animal studies in V1 that spatial frequency and orientation tuning of single neurons does not change with contrast levels (Albrecht and Hamilton, 1982; Li and Creutzfeldt, 1984; Movshon et al., 1978). Despite the difference in tuning curves between fMRI and MEG the spatial correspondence between the two modalities was very good. The results in Table 1 demonstrate that the peak gamma and BOLD responses were located in very similar areas, with an average error of 6.46 mm (range, 4.4–8.0 mm). This error is similar to previous reports of the consistency of MEG source co-localisation
1558
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
Table 1 Summary of the spatial locations for the peak BOLD and gamma (40– 60 Hz) MEG responses in medial visual cortex for each participant Participant Peak BOLD location
Peak MEG Difference gamma location (mm)
A B C D E F G H I J Average
138, 218, 156 113, 202, 128 105, 222, 148 109, 214, 136 117, 230, 136 117, 218, 148 121, 218, 148 117, 206, 148 117, 206, 140 121, 222, 148
133, 223, 154 116, 197, 125 105, 222, 156 108, 219, 137 113, 235, 139 116, 217, 155 121, 219, 155 116, 209, 145 117, 204,144 126, 220, 153
− 5, 5, −2 3, −5, −3 0, 0, 8 − 1, 5, 1 − 4, 5, 3 − 1, − 1, 7 0, 1, 7 − 1, 3, −3 0, −2, 4 5, −2, 5 − 0.4, 0.9, 2.7
3D separation (mm) 7.4 6.6 8.0 5.2 7.1 7.14 7.1 4.4 4.5 7.4 6.46 (2.87)
Data were averaged across all conditions prior to conducting the peak search. Co-ordinates are arbitrary co-registered voxel locations (sagittal R to L, coronal A to P, axial S to I). Because the anatomical images had 1 mm isotropic voxel resolution, units can also be considered as mm. The bold unbracketed number is the average 3D separation distance, whereas the bracketed bold number is the 3D separation of the mean differences.
with BOLD. For example, Moradi et al. (2003) showed a 3- mm spatial separation between the peak of the M70-evoked field component and the BOLD response in V1. For gamma band synchronisation using SAM, Brookes et al. (2005) report 9 ± 15 mm separation, hence our results lie somewhere between these two. There are a number of reasons why even if the same local population of neurons were causing both responses small errors in localisation might occur. Firstly, errors may occur in co-registration of both functional data sets to the anatomical image. The MEG may be more prone to these errors as it relies on accurate placement of the fiduciary coils. Displacement of, for example, the nasion coil by 1 mm may produce a levering effect error of 2 mm in posterior visual cortex (Singh et al., 1997). A further limitation on BOLD/ MEG co-localisation is the size of the voxels used in both the MEG and fMRI, in this case 3 mm. It should be noted that the Pythagorean distance across one functional voxel in the present study is 5.2 mm suggesting the mean error in our study could represent an error of one functional voxel. Another possibility is that due to vascularisation effects the peak of the BOLD response may not necessarily be at the peak of where the greatest oxygen demand is located, hence causing a discrepancy between the hemodynamic and neuronal measures. Finally, it is possible that the surface MEG is created by a more widespread assembly of neurons acting synchronously whereas the BOLD response by a more local focal population of neurons, not necessarily acting in synchrony (Nunez and Silberstein, 2000). Despite these theoretical and technical factors, it seems most likely that at a macroscopic (voxel) level the two modalities are recording responses from the same cortical area. While the different response functions for the two modalities is complex, this is not to be unexpected since it is both theoretically and empirically demonstrable that the two techniques can potentially be sampling activity from distinct subsets of the neuronal population in the same area (Nunez and Silberstein, 2000). For example, it has been traditionally thought (Creutzfeldt and Houchin, 1974; Lorento de No, 1947) and more recently empirically verified (Murakami and Okada, 2006) that spiny stellate cells with their symmetrical dendritic processes (and moreover populations of stellate cells) and hence closed field configurations such as those that exist in layer
4 of V1 (Tenke et al., 1993; Sincich and Horton, 2005) contribute little to the surface-evoked M/EEG (Murakami and Okada, 2006). Stellate cells represent approximately 15% of the neocortex (Braitenberg and Schuz, 1991) and have higher firing rates than pyramidal cells (presumably then creating greater hemodynamic demand) and hence the metabolic drive created by activity of these cells would be present in the BOLD signal but absent in the M/EEG signal. Evidence in rat somatosensory cortex suggests that spiny stellate cells in layer IV represent local signal processing modules whereas pyramidal cell activity the integration of columnar activity (Staiger et al., 2004). It should also be pointed out that M/EEG is particularly tuned for synchronous neuronal activity. For activity to summate to a surface field neurons in an open-field configuration must activate synchronously else cancellation will occur. Modelling evidence demonstrates that it would require only 1% of minicolumns in an area to be synchronous compared to the other 99% of local cortex acting asynchronously) to contribute 3 times more to the scalp EEG/MEG (Nunez, 1981). The surface field of this remaining 99% would only represent statistical fluctuations due to imperfect cancellation. Finally, the notion that synchronous gamma band activity could occur with no (or very subtle) BOLD effects is plausible given what is known about the gamma rhythm in cat visual cortex. Gamma oscillations in cat primary visual cortex have been demonstrated to be generated by only a small subset of the neuronal population in a cortical column in primary visual cortex; specifically complex cells which fire to edges (Gray and Singer, 1989). These gamma oscillations can be synchronised across a number of cortical columns (Gray et al., 1989). A question remains as to what property of the stimulus causes the difference in gamma rhythm between high and low spatial frequency stimuli but not hemodynamic response. One possibility is that there is a ceiling effect in the BOLD response caused by the high (maximum) contrast levels used in the experiment. If this were the case then it would indicate a non-linear aspect of the relationship between BOLD and gamma. However, we note that this discrepancy between spatial frequency, exists even at the 0 Hz temporal frequency and the BOLD response increases for the high spatial frequency stimulus with temporal frequency suggesting a ceiling level for BOLD had not been reached. Alternatively, it could be that sine wave modulated stimuli would more differentially activate the BOLD response (cf., Singh et al., 2000). The square wave stimuli in that sense is not a pure spatial frequency stimulus and the harmonic spatial frequencies in the lowfrequency stimuli may be elevating the BOLD in the low-frequency stimulus relative to that produced by a sine wave grating. It could be useful and informative to know how the gamma rhythm would react to this stimulus property as well. One likely possibility based on animal data is that the gamma rhythm is particularly tuned to continuous edges (Gray et al., 1989). The MEG gamma response exhibits a drop in power for the 6 Hz temporal frequency for both spatial frequencies. While this effect was consist and difficult to explain it is not without precedent. In the MEG temporal frequency tuning experiment of Fawcett et al. (2004), their gamma band response (41–43 Hz) exhibited a similar decrease (at 8 Hz temporal frequency). Fawcett et al. used square wave checkerboard stimuli with 15 s block sizes and here we have found a similar effect for two different spatial frequency gratings presented for a shorter temporal period. An earlier MEG study by Fylan et al. (1997) found a similar decrease in evoked global field power for a 5 Hz temporal frequency grating. This suggests that this effect is robust to differences in stimulus parameters and analysis methodologies but the physiological reason for this decrease is not clear.
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
The present data demonstrate that MEG and BOLD show differences in spatial and temporal frequency tuning, suggesting that potentially different neuronal populations contribute to their signal and/or that a complex non-linear relationship exists between the BOLD effect and the surface MEG. However, it should be pointed out that other experimental manipulations such as visual contrast tuning may yield less complex results (although this has not been tested directly). It has been found with MEG, using similar techniques to those here, that gamma oscillations increase linearly with contrast enhancement (Hall et al., 2005). This linear increase is similar to that seen in multi-unit recordings of the macaque (Logothetis et al., 2001). Further, it has been shown in cat visual cortex with simultaneous recording that gamma local field potential oscillations and hemodynamic responses measured with optical imaging show a correlation and both increase with stimulus contrast (Niessing et al., 2005). Therefore, it seems that both BOLD and gamma would linearly increase with contrast. In summary, the present experiment examined and compared the spatial and temporal frequency tuning characteristics of the BOLD response with gamma band MEG activity in primary visual cortex. The two responses showed a high degree of congruence in their spatial characteristics but distinctly different tuning characteristics. This suggests that the relationship between the two either contains a number of non-linear elements and/or that they measure distinct subsets of the total neuronal activity. Acknowledgments Krish Singh is supported by the Biotechnology and Biological Sciences Research Council (BBSRC) grant number BBS/B/08035. References Adjamian, P, Holliday, IE, Barnes, GR, Hillebrand, A, Hadjipapas, A, Singh, KD, 2004. Induced visual illusions and gamma oscillations in human primary visual cortex. Eur. J. Neurosci. 20 (2), 587–592. Albrecht, DG, 1995. Visual-cortex neurons in monkey and cat—effect of contrast on the spatial and temporal phase-transfer functions. Vis. Neurosci. 12 (6), 1191–1210. Albrecht, DG, Hamilton, DB, 1982. Striate cortex of monkey and cat— contrast response function. J. Neurophysiol. 48 (1), 217–237. Arieli, A, Grinvald, A, 2002. Optical imaging combined with targeted electrical recordings, microstimulation, or tracer injections. J. Neurosci. Methods 116, 15–28. Beckmann, CF, Jenkinson, M, Smith, SM, 2003. General multilevel linear modeling for group analysis in fMRI. NeuroImage 20 (2), 1052–1063. Braitenberg, V, Schuz, A, 1991. Anatomy of the Cortex, Statistics and Geometry. Springer, New York. Brookes, MJ, Vrba, J, Robinson, SE, Stevenson, CM, Peters, AM, Barnes, GR, Hillebrand, A, Morris, PG, in press. Optimising experimental design for MEG beamformer imaging. Neuroimage. Brookes, MJ, Gibson, AM, Hall, SD, Furlong, PL, Barnes, GR, Hillebrand, A, Singh, KD, Holliday, IE, Francis, ST, Morris, PG, 2005. GLMbeamformer method demonstrates stationary field, alpha ERD and gamma ERS co-localisation with fMRI BOLD response in visual cortex. NeuroImage 26 (1), 302–308. Cheyne, D, Gaetz, W, Garnero, L, Lachaux, J-P, Ducorps, A, Schwartz, D, Varela, FJ, 2003. Neuromagnetic imaging of cortical oscillations accompanying tactile stimulation. Cogn. Brain Res. 17, 599–611. Creutzfeldt, O, Houchin, J, 1974. Neuronal basis of EEG waves. In: Redmond, A. (Ed.), Handbook of Electroencephalography and Clinical Neurophysiology. Elsevier, Amsterdam, pp. 5–55.
1559
Fawcett, IP, Barnes, GR, Hillebrand, A, Singh, KD, 2004. The temporal frequency tuning of human visual cortex investigated using synthetic aperture magnetometry. NeuroImage 21 (4), 1542–1553. Foster, KH, Gaska, JP, Nagler, M, Pollen, DA, 1985. Spatial and temporal frequency selectivity of neurones in visual cortical areas V1 and V2 of the macaque monkey. J. Physiol. 365, 331–363. Fylan, F, Holliday, IE, Singh, IE, Anderson, SJ, Harding, GFA, 1997. Magnetoencephalographic investigation of human cortical area V1 using color stimuli. NeuroImage 6, 47–57. Gray, CM, Konig, P, Engel, AK, Singer, W, 1989. Oscillatory responses in cat visual-cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338 (6213), 334–337. Gray, CM, Singer, W, 1989. Stimulus-specific neuronal oscillations in orientation columns of cat visual-cortex. Proc. Natl. Acad. Sci. U. S. A. 86 (5), 1698–1702. Hall, SD, Holliday, IE, Hillebrand, A, Singh, KD, Furlong, PL, Hadjipapas, A, Barnes, GR, 2005. The missing link: analogous human and primate cortical gamma oscillations. NeuroImage 26 (1), 13–17. Hillebrand, A, Singh, KD, Holliday, I, Furlong, PL, Barnes, GR, 2005. A new approach to neuroimaging with magnetoencephalography. Hum. Brain Mapp. 25, 199–211. Holub, MA, Morton-Gibson, M, 1981. Response of visual cortical neurons of the cat to moving sinusoidal gratings: response-contrast functions and spatiotemporal interactions. J. Neurophysiol. 46 (6), 1244–1259. Huang, MX, Mosher, JC, Leahy, RM, 1999. A sensor-weighted overlapping-sphere head model and exhaustive head model comparison for MEG. Phys. Med. Biol. 44 (2), 423–440. Jarvis, MR, Mitra, PP, 2001. Sampling properties of the spectrum and coherency of action potentials. Neural Comput. 13 (4), 717–749. Jenkinson, M, Bannister, P, Brady, M, Smith, S, 2002. Improved optimization for the robust and accurate linear registration and motion correction of brain images. NeuroImage 17 (2), 825–841. Li, CY, Creutzfeldt, O, 1984. The representation of contrast and other stimulus parameters by single neurons in area-17 of the cat. Pflugers Arch. Eur. J. Phys. 401 (3), 304–314. Logothetis, NK, Pauls, J, Augath, M, Trinath, T, Oeltermann, A, 2001. Neurophysiological investigation of the basis of the fMRI signal. Nature 412, 150–157. Lorento de No, R, 1947. Action potential of the motoneurons of the hypoglossal nucleus. J. Cell. Comp. Physiol. 29, 207–287. Mentis, MJ, Alexander, GE, Grady, CL, Horwitz, B, Krasuski, J, Pietrini, P, et al., 1997. Frequency variation of a pattern-flash visual stimulus during PET differentially activates brain from striate through frontal cortex. NeuroImage 5 (2), 116–128. Mitra, PP, Pesaran, B, 1999. Analysis of dynamic brain imaging data. Biophys. J. 76 (2), 691–708. Moradi, F, Liu, LC, Cheng, K, Waggoner, RA, Tanaka, K, Ioannides, AA, 2003. Consistent and precise localization of brain activity in human primary visual cortex by MEG and fMRI. NeuroImage 18 (3), 595–609. Movshon, JA, Thompson, ID, Tolhurst, DJ, 1978. Spatial and temporal contrast sensitivity of neurons in areas-17 and areas-18 of cats visualcortex. J. Physiol.-London 283, 101–120 (Oct). Mukamel, R, Gelbard, H, Arieli, A, Hasson, U, Fried, I, Malach, R, 2007. Coupling between neuronal firing, field potentials, and fMRI in human auditory cortex. Science 309, 951–954. Murakami, S, Okada, Y, 2006. Contributions of principal neocortical neurons to magnetoencephalography and electroencephalography signals. J. Physiol.-London 575 (3), 925–936. Nichols, TE, Holmes, AP, 2002. Nonparametric permutation tests for functional neuroimaging: A primer with examples. Hum. Brain Mapp. 15 (1), 1–25. Niessing, J, Ebisch, B, Schmidt, KE, Niessing, M, Singer, W, Galuske, RAW, 2005. Hemodynamic signals correlate tightly with synchronized gamma oscillations. Science 309 (5736), 948–951. Nunez, PL, 1981. Electric Fields of the Brain: the Neurophysics of EEG. Oxford University Press, New York.
1560
S.D. Muthukumaraswamy, K.D. Singh / NeuroImage 40 (2008) 1552–1560
Nunez, PL, Silberstein, RB, 2000. On the relationship of synaptic activity to macroscopic measurements: does co-registration of EEG with fMRI make sense? Brain Topogr. 13 (2), 79–96. Ozus, B, Liu, HL, Chen, L, Iyer, MB, Fox, PT, Gao, JH, 2001. Rate dependence of human visual cortical response due to brief stimulation: an event-related fMRI study. Magn. Reson. Imaging 19 (1), 21–25. Parkes, LM, Fries, P, Kerskens, CM, Norris, DG, 2004. Reduced BOLD response to periodic visual stimulation. NeuroImage 21 (1), 236–243. Percival, DB, Walden, WT, 1993. Spectral Analysis for Physical Applications: Multitaper and Conventional Univariate Techniques. Cambridge University Press. Robinson, SE, Vrba, J, 1999. Functional neuroimaging by synthetic aperture magnetometry (SAM). In: Yoshimoto, T., Kotani, M., Kuriki, S., Karibe, H., Nakasato, N. (Eds.), Recent Advances in Biomagnetism. Tohoku University Press, Sendai, pp. 302–305. Sincich, LC, Horton, JC, 2005. The circuitry of V1 and V2: integration of color, form, and motion. Annu. Rev. Neurosci. 28, 303–326. Singh, KD, Holliday, IE, Furlong, PL, Harding, GFA, 1997. Evaluation of MRI-MEG/EEG co-registration strategies using Monte Carlo simulation. Electroencephalogr. Clin. Neurophysiol. 102 (2), 81–85. Singh, KD, Smith, AT, Greenlee, MW, 2000. Spatiotemporal frequency and direction sensitivities of human visual areas measured using fMRI. Neuroimage 12 (5), 550–564. Singh, KD, Barnes, GR, Hillebrand, A, Forde, EME, Williams, AL, 2002. Task-related changes in cortical synchronization are spatially coincident with the hemodynamic response. NeuroImage 16 (1), 103–114.
Singh, KD, Barnes, GR, Hillebrand, A, 2003a. Group imaging of taskrelated changes in cortical synchronisation using nonparametric permutation testing. NeuroImage 19 (4), 1589–1601. Singh, M, Kim, S, Kim, T-S, 2003b. Correlation between BOLD-fMRI and EEG signal changes in response to visual stimulus frequency in humans. Magn. Reson. Med. 49, 108–114. Smith, SM, 2002. Fast robust automated brain extraction. Hum. Brain Mapp. 17 (3), 143–155. Staiger, JF, Flagmeyer, I, Schubert, D, Zilles, K, Kotter, R, Luhmann, HJ, 2004. Functional diversity of layer IV spiny neurons in rat somatosensory cortex: quantitative morphology of electrophysiologically characterized and biocytin labeled cells. Cereb. Cortex 14 (6), 690–701. Tenke, CE, Schroeder, CE, Arezzo, JC, Vaughan, HG, 1993. Interpretation of high-resolution current source density profiles— a simulation of sublaminar contributions to the visual-evoked potential. Exp. Brain Res. 94 (2), 183–192. Thomas, CG, Menon, RS, 1998. Amplitude response and stimulus presentation frequency response of human primary visual cortex using BOLD EPI at 4 T. Magn. Reson. Med. 40 (2), 203–209. Thomson, DJ, 1982. Spectrum estimations and harmonic analysis. Proc. I.E.E.E. 70 (9), 1055–1096. Tolhurst, DJ, Movshon, JA, 1975. Spatial and temporal contrast sensitivity of striate cortical neurones. Nature 257 (5528), 674–675. Vrba, J, Robinson, SE, 2001. Signal processing in magnetoencephalography. Methods 25 (2), 249–271. Woolrich, MW, Behrens, TEJ, Beckmann, CF, Jenkinson, M, Smith, SM, 2004. Multilevel linear modelling for FMRI group analysis using Bayesian inference. NeuroImage 21 (4), 1732–1747.