Medial prefrontal cortex generates frontal midline theta rhythm

Cognitive Neuroscience

NeuroReport NeuroReport 10, 675±679 (1999)

FRONTAL midline theta rhythm (Fmè) is a distinct theta activity of EEG in the frontal midline area that appears during concentrated performance of mental tasks in normal subjects and re¯ects focused attentional processing. To tomographically visualize the source current density distributions of Fmè, we recorded Fmè by using a 64-channel whole-head MEG system from four healthy subjects, and applied a new analysis method, synthetic aperture magnetometry (SAM), an adaptive beam forming method. Fmè was observed in the MEG signals over the bilateral frontal regions. SAM analysis showed bilateral medial prefrontal cortices, including anterior cingulate cortex, as the source of Fmè. This result suggests that focused attention is mainly related to medial prefrontal cortex. NeuroReport 10:675±679 # 1999 Lippincott Williams & Wilkins.

Key words: Anterior cingulate cortex; Attention; Brain mapping; Calculation; Frontal midline theta rhythm; Magnetoencephalography; Medial prefrontal cortex; SAM; Theta wave

Introduction Frontal midline theta rhythm (Fmè) is a train of rhythmic waves, observed at a frequency of 6±7 Hz, having a focal distribution with maximum around the frontal midline in the EEG of normal subjects. The train usually lasts for several seconds and tends to wax and wane [1,2]. Fmè is associated with performance of various mental tasks (e.g. mental arithmetic, tracing a maze, counting the number of cubes piled in a 3-dimensional representation, imaging a scene etc.). For generating marked Fmè, it is necessary to concentrate on the tasks during an extended period of time, thus maintaining focused attentional processing. There have been previous EEG and MEG studies to determine the generators of Fmè [3±8]. By applying optical ¯ow detection techniques in imaging processing, Inouye et al. [3] found that the potential distributions of Fmè showed bilaterally symmetric movements in frontal area, and postulated the presence of two different source areas in each hemisphere. Gevins et al. [4,5] analyzed generators of Fmè by dipole modeling in EEG, and speculated that Fmè might re¯ect engagement of the anterior cingulate cortex. Sasaki et al. [6,7] recorded Fmè with MEG and estimated the equivalent cur0959-4965 # Lippincott Williams & Wilkins

Medial prefrontal cortex generates frontal midline theta rhythm Ryouhei Ishii,CA Kazuhiro Shinosaki, Satoshi Ukai, Tsuyoshi Inouye, Tsutomu Ishihara, Toshiki Yoshimine,1 Norio Hirabuki,2 Hiroshi Asada,3 Taizo Kihara,4 S. E. Robinson5 and Masatoshi Takeda Departments of Neuropsychiatry, 1 Neurosurgery and 2 Radiology, Osaka University Medical School, Yamada-oka 2-2, Suita City, 565-0871; 3 Department of Earth and Life Sciences, Osaka Prefecture University; 4 Aloka. Co. Ltd., Japan; 5 CTF Systems Inc., Canada CA

Corresponding Author

rent dipoles of Fmè in various parts of the lateral frontal lobes of both hemispheres. The aim of this paper is clari®cation of the speci®c role of prefrontal cortex in focused attentional processing. To accomplish this, we used MEG to measure the electrophysiological activity of subjects who generated marked Fmè during mental arithmetic tasks. The relationship of the MEG activity to functional anatomy of human was estimated using a new functional neuroimaging analysis, SAM [9].

Materials and Methods Subjects and tasks: Four healthy volunteers (three males and one female, age 28  3 years) were studied. All subjects were right-handed. Subjects gave their informed consent for the experimental procedures. Subjects were seated in a relaxed position with their heads positioned within a helmet-shaped MEG sensor array. During MEG recording, subjects were instructed to perform mental calculation for 2± 3 min. One subject calculated the power of three successively, i.e., 32 , 33 , 34 and so on, and the others serially subtracted 7 from 1000 or added 4 ®gures presented in front of the subjects. Fast and continVol 10 No 4 17 March 1999

675

NeuroReport uous calculation was required in order to maintain focused attentional processing and produce marked Fmè. MEG and EEG recordings: MEG recordings were obtained using a helmet-shaped 64-channel SQUID sensor array (NeuroSQUID Model 100, CTF Systems Inc.), within a magnetically shielded room. Each of the 64 primary sensors used a ®rst-order axial gradiometer ¯ux transformer. Ambient magnetic noise was reduced further by synthesizing third-order gradiometer response, in ®rmware, using the reference SQUID sensor array [10]; this was especially effective at reducing low-frequency noise. In one subject, EEG was recorded from 32 locations over the surface of the head simultaneously, corresponding to points de®ned by the international modi®ed 10-20 system. Electrodes were referenced to the linked ears. All electrodes were Ag/AgCl. Both MEG and EEG signals were digitized at 250 Hz, and ®ltered using a 60 Hz notch ®lter and 100 Hz low pass ®lter. The resulting data were recorded on disk and analyzed of¯ine. During recording, subjects were instructed to concentrate on mental calculation, thereby producing marked Fmè. When Fmè was observed for . 10 s, subjects were instructed to stop calculating, thus providing a non-Fmè control MEG signal. The MEG data recording consisted of alternating 10 s periods of Fmè generation (active-state) followed by 10 s of control-state activity. More than eight epochs of 20 s duration were recorded for each subject. Data analysis: Topographic maps of Fmè were calculated by plotting the averaged spectral power of EEG and MEG activity, in a 6±7 Hz band, onto a ¯attened 2-D view of the head. Spectra were estimated by fast Fourier transform (FFT) for activity over the marked Fmè trials. The tomographic distribution of Fmè was determined from the unaveraged MEG measurements, using SAM (synthetic aperture magnetometry) [9]. SAM estimates the three-dimensional distribution of electrophysiological source currents from MEG measurements. The SAM images are fused with each subject's MRI, resulting in a functional image, relating brain anatomy to function. The signal detected by any single MEG sensor results from the linear superposition of magnetic ®elds generated over a substantial cortical area. Since each individual source contributes differently to the signals observed from multiple sensor positions, the superposition of sources can be partially resolved by forming a linear combination of sensors; that is, one forms combinations such that a desired source sums constructively, while signals from unwanted sources sum destruc676

Vol 10 No 4 17 March 1999

R. Ishii et al. tively. This linear sensor combination, computed by SAM, may be thought of as a synthetic sensor, and is based upon adaptive beam-forming analysis [11,12]. A tomographic image is generated by applying the SAM beamformer to an array of coordinates, de®ning some region of interest. Regions of signi®cant cortical activation are determined by computing the Student's t-statistic from the source power of active and control states and their noise content, on a voxel-by-voxel basis. Since it is the t-statistic and not the source power that is mapped, the result is a statistical parametric map (SPM) [13]. SAM is not an inverse solution but a spatial ®lter, so the results of this analysis are unique. First, data were ®ltered into six frequency bands: 0.5±4 Hz, 4±8 Hz, 8±13 Hz, 13±25 Hz, 25±60 Hz and 0.5±100 Hz. Second, SAM was used to generate a 16 3 12 3 12 cm volumetric image of root-mean squared (RMS) source activity from the ®ltered MEG signals, with a 2.5 mm voxel resolution. Finally, the Student's t was computed, on a voxel-byvoxel basis, as the difference between the source activity of the active (eight epochs of Fmè-active 10 s) and the control (eight epochs of successive control 10 s) states, divided by their ensemble standard error, which included both instrumental (SQUID sensor) noise and experimental variance. We referred to the resulting image as a Student's tstatistic parametric map, which was then fused with the corresponding MRI.

Results There were marked Fmè at the frontal channels, during the 10 s calculating periods (Fig. 1). Theta band activity was detected continuously in the calculating state, whereas there was no frontal theta activity in the resting state. In Fig. 2, contour maps of spectral amplitude for theta band activity both of EEG and MEG of same subject were shown on a ¯attened 2-D view of the head. In the calculating state, there were some peaks of frontal theta band activities that, on the contrary, could not be found in the resting state. Using SAM analysis, the SPM image of the Student's t-statistic computed by the source current density of Fmè, was visualized tomographically over the large area of bilateral medial prefrontal cortices (Fig. 3). The most prominent and robust activities were found in cortex between superior frontal gyrus and anterior cingulate gyrus. Other subjects showed the same localization of current density distributions, but the maximum Student's t-value of the source current density were different among subjects. Other fre-

Medial prefrontal cortex generates frontal midline theta rhythm

1

Calculating state

3

2

4

NeuroReport Resting state

1

2 MEG 3

Fz

4

Fz

EEG

1pT 100uV 1s

FIG. 1. Simultaneous records of EEG and MEG in calculating and resting states. The sites for EEG and MEG channel are indicated in the left ®gures. Band pass ®lter was 2±50 Hz for EEG and MEG.

fT rms/√Hz uV/√Hz 200 40 185

37

171

34

157

31

143

28

128

25

114

22

100

20

85

17

71

14

57

11

42

8

28

5

14

2

0

0

FIG. 2. Contour maps of spectral amplitude for theta band activity (5.9±7.5 Hz) both EEG and MEG of subject 1 in calculating and resting states. Spectra were estimated by fast Fourier transform (FFT).

Vol 10 No 4 17 March 1999

677

NeuroReport

R. Ishii et al.

FIG. 3. The SPM images of the Student's t-statistic computed by the source current density of Fmè in four subjects. The most prominent and robust activities were found in bilateral prefrontal cortices, including anterior cingulate cortex.

quency band activities did not show any discrete current sources.

Discussion We investigated the source current density distribution of Fmè by using whole head MEG system and the new functional neuroimaging method, SAM. The main ®nding is that the source of Fmè is distributed over a large area of bilateral medial prefrontal cortices, including anterior cingulate cortex. It has already been suggested that Fmè re¯ects focused attentional processing, so the present result shows that focused attention is mainly related to bilateral medial prefrontal cortices. The present MEG study did not show a marked magnetic theta activity in the parieto-occipital area of either the left or right hemisphere, which seemed to correspond to one of the areas of the `calculation center' reported by neuropsychological studies [14]. Wave form and distribution of Fmè in MEG were insensitive to the type of information being processed, suggesting that the extracted Fmè compo678

Vol 10 No 4 17 March 1999

nents in MEG were responsible for the common situation rather than the speci®c processing activity that the subjects were engaged in. Many positron emission tomography (PET) studies [15±18] pointed out the role of the medial prefrontal cortex, including anterior cingulate cortex, in attention. Bench et al. [15] measured regional cerebral blood ¯ow (rCBF) using PET during the performance of the Stroop task and found that the most robust responses occurred in the medial prefrontal cortex. They concluded that the medial prefrontal cortex was mainly involved in attentional processing through the selection and recruitment of task execution. Siegel et al. [16] showed that medial prefrontal region was important in the normal performance of the continuous performance task, a test of sustained attention. Our present result is supported by these rCBF studies. Sasaki et al. [6,7] recorded Fmè with MEG and estimated the equivalent current dipoles (ECDs) of Fmè in various parts of the lateral frontal lobes of both hemispheres. In this study, we got different experimental results with their ®ndings. Explanation

Medial prefrontal cortex generates frontal midline theta rhythm for such discrepancy between these two results may be as follows. First, the Fmè source may not be dipolar. The ECD analysis is valid only for discrete sources with small spatial extent, and having high signal-to-noise ratio (SNR). The source of the Fmè may extend over a large cortical area, and cannot be averaged to improve its SNR. Second, the SAM analysis emphasizes the distribution of the multiple discrete sources oscillating asynchronously at a theta-band frequency, and rejects that portion of the MEG signal arising from synchrony over an extended cortical area. Although the dominant MEG Fmè signal is due to synchrony that appears to be from extended source, a portion of that signal is due to independent discrete sources. Since SAM emphasizes these discrete sources, it may identify different generators than would be deduced from ECD analysis. Third, axial gradiometer sensors may provide better SNR for the Fmè source than do planar gradiometer sensors. In this study, we recorded Fmè using a 64-channel whole cortex covering magnetometer system with SQUID-based ®rst-order axial gradiometer sensors. This type of sensor could detect the magnetic ®eld generated from slightly deeper sources in the brain than the planar gradiometer sensors [19]. Fourth, the environmental magnetic noise in theta-band was lower for the present study. The system which we used in this study had the software constructed third-order spatial gradients [10], providing exceptionally good low frequency performance. In particular, for studies of spontaneous brain activities, such as Fmè, which could not be averaged to enhance SNR suf®ciently, the software constructed third-order spatial gradients showed its ability to cancel environmental noise and extract the weak signal of the brain. Our data might be less contaminated by environmental noise in measured signals of low frequency band. Nevertheless, by employing the new analysis method, SAM, it is possible to obtain a best estimate for spatial distribution of cortical current power of Fmè. The 4±7 Hz wave was named the theta wave after the thalamus, which was proved by clinical studies to generate 4±7 Hz frequency band activities [20]. Sleep spindle studies [21] revealed that subcortical structures, such as thalamus, operate as pace-makers of the rhythmic activities, and the cortex receives projections from subcortical structures, thus generating the observed brain waves. Medial prefrontal cortex, including anterior cingulate cortex, has tight

NeuroReport

and functional connections with the thalamus [22,23], so we can assume that medial prefrontal cortex might be driven by the pacemakers of thalamus and work as generators of the potential of Fmè. The stereotypical experimental paradigm for MEG and EEG has employed the averaged evoked response to enhance SNR suf®ciently to ®t the observations at a speci®ed latency to an ECD. This new paradigm, based upon statistical parameteric imaging by SAM, extends the usefulness of MEG beyond that of localizing solely primary sensory or motor cortical centers. This method may eventually expand the application of MEG into clinical and research areas that were previously occupied by PET and functional MRI.

Conclusion Using whole head MEG system and the new analysis method, SAM, we have demonstrated that the large area of bilateral medial prefrontal cortices, including the anterior cingulate cortex, generates Fmè. We conclude that focused attention is mainly related to bilateral medial prefrontal cortices. References 1. Ishihara T and Yoshii N. Electroencephalogr Clin Neurophysiol 33, 71ÿ80 (1972). 2. Futagi Y, Ishihara T, Tsuda K et al. Electroencephalogr Clin Neurophysiol 106,392ÿ399 (1998). 3. Inouye T, Shinosaki K, Iyama A et al. Neurosci Lett 169,145ÿ148 (1994). 4. Gevins A, Leong H, Smith ME et al. Trends Neurosci 18, 429ÿ436 (1995). 5. Gevins A, Smith E, McEvoy L et al. Cerebr Cortex 7, 374ÿ385 (1997). 6. Sasaki K, Tsujimoto T, Nambu A et al. Neurosci Res 19, 229ÿ233 (1994). 7. Sasaki K, Tsujimoto T, Nishikawa S et al. Neurosci Res 26, 79ÿ81 (1996). 8. Iramina K, Ueno S and Matsuoka S. Brain Topogr 8, 329ÿ331 (1996). 9. Robinson SE and Cheyne D. J Jpn Biomag Bioelectromag Soc 10, 180ÿ183 (1997). 10. Vrba J, Betts K, Burbank M et al. IEEE Trans Appl Supercond 3, 1878ÿ1882 (1993). 11. Frost OL. Proc. IEEE 60, 926ÿ935 (1972). 12. Robinson SE and Rose DF. Current source image estimation by spatially ®ltered MEG. In: Hoke M, Eme SN and Okada YC, eds. Biomagnetism: Clinical Aspects: Proceedings of the 8th International Conference on Biomagnetism. New York: Elsevier, 1992: 761±765. 13. Friston KJ, Worsley KJ, Frackowiak RSJ et al. Hum Brain Mapp 1, 214ÿ220 (1994). 14. Kahn HJ and Whitaker HA. Brain Cogn 17, 102ÿ115 (1991). 15. Bench CJ, Frith CD, Grasby PM et al. Neuropsychologia 31, 907ÿ922 (1993). 16. Siegel BV Jr, Nuechterlein KH, Abel L et al. Schizophr Res 17, 85ÿ94 (1995). 17. Nobre AC, Sebestyen GN, Gitelman DR et al. Brain 120, 515ÿ33 (1997). 18. Carter CS, Mintun M, Nichols T et al. Am J Psychiatry 154, 1670ÿ5 (1997). 19. Hari R. Magnetoencephalography as a Tool of Clinical Neurophysiology. In: Niedermeyer E and Lopes da Silva FH, eds. Electroencephalography. Basic Principals, Clinical Applications, and Related Fields. 3rd edn. Baltimore: Williams & Willkins, 1993: 1035±1061. 20. Walter WG and Dovey VJ. J Neurol Neurosurg Psychiatry 7, 57±65 (1944). 21. Striade M. Cellular substrates of brain rhythms. In: Niedermeyer E and Lopes da Silva FH, eds. Electroencephalography. Basic Principals, Clinical Applications, and Related Fields. 3rd edn. Baltimore: Williams & Willkins, 1993: 27±62. 22. Masterman DL and Cummings JL. J Psychopharmacol Oxf 11, 107ÿ114 (1997). 23. Hsu MM and Shyu BC. NeuroReport 18, 2701ÿ2707 (1997).

Received 2 December 1998; accepted 4 January 1999

Vol 10 No 4 17 March 1999

679