 |
||||
|
||||
|
||||
|
||||
Previous Article | Next Article
All surgical procedures were performed according to the National Institutes of Health guidelines. Anesthesia was induced with an intramuscular injection of ketamine hydrochloride (10-15 mg/kg), supplemented in some experiments with xylazine hydrochloride (0.5 mg/kg). Atropine sulfate (0.05 mg/kg, i.m.) was also given at that time to decrease salivation. After tracheal cannulation, the animal was placed in a conventional stereotaxic apparatus and was artificially respirated. Local anesthetic (lidocaine hydrochloride) was applied to wound and pressure points. Sustained anesthesia was achieved by a continuous intravenous infusion of sodium pentothal (4 mg · kg
The Journal of Neuroscience, September 1, 2001, 21(17):6820-6835
The Cortical Representation of the Hand in Macaque and Human Area S-I: High Resolution Optical Imaging
Doron Shoham and Amiram Grinvald Department of Neurobiology, Weizmann Institute of Science, Rehovot 76100, Israel
ABSTRACT
TOP
ABSTRACT
INTRODUCTION
High-resolution images of the somatotopic hand representation in macaque monkey primary somatosensory cortex (area S-I) were obtained by optical imaging based on intrinsic signals. To visualize somatotopic maps, we imaged optical responses to mild tactile stimulation of each individual fingertip. The activation evoked by stimulation of a single finger was strongest in a narrow transverse band (~1 × 4 mm) across the postcentral gyrus. As expected, a sequential organization of these bands was found. However, a significant overlap, especially for the activated areas of fingers 3-5, was found. Surprisingly, in addition to the finger-specific domains, we found that for each of the fingers, weak stimulation activated also a second "common patch" of cortex, located just medially to the representation of the finger. These results were confirmed by targeted multiunit and single-unit recordings guided by the optical maps. The maps remained very stable over many hours of recording. By optimizing the imaging procedures, we were able to obtain the functional maps extremely rapidly (e.g., the map of five fingers in the macaque monkey could be obtained in as little as 5 min). Furthermore, we describe the intraoperative optical imaging of the hand representation in the human brain during neurosurgery and then discuss the implications of the present results for the spatial resolution accomplishable by other neuroimaging techniques, relying on responses of the microcirculation to sensory-evoked electrical activity. This study demonstrates the feasibility of using high-resolution optical imaging to explore reliably short- and long-term plasticity of cortical representations, as well as for applications in the clinical setting.
Key words: monkey; macaque; somatosensory cortex; somatotopy; area S-I; area 1; functional architecture; cortical maps; optical imaging; intrinsic signals; f-MRI; PET; neurosurgery
INTRODUCTION
Research over the past few decades has established that many sensory stimuli are represented in primary sensory cortices in the form of functional maps. Thus, cells with similar stimulus preferences are clustered in cortical columns, forming two-dimensional maps of stimulus properties on the surface of the cortex (Mountcastle, 1957
: Hubel and Wiesel, 1965
Traditionally, studies of cortical spatial representations have relied on the unit recording technique as a mapping tool (Merzenich et al., 1978
Several other reasons also motivated us to implement the intrinsic optical imaging method to monkey primary somatosensory cortex. First, cortical plasticity has been extensively studied in this brain region; a lot of data has accumulated regarding the reorganization of somatotopic maps in response to injury or experience (for review, see Kaas, 1991
Having accomplished high-resolution functional maps in the somatosensory cortex of the macaque, we were motivated to assess how difficult it is to obtain similar high-resolution functional maps of the human brain during neurosurgery. The pioneering efforts by MacVicar et al. (1990)
MATERIALS AND METHODS
The methods for preparing and maintaining the animals and the imaging setup and procedures were primarily similar to those described previously (Ts'o et al., 1990
Experimental procedures
Animals. Optical data for this study were obtained from five adult macaque monkeys (Macaca fascicularis) weighing 7-8 kg. In the first two monkeys we applied only median nerve stimulation, and in the other three we also used the mechanical stimulation paradigm as described below.1 · hr
at 1 kHz) glass-coated tungsten microelectrodes with tip diameter of ~15 µm. We tried to make the penetration angles as close to perpendicular to the cortical surface as possible. Single units were detected and isolated using a window discriminator. All of the spikes with an amplitude higher than approximately three times the noise level were recorded as multiunit activity. The recordings were made in the superficial layers (~300-800 µm) at three different depths for each penetration. The stimuli and the stimulus application protocol were the same as for the optical recording. In each recording the stimulus set was repeated eight times, with an interstimulus interval of 10 sec.
Data analysis
Computing the functional maps. As described above, the optical data from each block (sum of 5-16 trials) were kept as temporal sequences of 5-16 frames for each stimulus condition. In the first step of data analysis, we summed for each stimulus the frames from all of the blocks and all (or a range) of the temporal frame sequence. This summation resulted in a single cortical image per stimulus condition. The functional maps were produced from these images by dividing the image from one stimulus condition (or an average of two or more conditions) by the image for another stimulus condition (or average of a few conditions). This division removes the effect of uneven illumination and produces a map that depends only on the difference in cortical reflectance between the two conditions (or sets of conditions). In all of the imaging sessions, we included a "blank" (no-stimulus) condition. Most of the maps in this study were produced by dividing the image from a particular stimulated condition by the image from the blank condition ("single-condition maps"). In addition we also computed maps by dividing images from different stimuli ("differential maps"). We also examined the development of the optical signals in time. This was done by dividing each of the frames from the temporal frame sequence of a given stimulus condition by the corresponding frame from another stimulus condition (usually the blank). When attaching the stimulators to the different skin locations, we attempted to make all stimuli as identical as possible. However, there is still a possibility that some stimuli were more effective than others. The imaging results showed that the amplitudes of the optical signals evoked by equivalent stimuli to different fingers were quite similar in most cases. However, in some cases there were significant differences in these amplitudes. Thus, when the maps for different stimulus locations were compared or combined, we first normalized each of the maps. Normalization was done by subtracting from each map its median value and dividing the result by the amplitude of maximal activation found in the maps. The maximal activation value was computed from a smoothed version of the map (low-pass filtering with a Gaussian filter,= 120 µ) so that high-frequency noise did not affect the obtained value. Composite finger maps. We combined the information obtained by stimulating each finger separately into a single somatotopic map using two different methods. The first method uses contour lines surrounding a given signal level. The single-finger maps were smoothed with a Gaussian low-pass filter (
Figure 1B illustrates the remarkable improvement of functional maps that this approach can yield. This is the same sequence shown in Figure 1A, after subtracting the first frame. Here the evoked signal is evident, and the somatotopic activation is revealed. The case shown here is an extreme (but not atypical) example of the common slow-noise effect. Some of the recordings were much less affected by this type of noise, but for uniformity of the display, we used first-frame subtraction in all images shown below. Despite the impressive improvements, which can often be achieved with first-frame subtraction, this procedure also has its disadvantages, the most important of which is the introduction of high-frequency noise into the maps. This results from the fact that a single (noisy) image is subtracted from all other frames, and therefore the noise of the first frame is introduced into the analysis in a way that subsequent averaging over frames will not cancel out. Therefore, we used up to four no-stimulus frames to reduce this high-frequency noise. As noted above, the relative effect of the slow noise varies across experiments and even during a single experiment. This variability is possibly related to factors like the physiological state of the animal, the anesthesia level, etc. We have also discovered that quite often the body posture has a very strong effect on this signal. Using the enhanced video system, we could visualize these signals in real time. They often appeared as large (few millimeters) "clouds" that slowly scan the cortex or move in circles. Sometimes we could greatly improve the situation by simply elevating or lowering the body by a few centimeters (relative to the head) or by other changes to the body posture. Analysis of electrical recordings. The spike trains from each recording (eight trials) were used to generate peristimulus time histograms (PSTHs) with 10 msec bins. The PSTHs were displayed along with the raw data ("raster display"). In addition we computed the tuning curves for each recording by counting the spikes (summing over trials) in the 2 sec stimulation period. The baseline firing rate was computed by averaging the spike counts from the 0.5 sec before stimulus onset over all stimuli. The tuning curves from the different penetrations were normalized to have a common maximal value. Human subjects. The 15 subjects who participated in this study were adult patients who had to undergo brain surgery for tumor or AVM resection. They all signed informed consent forms according to regulations of the internal review board of the hospital. Intraoperative optical imaging. All of the experimental procedures have been conducted in conformance with the policies and principles contained in the federal policy for the protection of human subjects, in the declaration of Helsinki, and the protocol approved by the hospital internal review board. Both surgery and imaging were done under general anesthesia using isofluorane. A transparent perspex window was placed on the exposed cortical surface to stabilize it. The cortex was illuminated by two adjustable light guides attached to a Zeiss tungsten-halogen lamp (100 W). Cortical images were obtained using a single macrozoom lens attached to a digital imaging system. Typically we imaged an area of ~50 × 50 mm. To obtain high-contrast images of the cortical surface and its vasculature, we used green light (540 ± 15 nm) and focused the camera on the surface. To image the functional maps, we used red light (605 ± 5 or 630 ± 5 nm) and focused the camera 300-600 µm below the cortical surface. The camera was mounted on a fine x-y-z translator that was attached to a heavy Zeiss surgical binocular stand, replacing the binocular. It was essential to reduce camera vibrations relative to the patient cortical surface. Therefore we built a special stabilizing construction around the stand and its flexible arms. With this construction the camera could be first manipulated coarsely using the degrees of freedom of the arm of the Zeiss stand, and then its position was locked to prevent further movements or vibrations. Additional fine positioning could be performed after locking the system using the x-y-z translator and the z translator attached to the binocular stand. To reduce vibrational noise further, the camera itself was rigidly attached and tightly locked to the patient head holder, after focusing on the target area. The electrical and mechanical stimuli used were very similar to those we optimized in the macaque studies as described above. Data acquisition. The stimulus duration was 2 sec in most recordings, and the interstimulus interval was 7-15 sec. The acquisition of cortical images typically started ~0.5 sec before stimulus onset and lasted for 4-7 sec. During this period we collected 5-16 frames. Each of these frames was the result of an on-line summation of 10-15 video frames. To reduce the noise associated with respiration, in some patients stimulus onset and data acquisition were synchronized to the patient respiration cycle. The complete stimulus set was presented 5-16 times, each time in a different randomized sequence, to average out stimulus-dependent aftereffects. To optimize the off-line image processing, each of the individual responses was saved separately on the disk of the computer. Electrical recording of human surface-evoked potentials. To confirm the functional map obtained by optical imaging, we followed the imaging session in some patients with electrocorticographic recordings. We used surface electrode arrays (Ad-Tech; up to 4 × 5 electrodes; 1 cm interelectrode spacing) to record cortical surface-evoked potentials (SEP) from multiple sites. In addition to the electrode array, we inserted a strip of four electrodes under the dura and the skull lateral to the exposed cortex. These electrodes (electrically connected to each other) served as a common reference to all of the recordings shown in one figure (see Fig. 15). We used the same stimuli (both electrical and tactile) that was used for the imaging, applying single pulses at a rate of 2-5 Hz. The signals were amplified and filtered (10-1000 Hz) using Grass amplifiers and recorded on a personal computer after 12-bit digitization (1700 Hz). The responses to 128-256 repetitions of each stimulus were averaged on-line and saved to disk. At least two such sets of data were collected for each recording.
View larger version (93K):[in this window]
[in a new window]
Figure 1. Improvement of functional maps using first-frame subtraction to remove common slow vascular noise. A, Image sequence obtained with mechanical stimulation of a single fingertip (finger 1) in monkey M3. Each frame was divided by the corresponding frame from the blank condition and scaled such that the full gray scale corresponds to a fractional change of 1.7 × 10
RESULTS
The area explored in this study was the surface of the left postcentral gyrus of the macaque monkey focusing on the somatosensory hand representation. In the macaque monkey, the exposed part of the gyrus contains most of Brodmann's areas 1 and 2. This region is outlined on the schematic macaque brain in Figure 2A, which also shows the vascular pattern imaged in one of the experiments. All subsequent images are at the same orientation or slightly rotated so that the central sulcus parallels the bottom of the image.
View larger version (83K):[in this window]
[in a new window]
Figure 2. Time course of optical response to mechanical stimulation of a single finger. A, Left, A schematic dorsal view of the macaque brain outlining the cortical territory explored in a typical somatosensory imaging session. Right, Image of the cortical surface (left hemisphere, monkey M3) illuminated with green light (540 nm) to emphasize the blood vessel pattern. The somatosensory and motor areas are marked. The anterior part of the somatosensory strip is Brodmann's area 1, and the posterior part is area 2. B, An image series of the left hemisphere in monkey M3 showing the temporal development of the optical response to a weak mechanical indentation stimulation of finger 3 of the right hand. Each frame represents 500 msec of summation of collected video frames. The image at the bottom right shows the corresponding vascular image taken with green light (for general orientation, see A). The finger was stimulated at the distal phalanx using the pneumatic stimulator described in the text (Materials and Methods). Air pressure pulses (10 msec) were delivered for 2 sec at a rate of 10 Hz. The stimulus started concurrently with the acquisition of the second frame (stimulus duration is marked by a black horizontal bar). To obtain these images, we divided each image from the sequence obtained during stimulus application by the corresponding image from the blank (no-stimulus) condition. This procedure, in addition to normalizing for the illumination pattern, nearly eliminates the signals coming from respiration and heart pulsation because data acquisition was always synchronized to both of these cycles. First-frame subtraction was applied. Here and in the following figures, activity shows up as darkening in the image. The full gray scale corresponds to a fractional change of 1 × 10
Single-finger time course
To study the detailed somatotopy of the hand area, we used the pneumatic tactile stimulation system (see Materials and Methods) in three of the monkeys. The series of frames in Figure 2B shows the time course of the intrinsic signals evoked by stimulation of a single finger (finger 3). Two major components are evident in the response. The first component is darkening in a finger-specific stripe of cortex traversing the gyrus (the finger domain, black arrow in Fig. 2B, 2s) and in a second, nonspecific patch [white arrow, same as in Fig. 2B, bottom right for finger 1 activation; further details about this domain are below (see Fig. 7)]. This darkening peaks at the fifth frame (Fig. 2B, 1.5-2 sec after stimulus onset) and decays slowly. In addition there is slow brightening of the whole imaged area, which is strongest in the large blood vessels (see Fig. 2B, bottom right, the vascular pattern). This brightening rises steadily until the end of the recording. This time course appears approximately similar to that observed in our previous studies of macaque V1. Stimulation of the other fingers resulted in a similar activation pattern, with a different specific strip for each of the fingers, and a similar global brightening (Fig. 1B showing the response to finger 1 stimulation in the same recording session). It is well established that such darkening corresponds to electrically active areas, whereas the diffused brightening in the parenchyma and in the large vessels corresponds to enhanced blood flow during the hyperoxygenation phase, which spreads well beyond the electrically active area (Frostig et al., 1990
Single-finger maps
To emphasize the specific finger activation, we summed frames 4-9 (1-4 sec after the stimulus) from sequences like the one shown in Figure 2B. The results are shown in Figure 3. Here and in the following maps, we focused on the activated part of the postcentral gyrus. [Similar maps from two other monkeys are shown below (see Figs. 4, 12).] These maps show that the optical signals evoked by the tactile stimulation of a single finger were strongest in a narrow transverse strip (~1 × 4 mm) across the postcentral gyrus. As expected, we found a sequential organization of the finger representations, with the thumb most anterolateral and the little finger most posteromedial along the gyrus. In addition to these finger-specific domains, the maps show that another patch of cortex just medial to the hand representation was activated by all stimuli. This issue is discussed below (see Fig. 8 for additional details).
View larger version (63K):[in this window]
[in a new window]
Figure 3. Cortical representation of single fingers. Image sequences like the one in Figure 2B were obtained for stimulation of each of the five fingertips. Frames 4-9 of each series (1-4 sec after the stimulus) were averaged to produce the single-finger activity maps shown here. The resulting images were rotated and cropped to focus on the activated part of the postcentral gyrus. The white rectangle on the vascular imaging at the bottom outlines the cropped area. The maps in all subsequent figures are shown at this orientation. The five images surrounding the hand drawing show the areas activated by mechanical stimulation of the five fingers. A somatotopic organization of the hand representation is evident, with the thumb on the right (anterolateral) and the little finger on the left (posteromedial). In addition to the finger-specific activations, there is a patch of cortex just medial to the hand representation that is activated by all stimuli, including a weak activation by finger 5 (see also Figs. 4, 6, in another monkey). Further discussion of this domain is given below (see Fig. 8). The full gray scale corresponds to a fractional change of 1 × 10
Although the centers of the different finger domains formed a clear somatotopic succession along the gyrus, there was considerable spatial overlap between the optical signals evoked by stimulation of neighboring fingers. The largest overlap was between fingers 3, 4, and 5, and the smallest was between fingers 1 and 2. The response to finger 1 stimulation was consistently almost isolated, with very little spatial overlap with the response to finger 2 stimulation.
The postcentral gyrus in the macaque monkey contains most of Brodmann's areas 1 and 2. Area 1 occupies the anterior part of the gyrus (approximately one-half to two-thirds of it), and area 2 occupies the remaining posterior part. In our experiments we did not use independent measures to determine the exact location of the border between areas 1 and 2. Still, some features of the optical maps fit nicely with the presumed areal border. In most cases the single-finger domains were not uniform across the gyrus. The activation was generally strongest in the anterior part (bottom of the images) and weaker in the posterior part. In some cases there was even a gap between the anterior and the posterior activations (e.g., see Fig. 6), and in some cases there was a difference in the mediolateral position of the activation by the same finger in the two parts of the gyrus.
Composite finger maps
Some aspects of the overall organization of the mapped area are easier to study when the information is condensed into a single map. We used two different methods to combine the information from the five single-finger maps into a single image, as explained in Materials and Methods (here we assumed that the response at the peak of the activated area should be the same, and such a simplification may not always be justified). Figure 4 shows these composite maps from the three monkeys in which we mapped the responses to tactile finger stimulation. The left images show the contour-line maps, and the right images show the corresponding WTA maps. The two types of presentations emphasize complementary aspects of the data. The contour maps show the overlap between the evoked signals, and the WTA maps parcel the cortex according to which stimulus was most effective in each part. The somatotopic organization is clear in both types of presentations.
View larger version (88K):[in this window]
[in a new window]
Figure 4. Somatotopic organization of the hand area in three monkeys. Left hemisphere finger maps for right-hand stimulation were obtained in a way similar to the way in which the ones in Figure 3 were obtained. The maps from all three monkeys are summarized here in two formats. Left, The images show contour maps from the three monkeys. Contour lines at the level of 30% of peak activation for each finger are superimposed on the surface vasculature image (imaged with green light). The thin contour lines were computed from a smoothed version of the maps (low-pass filtered with a Gaussian filter,
Because of the (arbitrary) threshold that we selected, the two sets of maps reflect mainly the organization of the anterior part of the gyrus (presumed area 1), where the signals had higher amplitude. The fainter patches of activity in the posterior part of the gyrus (presumed area 2) were not fully captured by the contours or the WTA maps at this threshold level. The relation between the contour lines and the original activation maps can be seen (see Fig. 6) when the individual contours are superimposed on the raw finger maps.
The WTA maps show that, in general, finger 1 had the widest territory and fingers 3-5 had the narrowest. It is important to note that this does not necessarily imply that the stimulation of finger 1 evoked activity over a larger cortical area. In fact, the contour maps (Fig. 4) and the raw finger maps (e.g., see Figs. 3, 6, 12) show that the size of the activated area was approximately the same for all stimuli. The domains of fingers 3-5 appear as narrow stripes in the WTA maps not because of a narrower activation area but because of the larger overlap between the activated areas. This is also demonstrated by the differential finger maps that are described below.
Differential finger maps
It is common practice in optical imaging studies to use differential maps in which the raw images from pairs of "orthogonal stimuli" are divided to show the difference in activation. Equivalently, blank-divided images can be subtracted (Bonhoeffer et al., 1995
View larger version (61K):[in this window]
[in a new window]
Figure 5. Differential finger maps. The five single-finger maps from monkey M4 were used to generate the differential maps for each pair of fingers. The top row shows the images resulting from subtracting the maps of fingers 2-5 from the finger 1 map; the second row shows the finger 2 map minus the maps of fingers 3-5, and so on. Before the images were subtracted, each map was normalized as explained in Materials and Methods, so that the maximal activation (defined as the minimal value in the smoothed version of the map) was mapped to 1.0 and the median value was mapped to 0.0. Next, the resulting differential maps were all scaled such that the range ( 1 to 1) would span the full gray scale.
Reproducibility of the maps
To assess the reliability of the optical maps, we performed reproducibility tests. A typical example is shown in Figure 6 that shows the single-finger maps and the WTA maps from monkey M4. The maps in the right column are from a full recording session of 110 trials. The same data were split into two sets of alternating independent blocks, and the two sets of corresponding maps are shown in the left and middle columns. As expected, these maps are somewhat noisier than the average over all blocks, but the similarity between the two independent sets of maps is evident, confirming the reproducibility of the maps.
View larger version (128K):[in this window]
[in a new window]
Figure 6. Reproducibility of the finger maps. The left and middle columns show two sets of single-finger maps (rows 1-5 from the top) and WTA maps (bottom row) from independent trials in the same cortex (monkey M4). The finger 1 map is at the top, and the finger 5 map is on row 5. These data were collected in 24 blocks of five trials each. For the reproducibility test, these images were divided into two sets of alternating blocks, and the functional maps were computed separately from each of these sets, containing 12 blocks. The average of the two left maps in each row creates the map in the right column. To aid in the comparison, the contour lines that were computed from the full data set were superimposed on all the maps. The color code is the same as that in Figure 4. The full gray scale corresponds to a fractional change of 1.2 × 10
Electrophysiological confirmation of the optical maps
Targeted single-unit and multiunit recordings were used to confirm the optical results. At the end of the imaging session, we inserted microelectrodes into several identified loci in the imaged area. In each locus we recorded single-unit and multiunit responses to the same stimuli that were used for imaging (pneumatic stimulation of each of the finger tips). Multiunit data from such recordings are shown in Figure 7.
View larger version (29K):[in this window]
[in a new window]
Figure 7. Targeted electrophysiological confirmation of the optical maps. A, The optical maps were used to guide electrode penetrations into optically identified loci. The data in this figure are from a recording in the finger 1 area (as determined by optical imaging) of monkey M3. [The penetration site is marked on the vascular image (bottom left) by a red X.] The contours that summarize the imaging results (see Fig. 4) are also superimposed on the vascular image (below the left spike train). The stimuli used were the same as in the imaging session (at each of the 5 fingertips, 2 sec at 10 Hz). The data recorded from the site representing digit 1 show multiunit responses to stimulation of fingers 1-5 (right to left, counterclockwise). The spikes collected in eight trials for each stimulus are shown in the raster displays (bottom), and the resulting PSTHs (10 msec bins) are shown on the top. Stimulation of the thumb elicited a strong response with clear entrainment of the spikes to the stimulus pulses (right spike train). Stimulation of the other fingers elicited no response or even slight inhibition (e.g., digits 2-5 particularly during the response onset). The specificity of the response to finger 1 stimulation is very clear, confirming the optical results. B, The multiunit data from A are summarized here together with data from the three other penetrations, one on the border of digits 2 and 3 (second histogram from the right), one for digit 5, and one from the common patch. The contour finger map from Figure 4 is reproduced here, with marks (X symbols) at the four electrode penetration sites. The tuning curves obtained in each recording site are shown above as color-coded histograms. Thus despite the overlap, the responses are rather tuned. The vertical bars show the normalized spike counts from the 2 sec stimulation period. The dashed horizontal lines show the spontaneous firing level that was computed from the periods with no stimulation. Note that the small inhibition mentioned above is apparent in all three recording sites that correspond to digits 1, 2/3, and 5. Each of the tuning curves was normalized to have a common maximal value.
The raw data from one penetration are shown in Figure 7A. The electrode here was placed into the optically identified finger 1 area. The spike trains shown in the raster displays show the clear selective response to stimulation of finger 1, as predicted from the optical maps. There was no response, or even slight inhibition, to stimulation of the other fingers. Furthermore, the response to the finger 1 stimulation showed a strong entrainment of the spikes to the stimulus pulses. The spikes were tightly locked to the stimulus, and almost all of the spikes during the response fell into single 10 msec bins of the PSTH.
These multiunit data together with recordings from three other penetrations in the same monkey (M3) are summarized as tuning curves in Figure 7B. The second penetration was around the border of the finger 2 and finger 3 domains. The spike data, like the corresponding optical data, show a clear response to both fingers at approximately the same amplitude, with little response to the other fingers.
The third penetration is approximately at the center of the finger 5 domain. The cells recorded here responded best to stimulation of finger 5, but there was also a notable response to stimulation of the neighboring finger 4. This is also consistent with the overlapping activation seen in the optical maps.
The last recording was in the medial patch of cortex, which was activated by stimulation of all of the fingers. This patch and the results from this recording are discussed below.
The single-unit data were very similar to the multiunit data (obviously with fewer spikes), including the multifinger responses.
The "common patch"
As described above, the tactile stimulation of single fingertips usually activated two separate patches of cortex along the postcentral gyrus. The more lateral part, the specific finger domain, was different for each finger. These patches make up the somatotopic map of the fingers that was described in detail above. The secondary activated patch was common to all of the fingers and was located posteromedially to the finger 5 representation. This common patch was observed in each of the three monkeys in which we mapped the responses to the rather weak tactile stimulation (e.g., see Figs. 3, 6, 12). The activation of this patch was strongest when finger 1 or 2 was stimulated and weaker for the other fingers. This is seen most clearly in the differential maps (Fig. 5); the common patch appears black (more active) in all of the maps in which the images from stimulation of fingers 3-5 were subtracted from the fingers 1 and 2 images (Fig. 5, top two rows, apart from the rightmost image on the top) and gray (equal activation) in the other maps. In one animal, activation was evoked almost exclusively by finger 1 stimulation (see Fig. 12). To illustrate the similarity of the architecture of this domain in the three studied animals, we aligned the finger 1 maps from the three experiments (Fig. 8, top left). The size of this patch was comparable with the areas of the individual finger domains. Also like many of the single-finger domains, the common activation in the anterior part of the gyrus was stronger than that in the posterior part.
View larger version (56K):[in this window]
[in a new window]
Figure 8. The common patch. Top left, The three images show the cortical activation by stimulation of the thumb in three monkeys. In addition to the finger-specific domain on the lateral (right) side, there is another activated patch on the medial part (left part of the image). This common patch was activated also by stimulation of the other fingers (data not shown). Top right, The map (monkey M4) shows the activity evoked by stimulation of the wrist using the same pneumatic stimulator that was used for the fingers. This activation pattern also includes the common patch. To assist in comparison with the fingers somatotopy, the contour lines of fingers 1-5 from Figure 4 were superimposed on the map. The full gray scale corresponds to a fractional change of 1 × 10
As part of the confirmation of the optical data with single-unit and multiunit recordings in monkey M3, one of the penetrations to the common patch area was targeted. By using the same stimuli that were used for the imaging, we confirmed the imaging data, including the difference in response amplitude between stimulations of fingers 1 and 2 and stimulations of the other fingers. The multiunit tuning curve is shown in Figure 7B, left histogram, and the spike trains and PSTHs are shown in the bottom of Figure 8. These data confirmed the optical maps. Furthermore, they showed interesting differences in the temporal pattern of response. Although the total number of spikes in the responses to the stimulation of fingers 1 and 2 was similar (Fig. 7B), there was a clear difference in the distribution of these spikes in time (Fig. 8). The spikes evoked by finger 1 stimulation were more tightly locked to the stimulus pulses, and those evoked by finger 2 stimulation were more spread over the stimulus period. This is reflected also in the difference in the amplitude of the PSTHs. The responses to the other fingers (3-5) were not only weaker than that of fingers 1 and 2 but also different in the temporal pattern; the responses to the first few pulses were much more time-locked than were those to the last pulses.
In this penetration, after completing the recording of the response to the computer-controlled pneumatic stimuli, we also mapped the receptive field of the recorded units manually by lightly touching the skin at different locations. We found that the units in this recording site were strongly activated by slight touch of all parts of the hand, the wrist, or the forearm and also weakly activated by upper arm stimulation.
In another experiment (M4), we attached a pneumatic stimulator, like those used to stimulate the fingers, to the wrist (approximately above the median nerve) and imaged the optical signals. The optical map shows that this stimulus also activated the common patch (Fig. 8, top right). However with the wrist stimulus, the activation was more uniform across the gyrus, unlike the activation caused by stimulating the fingers. In addition the posterior part of the gyrus (top of the image; presumably area 2) was activated almost to the same level as the anterior part (area 1).
Median nerve stimulation
At the first stage of this study, we mapped the cortical response to an electrical stimulation of the median nerve at the wrist. This was an effective stimulus that resulted in localized activation in the somatosensory cortex (~10 mm along the gyrus). In those experiments we could not examine the correspondence of median nerve activation to the finger maps because we did not map the individual fingers. In one of the latest experiments, when we also mechanically stimulated individual fingers, we could compare the activation from the two types of stimuli in the same cortex. The time course of cortical activation resulting from the median nerve stimulation is shown in Figure 9. The amplitude of the optical signal was similar to that of the signal evoked by the tactile stimulus. The pattern of cortical activation resulting from the electrical stimulus is shown in Figure 10, in which the images obtained in the period 1-4 sec after stimulus onset were integrated. The contour lines from the finger maps are superimposed on the image. The comparison of the median nerve map with the finger maps shows that in this case the strongest activation was in the common patch area, medial to the hand representation. The pattern is quite similar to that evoked by a tactile stimulation of the wrist at the same site (see Fig. 8, top right). This suggests that some of the activation we saw after electrical stimulation in this case did not result from stimulation of the medial nerve but rather came from direct stimulation of the receptors at the wrist below the electrodes.
View larger version (93K):[in this window]
[in a new window]
Figure 9. Time course of optical response to electrical stimulation of the median nerve. An image series shows the temporal development of the optical response to a right median nerve stimulation in monkey M4. Each frame represents 500 msec. The median nerve was stimulated at the wrist via a pair of EKG electrodes. Current pulses of 1 msec were delivered for 2 sec at a rate of 10 Hz. The stimulus started concurrently with the acquisition of the second frame (stimulus duration is marked by a black horizontal bar). The full gray scale corresponds to a fractional change of 1.5 × 10
View larger version (115K):[in this window]
[in a new window]
Figure 10. Cortical activation by electrical stimulation of the median nerve. An activity map produced by averaging the last six frames from Figure 9 (corresponding to the period 1-4 sec after stimulus onset) is shown. The contour lines of fingers 1 and 5 from Figure 4 are superimposed on the map to aid in comparison with the fingers somatotopy. Note that a much larger area was activated (light gray area darker than the top of the image), but the common patch was the strongest area activated and occupied much of the dynamic range of this figure. Note that median nerve stimulation in different experiments on different monkeys or human subjects often yields highly variable results, presumably because of the difficulty in reproducing exactly the same activation of the nerve in different subjects (data not shown). The full gray scale corresponds to a fractional change of 1.5 × 10
Stability of the maps
Some experimental questions require repeated imaging over many hours or days or imaging of the same cortical area before and after a time-consuming experimental manipulation. It is then important to verify the stability of the maps over many hours. To assess the stability of the optically imaged somatotopic maps in our preparation, we compared the maps obtained at the early stage of the experiment with those obtained toward the end, ~28 hr later. Figure 11 shows that the maps were very stable over this period. The maps shown here are differential maps obtained by simultaneously stimulating either fingers 1 and 2 or fingers 3-5. We often used this pair of stimuli in the beginning of the experiment, because it allowed us to delineate the region of interest and to assess the quality of the recording very quickly.
View larger version (101K):[in this window]
[in a new window]
Figure 11. Stability of cortical maps. Top, The images are differential maps in which the image obtained during simultaneous stimulation of fingers 1 and 2 was divided by the image from stimulation of fingers 3-5. The data for the left map were collected at the beginning of the experiment, and the map on the right was obtained ~28 hr later (in 28 hr, ~50 different maps can be accumulated under different stimulation conditions). The full gray scale corresponds to a fractional change of 0.8 × 10
Rapid functional imaging
Because the optical signals are small and susceptible to various noise sources, extensive averaging is usually needed to obtain high-quality maps. There are, however, situations in which the recording time is limited and does not allow much averaging. In such situations it is crucial to be able to obtain at least an approximate version of the functional maps very quickly. In the current series of experiments, under optimal conditions for the animal, the functional maps could be obtained extremely rapidly. In the best recordings an experienced observer could detect the map even after a single trial. A clear map was often visible after a single block of only a few trials. This is demonstrated in Figure 12. The maps on the right were obtained by averaging the responses to only five stimulus presentations (and five blanks). A comparison of these maps with those obtained with more extensive averaging (left) shows that the functional borders are quite visible with so little averaging. Because we presented approximately five stimuli per minute, the imaging of all five finger maps took only ~6 min. The complete data set (55 trials) was obtained in a little over an hour.
View larger version (117K):[in this window]
[in a new window]
Figure 12. Rapid functional imaging. Single-finger maps from monkey M5, with finger 1 at the top and finger 5 at the bottom. Right, The images are from a single block of five trials. Left, The corresponding images are averages over 11 such blocks. The data collection for these maps took ~6 min per block. All images were autoclipped to occupy the full dynamic range of each image.
Having optimized the technique to such an extent, we attempted to apply this methodology in the operating room (OR) to assist neurosurgeons in delineating functional borders in the human brain before surgical procedures.
Intraoperative optical imaging of human patients
We used optical imaging intraoperatively in 15 patients undergoing tumor or AVM resection. In most of these patients (13), the surgery involved exposing the central region of the cerebrum, close to the sensorimotor areas. Many imaging sessions were used to adapt the imaging setup and procedures to the special needs of the intraoperative use. In some of these initial studies, the noise level was so high that it was expected to mask any evoked signals of the amplitude found in the present monkey studies. By optimizing various aspects of the imaging apparatus and procedures as described in Materials and Methods, we managed to reduce the noise to a level that should allow clear visualization of the expected evoked optical signals. In some of these relatively low-noise imaging sessions, we thought that we detected evoked signals. On closer inspection, however, we found that the exact pattern of these responses varied from trial to trial even if the same stimulus was used. Thus, in all cases except one, we were not able to exclude that these responses were contaminated by vascular activity rather than neuronal activity. We concluded that it is important to understand the nature of the spontaneous vascular activity in anesthetized human subjects before optimal, noise-free, and reproducible functional maps can be readily obtained. Therefore, we started exploring the behavior of the cortical microcirculation without a sensory stimulus.
Spontaneous fluctuations in blood flow/volume in anesthetized patients
Here we focus on one case we studied extensively, although similar spontaneous fluctuations in flow were seen in most subjects. This paragraph describes the behavior of spontaneous vascular activity. The next paragraph then shows optical imaging of the hand representation of this subject, obtained by signal averaging, which was subsequently confirmed by electrical mapping. The case we report involved a 44-year-old male with a low-grade glioma in the left hemisphere that was estimated to be closely posterior to the somatosensory cortex. In a single-trial recording of 6.5 sec sessions, we observed two different types of spontaneous vascular activity. First, slow changes in the oxygenation levels were readily seen in the large vessels as well as the parenchyma (e.g., Fig. 13A-C). Second, these vascular changes were often followed by a considerable darkening (Fig. 13C-F, white arrows) or whitening (Fig. 13D-F, black arrows) of the entire parenchyma in much smaller and well defined cortical areas, presumably "serviced" by a defined set(s) of an artery and veins. This vascular activity is presumably related to the well known vasomotion fluctuations (Schiff, 1854
View larger version (138K):[in this window]
[in a new window]
Figure 13. Visualization of spontaneous vascular activity in the human cortex. Each row shows spontaneous changes (without a stimulus) in cortical reflection as a function of time. Frame duration is 500 msec. A, B, These rows depict an increase in vascular oxygenation level (vessels brighten) over a large cortical area. The full gray scale (clipping range) corresponds to a large fractional change of 2.0 × 10
Imaging the human hand representation and its electrical confirmation
Obviously, the very large fluctuations can be automatically rejected by preset criteria. The smaller type of noise (e.g., Fig. 13D,E) could be minimized with signal averaging. In this patient we imaged optical- and electrical-evoked responses using an electrical stimulation of the median nerve at the wrist. This stimulus produced larger evoked activity then did tactile stimulation and should reveal the entire hand representation as shown in Figure 14. The left panel of Figure 14 shows the imaged cortical region with the motor and sensory strips marked as identified by the imaging and by the electrical recording. The tumor was just posterior to the imaged area. The optical map of the intrinsic signals evoked by the median nerve stimulation is shown in the middle panel of Figure 14, averaging over 16 repetitions that were summed and divided by the average image from the no-stimulus condition (also 16 repetitions). The dark triangle-like shape at the top is the somatosensory hand area. The right panel in Figure 14 shows the flat control map obtained by comparing two sets of no-stimulus conditions (16 repetitions in each). Evidently the noise was averaged out.
View larger version (65K):[in this window]
[in a new window]
Figure 14. Intraoperative imaging of the somatosensory hand area using median nerve stimulation. Left, Image of the cortical surface (left hemisphere) illuminated with green light (540 nm) to emphasize the blood vessel pattern. The somatosensory and motor stripes as determined by optical imaging and confirmed with an electrocorticogram (see Fig. 15) are marked. Middle, Optical map of the area activated by median nerve stimulation. To minimize the noise, trials showing large vascular noise (such as that seen in Fig. 13A,B) above a fixed preset value were autorejected by the analysis program. To obtain this functional map, the average image (16 trials) from the stimulated conditions was divided by the average image from the blank condition (no stimulus, 16 trials). Right, Flat map showing the control map obtained by dividing the sum of the same blank condition by another independent set of the second blank condition we used as a control (16 trials each).
Stimulation of individual fingers always resulted in much weaker maps in the same area. Because of the current signal-to-noise ratio, we were unable to localize the individual finger representation within the general area of the hand, in a reproducible manner.
Figure 14 also demonstrates some of the problems associated with trying to estimate visually the location of even the major sulci during surgery; because of the small amount of cortex exposed, it was difficult to determine the identity of the sulci seen. Furthermore, in this case it was impossible to determine even the course of the sulci in the exposed area, because they were hidden below large blood vessels. Looking at the cortex in this particular case, we suspected that the large blood vessel, which runs diagonally at the top left of the image, parallels the course of a sulcus. As the imaging data showed, this sulcus (identified by the imaging as the postcentral sulcus) coincides with the blood vessel only in the bottom of the image, and where the blood vessel continues along a straight line across the postcentral gyrus, the sulcus takes a left turn.
To confirm the optical map, we used the well established method of recording SEP from the cortical surface. We used an array of 4 × 5 electrodes (Fig. 15, left) and recorded each time from a selected group of 8 electrodes simultaneously. The results from two such recordings are shown in Figure 15, right. The electrodes from which these recordings were done are colored yellow on the photographic image of the brain (Fig. 15, left). The median nerve stimulation (Fig. 15, right, top) resulted in a large signal over a large cortical area nearly coinciding with the area that was identified as the hand area in the imaging session. The tactile stimulation of a single finger resulted in the SEP signal, which had a lower amplitude than did the median nerve stimulation (note the calibration bars) but was much more localized, almost limited to a single electrode (labeled by a green circle; Fig. 15). These results indicate that the resolution of the surface-evoked potential is better than traditionally believed. It is difficult to assess whether the pioneering OR studies by optical imaging offered a better signal-to-noise ratio. In our hands it appears that further improvements are needed before high-resolution optical imaging can be obtained from anesthetized patients.
