- Email: [email protected]
A new approach to optical imaging applied to rat barrel cortex
JOURNALOF NEUROSClENCE METHODS
ELSEVIER
Journal of Neuroscience Methods 54 (1994) 39-47
A new approach to optical imaging applied to rat barrel cortex Bret E. Peterson a,,, Daniel Goldreich a
b
Bioengineering Group, University of California, Berkeley and San Francisco, San Francisco, CA, USA; b Keck Center for Integrative Neuroscience, University of California, San Francisco, CA, USA Received 10 August 1993; revised 14 March 1994; accepted 18 April 1994
Abstract
Several groups have described using intrinsic optical imaging to form images of activity patterns in the cortex. Because the signal is small, the general approach has been to use expensive camera equipment with a high dynamic range to make these measurements. However, by using signal averaging to compensate for lower dynamic range, images can be obtained using equipment already available in many laboratories. This modified technique has been used for imaging activity in 'barrel' cortex of the rat. A map of the representation of a single whisker as determined by the imaging technique corresponded well with a similar map made using standard electrophysiology. A map of several whiskers was made by overlaying images of single-whisker representations. The details of the images differ from those previously described. Possible mechanisms for the signal are discussed.
Keywords: Optical imaging; Intrinsic imaging; Vibrissa; Barrels; Cortex; (Rat)
I. Introduction
In 1986, Grinvald and colleagues described a slow time course signal that interfered with their voltagesensitive dye measurements. They found that they could use these 'intrinsic optical signals' to image activity correlates in the brain. The technique has since been used to form images that resemble physiological maps in somatosensory and visual cortices (Grinvald et al., 1986, 1991; Ts'o et al., 1990; Bonhoeffer and Grinvald, 1991; Gochin et al., 1992), and to image language centers in human cortex (Haglund et al., 1992). The intrinsic imaging technique uses the optical absorption properties of hemoglobin to detect changes in the local cortical blood supply that occur when neuronal activity increases. Feedback systems in the vasculature can respond to the increased metabolic demand of active cells by increasing local blood flow (Fox and Raichle, 1986). The resulting increased tissue hemoglobin content can be imaged as a change in light absorbency. Furthermore, the amount of deoxygenated
* Corresponding author. Elsevier Science B.V.
SSDI 0 1 6 5 - 0 2 7 0 ( 9 4 ) 0 0 0 5 9 - P
hemoglobin leaving the capillary beds increases as neuronal tissue becomes metabolically activated. Because deoxygenated hemoglobin has a different absorption spectrum than does its oxygenated counterpart, a change in the amount of deoxygenated hemoglobin results in a change in the optical properties of the activated tissue (Frostig et al., 1990). A variety of other techniques are used for imaging the metabolic activity of electrically active brain regions. These include positron emission tomography (PET), single-photon emission computed tomography (SPECT), functional magnetic resonance imaging (MRI), and 2-deoxyglucose autoradiography (2-DG). PET, SPECT and MRI are high-cost, non-invasive techniques with relatively low spatial resolution that have been designed for use on humans. Because they require large expensive equipment, they are not readily available to most laboratories. 2-DG is an animal research technique with high spatial resolution. However, it requires postmortem processing and therefore can be used to obtain only a single image of a brain region per animal. Optical imaging can be used on both research animals (e.g., Grinvald et al., 1986) and humans (Haglund et al., 1992). It is more invasive than the other metabol-
40
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
ically based imaging techniques such as PET, SPECT, and MRI in that it requires an open cranium for optimal imaging (but see Chance et al., 1993), but it offers greater spatial resolution comparable to 2-DG imaging. Unlike 2-DG imaging, it can be used to make repeated measurements in a single animal. Unfortunately, its reported expense and complexity (e.g., Optical Imaging, $75,000) have apparently dissuaded laboratories from using this powerful technique. We describe here a new approach to optical imaging that allows it to be applied inexpensively and simply, using equipment that is readily available to many research laboratories. We have developed our modified version of the technique in the somatosensory whisker 'barrel' cortex of the rat. However, it should prove applicable to other species and modalities.
Experimental Set Up
Camera and
Dissecting Scope
-j Macintosh Ilci
Control Box
2. Preparation of the barrel cortex Sprague-Dawley rats were anesthetized with sodium pentobarbital (Nembutal, 55 m g / k g , i.p.) and given 0.1 ml of atropine sulfate i.m. Rectal temperature was monitored and maintained at 37.5°C with a thermostatically controlled heating pad. Femoral venous and tracheal cannulae were introduced. The atlanto-occipital membrane covering the cisterna magna was transected to reduce cerebrospinal fluid pressure. A unilateral craniotomy was performed over the barrel cortex of SI. The dura over the cortex was reflected, and the exposed cortical surface was covered with silicon oil. Intravenous pentobarbital was administered as necessary to maintain an areflexic level of anesthesia. Physiological recordings were often made to verify the location of key barrels. These recordings were made using 10 p.m carbon-fiber electrodes (ArmstrongJames and Millar, 1979) that were introduced orthogonal to the surface and inserted 580 ~ m beneath the pia with a microdrive. Neuronal responses were amplified and connected to a speaker. The response properties of each penetration site were characterized qualitatively by tapping lightly on each whisker with a glass probe and noting the neuronal response. A map of the whisker representation ('barrel' cortex in SI) was created by recording the characterization of penetration sites made at 200 ~ m intervals.
3. Basic imaging technique The optical imaging setup is shown in Fig. 1. The cortex was illuminated with narrow band red light (mean wavelength: ~ 660 nm) from high-intensity light-emitting diodes (LEDs). These LEDs provide a convenient, inexpensive source of a wavelength of light that is differentially absorbed by oxygenated and de-
Fig. 1. Experimental set-up. The camera is used to image the cortex via the dissecting microscope. The images are captured by the Macintosh through an add-on board. The Macintosh also controls whisker stimulation by sending signals to the bimorph control box. The control box converts these signals to high voltages, causing the bimorph to move the whisker. The software creates an average image when the stimulus is on and another when the stimulus is off. The difference between these 2 images indicates a change in the intrinsic optical properties of the cortex as a result of the cortical r 'sponse to the whisker stimulation.
oxygenated hemoglobin (Fig. 8). A charge-coupled device (CCD) camera (Cohu 4815; $800) connected to a dissecting microscope (Zeiss) was focused on the area of interest based on the physiological map. Custom software (contact author) for the Macintosh was used to capture frames from the camera via a QuickCapture card (Data Translation, $1300). The software also controlled a Lab-NB board (National Instruments, $600) that was used to control bimorphs. The bimorphs (Telesensory Systems; $1) are piezoelectric devices that were used to deflect the whiskers. Experiments were based on the coordination of frame capturing from the camera with the application of stimuli that effectively excited the region of cortex being imaged. For each paradigm, two image averages were created. The first average was made from frames captured before the stimulus was delivered; the second was made during and after the whisker stimulation (see description of trials in Fig. 2). Average images were normalized based on the number of frames used to create them, and the percent change of each pixel from the unstimulated average to the stimulated average was calculated and used to form an image. Noise in the equipment made the effective measurement resolution about 3% ( ~ 5 bits). Since the change in r e f e c t a n c e for the optical signal is on the order of 0.1%, a measurement 100 times better than can be
41
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
obtained from a single frame is required. The noise in the system serves to m a k e the digitization limit (8 bits) inconsequential to the averaged m e a s u r e m e n t (signalto-noise is improved by averaging; in a noiseless system, nothing is gained by averaging, and the resolution is determined by the n u m b e r of bits digitized) Since the measured mean value approaches the actual mean value as one over the square root of the n u m b e r of frames which have been averaged, signal m e a s u r e m e n t would theoretically require about 10,000 frames for each average. In the current experimental set-up, about 40 frames are averaged during each trial. Since each trial takes 20 s, it would take well over 1 h to obtain this many averages. To circumvent this problem, we use a low-pass spatial filter (approximate gaussian: o-~ 1.5 pixels) to remove spatial high-frequency noise. In essence, our images are averaged spatially as well as temporally. We consistently obtain high-quality images in 30 min with this procedure. In some cases, clear images are formed in under 10 min. Image resolution is dependent upon microscope magnification. Pixels in most of the images shown here are 1 6 / z m on a side.
4. Image quality over time To maximize the efficiency of the technique, we measured the development of the signal over time both within a trial and over many trials. Fig. 3 shows the intra-trial signal for different time intervals over the first 14 s after stimulation begins. Each image was made by averaging frames over a 2-s interval at a constant time relative to the stimulus. Frames were averaged for 60 min. The optical signal is not clearly apparent until the third second after the stimulation begins. It is then observable for several seconds after the stimulus has stopped. Based on these images, frames were subsequently routinely averaged from 3.5 to 8 s after the stimulus initiation. The time course of this signal is slower than the signal previously described by Grinvald. Fig. 4 illustrates the development of an image over many trials. For this figure, frames were averaged during the 2-s interval just after the stimulus stopped. The signal steadily increases as the noise is averaged out.
agreement between the optical image and the electrophysiological response. In 2 other experiments, optical imaging was used at low magnification to locate the cortical representation of a particular whisker before microelectrode penetrations were made. The first electrode penetration was then made within the region of the images. In both cases, the physiological responses verified that the whisker used to make the images was the most effective for evoking neuronal responses at those cortical sites.
6. Imaging multiple whisker representations To ensure that the technique was not being influenced by the position of the camera or the light source, several different whisker representations in the rat barrel cortex were imaged without moving the camera or the lights. Each image was gathered over a 30-min period. The images were filtered and then displayed in a pseudocolor representation. Polygons were drawn by following a high-contrast border that distinguished the signal from the background. Compensation was made to prevent signals from veins which were in the return pathway from being included in the activated region. A map of the whisker representation based on the acquired images is shown in Fig. 6.
7. Agarose, trans-dural, and trans-cranial images One advantage of optical imaging is its insensitivity to electrical noise. This feature makes it an attractive WHISKER
................................
2 s e c ',. . . .
I
5sec
20 sec per trial
. . . . :. . . . . . . . . . . . . . .
~ l [ ~ l ~ l l ~ 2 4 6
icollecti
: no : stimulus
STIMULATION
:
,
I 8
...............................
13sec
I 10
I
I 12
I
•
..................
I 14
I
I 16
I
I 18
I
I 20
collect
stimulus
frames
frames
5. Comparison of images to qualitative physiological maps Fig. 5A,B shows an image formed while stimulating a single whisker (whisker D1; see Waite, 1973 for whisker nomenclature). Comparison of B and C (blood vessels highlighted to aid in this process) reveals strong
Fig. 2. Each trial is 20-s long. During the first 2 s, frames are collected and added to the 'no-stimulus' average. The bimotph is then turned on and moves the whisker at 4 hz. After 3 s of stimulation, frames are collected and added to the 'stimulus' average. The stimulus is turned off after it has been presented for 5 s. The 'stimulus' average frames are collected for another 2 s after the stimulus has been turned off. For the 13 s after the stimulus stops, the cortex is allowed to recover.
42
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
prospective tool for monitoring cortical activity in awake free-moving animals. In order to use it for such purposes, the invasiveness of the technique will have to be minimized. We have tried several modifications to the technique which reduce exposure of the cortex. The first approach was to cover the exposed cortex with agarose. When the agarose was applied smoothly with no underlying optical flaws, images were comparable to normal images through silicon oil. The second approach involved imaging the cortex before removing the dura. These images were of similar quality to those produced after the dura was removed. Because the underlying pial vasculature was more difficult to see, the images were slightly more difficult to compare with
respect to their absolute location. The final approach that was tried involved imaging through thinned skull. The skull was planed down to a few hundred micrometers with a dental drill. Images were of lesser quality, but a detectable signal was observed (Fig. 7).
8. Possible origins of the optical signal The images show a dark area where the activity is presumably centered. They also show a darkening of the veins that presumably collect blood from the activated area. Because deoxygenated hemoglobin is more absorbent in the wavelength of light used ( ~ 660 nm)
I-3 sec
3-5 sec
5-7 sec
7-9 sec
I0-12 sec
12-14 sec
Fig. 3. Signal development within a trial. The first image was created by capturing frames in the interval 1-3 s after the 4 Hz stimulus began. The signal reaches a maximum in the last 2 s of the stimulus (3-5 s) and during the 2 s after the stimulus has stopped (5-7 s). The signal slowly returns to the comparison state which is captured during the 18-20-s interval. The 9-10-s interval was not captured because of memory constraints.
B.E. Peterson, D. Goldreich / Journal of Neuroscience Methods 54 (1994) 39-47
than is oxygenated hemoglobin (Fig. 8), we conclude that the signal could be substantially accounted for by an increase in the relative amount of deoxygenated hemoglobin in the area. Such an explanation is supported by the darkening of veins in the return pathway. This explanation is also consistent with findings based on other techniques which reveal correlation between neuronal activity and cellular aerobic metabolism. Since an increase in cellular metabolism would require an increase in oxygen uptake, oxygen removal from the local blood supply is an expected consequence of local cortical activation. Also in agreement with this conclusion is the relative invisibility of large arteries against the cortical background at this wavelength of light. Increased oxygen uptake is unlikely to change the composition of arterial blood. Arterial dilatation resulting in greater blood flow has been measured at up to 30% in awake humans, whereas changes in oxygen uptake were only 5% (Fox and Raichle, 1986). Arterial changes do not manifest themselves as signal in this technique presumably because the composition of the arterial blood does not change, and therefore changes in arterial diameters and blood flow remain invisible against the cortical
43
background. Only after the blood passes through the capillary beds where oxygen is extracted does the signal appear. If blood flow increases are larger than increases in oxygen uptake, as reported in human (Fox and Raichle, 1986), then the venous deoxygenated hemoglobin concentration could decrease rather than increase. According to Beer's law, absorption should therefore decrease as well. It is possible that the flow rate increase is not greater than the increase in oxygen uptake in the anesthetized rat. It is also possible that increases in flow rate, although having a rapid onset, may have a slow decay, and therefore come to an equilibrium over the course of repeated stimuli. At that point, the flow rate would be relatively constant, and the observable signal would be the change in the amount of deoxygenated hemoglobin. This effect may explain the differences between the images described here and those obtained by Grinvald. Because he employed a more sophisticated camera, he needed to average only a few frames ( ~ 36). All of these frames might be captured before the arterial dilatation equilibrated. Thus, the changes he recorded were in the arterioles. Because our averages are collected over tens of minutes, most
900 frames
1800 frames
3600 frames
5400 frames
Fig. 4. Developmentof image as frames are averaged.The numbers above each image show the approximate number of frames in both the signal average (5-7 s after stimulus began) and the unstimulated average (13-15 s after stimulus stopped) used to create the images.
44
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
of o u r frames may be g a t h e r e d after arteriole diameters have stabilized. T h e r e are several o t h e r i m p o r t a n t differences between the e x p e r i m e n t s described here a n d those described by G r i n v a l d that might explain some of the differences in the details of the results. T h e s e i n c l u d e differences in the spectra of the light sources, differences in the species a n d sensory systems, a n d differ-
ences in the stimulus p r e s e n t a t i o n s a n d the corres p o n d i n g collection intervals. F o r example, we observe a signal associated with large pial veins, p r o b a b l y because we use a n o n - s t i m u l u s r e f e r e n c e condition. By contrast, G r i n v a l d is able to use a physiologically ort h o g o n a l visual stimulus as a reference, so that vessels serving both activated areas p r e s u m a b l y cancel in the difference image.
A
Fig. 5. Comparison between optical imaging and an electrophysiologymap. A: image created while stimulating whisker D1 at 4Hz. B: same image with vessels outlined. C: map of some of the whisker representations created by characterizing the neuronal response at each marked penetration site. The 3 sites that responded best to whisker D1 are circled. Vessels are outlined in C which correspond to those outlined in B. The locations of the dark patches in images in A and B (corresponding to neuronal activity) match well with the location of the circled penetration sites in C relative to the vessels. The 2 techniques produce similar locations of the D1 whisker representation.
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
One aspect of the technique that remains unexplained is that image quality often dramatically improves several hours after initiation of the experiment.
IG
45
The most probable explanation is recovery of the cortex from the shock of the surgery and/or administered drugs. Other possibilities include changes in the ani-
Composite
Fig. 6. Response area of 6 different whiskers imaged from the same camera position under the same lighting conditions. The stimulated whiskers were delta (A), gamma (B), E1 (C), D1 (D), C1 (E), and D2 (F). A composite image (G) is made by lining up the vessels and then tracing the darkened region. The composite shows that there is a great deal of overlap in the optical signals, but also that the signal centers appear to be in the correct location.
B.E. Peterson, D. Goldreich /Journal ~)f Neuroscience Methods 54 (1994)39 47
46
1
% Change in Reflectance
-0.02
0.0
0.02
% Change in Reflectance
-0.04
0.0
0.04
Fig. 7. Trans-dural and trans-cranial images. Images to the left and right are identical and are shown on different absorption scales for comparison purposes. The legend shows the colors associated with the percent change in absorption. A and B: A typical image after the dura has been removed. C and D: an image taken through the dura. E and F: an image taken through thinned skull. Notice that this image is more diffused.
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
47
formed simply and with inexpensive equipment (computer, video capture card, and a modest CCD camera) already available in many laboratories. We are currently using the technique to study a number of physiological processes in the rat including thresholds and inhibitory interactions.
A.
E _J
Acknowledgements 200
300
400
500
'
j ;/~
.'
600
700
800
Wavelength (nm)
.
.
.
.
900
1000 1100
.
o
The authors wish to thank Dr. Michael M. Merzenich for his steadfast support of their efforts. They would also like to thank G. Rossman and P. Goldreich at Cal Tech for measuring the high-intensity LED spectrum. This work was supported by NIH Grants NS-10414 and 2T32GM07449. For information about using the Macintosh software, contact Bret Peterson.
References
200 ~o
~
~o
~o
i
700
800
~
10b0 It00
Wavelength (nm) Fig. 8. A: spectrum of the light source (high-intensity LEDs) used to illuminate the area of cortex being imaged (courtesy of G. Rossman and P. Goldreich). B: absorption curves for oxygenated (dashed) and deoxygenated (solid) hemoglobin (Tremper and Barker, 1986). The deoxgenated hemoglobin absorbs more red light than oxygenated hemoglobin so it appears more blue. The dashed vertical line shows that the center wavelength of the light source is near the wavelength at which the absorption between oxgenated and deoxygenated hemoglobin differ most and thus where the largest signal is likely to be produced.
mal's physiological state over time (e.g., change in blood gas percentages) and changes in the state of the brain (e.g., cerebral vasoconstriction or dilatation, reduction of intracranial fluid volume). Once the improvement begins, it occurs quickly and remains for the duration of the experiment. We are currently investigating this phenomenon.
9. Conclusion Optical imaging is a powerful technique for imaging in vivo correlates of neuronal activity. It is relatively non-invasive, it is insensitive to electrical noise, and it can image a broad area in a short period. Most importantly, it is easily accessible to neurophysiological laboratories because, as described here, it can be per-
Armstrong-James, M.A. and Millar, J.M. (1979) Carbon fibre microelectrodes. J. Neurosci. Methods, 1: 279-287. Bonhoeffer, T. and Grinvald, A. (1991) Iso-orientation domains in cat visual cortex are arranged in pinwheel-like patterns. Nature, 353: 429-431. Chance, B., Zhuang, Z., Unah, C., Alter, C. and Lipton, L. (1993) Cognition-activated low-frequency modulation of light absorption in human brain. Proc. Natl. Acad. Sci. USA, 90: 3770-3774. Fox, P.T. and Raichle, M.E. (1986) Focal physiological uncoupling of cerebral blood flow and oxidative metabolism during somatosensory stimulation in human subjects. Proc. Natl. Acad. Sci. USA, 83: 1140-1144. Frostig, R.D., Lieke, E.E., T'so, D.Y. and Grinvald, A. (1990) Cortical functional architecture and local coulping between neuonal activity and the microcirculation revealed by in vivo high-resolution optical imaging of intrinsic signals. Proc. Natl. Acad. Sci. USA, 87: 6082-6086.. Gochin, P.M., Bedenbaugh, P., Gelfand, J.J., Gross, C.G. and Gerstein, G.L. (1992) Intrinsic signal optical imaging in the forepaw area of rat somatosensory cortex. Proc. Natl. Acad. Sci. USA, 89: 8381-8383. Grinvald, A., Frostig, R.D., Siegel, R.M. and Bartfeld, E. (1991) High-resolution optical imaging of functional brain architecture in the awake monkey. Proc. Natl. Acad. Sci. USA, 88: 1155911563. Grinvald, A., Lieke, E., Frostig, R.D., Gilbert, C.D. and Wiesel, T.N. (1986) Functional architecture of cortex revealed by optical imaging of intrinsic signals. Nature, 324: 361. Haglund M.M., Ojemann G.A. and Hochman D.W. (1992) Optical imaging of epileptiform and functional activity in human cerebral cortex. Nature, 358: 668-671. Ts'o, D.Y., Frostig, R.D., Lieke, E.E. and Grinvald, A. (1990) Functional organization of primate visual cortex revealed by high resolution optical imaging. Science, 249: 417-420. Waite, P.M.E. (1973) Somatotopic organization of vibrissal responses in the ventro-basal complex of the rat thalamus. J. Physiol. (Lond.), 228: 527-540.
ELSEVIER
Journal of Neuroscience Methods 54 (1994) 39-47
A new approach to optical imaging applied to rat barrel cortex Bret E. Peterson a,,, Daniel Goldreich a
b
Bioengineering Group, University of California, Berkeley and San Francisco, San Francisco, CA, USA; b Keck Center for Integrative Neuroscience, University of California, San Francisco, CA, USA Received 10 August 1993; revised 14 March 1994; accepted 18 April 1994
Abstract
Several groups have described using intrinsic optical imaging to form images of activity patterns in the cortex. Because the signal is small, the general approach has been to use expensive camera equipment with a high dynamic range to make these measurements. However, by using signal averaging to compensate for lower dynamic range, images can be obtained using equipment already available in many laboratories. This modified technique has been used for imaging activity in 'barrel' cortex of the rat. A map of the representation of a single whisker as determined by the imaging technique corresponded well with a similar map made using standard electrophysiology. A map of several whiskers was made by overlaying images of single-whisker representations. The details of the images differ from those previously described. Possible mechanisms for the signal are discussed.
Keywords: Optical imaging; Intrinsic imaging; Vibrissa; Barrels; Cortex; (Rat)
I. Introduction
In 1986, Grinvald and colleagues described a slow time course signal that interfered with their voltagesensitive dye measurements. They found that they could use these 'intrinsic optical signals' to image activity correlates in the brain. The technique has since been used to form images that resemble physiological maps in somatosensory and visual cortices (Grinvald et al., 1986, 1991; Ts'o et al., 1990; Bonhoeffer and Grinvald, 1991; Gochin et al., 1992), and to image language centers in human cortex (Haglund et al., 1992). The intrinsic imaging technique uses the optical absorption properties of hemoglobin to detect changes in the local cortical blood supply that occur when neuronal activity increases. Feedback systems in the vasculature can respond to the increased metabolic demand of active cells by increasing local blood flow (Fox and Raichle, 1986). The resulting increased tissue hemoglobin content can be imaged as a change in light absorbency. Furthermore, the amount of deoxygenated
* Corresponding author. Elsevier Science B.V.
SSDI 0 1 6 5 - 0 2 7 0 ( 9 4 ) 0 0 0 5 9 - P
hemoglobin leaving the capillary beds increases as neuronal tissue becomes metabolically activated. Because deoxygenated hemoglobin has a different absorption spectrum than does its oxygenated counterpart, a change in the amount of deoxygenated hemoglobin results in a change in the optical properties of the activated tissue (Frostig et al., 1990). A variety of other techniques are used for imaging the metabolic activity of electrically active brain regions. These include positron emission tomography (PET), single-photon emission computed tomography (SPECT), functional magnetic resonance imaging (MRI), and 2-deoxyglucose autoradiography (2-DG). PET, SPECT and MRI are high-cost, non-invasive techniques with relatively low spatial resolution that have been designed for use on humans. Because they require large expensive equipment, they are not readily available to most laboratories. 2-DG is an animal research technique with high spatial resolution. However, it requires postmortem processing and therefore can be used to obtain only a single image of a brain region per animal. Optical imaging can be used on both research animals (e.g., Grinvald et al., 1986) and humans (Haglund et al., 1992). It is more invasive than the other metabol-
40
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
ically based imaging techniques such as PET, SPECT, and MRI in that it requires an open cranium for optimal imaging (but see Chance et al., 1993), but it offers greater spatial resolution comparable to 2-DG imaging. Unlike 2-DG imaging, it can be used to make repeated measurements in a single animal. Unfortunately, its reported expense and complexity (e.g., Optical Imaging, $75,000) have apparently dissuaded laboratories from using this powerful technique. We describe here a new approach to optical imaging that allows it to be applied inexpensively and simply, using equipment that is readily available to many research laboratories. We have developed our modified version of the technique in the somatosensory whisker 'barrel' cortex of the rat. However, it should prove applicable to other species and modalities.
Experimental Set Up
Camera and
Dissecting Scope
-j Macintosh Ilci
Control Box
2. Preparation of the barrel cortex Sprague-Dawley rats were anesthetized with sodium pentobarbital (Nembutal, 55 m g / k g , i.p.) and given 0.1 ml of atropine sulfate i.m. Rectal temperature was monitored and maintained at 37.5°C with a thermostatically controlled heating pad. Femoral venous and tracheal cannulae were introduced. The atlanto-occipital membrane covering the cisterna magna was transected to reduce cerebrospinal fluid pressure. A unilateral craniotomy was performed over the barrel cortex of SI. The dura over the cortex was reflected, and the exposed cortical surface was covered with silicon oil. Intravenous pentobarbital was administered as necessary to maintain an areflexic level of anesthesia. Physiological recordings were often made to verify the location of key barrels. These recordings were made using 10 p.m carbon-fiber electrodes (ArmstrongJames and Millar, 1979) that were introduced orthogonal to the surface and inserted 580 ~ m beneath the pia with a microdrive. Neuronal responses were amplified and connected to a speaker. The response properties of each penetration site were characterized qualitatively by tapping lightly on each whisker with a glass probe and noting the neuronal response. A map of the whisker representation ('barrel' cortex in SI) was created by recording the characterization of penetration sites made at 200 ~ m intervals.
3. Basic imaging technique The optical imaging setup is shown in Fig. 1. The cortex was illuminated with narrow band red light (mean wavelength: ~ 660 nm) from high-intensity light-emitting diodes (LEDs). These LEDs provide a convenient, inexpensive source of a wavelength of light that is differentially absorbed by oxygenated and de-
Fig. 1. Experimental set-up. The camera is used to image the cortex via the dissecting microscope. The images are captured by the Macintosh through an add-on board. The Macintosh also controls whisker stimulation by sending signals to the bimorph control box. The control box converts these signals to high voltages, causing the bimorph to move the whisker. The software creates an average image when the stimulus is on and another when the stimulus is off. The difference between these 2 images indicates a change in the intrinsic optical properties of the cortex as a result of the cortical r 'sponse to the whisker stimulation.
oxygenated hemoglobin (Fig. 8). A charge-coupled device (CCD) camera (Cohu 4815; $800) connected to a dissecting microscope (Zeiss) was focused on the area of interest based on the physiological map. Custom software (contact author) for the Macintosh was used to capture frames from the camera via a QuickCapture card (Data Translation, $1300). The software also controlled a Lab-NB board (National Instruments, $600) that was used to control bimorphs. The bimorphs (Telesensory Systems; $1) are piezoelectric devices that were used to deflect the whiskers. Experiments were based on the coordination of frame capturing from the camera with the application of stimuli that effectively excited the region of cortex being imaged. For each paradigm, two image averages were created. The first average was made from frames captured before the stimulus was delivered; the second was made during and after the whisker stimulation (see description of trials in Fig. 2). Average images were normalized based on the number of frames used to create them, and the percent change of each pixel from the unstimulated average to the stimulated average was calculated and used to form an image. Noise in the equipment made the effective measurement resolution about 3% ( ~ 5 bits). Since the change in r e f e c t a n c e for the optical signal is on the order of 0.1%, a measurement 100 times better than can be
41
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
obtained from a single frame is required. The noise in the system serves to m a k e the digitization limit (8 bits) inconsequential to the averaged m e a s u r e m e n t (signalto-noise is improved by averaging; in a noiseless system, nothing is gained by averaging, and the resolution is determined by the n u m b e r of bits digitized) Since the measured mean value approaches the actual mean value as one over the square root of the n u m b e r of frames which have been averaged, signal m e a s u r e m e n t would theoretically require about 10,000 frames for each average. In the current experimental set-up, about 40 frames are averaged during each trial. Since each trial takes 20 s, it would take well over 1 h to obtain this many averages. To circumvent this problem, we use a low-pass spatial filter (approximate gaussian: o-~ 1.5 pixels) to remove spatial high-frequency noise. In essence, our images are averaged spatially as well as temporally. We consistently obtain high-quality images in 30 min with this procedure. In some cases, clear images are formed in under 10 min. Image resolution is dependent upon microscope magnification. Pixels in most of the images shown here are 1 6 / z m on a side.
4. Image quality over time To maximize the efficiency of the technique, we measured the development of the signal over time both within a trial and over many trials. Fig. 3 shows the intra-trial signal for different time intervals over the first 14 s after stimulation begins. Each image was made by averaging frames over a 2-s interval at a constant time relative to the stimulus. Frames were averaged for 60 min. The optical signal is not clearly apparent until the third second after the stimulation begins. It is then observable for several seconds after the stimulus has stopped. Based on these images, frames were subsequently routinely averaged from 3.5 to 8 s after the stimulus initiation. The time course of this signal is slower than the signal previously described by Grinvald. Fig. 4 illustrates the development of an image over many trials. For this figure, frames were averaged during the 2-s interval just after the stimulus stopped. The signal steadily increases as the noise is averaged out.
agreement between the optical image and the electrophysiological response. In 2 other experiments, optical imaging was used at low magnification to locate the cortical representation of a particular whisker before microelectrode penetrations were made. The first electrode penetration was then made within the region of the images. In both cases, the physiological responses verified that the whisker used to make the images was the most effective for evoking neuronal responses at those cortical sites.
6. Imaging multiple whisker representations To ensure that the technique was not being influenced by the position of the camera or the light source, several different whisker representations in the rat barrel cortex were imaged without moving the camera or the lights. Each image was gathered over a 30-min period. The images were filtered and then displayed in a pseudocolor representation. Polygons were drawn by following a high-contrast border that distinguished the signal from the background. Compensation was made to prevent signals from veins which were in the return pathway from being included in the activated region. A map of the whisker representation based on the acquired images is shown in Fig. 6.
7. Agarose, trans-dural, and trans-cranial images One advantage of optical imaging is its insensitivity to electrical noise. This feature makes it an attractive WHISKER
................................
2 s e c ',. . . .
I
5sec
20 sec per trial
. . . . :. . . . . . . . . . . . . . .
~ l [ ~ l ~ l l ~ 2 4 6
icollecti
: no : stimulus
STIMULATION
:
,
I 8
...............................
13sec
I 10
I
I 12
I
•
..................
I 14
I
I 16
I
I 18
I
I 20
collect
stimulus
frames
frames
5. Comparison of images to qualitative physiological maps Fig. 5A,B shows an image formed while stimulating a single whisker (whisker D1; see Waite, 1973 for whisker nomenclature). Comparison of B and C (blood vessels highlighted to aid in this process) reveals strong
Fig. 2. Each trial is 20-s long. During the first 2 s, frames are collected and added to the 'no-stimulus' average. The bimotph is then turned on and moves the whisker at 4 hz. After 3 s of stimulation, frames are collected and added to the 'stimulus' average. The stimulus is turned off after it has been presented for 5 s. The 'stimulus' average frames are collected for another 2 s after the stimulus has been turned off. For the 13 s after the stimulus stops, the cortex is allowed to recover.
42
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
prospective tool for monitoring cortical activity in awake free-moving animals. In order to use it for such purposes, the invasiveness of the technique will have to be minimized. We have tried several modifications to the technique which reduce exposure of the cortex. The first approach was to cover the exposed cortex with agarose. When the agarose was applied smoothly with no underlying optical flaws, images were comparable to normal images through silicon oil. The second approach involved imaging the cortex before removing the dura. These images were of similar quality to those produced after the dura was removed. Because the underlying pial vasculature was more difficult to see, the images were slightly more difficult to compare with
respect to their absolute location. The final approach that was tried involved imaging through thinned skull. The skull was planed down to a few hundred micrometers with a dental drill. Images were of lesser quality, but a detectable signal was observed (Fig. 7).
8. Possible origins of the optical signal The images show a dark area where the activity is presumably centered. They also show a darkening of the veins that presumably collect blood from the activated area. Because deoxygenated hemoglobin is more absorbent in the wavelength of light used ( ~ 660 nm)
I-3 sec
3-5 sec
5-7 sec
7-9 sec
I0-12 sec
12-14 sec
Fig. 3. Signal development within a trial. The first image was created by capturing frames in the interval 1-3 s after the 4 Hz stimulus began. The signal reaches a maximum in the last 2 s of the stimulus (3-5 s) and during the 2 s after the stimulus has stopped (5-7 s). The signal slowly returns to the comparison state which is captured during the 18-20-s interval. The 9-10-s interval was not captured because of memory constraints.
B.E. Peterson, D. Goldreich / Journal of Neuroscience Methods 54 (1994) 39-47
than is oxygenated hemoglobin (Fig. 8), we conclude that the signal could be substantially accounted for by an increase in the relative amount of deoxygenated hemoglobin in the area. Such an explanation is supported by the darkening of veins in the return pathway. This explanation is also consistent with findings based on other techniques which reveal correlation between neuronal activity and cellular aerobic metabolism. Since an increase in cellular metabolism would require an increase in oxygen uptake, oxygen removal from the local blood supply is an expected consequence of local cortical activation. Also in agreement with this conclusion is the relative invisibility of large arteries against the cortical background at this wavelength of light. Increased oxygen uptake is unlikely to change the composition of arterial blood. Arterial dilatation resulting in greater blood flow has been measured at up to 30% in awake humans, whereas changes in oxygen uptake were only 5% (Fox and Raichle, 1986). Arterial changes do not manifest themselves as signal in this technique presumably because the composition of the arterial blood does not change, and therefore changes in arterial diameters and blood flow remain invisible against the cortical
43
background. Only after the blood passes through the capillary beds where oxygen is extracted does the signal appear. If blood flow increases are larger than increases in oxygen uptake, as reported in human (Fox and Raichle, 1986), then the venous deoxygenated hemoglobin concentration could decrease rather than increase. According to Beer's law, absorption should therefore decrease as well. It is possible that the flow rate increase is not greater than the increase in oxygen uptake in the anesthetized rat. It is also possible that increases in flow rate, although having a rapid onset, may have a slow decay, and therefore come to an equilibrium over the course of repeated stimuli. At that point, the flow rate would be relatively constant, and the observable signal would be the change in the amount of deoxygenated hemoglobin. This effect may explain the differences between the images described here and those obtained by Grinvald. Because he employed a more sophisticated camera, he needed to average only a few frames ( ~ 36). All of these frames might be captured before the arterial dilatation equilibrated. Thus, the changes he recorded were in the arterioles. Because our averages are collected over tens of minutes, most
900 frames
1800 frames
3600 frames
5400 frames
Fig. 4. Developmentof image as frames are averaged.The numbers above each image show the approximate number of frames in both the signal average (5-7 s after stimulus began) and the unstimulated average (13-15 s after stimulus stopped) used to create the images.
44
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
of o u r frames may be g a t h e r e d after arteriole diameters have stabilized. T h e r e are several o t h e r i m p o r t a n t differences between the e x p e r i m e n t s described here a n d those described by G r i n v a l d that might explain some of the differences in the details of the results. T h e s e i n c l u d e differences in the spectra of the light sources, differences in the species a n d sensory systems, a n d differ-
ences in the stimulus p r e s e n t a t i o n s a n d the corres p o n d i n g collection intervals. F o r example, we observe a signal associated with large pial veins, p r o b a b l y because we use a n o n - s t i m u l u s r e f e r e n c e condition. By contrast, G r i n v a l d is able to use a physiologically ort h o g o n a l visual stimulus as a reference, so that vessels serving both activated areas p r e s u m a b l y cancel in the difference image.
A
Fig. 5. Comparison between optical imaging and an electrophysiologymap. A: image created while stimulating whisker D1 at 4Hz. B: same image with vessels outlined. C: map of some of the whisker representations created by characterizing the neuronal response at each marked penetration site. The 3 sites that responded best to whisker D1 are circled. Vessels are outlined in C which correspond to those outlined in B. The locations of the dark patches in images in A and B (corresponding to neuronal activity) match well with the location of the circled penetration sites in C relative to the vessels. The 2 techniques produce similar locations of the D1 whisker representation.
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
One aspect of the technique that remains unexplained is that image quality often dramatically improves several hours after initiation of the experiment.
IG
45
The most probable explanation is recovery of the cortex from the shock of the surgery and/or administered drugs. Other possibilities include changes in the ani-
Composite
Fig. 6. Response area of 6 different whiskers imaged from the same camera position under the same lighting conditions. The stimulated whiskers were delta (A), gamma (B), E1 (C), D1 (D), C1 (E), and D2 (F). A composite image (G) is made by lining up the vessels and then tracing the darkened region. The composite shows that there is a great deal of overlap in the optical signals, but also that the signal centers appear to be in the correct location.
B.E. Peterson, D. Goldreich /Journal ~)f Neuroscience Methods 54 (1994)39 47
46
1
% Change in Reflectance
-0.02
0.0
0.02
% Change in Reflectance
-0.04
0.0
0.04
Fig. 7. Trans-dural and trans-cranial images. Images to the left and right are identical and are shown on different absorption scales for comparison purposes. The legend shows the colors associated with the percent change in absorption. A and B: A typical image after the dura has been removed. C and D: an image taken through the dura. E and F: an image taken through thinned skull. Notice that this image is more diffused.
B.E. Peterson, D. Goldreich /Journal of Neuroscience Methods 54 (1994) 39-47
47
formed simply and with inexpensive equipment (computer, video capture card, and a modest CCD camera) already available in many laboratories. We are currently using the technique to study a number of physiological processes in the rat including thresholds and inhibitory interactions.
A.
E _J
Acknowledgements 200
300
400
500
'
j ;/~
.'
600
700
800
Wavelength (nm)
.
.
.
.
900
1000 1100
.
o
The authors wish to thank Dr. Michael M. Merzenich for his steadfast support of their efforts. They would also like to thank G. Rossman and P. Goldreich at Cal Tech for measuring the high-intensity LED spectrum. This work was supported by NIH Grants NS-10414 and 2T32GM07449. For information about using the Macintosh software, contact Bret Peterson.
References
200 ~o
~
~o
~o
i
700
800
~
10b0 It00
Wavelength (nm) Fig. 8. A: spectrum of the light source (high-intensity LEDs) used to illuminate the area of cortex being imaged (courtesy of G. Rossman and P. Goldreich). B: absorption curves for oxygenated (dashed) and deoxygenated (solid) hemoglobin (Tremper and Barker, 1986). The deoxgenated hemoglobin absorbs more red light than oxygenated hemoglobin so it appears more blue. The dashed vertical line shows that the center wavelength of the light source is near the wavelength at which the absorption between oxgenated and deoxygenated hemoglobin differ most and thus where the largest signal is likely to be produced.
mal's physiological state over time (e.g., change in blood gas percentages) and changes in the state of the brain (e.g., cerebral vasoconstriction or dilatation, reduction of intracranial fluid volume). Once the improvement begins, it occurs quickly and remains for the duration of the experiment. We are currently investigating this phenomenon.
9. Conclusion Optical imaging is a powerful technique for imaging in vivo correlates of neuronal activity. It is relatively non-invasive, it is insensitive to electrical noise, and it can image a broad area in a short period. Most importantly, it is easily accessible to neurophysiological laboratories because, as described here, it can be per-
Armstrong-James, M.A. and Millar, J.M. (1979) Carbon fibre microelectrodes. J. Neurosci. Methods, 1: 279-287. Bonhoeffer, T. and Grinvald, A. (1991) Iso-orientation domains in cat visual cortex are arranged in pinwheel-like patterns. Nature, 353: 429-431. Chance, B., Zhuang, Z., Unah, C., Alter, C. and Lipton, L. (1993) Cognition-activated low-frequency modulation of light absorption in human brain. Proc. Natl. Acad. Sci. USA, 90: 3770-3774. Fox, P.T. and Raichle, M.E. (1986) Focal physiological uncoupling of cerebral blood flow and oxidative metabolism during somatosensory stimulation in human subjects. Proc. Natl. Acad. Sci. USA, 83: 1140-1144. Frostig, R.D., Lieke, E.E., T'so, D.Y. and Grinvald, A. (1990) Cortical functional architecture and local coulping between neuonal activity and the microcirculation revealed by in vivo high-resolution optical imaging of intrinsic signals. Proc. Natl. Acad. Sci. USA, 87: 6082-6086.. Gochin, P.M., Bedenbaugh, P., Gelfand, J.J., Gross, C.G. and Gerstein, G.L. (1992) Intrinsic signal optical imaging in the forepaw area of rat somatosensory cortex. Proc. Natl. Acad. Sci. USA, 89: 8381-8383. Grinvald, A., Frostig, R.D., Siegel, R.M. and Bartfeld, E. (1991) High-resolution optical imaging of functional brain architecture in the awake monkey. Proc. Natl. Acad. Sci. USA, 88: 1155911563. Grinvald, A., Lieke, E., Frostig, R.D., Gilbert, C.D. and Wiesel, T.N. (1986) Functional architecture of cortex revealed by optical imaging of intrinsic signals. Nature, 324: 361. Haglund M.M., Ojemann G.A. and Hochman D.W. (1992) Optical imaging of epileptiform and functional activity in human cerebral cortex. Nature, 358: 668-671. Ts'o, D.Y., Frostig, R.D., Lieke, E.E. and Grinvald, A. (1990) Functional organization of primate visual cortex revealed by high resolution optical imaging. Science, 249: 417-420. Waite, P.M.E. (1973) Somatotopic organization of vibrissal responses in the ventro-basal complex of the rat thalamus. J. Physiol. (Lond.), 228: 527-540.