NeuroImage 14, 977–994 (2001)doi:10.1006/nimg.2001.0897, available online at http://www.idealibrary.com on

Scattered-Light Imaging in Vivo Tracks Fast and Slow Processesof Neurophysiological Activation

David M. Rector,* Robert F. Rogers,† James S. Schwaber,† Ronald M. Harper,‡ and John S. George*,1

*Biophysics Group, Los Alamos National Laboratory, MS-D454, Los Alamos, New Mexico 87545; †Department of Pathology, Anatomyand Cell Biology, Thomas Jefferson University, Philadelphia, Pennsylvania 19107; and ‡Department of Neurobiology,

University of California at Los Angeles, Los Angeles, California 90095

Received November 15, 2000

or

We imaged fast optical changes associated with

evoked neural activation in the dorsal brainstem ofanesthetized rats, using a novel imaging device. Theimager consisted of a gradient-index (GRIN) lens, amicroscope objective, and a miniature charged-cou-pled device (CCD) video camera. We placed theprobe in contact with tissue above cardiorespiratoryareas of the nucleus of the solitary tract and illumi-nated the tissue with 780-nm light through flexiblefibers around the probe perimeter. The focus depthwas adjusted by moving the camera and microscopeobjective relative to the fixed GRIN lens. Back-scat-tered light images were relayed through the GRINlens to the CCD camera. Video frames were digitizedat 100 frames per second, along with tracheal pres-sure, arterial blood pressure, and electrocardiogramsignals recorded at 1 kHz per channel. A macroelec-trode placed under the GRIN lens recorded field po-tentials from the imaged area. Aortic, vagal, andsuperior laryngeal nerves were dissected free of sur-rounding tissue within the neck. Separate shocks toeach dissected nerve elicited evoked electrical re-sponses and caused localized optical activity pat-terns. The optical response was modeled by fourdistinct temporal components corresponding to pu-tative physical mechanisms underlying scatteredlight changes. Region-of-interest analysis revealedimage areas which were dominated by one or moreof the different time-course components, some ofwhich were also optimally recorded at differenttissue depths. Two slow optical components appearto correspond to hemodynamic responses to meta-bolic demand associated with activation. Two fast

1 To whom correspondence should be addressed at the BiophysicsGroup, P-21, MS-D454, Los Alamos National Laboratory, P.O. Box1663, Los Alamos, NM 87545. Fax: (505) 665-4507. E-mail:jsg@lanl.gov.

977

ptical components paralleled electrical evokedesponses. © 2001 Academic Press

INTRODUCTION

The function of neuronal networks depends on dy-namic spatial and temporal patterns of activationspanning many individual cells. Serial single-unit re-cordings have provided important insights into neuro-nal function, yet they are inherently limited in theirinability to probe real-time spatiotemporal aspects ofnetwork function. Studies of sensory, motor, or auto-nomic control regions typically require arduous prob-ing with microwire or sharp glass electrodes to recordfrom cell populations involved in such processes. Theseprocedures are selective in their sampling and maymiss important aspects of correlated function within anetwork. Locating cells for individual electrophysiolog-ical study is typically a blind and somewhat stochasticprocess that might be greatly enhanced through imag-ing procedures which detect neural-activity-dependentchanges in tissue properties.

Methods that allow simultaneous recording fromlarge cell populations while allowing resolution for sin-gle cells or small functionally related cell clusters are ofpotentially great significance. Electrode arrays havebecome increasingly sophisticated and have achievedhigher density, but still provide rather sparse sam-pling (Rousche et al., 1999; Jones and Barth, 1999;Chapin and Nicholelis, 1999). Arrays for studies ofintact tissue require vigorous insertion techniques(e.g., the pneumatic driver marketed by Bionics, Inc.,UT), and progressive changes in the tissue limit theutility of electrode arrays for chronic recording (Braginet al., 2000). The most dense electrode arrays providein-plane sampling with 100- to 150-mm increments,sometimes with greater density across depth. Althoughelectrical signals provide better signal to noise, sam-

1053-8119/01 $35.00Copyright © 2001 by Academic Press

All rights of reproduction in any form reserved.

Mlfvsgsvfi

eseae

M1

osgslratt

m1iatasntrt

sptis

978 RECTOR ET AL.

pling density with electrode arrays is typically muchless than the micrometer scale possible in principlewith optical techniques. With further technical im-provements in spatial resolution and sensitivity, opti-cal techniques may achieve recording quality compa-rable to microelectrode array techniques, with betterspatial resolution and without penetrating the tissue.

Imaging procedures based on intrinsic optical sig-nals for assessing neural activity in vivo have producedremarkable maps of the functional microscopic archi-tecture of neural networks. However, such studies aretypically limited to slow measurements dominated byhemodynamic changes recorded from tissue near thesurface (Grinvald et al., 1986, 1988). Functional dyeshave unique advantages for mapping fast neural re-sponses over a large area. Dyes have been developedwith fluorescence or absorbance properties that varywith cell membrane potential or cytoplasmic ions.Some of these can track the characteristic dynamics ofneural function with high contrast. For short-termstudies, voltage-sensitive dyes are particularly useful(Orbach et al., 1985). New developments have providedvoltage-sensitive dyes that are easier to administerand result in less tissue damage for longer recordingperiods (Tsodyks et al., 1999; Shoham et al., 1999).

ultiphoton fluorescence techniques also reduce prob-ems with photic damage (Xu et al., 1996). By usingunctional dyes, it is sometimes possible to label indi-idual neurons involved in neurophysiological re-ponses, perhaps achieving greater correlation to sin-le-unit electrical recordings. However, delivery ofuch indicator or contrast agents is problematic for inivo studies and precludes imaging for long periods inreely behaving animals (Rector et al., 1991). Thus, it ismportant to develop alternative methods.

Fast intrinsic optical responses, synchronous withlectrical spike and evoked activity, have been ob-erved in isolated nerve preparations, tissue slices, andxplants using single-channel photodiode detectors orrrays (Cohen, 1973; Tasaki and Byrne, 1992; Salzbergt al., 1985). Optical signals in vivo are more complex;

fast scattered light changes (or dye-dependent signals)are superimposed on hemodynamic responses, and op-tical influences from tissue outside the active area mayinfluence or obscure local signals. Fast optical signalshave been observed in vivo using a photodiode detectorwithout using dyes (Rector et al., 1997a), but previousstudies have not achieved the sensitivity and resolu-tion required for dynamic imaging.

To image rapid intrinsic optical changes in vivo, andto better visualize activity below the surface, we devel-oped a variable-focus, high-performance image probe(Rector et al., 1997b, 1999) and used it to detect low-level (1025 to 1024), rapid optical responses in rat dorsalbrain stem. Illumination around the perimeter of theimage field mimicked dark-field techniques, minimized

specular surface reflectance, and enhanced scatteredlight changes below the tissue surface (Stepnoski et al.,1991; Holthoff and Witte, 1996; Rector et al., 1999).

Neurons within the nucleus of the solitary tract(NTS) at the rostrocaudal level of the area postremaare situated at the first synapse of peripheral cardio-vascular primary afferent input and participate in car-diorespiratory control. These neurons receive mono-synaptic and multisynaptic input from receptorsenervating the heart and great vessels, lungs, larynx,pharynx, and carotid sinus regions (Lowey, 1990) andmay be excited and/or inhibited by single shocks to theaortic, vagal, superior laryngeal, or carotid sinusnerves (Donoghue et al., 1985; Mifflin et al., 1988;

ifflin, 1993; Rogers et al., 1993, 1996; Lipski et al.,975; Seagard et al., 1995).In these experiments, the nerves were dissected free

f surrounding tissue and were electrically stimulatedeveral centimeters away from the recording site. Sin-le shocks to individual nerves elicited electrical re-ponses within the dorsal brain stem, accompanied byocalized optical activity patterns. Dynamic imagingevealed familiar hemodynamic response componentss well as fast intrinsic optical signals that closelyrack the electrical dynamics of neurophysiological ac-ivation.

MATERIALS AND METHODS

General surgical and physiological preparationethods have been described previously (Rogers et al.,

993). Twenty-five male rats (250–450 g, 1 per exper-ment) were anesthetized with urethane (1.3 g/kg, ip),nd the femoral artery and vein were cannulated. Aracheotomy was performed for artificial ventilation,nd the right cervical vagus nerve bundle was dis-ected to identify and isolate the vagus (VN), aorticerve (AN), and superior laryngeal nerve (SLN) fromhe surrounding tissue. Animals were placed in a ste-eotaxic apparatus and suspended from a cervical ver-ebra. The head was ventroflexed ;60°, and the over-

lying dural sheet was removed. The AN, VN, and SLNeach were placed sequentially on bipolar stimulatinghook electrodes. To minimize potential movement ar-tifact from respiratory musculature, some animalswere paralyzed (vecuronium bromide, 0.5 mg/kg iv bo-luses, 40 min apart) and a pneumothorax was per-formed. We later determined that paralysis was notnecessary to observe the optical signals.

The endoscope probe, described in detail elsewhere(Rector et al., 1999), was placed on the brain-stemurface overlying the NTS (Figs. 1 and 2). The imagerobe was gently placed on the dural surface of theissue. We monitored images and optical dynamics dur-ng probe placement. Initial contact with the tissueurface at one or more locations was evident in the

sosc

wtt(wCtsdwmtp

iirtsmr

979SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

images. Typically an additional 100 mm of travel re-ulted in a uniform image field while suppressing largescillations associated with movement of vessels. Thistrategy ensured a stable interface, while minimizingompression and allowing continued tissue perfusion.

Illumination light from a 780-nm diode laser arrayas conveyed by plastic optical fibers that surrounded

he perimeter of the probe. Light returning from theissue entered a 3-mm-diameter gradient-indexGRIN) lens (Gradient Index Corp., Rochester, NY) andas focused through a microscope objective lens onto aCD camera. To change the plane of focus below the

issue, the camera and microscope objective lens as-embly was moved relative to the GRIN lens, withoutisturbing the tissue/probe interface. Camera outputas sampled by a computer-based digitizer which si-ultaneously collected physiological signals, including

he electrocardiogram (EKG), tracheal pressure, bloodressure, and end-tidal CO . A fine wire was inserted

FIG. 1. A three-dimensional and coronal view of the rat medullallustrates relevant structures and image probe placement. The areamaged by the gradient-index (GRIN) lens (outline circle) covered theight side of the dorsal medulla overlying the area postrema (AP),ractus solitarius (TS), nucleus of the solitary tract (NTS), hypoglos-al nucleus (XII), and dorsal motor nucleus of the vagus (DMV). Aacrowire (WIRE) was inserted underneath the image probe to

ecord evoked electrical activity from the region.

2

under the probe for recording evoked responses tonerve shocks.

Stability of the tissue/lens interface as well as thehealth of the preparation was monitored with onlineprocessing, analysis of blood pressure correlation se-quences, and continuous monitoring of physiology(EKG, blood pressure, etc.). Significant deviations inthe temporal or spatial distribution of the vascularpattern typically indicated problems with the record-ing. Additionally, online display of the optical re-sponses to neural stimulation during the experimentshowed the stability of the neural response to stimula-tion.

The signal-to-noise ratio of the CCD camera (TI-TC211) was better than 66 dB with a pixel clock of 5MHz. The CCD well size is 150 K electrons with asensitivity of 260 mV/lux. We maximized the dynamicrange for detecting small changes in scattered light byoptimizing the voltage input range of the digitizer sothat the darkest pixel within the imaged area had avalue 10% above 0 A/D units, and the brightest pixel

FIG. 2. A diagram of system components shows a 780-nm laserdiode array light source (L.S.) projecting light into flexible fiberoptics mounted around the perimeter of a GRIN lens. The GRIN lenswas placed in contact with the dorsal medulla, which collected back-scattered light from the rat dorsal medulla, and formed an imagethrough a microscope objective (M.O.) onto a CCD camera. A changein the focal plane was achieved by moving the CCD camera andobjective relative to the GRIN lens, which remained fixed in position.Video frames from the CCD camera and 16 channels of electrophys-iology were multiplexed and digitized continuously during the exper-iment. A Pentium host computer collected the raw frame sequencesand streamed them continuously onto a mass storage device. Thehost computer also searched image sequences for stimulus triggerevents, and epochs of 90 frames were averaged and transferred to adisplay processor via a fast ethernet connection for online analysisand visualization.

980 RECTOR ET AL.

was 90% of the digitizer full-scale range. This optimi-zation procedure allowed us to increase the amplifica-tion of the optical signal by a factor of approximately6.7. Results were converted from relative changes toabsolute changes through a calibration procedure thatinvolved illumination of the image with an LED drivenby a sine wave of known offset and peak-to-peak am-plitude, calibrated with a photodiode detector. The re-sulting transfer function was applied to the CCD in-tensity changes to obtain absolute values.

The laser diode array we used (OPC-A020-795-CS;Opto Power Corp., City of Industry, CA) was driven at6 to 12 A by a high-performance laser diode driver(SDL-820; SDL, Inc., San Jose, CA). The driver is es-sentially a highly regulated power supply and achieveda stable illumination with variations that were notdetectable above the noise levels of the imager. Wetried a number of other power supplies, includingswitching power supplies and high-current linearpower supplies, none of which gave us the stability werequired. Additionally, the laser diode array was cooledat a constant temperature of 10°C with a temperature-regulated circulating water bath (Lauda RM3; Brink-mann Instruments, Inc., Westbury, NY).

In these experiments, laser speckle was not a signif-icant issue for several reasons. The coherence of thelight was reduced through numerous interconnectionsthrough multimode fiber optics. However, in our expe-rience, multimode fiber optics alone are not sufficientto eliminate speckle. Scattering during photon migra-tion through tissue would also reduce coherence. Slightmovements of the tissue though cardiac and respira-tory pulsation may help to randomize the speckle dur-ing the 10-ms integration period of the CCD camera.Finally, in most experiments, we had adequate lightfrom the diode array in low coherence LED operationbelow the transition to lasing.

After a 30-min stabilization period, imaging experi-ments were initiated. The depth of focus was initiallyadjusted to 0.40-mm below the surface. Back-scatteredlight images from the NTS were digitized during singleshocks to the AN, VN, or SLN nerve trunks. Currentwas adjusted to an amplitude twice that required toevoke a change in arterial pressure in response to a 1-s,100-Hz pulse train used to test the viability of brain-stem reflexes. Single-pulse stimuli were given in ran-dom phase relative to the cardiac and ventilatory cy-cles (typically 0.04-mA, 0.1-ms pulse, 1–2 s uniformrandom interval) and did not produce detectablechanges in heart rate or blood pressure. Cameraframes were digitized continuously (100 fps) along withphysiological signals (1 kHz per channel) and stored ininterleaved file format for physical data (IFFPHYS), anextensible, internationally recognized standard formultisource data in a header/data form.

Control experiments were performed in two animalswith the probe placed over the dorsal medulla 3 mmcaudal to the NTS, in an area not enervated by thevagus. The descending portion of the vagus was cut inone animal to control for possible direct cardiac influ-ences from single-pulse stimulation. We cut the as-cending portion of the vagus in another animal to con-firm the specificity of the optical response tostimulation. Image sequences were also collected fromtwo animals after euthanasia to control for potentialelectronic artifacts or cross talk.

Several systematic noise sources can affect the opti-cal signal, including cardiac and respiratory pulsationand a 0.1-Hz oscillation in vasomotion (Mayhew et al.,1996). Had we explicitly considered the phase of suchsignals during stimulation, fewer averages may havebeen required since such oscillatory signals are a majorsource of systematic noise. However, stimulating anyof the nerves in the vagal bundle at certain frequenciescan cause entrainment of cardiac and respiratory cy-cles. Thus, the stimuli were randomized between 1 and2 s, specifically to avoid potential cardiac and respira-tory entrainment by stimulation. Additionally, sincethe 0.1-Hz oscillations can appear synchronous withcardiorespiratory cycles, it is possible that triggeringthe stimulus off the 0.1-Hz cycle might also introduceadditional cardiac-related artifact. Rather than trig-gering at a constant phase of the cardiorespiratorycycle, or relative to other potential noise sources suchas the 0.1-Hz rhythm, we utilized additional averagingto reduce systematic noise sources.

A Pentium-based computer displayed images andphysiological signals as well as image sequences aver-aged relative to the stimulator pretrigger or other tim-ing pulse. The computer also stored continuous imagesand physiological records on a mass storage device forsubsequent analysis. Image sequences were generatedby subtracting the average of a sequence of prestimu-lus images from each of the poststimulus images.Time-averaged evoked responses to individual stimuliwere obtained by averaging image sequences collectedrelative to the stimulus pretrigger. In these experi-ments, we required at least 100 stimuli in an averagebecause the evoked optical response is small in com-parison to the total noise of the pixel intensity data.For better visualization of the response, some plotsused as many as 2000 averages to obtain a bettersignal-to-noise ratio.

Images in the average sequence were pseudocoloredfor display such that cool colors (blue to purple) repre-sent increased light reflectance from the tissue (typi-cally correlated with decreased neural activity), warmcolors (yellow to red) represent decreased light reflec-tance from the tissue (associated with increased neuralactivity), and green represents no significant changefrom the baseline image. For comparison of neural

af

981SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

electrical patterns to hemodynamic effects, image se-quences were generated using the cardiac R wave asthe trigger and correlated on a pixel-by-pixel basis tothe blood pressure temporal signal, thereby visualizingthe vasculature.

Since the random stimulations were asynchronouswith the image acquisition timing, it was possible toconstruct image sequences with temporal resolutionfiner than the 10-ms frame interval. Taking into ac-count the timing of the stimulation recorded at 1000Hz, then summing each image into 10 properly aligned1-ms bins rather than arbitrary 10-ms image frames,we achieved better time resolution without sacrificingsignal to noise. For some of the plots, time traces withhigher resolution are used to illustrate the faster tim-ing components.

By inspection of dynamic optical maps and responsewaveforms, we identified at least four temporal com-ponents within the optical response; image subregionsoften contained one or more of the four components. Tomodel the optical responses and characterize the spa-tial distribution of each response component, we ap-plied linear decomposition strategies to average imagesequences. First we averaged the back-scattered lightsignal across all trials (Fig. 8a). The average signal wasmodeled using four separate time-series basis func-tions (Fig. 8b) corresponding to putative physicalmechanisms underlying the optical changes and con-sistent with components resolved by principal compo-nents analysis (PCA) (Fig. 11). Region-of-interest in-formation was also used to tailor time-seriestemplates, since some regions showed one or more re-sponse components in relative isolation.

Using IDL (Interactive Data Language; ResearchSystems, Boulder, CO), we fit a linear combination ofthe four-component temporal basis functions (CURVE-FIT algorithm) to the time-intensity profile of eachpixel in an evoked image sequence (Fig. 8c) and gener-ated four coefficients for each pixel. The resulting basiscoefficients represented the extent to which each time-series basis function was represented in each pixel’stime-intensity profile and thereby provided an esti-mate of the magnitude of each component process foreach pixel. Using basis coefficient values for each pixel,we created four basis coefficient images (Fig. 8d) rep-resenting the magnitude of each temporal componentwithin the evoked image sequence. We used the fourbasis coefficient images together with the time seriesbasis functions to reconstruct the evoked image se-quence (Fig. 8e). We evaluated the residual variance bysubtracting the original image sequence from the re-constructed sequence. Each pixel also had a Z scoressociated with the simultaneous fit of the four basisunctions. When the Z scores for each pixel were as-

sembled into an image, we obtained a parameter mapof how well the four basis functions fit each pixel.

RESULTS

Shocks to each nerve elicited an electrical presynap-tic population spike ;30 ms after the stimulus and apostsynaptic population-evoked potential whichpeaked ;80 ms after the stimulus. Electrical stimula-tion also produced consistent spatiotemporal patternsof optical changes as recorded with the imaging probe.A typical optical image sequence to aortic nerve stim-ulation at 400 mm depth is illustrated in Fig. 3. Aprominent decrease in reflected light (yellow to red)appears at the bottom of the image within 3 frames (30ms) after the stimulus and fades away after 14 frames(140 ms). A long-lasting increase in reflected light(blue) appears in the bottom left of the image after 11frames (110 ms) and lasts for nearly the entire se-quence.

A typical optical response to vagal stimulationshown in Fig. 4A is not present when the probe isplaced over the dorsal medulla 3 mm caudal to the NTS(Fig. 4B). Moving the probe back to the original posi-tion shows a return of the optical response (Fig. 4C).The small reduction in the amplitude of the response inFig. 4C could be a result of a slightly different place-ment. When the ascending portion of the vagus wascut, no optical response was observed (Fig. 4D); how-ever, cutting the nerve between the site of stimulationand the heart had no detectable effect on the opticalresponse. Pseudorandom image sequences, during non-stimulated epochs, analyzed in the same manner asstimulated sequences, produced image sequences thatwere similar to the prestimulus baseline images shown(Fig. 3). No responses were observed when stimulatedimage sequences were collected from a euthanized an-imal.

Similar optical and electrical responses were ob-served for all animals studied (Fig. 5), with slight dif-ferences in the spatial organization of the responsewhen stimulating different nerves (Fig. 6). Averagedimage sequences from the same data sets, triggered bythe cardiac R wave instead of the stimulus pretrigger,were correlated to the blood pressure signal andshowed patterns different from early components ofthe stimulus sequences (Fig. 6). At some lag times,these correlation images clearly disclosed the micro-vascular structure; animated image sequences of car-diac R-wave-triggered image averages illustrated bloodflow through the imaged region. Similar patterns couldbe visualized at much higher contrast by illuminatingwith green light. The time-course plot of the transientoptical response from four regions of interest duringvagus nerve stimulation can be seen in Fig. 7. Differentregions of the sampled area revealed different tempo-ral responses.

Four major temporal components were identified,based on inspection of average waveforms and region of

982 RECTOR ET AL.

interest analysis (Fig. 8): an early positive-going responsecorresponding to the presynaptic population spike (P30),an intermediate negative response corresponding to thepostsynaptic population EPSP (N80), a late and long-lasting negative response that appears to be related to

FIG. 3. A sample average image sequence during aortic nerve simaged area (1000 trials). Images are pseudocolored such that decincreases in light intensity are colored with cool colors (blue to purpscale is notably nonlinear in order to highlight both large and small cthe left edge is on the midline. The stimulus begins within the red-obe seen toward the bottom of the image within 30 ms (3 frames) aftbegins on the 18th frame and persists for nearly the entire sequence.and electrical evoked response (ERP) are plotted for a 100-ms periodand some other time-course calculations, average optical images wephase of stimulus delivery within the 10-ms CCD frame.

the hemodynamic “mapping” signal (N300), and a slowpositive response which was similar to the hemodynamicblood oxygen level-dependent (BOLD) signal employedfor fMRI (P800). Table 1 lists the magnitudes of the fourresponses across all animals.

ulation shows the temporal sequence of optical changes across theses in light intensity are colored with warm colors (yellow to red),and no significant change from baseline is colored green. The color

nges. Images are oriented so that the top of the image is rostral, andined frame and is 100 ms in duration. A rapid optical response canhe stimulus, and a long-lasting light intensity increase (blue color)order to highlight the fast changes, the average image intensity (DI)ter the stimulus corresponding to images outlined in cyan. For thisonstructed at 1 ms resolution by exploiting the random but known

timreale),hautler tInaf

re c

983SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

The response components described in Fig. 8 wereused for a linear decomposition of the intensity time-course plots. Figure 9A shows an original image se-quence after stimulating the vagus, the corresponding

FIG. 6. Optical responses from three animals after stimulatinpatterns. Only the first 100 ms after the stimulus is shown for each seR wave. Images representing the correlation coefficient of reflected liglag times. Cardiac-triggered image sequences typically show vasculapatterns seen in the early frames of the stimulus image sequences. Ffor further analysis. Nerve stimulus image sequences are pseudocoloindicate a light intensity increase, and green indicates no significansuch that warm colors indicate positive correlation, cool colors indicEach image sequence represents an average of 400 trials.

basis coefficient images for each component (Fig. 9C),and a reconstruction using the basis coefficient imagesand time-series basis functions (Fig. 9D). The basiscoefficient images reveal more widespread optical re-

hree different nerves show significant differences in the responsence. Image sequences were also averaged with respect to the cardiacorrelated with the blood pressure signal are shown at three differenterfusion, with clear outlines of vessels. These do not correspond to

each animal, four regions of interest were selected within the imagesuch that warm colors indicate a light intensity decrease, cool colors

hange from baseline. Blood pressure correlation images are colorednegative correlation, and green indicates no significant correlation.

g tqueht cr por

redt c

ate

984 RECTOR ET AL.

sponses across the image for the two slower compo-nents (N300 and P800), which also disclosed structuresimilar to the blood perfusion patterns derived fromcardiac-triggered correlation images (Fig. 9B). The re-constructed image sequence based on the four basisfunctions accounted for 91 6 5% of the signal variance.The same four basis functions accounted for 85 6 9% ofthe total variance in image sequences from all animalsstudied, although in many cases, the model agreementfor an individual animal was improved by minor timingadjustments of the process time courses. Basis coeffi-cient images and their associated Z score fit and bloodpressure correlation image are shown for four animalsin Fig. 10.

PCA was applied to the average image sequences inorder to separate components using a blind decompo-sition technique. The first temporal component (C0)resembled the mean response across the entire image.The next two components, C1 and C2, isolated twowaveforms from the response with distinct spatial pat-terns in the image (Fig. 11). The temporal profile andspatial distribution of C1 corresponded to the N80 ba-sis component, and C2 resembled the N300 component.The other identified PCA components appeared to beadmixtures of all components.

FIG. 4. Average image intensity change of the NTS (A) duringstimulation of the vagus nerve shows a typical optical response overa period of 800 ms after the stimulus onset (dotted line). When theimage probe was moved 3 mm caudal to the NTS over the dorsalmedulla, no optical response to stimulation of the vagus was ob-served (B). The optical response returns when the image probe wasplaced back over the NTS (C). When the ascending portion of thevagus is cut, both the optical and the electrical response disappear(D). The lower trace (E) represents the electrical evoked response asrecorded by a macrowire placed under the imaged area of the NTS.Each trace represents an average of 2000 trials.

Varying the focus depth also revealed some consis-

tent patterns. Time-course plots of the optical responseafter vagus stimulation across depths show a signifi-cant decrease in the N300 response at deeper levels(Fig. 12). The values for the basis coefficients for allanimals were averaged and compared across depth.The most striking difference appeared between theN80 and the N300 components, such that the N80component tended to increase, and the N300 compo-nent decreased with increasing depth, especially forvagus nerve stimulation. The P800 response showed no

FIG. 5. Average image intensity change of the NTS during stim-ulation of the vagus for many animals shows the variability of theresponse. Optical responses to superior laryngeal (401SN) and aortic(401AN) nerve stimulation are plotted with higher time resolutionfor one animal for comparison. A typical event-related (evoked) elec-trical response is plotted in the lower trace (ERP). Each trace rep-resents an average of 1000 trials.

FIG. 7. Average pixel intensity for four regions of interest, A, B,C, and D, outlined in Fig. 6 are plotted across time for vagus nervestimulus for three animals (400 trials averaged per trace). Theseregions were selected to illustrate the named components. A sampleevoked potential is displayed at the bottom. Each region of interestdisplays a unique temporal pattern, depending on its position withinthe image.

VAS

tttlnt

985SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

significant difference as a function of depth, and theP30 response showed a small increase within thedeeper levels for VN stimulation. The probe used inthis experiment was not confocal and thus had limitedresolution in depth, so that statistically significant dif-ferences were seen in only a few comparisons, illus-trated in Fig. 12.

Stimulating the branches of the vagal bundle hadthe potential to affect arterial pressure and thus pro-duce optical responses by modulating heart rate andaortic tone; however, these changes would be muchslower than the observed responses (greater than 3 s)and relatively consistent during stimulation at 0.5–1.0Hz. Pulse trains which we used to affect heart rate andblood pressure were much more rapid and long-lasting(100 Hz for 1 s) than our stimulus paradigm. We didnot observe a change in heart rate during normal stim-ulation compared to prestimulus conditions (396 6 30

FIG. 8. Average intensity change after stimulation averagedacross all animals, depths, and stimulated nerves produced a typicalresponse curve (A). Four components were identified based on puta-tive physical processes for each optical component. The earliest com-ponent (P30) is a positive-going response which peaks 30 ms after thestimulus, corresponding to the presynaptic population spike. Thesecond response (N80) is a negative response which peaks 80 msafter the stimulus, corresponding to the postsynaptic populationpotential. A slower negative response (N300) peaks roughly 300 msafter the stimulus and may correspond to the hemodynamic mappingsignal. The longest lasting positive-going response (P800) peaksmore than 800 ms after the stimulus and may correspond to thehemodynamic recovery, blood oxygen level-dependent signal. Eachtemporal component was isolated into time-series basis functions ofunit peak amplitude (B). To determine the spatial representation ofeach temporal component, evoked image sequences (C) were fit on apixel-by-pixel basis, producing an image of coefficients (D) for eachtime-series basis function. The coefficients represent the fraction ofeach image perturbed by the four component processes. A recon-structed image sequence (E) was produced by multiplying the time-series basis functions by their respective basis coefficient images.

bpm vs 397 6 29 bpm). Additionally, blood pressurecorrelation image sequences generated during both thestimulated and the nonstimulated period disclosed nosignificant difference between the two conditions.

DISCUSSION

Fast scattered light changes were observed in syn-chrony with evoked potentials in the NTS. We observedconsistent spatiotemporal patterns in image sequencesassociated with dorsal medulla activation after nerveshocks. The responses could be resolved into at leastfour temporal components with distinct time signa-tures and spatial representations. When imaged atdifferent depths, the slower, N300 hemodynamic com-ponent occurred in the superficial levels, and thefaster, N80 component was largest at deeper levels,especially during VN stimulation. Stimulating differ-ent vagal bundle branches also produced slightly dif-ferent response patterns.

Comparison with Other Techniques

While fast optical signals have been seen in isolatednerve preparations, most other groups have not re-ported such fast optical changes in an in vivo prepara-tion. Intrinsic optical responses usually appear 1–2 spoststimulation. We believe there are many reasonsthat investigators see primarily the slow signals. First,many image recording systems are relatively slow (30Hz) and have limited signal to noise and dynamicrange (50 dB at best) (Dowling et al., 1996). Manystandard video digitizers are limited to 6 or 8 bits andalso cannot stream the data continuously to the hostfor analysis and archiving. Archiving to videotape canprovide continuous image sequences, but with a signif-icant reduction in signal to noise and dynamic range(46 dB with the best sVHS recorders). Digital video-tape devices could potentially overcome some of thenoise limitations of standard analog video recorders;however, digital videotape recorders rely heavily oncompression techniques. We find that any type of lossy

TABLE 1

P30 N80 N300 P800

N 1.16 6 0.08 26.99 6 0.69 23.05 6 0.48 6.05 6 0.73N 0.22 6 0.06* 24.39 6 0.32 21.17 6 0.54 1.34 6 0.13N 0.34 6 0.12* 23.64 6 0.58 21.06 6 0.36 0.67 6 0.26

Note. The optical response, averaged across the entire image (100o 1000 trials each) and across 25 animals, was divided into fouremporal components (P30, N80, N300, and P800). The mean ampli-udes for each of the components are tabulated for the three stimu-ated nerves (vagus nerve, VN; aortic nerve, AN: superior laryngealerve, SN). All values are 6SEM and 31025; *indicates comparisonshat were not significantly different from baseline (a 5 0.1).

sbbtpcmtsf

986 RECTOR ET AL.

compression (e.g., MPEG) will introduce significant ar-tifact into small changes in video images.

For better dynamic range, some investigators use pho-todiode arrays which provide both high speed (at limitedresolution) and good dynamic range. However, in ourexperience, photodiode arrays have limited sensitivity,making them good for high-level illumination. For lowlight levels, CCD technology is more appropriate, offering

FIG. 9. Average image sequences synchronized to electrical stimactivity in the dorsal medulla (1000 trials). (A) 90 frames at 10-ms intto the ipsilateral vagus nerve on the frame outlined in red (frame 10reflectance decrease (to 25 3 1024), and cool colors represent percignificant change from baseline. (B) A correlation image of the blooy the peak of the cardiac R wave. Different regions of the stimuluse separated into four major components which model four physiemporal basis functions, four basis coefficient images were createdrofile accounted for by each basis function (C). Since the P30 anorresponding basis function images were inverted to parallel the reost of the variance in the evoked image sequence can be accounted

op edge of the image is rostral, and the left edge of the image is on theo that warm colors represent positive correlation or positive fractiractional coefficients, and green represents no correlation or a fract

better quantum efficiency. For optimum signal to noise,many investigators have used slow-scan 16-bit CCD cam-eras, but with severe limitations on the acquisition speed(.300 ms per frame). Recording these signals requiresfast, low-noise cameras and attention to optimize dy-namic range of the recording system. By carefully adjust-ing offset and gain, we can employ greater amplificationof the video signal. We believe that the illumination strat-

ation of the vagus nerve show the spatial representation of evokedals representing 100 ms before and 800 ms after a shock is deliveredmages are pseudocolored so that warm colors represent percentage

tage reflectance increase (to 5 3 1024), while green represents noressure signal compared with an average image sequence triggeredoked sequence in (A) show unique response time courses which canprocesses associated with optical changes. Using four componentpixel-by-pixel basis representing the fraction of the pixel intensity

P800 basis functions are opposite the N80 and N300 curves, thectance change. A reconstruction of the original data (D) shows thatr by the four basis functions. All images are oriented such that theidline. The correlation and basis coefficient images are pseudocoloredl coefficients, cool colors represent negative correlation or negativeal coefficient of 0.

ulerv). Iend p-evcalon adflefom

onaion

987SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

mdsGrTicgpswNrttwa

irhtcHBtsspbdfin

Td

o

rcd

988 RECTOR ET AL.

egy described in the following section is another key toobserving the fast optical signals.

Optical techniques have limitations of their own atthis point, but they do provide advantages over othertechniques. Both electrode methods and dye imagingrequire penetration of the tissue with a foreign sub-stance and have limited utility for chronic recordings(e.g., hours, days, or weeks). Our experiments cur-rently require removal of the skin and bone and contactwith the tissue surface of interest. However, severalinvestigators have observed optical signals noninva-sively through thinned bone in rat (Masino et al.,

FIG. 10. Blood pressure correlation images (BP) and basis coeffiche basis function-fitting algorithm returns a x2 value for each pixisplayed in the Z score column show that most of the pixels fit the

showed increased variability. The correlation and basis coefficientcorrelation or positive fractional coefficients, cool colors represenrepresents no correlation or a fractional coefficient of 0. The Z scoref 0.0, green represents a x2 of 0.10, and white represents 0.20.FIG. 11. A trace of a light-scattering response averaged across t

esponse functions, termed P30, N80, N300, and P800. PCA revealomponents and had distinct spatial representations in the images.isclosed in blood pressure correlation images for this image sequen

FIG. 12. Average pixel intensity across the entire image plottedacross time after vagal stimulation (vertical line) shows optical sig-nals as a function of depth from neural activation from three ani-mals. Each trace represents an average of 400 trials. With increasingdepth, the N300 component decreased in amplitude. We would ex-pect a hemodynamic mapping response to decrease with depth sincethe vessels are generally more superficial. The average basis coeffi-cient values for the N80 and N300 responses across all animals ateach depth show a significant and opposite depth dependence of thetwo responses. An asterisk indicates a significant difference from themost superficial value (a 5 0.1).

1993), and through bone in the mouse (Prakash et al.,2000), and through skin and bone in human subjects(Steinbrink et al., 2000; Gratton et al., 1997). Lessinvasive procedures come at the price of lower spatialresolution.

We have successfully implanted image probes in catsand goats and recorded slower optical signals continu-ously for weeks at a time (Rector et al., 1997b). The

ost successful implantable probes utilized image con-uit for transmitting back-scattered light from the tis-ue to the CCD camera. However, we have developed aRIN lens probe small enough to implant into cats and

ats for freely behaving studies (Rector et al., 2000).he GRIN lens numerical aperture is low (0.10) and it

s not nearly as light efficient as the image conduitonfiguration so that adequate illumination is areater problem. Fixed-focus versions of the imagerobe, using GRIN lens or image conduit, are welluited to chronic implantation in freely behaving ratsithin cortical or subcortical structures. However, theTS does not reside within a rigid region of the skull;

ather it is within a highly mobile region just outsidehe foramen magnum in the neck. Probe placement inhis region for studies in a freely behaving animalould require vertebrae fusing procedures for stabilitynd may not be well tolerated by the preparation.

Rapid Light-Scattering Changes

Previous imaging studies of scattered-light changesn vivo have failed to observe the fast optical changeseported here. A number of technical improvementsave allowed such measurements. Dark-field illumina-ion methods increased our ability to detect smallhanges in scattered light (Stepnoski et al., 1991;olthoff and Witte, 1996; Rector et al., 1997b, 1999).right-field illumination (Grinvald et al., 1986) tends

o be dominated by reflected light directly from theurface or scattered in shallow layers; thus, only amall proportion of the imaged light is perturbed byhysical processes associated with activation occurringelow the tissue surface. Additionally, blood vesselsominate the tissue surface; thus, more of the bright-eld signal is derived from hemodynamic responses toeural activity. Dark-field measurements record

t images (P30, N80, N300, and P800) for four animals are displayed.indicating how well the basis functions fit the data. The x2 valuessis functions with a x2 of 0.02 or better, with a few “hot spots” thatages are pseudocolored such that warm colors represent positiveegative correlation or negative fractional coefficients, and greenges are pseudocolored such that black represents a perfect x2 value

ntire image for animal 401 (1000 trials) illustrates four componenttwo strong components that roughly paralleled the N80 and N300C2 component has a spatial distribution similar to the vasculature

(see Fig. 9).

ienel,baim

t nima

he eedThece

989SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

deeper changes since light must first penetrate thetissue before returning to the camera.

A rapid frame acquisition rate allowed us to collectimages of optical changes with finer temporal and spa-tial resolution for recording transient responses. Ahigh-sensitivity CCD camera and analog amplifierswith high signal-to-noise resolution allowed 12-bit dig-itizing over the effective dynamic range of the signal,providing greater sensitivity to small signal changes.Signal gain at the camera head reduced the conse-quences of noise on the transmission cable. This strat-egy provided 10 to 11 bits of usable signal in single-pass data. Finally, placement of the image-collectinggradient index lens in contact with the tissue withillumination around the perimeter of the imaged areastabilized the tissue interface, suppressing some move-ment artifacts, and optimized scattered light changesfrom tissue below the surface. Focal depth adjustmentsalso allowed optimization of signals originating fromdeep sources.

A number of studies have demonstrated a close re-lationship between light scattering changes and neuralactivity (Hill, 1950; Cohen, 1971; Lipton, 1973; Grin-vald et al., 1986; MacVicar and Hochman, 1991; Rectoret al., 1997a; Haglund et al., 1992). Increased neuralactivity (as measured by action potential discharges)typically causes a decrease in the opacity of the tissue,especially to red and infrared light. Poe et al. (1996)showed that administering water into the interstitialspace to initiate cellular swelling has an optical effectcomparable to neural discharge, decreasing theamount of back-scattered light. Similarly, infusion of asucrose solution caused concentration-dependent cellshrinkage and increased back-scattered light. Studiesby MacVicar and Hochman (1991), using transmittedlight, showed that local administration of sodium–po-tassium pump inhibitors prevented the flow of waterinto the cells with activation and abolished action po-tentials along with most of the optical response.

Although near-IR wavelengths have several advan-tages for studies of intact tissue, the wavelength thatwe used was partly out of practicality rather thandesign choice. In our fiber-optic image conduit probes,virtually any wavelength of light could be used sincethe image conduit was very efficient at collecting back-scattered light from the tissue and transmitting it tothe CCD camera, allowing us to use very small LEDsfor illumination in implantable camera designs. Thefiber-optic image conduit design, however, precludedfocusing deeper into the tissue. The GRIN lens designis not very efficient at transmitted back-scattered lightfrom neural tissue and required more powerful illumi-nation sources. We switched to a 780-nm laser diodearray that we have previously used for other purposes.Near-infrared wavelengths also allowed deeper pene-tration into tissue (Rector et al., 1999) and this wave-

length provided spectral sensitivity to hemoglobinoxygenation. We consider multiple-wavelength illumi-nation a critical strategy for interpreting the origin ofthese signals (Malonek and Grinvald, 1996) and arecurrently developing experiments to use spectral im-aging.

Temporal Components

The four temporal response components had differ-ent time courses and spatial distributions. The twofaster responses (P30, N80) were limited in their spa-tial distribution and appeared to reflect direct conse-quences of neural activation. We hypothesized that theearly responses were tightly linked to the populationelectrical responses, and therefore we patterned thefirst two basis functions (P30 and N80) after the ob-served electrical signals. Such fast responses might beattributed to a number of cellular processes whichchange the light-scattering properties of neural tissuewith activation. Neural swelling accompanies activa-tion (Hill, 1950; Van Harreveld, 1958; Cohen, 1973;Lipton, 1973; Tasaki and Byrne, 1992) with a timecourse that can parallel the action potential. The earlyP30 response appears to correlate with presynapticactivation, although a small population of fast “A” fi-bers might produce postsynaptic activation on thistime scale (Rogers et al., 1993).

The two slower components (N300, P800) were moredistributed and overlapped with regions showing dis-tinct structure in the blood pressure correlation im-ages. The time course of the slower componentsmatches the hemodynamic responses reported in ear-lier studies (Frostig et al., 1990; Malonek and Grin-vald, 1996, 1997). The N300 response appears consis-tent with the hemodynamic “mapping” signaldescribed by Grinvald and colleagues. This signal isbelieved to represent a fall in hemoglobin oxygenation,perhaps triggered by increased metabolic demand dur-ing activation. The P800 response is similar to thehemodynamic BOLD signal observed with fMRI, whichtypically peaks 8–10 s after stimulation. Magnetic res-onance imaging changes are consistent with an in-crease in venous oxygenation driven by an (over)com-pensatory increase in blood flow (Kleinschmidt et al.,1996).

Spatial Component Patterns

The spatial specificity of the activation patternswithin the NTS were not as well defined as we antici-pated. However, spatial differences emerged both be-tween the components and as a result of stimulatingdifferent nerves. In general, the N300 and P800 re-sponse components reveal spatial patterns that closelyresemble the vasculature observed in the blood pres-sure correlation images, and the P30 and N80 response

990 RECTOR ET AL.

components appear predominantly in different regionswithin the image. The N80 response partially overlapswith the vascular reaction, possibly due to incompleteseparation of the N300 response or perhaps resultingfrom the presence of activated cells within regions ofthe vasculature.

Unfortunately, the spatial organization of the NTS isnot as well laid out as is the case in other structuressuch as the cortex. Since the spatial distribution ofactivity in the NTS is so diffuse, our initial aims forthis study were to use optical signals to precisely local-ize functional organization. In this study, the spatialresolution of the scattered light signals is limited byseveral factors, including significant blurring by lightscattering and inadequate sensitivity of the detectorsto observe optical changes from smaller neural popu-lations. We hope to address these issues in future ex-periments through the use of confocal techniques toreduce contributions from out-of-plane light andthrough development of better detectors for improvedsensitivity.

Presynaptic cells in the orthodromically stimulatedpathway have cell bodies in the animal’s thorax, andthus we expect the synapses to be distributed over thetarget cells. Cells antidromically driven by stimulationmight contribute to the P30, but presumably would notcontribute to the N80 response. In any case, the distri-bution of such cells is not known. There is some evi-dence for different spatial localization of the P30, N80,and N300 responses since the basis component imagesreveal different structure. In general, the N300 andP800 appear as changes in a superset of the area ap-pearing in the N80, presumably reflecting an addi-tional venous component.

The P30 Response

There is evidence that the earliest optical responseproduced by electrical stimulation may precede electri-cal signal arrival at presynaptic terminals by at leastone image frame (10 ms). The exact timing of thisresponse is difficult to ascertain, since the 10-ms frameinterval does not adequately resolve the response andtemporally smears the signal. This surprising result,also noted by others (Landowne, 1985; Salzberg et al.,1985), might result from voltage-dependent channelconformational changes that precede the ionic flux;however, channel conformational changes would be ex-pected to precede the electrical signal at the site ofmeasurement by ,0.1 ms. Alternatively, we hypothe-size that a compression wave may arise from ion andwater influx at the stimulus site and as excitationmoves along the axon. Such a compression wave wouldbe expected to travel at the speed of sound, faster thannerve conduction velocities, and might explain theseearly signals, which occur 10 ms or more before elec-

trical events. Faster acquisition rates coupled withother physical measurements will allow us to test thishypothesis.

The P30 response does not appear as robust as theother components in the images, presumably becausethe fastest neural component is one-fifth the amplitudeof the other components and may be more diffuselyrepresented across the NTS. Also, the P30 responsewas significant over baseline only after stimulation ofthe vagus nerve, suggesting that a bigger population ofnerves was required to produce an observable re-sponse. There may also be some P30 contribution fromP800 wraparound, though this does not explain thesharp peak in the time course or the focal hot spotsapparent in many P30 maps.

The traces depicted in Fig. 7 are of selected subre-gions within the image to illustrate the point that notall parts of the image show the same optical responsecomponents. Some regions include one or more of thedifferent components. Thus, examination of all the sub-regions within the same animal revealed the P30 re-sponse in some regions, but not others; however, allanimals in this experimental ensemble have some re-gion where the P30 response was observed.

The N80 Response

Light-scattering changes associated with populationpostsynaptic potentials probably contribute to the N80response. Additionally, surrounding glial cells mayalso respond to tissue activity and produce opticalchanges (Andrew and MacVicar, 1994), although mostinvestigators have concluded that glial responses havea slower time course. The P30 and N80 responses ap-peared consistently in the intensity traces from aver-aged image sequences. In these experiments, the N80response was often the largest component. Althoughthe N80 response was overlapping in time and spacewith the N300 response, we were often able to identifyregions of interest with minimal apparent contributionfrom the N300 (Fig. 7).

The N300 Response

The characteristic N300 (and P800) response pat-terns are highly localized to the vasculature identifiedthrough comparison with blood pressure correlationimages and appear prominently in the images. Basisimages of the N80 component show some spatial over-lap with the N300 component. We would anticipatethat focal neural activation would produce localizedchanges in oxygenation within the overlying capillarybed as well as slower changes in the microvenousdrainage. However, the temporal and spatial overlapbetween the N80 and the N300 components might re-sult from the somewhat arbitrary assignment of the

991SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

basis functions, which could lead to a failure of thedecomposition scheme.

In order to seek additional justification for discrimi-nation of the N80 and N300 responses, we utilizedPCA. PCA is a blind decomposition technique driven bymathematical criteria. Although PCA cannot becounted on to separate biologically significant compo-nents, in some cases it does, in particular when thecomponents are strong and orthogonal. PCA revealedsignificant temporal and spatial differences in compo-nents that correspond to the N80 and N300 compo-nents that we assigned based on the electrical andhemodynamic responses. The spatial pattern associ-ated with the C2 PCA component was consistent withpatterns of vasculature, supporting the assignment ofthe N300 response as a hemodynamic component.

Several investigations have observed “early” hemo-dynamic effects on the order of 150 to 250 ms (Lindaueret al., 1993; Sandman et al., 1984) that may explain theN300 component. These early hemodynamic signalswere observed in cortex and may be most apparentduring stimulation at a particular phase of the cardiaccycle. In our experiments, the apparent peak at 300 msmight reflect a superposition of multiple processes. Afast deoxygenation response may overlap with thestart of the BOLD signal. Other secondary neural in-teractions as well as glial cells could also respond withlonger time courses. Spectrally resolved measurementsmay allow dissection of component processes morecompletely (Malonek and Grinvald, 1996).

The N300 response represents a decrease in reflec-tance such as would correspond to an increase in ab-sorbance at 780-nm wavelength due to deoxygenationof hemoglobin. Difference spectra for hemoglobin as afunction of oxygenation have a peak at 780 nm and anisosbestic point at 800 nm. However, the underlyingabsorption peak is much smaller than the Soret bandsin the visible range. By obtaining simultaneous opticalmeasurements at multiple wavelengths (e.g., at 780and 800 nm) hemodynamic oxygenation can be directlyassessed or these effects may be removed if desired.

The P800 Response

The spatial representations of the P800 componentecho the patterns of the vasculature seen in the bloodpressure images. The polarity of the signal is consis-tent with a rise in blood oxygenation, although quan-tization is confounded by measurements at a singlewavelength. We believe the P800 component corre-sponds directly to the late and long-lasting hemody-namic optical response that most investigators observefrom the surface of the cortex. These responses canbegin as early as 1 s and last 10 to 20 s.

In our experiments, the interstimulus interval (ISI;uniformly distributed between 1 and 2 s) was not suf-

ficiently long to avoid wraparound of the P800 responsefrom the previous stimulus. Some of the image se-quences show a small downward shift in the baselineimage intensity that may represent this wrap-aroundeffect from the P800 component. Experimental limita-tions precluded an ISI that was adequate for recoveryof baseline conditions: to adequately resolve the dy-namics of the P800 response, along with other compo-nents, would require much longer recording periods.Concern for the stability of the signal over such a longperiod precluded such a long ISI. If the P800 responseresulted from a BOLD response recorded in isolation,we would expect it to be the largest signal; however,our P800 signal is probably suppressed (saturated) bythe rapid stimulus rate and baseline correction strat-egy employed. Had there been longer intervals be-tween stimuli, we would expect the P800 componentwould be more significant in amplitude. However, sincethe primary aim was to characterize the fast compo-nents, we chose to decrease the ISI to achieve an ap-propriate number of averages over a shorter time pe-riod.

Control Experiments

Several controls were utilized to eliminate the pos-sibility of artifacts producing the observed signals.When the probe was placed caudal to the NTS, whenthe ascending portion of the nerve was cut, or when theanimal was no longer alive, the signal was not present(Fig. 4) and image sequences resembled the prestimu-lus baseline images. Cutting the nerve downstream ofthe stimulus had no effect on the optical response.Additionally, many regions within the imaged area didnot show a significant response to stimulation, suggest-ing that the response was localized to subregionswithin the dorsal medulla. Pharmacological agentssuch as glutamate, tetratotoxin (TTX), and magnesiumwould be excellent control paradigms for isolatednerves or tissue slices or cultures. However, theseagents are inappropriate for in vivo use in these exper-iments. We previously observed an apparent hyperex-citation of cortical and hippocampal tissue in responseto TTX in vivo, observed as a dramatic decrease inback-scattered light across the entire image (Poe et al.,1996). This may reflect a more pronounced effect of theagent on inhibitory interneurons.

Stimulus amplitudes considerably larger than thoseused to generate these evoked responses sometimesproduced spreading currents which induced muscletwitches and caused movement artifact in the imagesequences. Under these conditions, the optical signalhad a large monophasic response that was uniformacross the entire field of view. This signal had a longerrise time and was longer lasting than neural optical

waa

992 RECTOR ET AL.

signals and was not correlated with the evoked electri-cal or hemodynamic responses reported here.

Depth-Dependent Changes

The depth dependence of the response revealed sev-eral aspects of the optical signals. Since the slower,N300 hemodynamic response was recorded more su-perficially, the present results are consistent with his-tological evidence that vascularization and major ves-sels most commonly appear near the tissue surface.Since the more rapid N80 component was visualized atdeeper levels, where NTS neural cell bodies and den-drites exist, we suspect that the faster optical re-sponses result from physical neuronal changes associ-ated with electrical activation of these cells. The fitcoefficients for the N80 and N300 responses signifi-cantly differ with depth, additional evidence that theN80 and N300 responses result from different pro-cesses (Fig. 12).

Not all of the nerves stimulated showed a clear depthdependence. Significant differences were primarilyseen when stimulating the vagus nerve, which is alsothe largest nerve of the three stimulated and thusprovided a more robust response. Interpretation of thedepth dependence is limited in these experiments sinceout-of-plane light contributes significantly (35% at0.400 mm) to images at each level partly due to the lownumerical aperture of the GRIN lens (0.10). Confocalprocedures are needed to help exclude out-of-planelight and better resolve depth.

Stimulated-Nerve-Dependent Changes

The imaged area in the rat is known to containneurons that are driven by inputs from the nervesstimulated in the present study (Rogers et al., 1993;Scheuer et al., 1996). VN stimulation elicited a strongN80 response in the deeper layers, which correspondedto the region of the ventral NTS and the dorsal motornucleus of the vagus (Paxinos and Watson, 1986). Ofthese two structures, the latter would be expected togive the most temporally coherent response since thesemotor neurons, whose axons course through the vagus,would be antidromically activated during VN stimula-tion. The latency of response (;30 ms) is consistent

ith the observation that these motor neurons have, bynd large, nonmyelinated or slowly conducting myelin-ted axons (C and B fibers, respectively; Jones et al.,

1995).The number and density of NTS neurons may not be

sufficiently large to produce an observable early opticalresponse for some stimulation paradigms. This techni-cal obstacle might be overcome by focusing the imagefield to a smaller region of interest and increasing thesignal amplitude through the use of polarized light orvital dyes. Improved sensitivity and reduced toxicity of

voltage-sensitive dyes may make such procedures moreuseful for detailed analysis of neural networks (Tso-dyks et al., 1999; Shoham et al., 1999). The use ofmultiphoton microscopy techniques should decreasethe photodamage (Svoboda et al., 1999). However, forlong-term recordings in freely behaving animals, min-imally invasive measurements based on intrinsic sig-nals have significant advantages.

Due to issues with the spatial resolution and limitedprior knowledge of functional organization, interpreta-tion of the spatial activation patterns across nerves islimited at best. Observed spatial activation patternswere similar, but not identical, across animals. Stim-ulation of different nerves produced activation pat-terns that differ slightly but consistently. Since theprojections of different nerves to the NTS are not seg-regated spatially, dramatic differences in activity pat-terns did not emerge. Also, it is known that many cellsin the NTS receive convergent inputs from more thanone nerve. Since the structure of the NTS is not as welldefined as other CNS structures, we did not expect tosee dramatic differences in the spatial representationsof the different nerves.

We illustrated the responses of the three nervesacross three animals (Fig. 6) to show relative spatialpatterns across animals. Inconsistency in the exactspatial distribution of activity between animals isprobably due to two factors. First, the NTS is notorganized in a laminar fashion, as is the case in thelateral geniculate, hippocampus, and inferior olive, orwith functional columns as in the visual or somatosen-sory cortex. Clusters of functional elements intermin-gle within the nucleus, and substantial interactionstake place between those elements. Second, stereotaxicidentification and placement of the image probe overthe NTS was not feasible due to limitations of atlasesfor this region. Since the NTS resides outside the rigidstructure of the skull, precise localization of the NTSwas not possible. Probe placement was achievedthough identification of gross features such as the areapostrema and fourth ventricle. However, with consis-tent procedures (see data from animals 401 and 402),similar patterns of venous anatomy and activation pat-terns could be observed.

CONCLUSIONS

The present study demonstrates the feasibility ofimaging fast intrinsic optical signals that are tightlycoupled with neurophysiological responses. Experi-mental observations are consistent with the knownphysiological response properties and microanatomy ofthe dorsal brainstem structures. Dynamic activationpatterns were imaged in three dimensions with reso-lution and coverage not previously achieved. A numberof potential technical advances for the imaging probe

C

D

D

F

F

G

G

G

H

H

H

J

J

K

L

L

L

L

L

L

M

M

M

M

993SCATTERED LIGHT TRACKS FAST AND SLOW PROCESSES

should increase the spatial and temporal resolution ofimaging studies; the probe can be configured to createconfocal, spectrally resolved images of the tissue forbetter spatial resolution, biochemical and physiologicalspecificity, and 3-D reconstruction of tissue activationvolume. High-speed and high-sensitivity optical re-cordings have exciting potential for studying dynamictemporal and spatial activity patterns which may un-derlie high-level processing within local neural net-works during autonomic (Lindsey et al., 1997), sensory(Singer and Gray, 1995), and cognitive (Fries et al.,1997) functions. Using near-infrared light, it mayeventually be possible to perform imaging or opticaltomography of cortical activation noninvasively (Ma-sino et al., 1993), perhaps in human subjects (Grattonet al., 1997; Steinbrink et al., 2000).

ACKNOWLEDGMENTS

This study was funded in part by grants from the NIH (HL-22418and MH 60993-04) and the U.S. DOE Technology Transfer Initiative.D.R. was a Director’s Fellow at Los Alamos National Laboratoryduring this work. R.R. is supported by NIH Grant 1-R01-HL-54194.MH 60993-04 is a Human Brain Project/Neuroinformatics ResearchProject funded jointly by the National Center for Research Re-sources, National Institute of Mental Health, National Institute onDrug Abuse, and National Science Foundation.

REFERENCES

Andrew, R. D., and MacVicar, B. A. 1994. Imaging cell volumechanges and neuronal excitation in the hippocampal slice. Neuro-science 62: 371–383.

Bragin, A., Hetke, J., Wilson, C. L., Anderson, D. J., Engel, J., Jr.,and Buzsaki, G. 2000. Multiple site silicon-based probes forchronic recordings in freely moving rats: Implantation, recordingand histological verification. J. Neurosci. Methods 98: 77–82.

Chapin, J. K., and Nicolelis, M. A. 1999. Principal component anal-ysis of neuronal ensemble activity reveals multidimensional so-matosensory representations. J. Neurosci. Methods 94: 121–140.

Cohen, L. B. 1973. Changes in neuron structure during action po-tential propagation and synaptic transmission. Physiol. Rev. 53:373–413.ohen, L. B., and Keynes, R. D. 1971. Changes in light-scatteringassociated with the action potential in crab nerve. J. Physiol. 212:259–275.onoghue, S., Felder, R. B., Gilbey, M. P., Jordan, D., and Spyer,K. M. 1985. Post-synaptic activity evoked in the nucleus tractussolitarius by carotid sinus and aortic nerve afferents in the cat.J. Physiol. 360: 261–273.owling, J. L., Henegar, M. M., Liu, D., Rovainen, C. M., and Wool-sey, T. A. 1996. Rapid optical imaging of whisker responses in therat barrel cortex. J. Neurosci. Methods 66: 113–122.

ries, P., Roelfsema, P. R., Engel, A. K., Konig, P., and Singer, W.1997. Synchronization of oscillatory responses in visual cortexcorrelates with perception in interocular rivalry. Proc. Natl. Acad.Sci. USA 94: 12699–12704.

rostig, R. D., Lieke, E. E., Ts’o, D. Y., and Grinvald, A. 1990.Cortical functional architecture and local coupling between neuro-nal activity and the microcirculation revealed by in vivo high-

resolution optical imaging of intrinsic signals. Proc. Natl. Acad.Sci. USA 87: 6082–6086.ratton, G., Fabiani, M., Corballis, P. M., and Gratton, E. 1997.Noninvasive detection of fast signals from the cortex using fre-quency-domain optical methods. Ann. N. Y. Acad. Sci. 820: 286–298.rinvald, A., Frostig, R. D., Lieke, E., and Hildesheim, R. 1988.Optical imaging of neuronal activity. Physiol. Rev. 68: 1285–1368.rinvald, A., Lieke, E., Frostig, R. D., Gilbert, C. D., and Wiesel,T. N. 1986. Functional architecture of the cortex revealed by opti-cal imaging of intrinsic signals. Nature 324: 361–364.aglund, M. M., Ojemann, G. A., and Hochman, D. W. 1992. Opticalimaging of epileptiform and functional activity in human cerebralcortex. Nature 358: 668–671.ill, D. K. 1950. The volume change resulting from stimulation of agiant nerve fibre. J. Physiol. 111: 304–327.olthoff, K., and Witte, O. W. 1996. Intrinsic optical signals in ratneocortical slices measured with near-infrared dark-field micros-copy reveal changes in extracellular space. J. Neurosci. 16: 2740–2749.

ones, J. F. X., Wang, Y., and Jordan, D. 1995. Heart rate responsesto selective stimulation of cardiac vagal C fibers in anaesthetizedcats, rats, and rabbits. J. Physiol. 489: 203–214.

ones, M. S., and Barth, D. S. 1999. Spatiotemporal organization offast (.200 Hz) electrical oscillations in rat vibrissa/barrel cortex.J. Neurophysiol. 82: 1599–1609.leinschmidt, A., Obrig, H., Requardt, M., Merboldt, K. D., Dirnagl,U., Villringer, A., and Frahm, J. 1996. Simultaneous recording ofcerebral blood oxygenation changes during human brain activa-tion by magnetic resonance imaging and near-infrared spectros-copy. J. Cereb. Blood Flow Metab. 16: 817–826.

andowne, D. 1985. Molecular motion underlying activation andinactivation of sodium channels in squid giant axon. J. Membr.Biol. 88: 173–185.

indauer, U., Villringer, A., and Dirnagl, U. 1993. Characterizationof CBF response to somatosensory stimulation: Model and influ-ence of anesthetics. Am. J. Physiol. 264: H1223–1228.

indsey, B. G., Morris, K. F., Shannon, R., and Gerstein, G. 1997.Repeated patterns of distributed synchrony in neuronal assem-blies. J. Neurophysiol. 78: 1714–1719.

ipski, J., McAllen, R. M., and Spyer, K. M. 1975. The sinus nerveand baroreceptor input to the medulla of the cat. J. Physiol. 251:61–78.

ipton, P. 1973. Effects of membrane depolarization on light scat-tering by cerebral cortical slices. J. Physiol. 231: 365–383.

oewy, A. D. 1990. Central autonomic pathways. In Central Regu-lation of Autonomic Functions (A. D. Loewy and K. M. Spyer, Eds.),pp. 88–103. Oxford Univ. Press, New York.acVicar, B. A., and Hochman, D. 1991. Imaging of synapticallyevoked intrinsic optical signals in hippocampal slices. J. Neurosci.11: 1458–1469.alonek, D., and Grinvald, A. 1996. Interactions between electricalactivity and cortical microcirculation revealed by imaging spec-troscopy: Implications for functional brain mapping. Science 272:551–554.alonek, D., and Grinvald, A. 1997. Vascular regulation at submillimeter range: Sources of intrinsic signals for high resolutionoptical imaging. Adv. Exp. Med. Biol. 413: 215–220.asino, S. A., Kwon, M. C., Dory, Y., and Frostig, R. D. 1993.Characterization of functional organization within rat barrel cor-tex using intrinsic signal optical imaging through a thinned skull.Proc. Natl. Acad. Sci. USA 90: 9998–10002.

R

R

R

R

R

R

R

R

S

S

S

S

S

S

S

S

S

T

T

V

X

994 RECTOR ET AL.

Mayhew, J. E., Askew, S., Zheng, Y., Porrill, J., Westby, G. W.,Redgrave, P., Rector, D. M., and Harper, R. M. 1996. Cerebralvasomotion: A 0.1-Hz oscillation in reflected light imaging of neu-ral activity. NeuroImage 4: 183–193.

Mifflin, S. W. 1993. Laryngeal afferent inputs to the nucleus of thesolitary tract. Am. J. Physiol. 265: R269–R276.

Mifflin, S. W., Spyer, K. M., and Withington-Wray, D. J. 1988.Baroreceptor inputs to the nucleus tractus solitarius in the cat:Postsynaptic actions and the influence of respiration. J. Physiol.399: 349–367.

Orbach, H. S., Cohen, L. B., and Grinvald, A. 1985. Optical mappingof electrical activity in rat somatosensory and visual cortex.J. Neurosci. 5: 1886–1895.

Paxinos, G., and Watson, C. 1986. The Rat Brain in StereotaxicCoordinates, 2nd ed. Academic Press, New York.

Poe, G. R., Nitz, D. A., Rector, D. M., Kristensen, M. P., and Harper,R. M. 1996. Concurrent reflectance imaging and microdialysis inthe freely behaving cat. J. Neurosci. Methods 65: 143–149.

Prakash, N., Vanderhaeghen, P., Cohen-Cory, S., Frisen, J., Flana-gan, J. G., and Frostig, R. D. 2000. Malformation of the functionalorganization of somatosensory cortex in adult ephrin-A5 knock-outmice revealed by in vivo functional imaging. J. Neurosci. 20: 5841–5847.ector, D. M., and Harper, R. M. 1991. Imaging of hippocampalneural activity in freely behaving animals. Behav. Brain Res. 42:143–149.ector, D. M., Poe, G. R., Kristensen, M. P., and Harper, R. M. 1997a.Light scattering changes follow evoked potentials from hippocam-pal Schaeffer collateral stimulation. J. Neurophysiol. 78: 1707–1713.ector, D. M., Poe, G. R., Redgrave, P., and Harper, R. M. 1997b. Aminiature CCD video camera for high-sensitivity light measure-ments in freely behaving animals. J. Neurosci. Methods 78: 85–91.ector, D. M., Rogers, R. F., and George, J. S. 1999. A focusing imageprobe for assessing neural activity in-vivo. J. Neurosci. Methods91: 135–145.ector, D. M., Richard, C. A., Staba, R. J., and Harper, R. M. 2000.Sleep states alter ventral medullary surface responses to bloodpressure challenges. Am. J. Physiol. Regul. Integr. Comp. Physiol.278: R1090–1098.ogers, R. F., Paton, J. F. R., and Schwaber, J. S. 1993. NTS neuro-nal responses to arterial pressure and pressure changes in the rat.Am. J. Physiol. 265: R1355–R1368.ogers, R. F., Rose, W. C., and Schwaber, J. S. 1996. Simultaneousencoding of carotid sinus pressure and dP/dt by NTS target neu-rons of myelinated baroreceptors. J. Neurophysiol. 76: 2644–2660.

ousche, P. J., Petersen, R. S., Battiston, S., Giannotta, S., andDiamond, M. E. 1999. Examination of the spatial and temporaldistribution of sensory cortical activity using a 100-electrode ar-ray. J. Neurosci. Methods 90: 57–66.

alzberg, B. M., Obaid, A. L., and Gainer, H. 1985. Large and rapidchanges in light scattering accompany secretion by nerve termi-nals in the mammalian neurohypophysis. J. Gen. Physiol. 86:395–411.

andman, C. A., O’Halloran, J. P., and Isenhart, R. 1984. Is there anevoked vascular response? Science 224: 1355–1357.

cheuer, D. A., Zhang, J., Toney, G. M., and Mifflin, S. W. 1996.Temporal processing of aortic nerve evoked activity in the nucleusof solitary tract. J. Neurophysiol. 76: 3750–3757.

eagard, J. L., Dean, C., and Hopp, F. A. 1995. Discharge patterns ofbaroreceptor-modulated neurons in the nucleus tractus solitarius.Neurosci. Lett. 191: 13–18.

hoham, D., Glaser, D. E., Arieli, A., Kenet, T., Wijnbergen, C.,Toledo, Y., Hildesheim, R., and Grinvald, A. 1999. Imaging corticaldynamics at high spatial and temporal resolution with novel bluevoltage-sensitive dyes. Neuron 24: 791–802.

inger, W., and Gray, C. M. 1995. Visual feature integration and thetemporal correlation hypothesis. Annu. Rev. Neurosci. 18: 555–586.

teinbrink, J., Kohl, M., Obrig, H., Curio, G., Syre, F., Thomas, F.,Wabnitz, H., Rinneberg, H., and Villringer, A. 2000. Somatosen-sory evoked fast optical intensity changes detected non-invasivelyin the adult human head. Neurosci. Lett. 291: 105–108.

tepnoski, R. A., LaPorta, A., Raccuia-Behling, F., Blonder, G. E.,Slusher, R. E., and Kleinfeld, D. 1991. Noninvasive detection ofchanges in membrane potential in cultured neurons by light scat-tering. Proc. Natl. Acad. Sci. USA 88: 9382–9386.

voboda, K., Helmchen, F., Denk, W., and Tank, D. W. 1999. Spreadof dendritic excitation in layer 2/3 pyramidal neurons in rat barrelcortex in vivo. Nat. Neurosci. 2: 65–73.

asaki, I., and Byrne, P. M. 1992. Rapid structural changes in nervefibers evoked by electrical current pulses. Biochem. Biophys. Res.Commun. 188: 559–564.

sodyks, M., Kenet, T., Grinvald, A., and Arieli, A. 1999. Linkingspontaneous activity of single cortical neurons and the underlyingfunctional architecture. Science 286: 1943–1946.an Harreveld, A. 1958. Changes in the diameter of apical dendritesduring spreading depression. Am. J. Physiol. 192: 457–463.

u, C., Zipfel, W., Shear, J. B., Williams, R. M., and Webb, W. W.1996. Multiphoton fluorescence excitation: New spectral windowsfor biological nonlinear microscopy. Proc. Natl. Acad. Sci. USA 93:10763–10768.

INTRODUCTIONMATERIALS AND METHODSFIG. 1FIG. 2

RESULTSFIG. 3FIG. 6FIG. 4FIG. 5FIG. 7FIG. 8

DISCUSSIONTABLE 1FIG. 9FIG. 10FIG. 11FIG. 12

CONCLUSIONSACKNOWLEDGMENTSREFERENCES