Spatiotemporal precision and hemodynamicmechanism of optical point spreads in alert primatesYevgeniy B. Sirotina, Elizabeth M. C. Hillmanb,c, Clemence Bordierb, and Aniruddha Dasa,b,d,e,1

Departments of aNeuroscience, bBiomedical Engineering, cRadiology, and dPsychiatry, Columbia University, New York, NY 10027; and eNew York StatePsychiatric Institute, 1051 Riverside Drive, Unit 87, New York, NY 10032

Edited by Leslie G. Ungerleider, National Institutes of Health, Bethesda, MD, and approved September 4, 2009 (received for review May 18, 2009)

In functional brain imaging there is controversy over which hemo-dynamic signal best represents neural activity. Intrinsic signal opticalimaging (ISOI) suggests that the best signal is the early darkeningobserved at wavelengths absorbed preferentially by deoxyhemoglo-bin (HbR). It is assumed that this darkening or “initial dip” reports localconversion of oxyhemoglobin (HbO) to HbR, i.e., oxygen consump-tion caused by local neural activity, thus giving the most specificmeasure of such activity. The blood volume signal, by contrast, isbelieved to be more delayed and less specific. Here, we used multi-wavelength ISOI to simultaneously map oxygenation and bloodvolume [i.e., total hemoglobin (HbT)] in primary visual cortex (V1) ofthe alert macaque. We found that the hemodynamic ‘‘point spread,’’i.e., impulse response to a minimal visual stimulus, was as rapid andretinotopically specific when imaged by using blood volume as whenusing the initial dip. Quantitative separation of the imaged signal intoHbR, HbO, and HbT showed, moreover, that the initial dip wasdominated by a fast local increase in HbT, with no increase in HbR. Wefound only a delayed HbR decrease that was broader in retinotopicspread than HbO or HbT. Further, we show that the multiphasic timecourse of typical ISOI signals and the strength of the initial dip mayreflect the temporal interplay of monophasic HbO, HbR, and HbTsignals. Characterizing the hemodynamic response is important forunderstanding neurovascular coupling and elucidating the physio-logical basis of imaging techniques such as fMRI.

fMRI � imaging � macaque � visual � neurovascular coupling

Cerebral hemodynamics respond quickly and specifically to localneural activity (1, 2). Hemodynamic signals are thus used

extensively as proxies for such activity in functional neuroimagingtechniques like fMRI and intrinsic signal optical imaging (ISOI).There has been considerable debate, however, as to which of thepossible hemodynamic signals, e.g., changes in local blood oxygen-ation, volume, or flow, with their distinct response properties,constitutes the ‘‘best’’ signal for inferring neural activity (1, 3–5).

The debate regarding the best signal sharpened with reports ofinitial dips in both ISOI and fMRI signals. ISOI studies consistentlyfound a brief stimulus-evoked darkening followed by a strongbrightening at imaging wavelengths preferentially absorbed indeoxyhemoglobin (HbR) [e.g., at 605 nm (6)]. The darkening, latertermed the initial dip*, was interpreted as a local conversion ofoxyhemoglobin (HbO) to HbR caused by increased oxygen con-sumption by local neurons before any active vascular response (1,7) The subsequent brightening was taken to measure the ‘‘rebound’’in [HbO]† caused by a delayed, stimulus-triggered increase incerebral blood flow (7).

In parallel, some fMRI studies reported finding an initial dip beforethe rise in the blood oxygen level-dependent (BOLD) signal (8).Because theBOLDsignalmeasureschanges in [HbR]alone‡, the fMRIinitial dip was seen as clear evidence for an initial increase in [HbR]. Itwas equated to the ISOI dip (8) and was argued to be the best markerof neural activity, free of the venous artifacts seen in positive BOLD (8).By extension, positive BOLD was equated to the ISOI rebound (9). Akey implication of these studies was the presence of a significant delaybetween the start of neural activity-triggered oxygen consumption,believed to underlie the initial dip, and the subsequent active vascular

responsegenerating thereboundinfMRIorISOI(4,7,9,10).TheISOIstudies also suggested that the rebound was more spatially diffuse andless stimulus-specific than the initial dip (4, 7, 10). The initial dip thusbecame the signal of choice in a number of functional imaging studies,not only in ISOI (11–15) but also in fMRI (16).

There are a number of unresolved issues, however, questioningthe interpretation of the ISOI initial dip as local deoxygenationpreceding the active vascular response. First, the initial dip, whilerobust and reliable in ISOI, has proved to be elusive in fMRI (3, 9).Next, oxygen consumption preceding increased blood flow shouldreduce [HbO] in addition to increasing [HbR]. However, evenstudies proposing an initial deoxygenation failed to see the expected[HbO] decrease (7). Finally, a number of ISOI studies in the rodentthat quantitatively separated the imaging signal into [HbR] and[HbO] either failed to see any transient [HbR] increase (17–20) orfound a simultaneous rise in [HbO] consistent with a stimulus-evoked increase in total hemoglobin [HbT] (21, 22).

The ISOI dip has been successfully used to map cortical columns(e.g., orientation or ocular dominance) where activated columnsdarken more than inactive ones. However, despite only a transientinitial dip in the global signal, the mapping signal (i.e., the additionaldarkening in active columns) remains throughout the entire dura-tion of the hemodynamic response despite the rebound observed inthe global signal. The mapping signal is typically interpreted as amaintained increase in [HbR] in the activated columns, persistingthrough the presumed spatially diffuse rebound of oxygenatedblood (7, 23). Such a maintained [HbR] increase demands asustained elevation of local oxygen consumption for many seconds(�10 s) after cessation of neural activity (23). This interpretation isnot supported by fMRI. The BOLD mapping signal shows a tunedincrease in activated columns, indicating a spatially specific de-crease of [HbR], during the rebound (24). The observed orienta-tion-tuned decrease in [HbR] suggests that the prolonged mappingsignal is not caused by deoxygenation and, by extension, also raisesdoubts about the standard interpretation of the initial dip.

These conundrums might be resolved by considering alternateexplanations, other than deoxygenation, for the darkening seen attypical imaging wavelengths (e.g., 605 nm). Darkening at thesewavelengths may result from a rise in local blood volume, increasing

Author contributions: Y.B.S., E.M.C.H., and A.D. designed research; Y.B.S., E.M.C.H., andA.D. performed research; Y.B.S., E.M.C.H., and C.B. contributed new reagents/analytictools; Y.B.S., E.M.C.H., and A.D. analyzed data; and Y.B.S. and A.D. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

1To whom correspondence should be addressed. E-mail: ad2069@columbia.edu.

*The term initial dip has been used in different senses in the literature. Here, we use it torefer to the observed phenomenon of a dip, i.e. the observed initial darkening in ISOI (e.g.,ref. 30) or transiently reduced BOLD signal in fMRI (e.g., ref. 8). We are not using the termto refer to the calculated increase in [HbR] (e.g., ref. 3).

†The square brackets [] indicate concentration per volume of imaged cortical tissue. Thismeasure combines the concentration of a species in blood with the fraction of the corticalpixel occupied by blood vessels. This measure of concentration is similar to the measure ofconcentration per voxel in fMRI.

‡Unlike in fMRI BOLD, both [HbR] and [HbO] contribute to ISOI absorption at all wavelengths.

This article contains supporting information online at www.pnas.org/cgi/content/full/0905509106/DCSupplemental.

18390–18395 � PNAS � October 27, 2009 � vol. 106 � no. 43 www.pnas.org�cgi�doi�10.1073�pnas.0905509106

Dow

nloa

ded

by g

uest

on

Aug

ust 1

9, 2

020

absorption by both HbO and HbR, or even an increase in [HbO]alone. Most hemoglobin in cerebral blood is highly oxygenated[saturation between 95% and 60% (25)] and HbO is a significantabsorber at all imaging wavelengths. Even small fractional increasesin [HbO] could yield increased absorption. The current study wasdesigned to test these distinct interpretations of the ISOI signalswith a focus on the initial dip.

ResultsWe analyzed stimulus-evoked hemodynamic signals in the brain byusing dual-wavelength ISOI in primary visual cortex (V1) of thealert behaving macaque (ref. 26; two animals V and S). With thistechnique we imaged the hemodynamic point spread, i.e., theimpulse response to a minimal visual stimulus (a 1⁄4 deg line) flashed(1 s) while the animal held its gaze steady in a fixation task (4-sfixation periods, trial length 10 s). Stimulated trials alternated withnonstimulated trials (giving 20 s between stimuli for hemodynamicresponse to return to baseline). The imaging wavelengths werechosen pairwise from three light-emitting diode (LED) sources(Fig. 1A): a ‘‘volumetric’’ LED (green, centered at 530 nm) that wasabsorbed equally in HbO and HbR giving a measure of [HbT] orblood volume§; and ‘‘oximetric’’ LEDs (orange: center, 605 nm; red,center 630 nm), absorbed more strongly in HbR than HbO, and thusbeing sensitive to blood oxygenation changes. Consistent with

earlier findings, the signals at oximetric wavelengths showed arobust initial dip followed by a rebound and then a delayedundershoot (secondary darkening). Notably, however, volumetricsignals started at the same time as the initial dip and showed similarspatial specificity.

To get a quantitative measure of the responses as blood oxygen-ation and volume we used the absorption spectra of HbR and HbOto spectrally decompose the measured imaging signals into [HbO],[HbR], and [HbT]. Notably, we used the Beer-Lambert Lawmodified to account for the wavelength-dependent optical pathlengths in cortical tissue at different imaging wavelengths (seeMethods and refs. 17 and 27). Path length correction, althoughmissing from prior work in monkeys and cats, is necessary to avoiderrors in the spectral decomposition and interpretation of ISOIsignals (17, 27) This quantitative analysis showed that the earlieststimulus-evoked response was driven by a rapid increase in [HbT]with no change in [HbR]. The full multiphasic time course of typicalimaging signals, including the initial dip, rebound, and undershoot,could be well explained by a fast increase in [HbT] preceding a moretransient decrease in [HbR].

Spatial and Temporal Properties of Absorption Point Spreads Mea-sured at Different Wavelengths. The point spread at oximetricwavelengths (605 and 630 nm) had a triphasic time course with an

§HbT is equivalent to cerebral blood volume (CBV) as measured in fMRI for fixed hematocrit.Because CBV in MR studies often measures plasma volume (e.g. via a contrast agent) the twodiffer if the hematocrit changes.

-0.5

0

0.5

1

Nor

mal

ized

Res

pons

e LED 605 nmLED 630 nmLED 530 nm

A

D E F

B

Nor

mal

ized

Res

pons

e

0

1

'Oximetric''Volumetric'

0 2.5 5

0.5

0

1

0 2.5 5

0.5

Radius (mm)

0

1

0 2.5 5

0.5

-0.03

-0.02

-0.01

0

0.01

dR/R

1.1

0.0

75%

HWC

'Initial dip' 'Rebound' 'Undershoot'

Tick spacing: 2 sec

'Initial dip'

'Rebound'

'Undershoot'green = fixatered = stimulus

Fig. 2. Imaged point spreads at oximetric and volumetric wavelengths suggesta blood volume contribution to the initial dip. (A) Temporal profiles of signalchanges measured at the point spread center, for the three LEDs used. Note thedistinct triphasic signal at 605 and 630 nm vs. the monophasic signal at 530 nmaveraged over all experiments. Error bars show the SEM. Dotted vertical lineindicates stimulus onset. Arrowheads indicate time points used in the analysis ofspatial profiles in C–E for the initial dip, rebound, and undershoot. (B) Same dataas in A, but scaled to maximum for each session before averaging. Note the muchlarger initial dip and undershoot observed with 605 nm than with 630 nm. Alsonote that scaling the image makes it appear artifactually as if the initial dip at 605and 630 nm starts earlier than changes at 530 nm. (C) Time courses of 75%half-widths for oximetric (red) and volumetric (green) point spreads. Time axis isaligned to A. (D–F) Radial profiles normalized to the center point, for oximetric(red, combining 605 and 630 nm) and volumetric (green, 530 nm) point spreadsduring the three response phases calculated by averaging the signal on radialannuli after masking off the vasculature. (D) Profiles early into the initial dip (t �0.8 s) are shown. Note the similar spatial extent of both signals. Dashed linesindicate 75% response level. (E) Same as D, but for the rebound (t � 3 s). Note themuch wider oximetric signal with an extended and rounded peak (red) vs. thesharp and narrow profile of the volumetric image (green). (F) Same as D, but forthe undershoot (t � 9 s). Note the similarity of point spreads across wavelengthsduring the initial dip and undershoot.

A

4e-3

-4e-3

'Rebound(2.0-4.0 sec)

dR/R

1e-3

-1e-3

'Undershoot(7.3-8.7 sec)

3.0e-2

-3.0e-2LED

530

nm 2e-3

-2e-3

LED

630n

m

5e-4

-5e-4

'Initial dip'(0.7-1.3 sec)

7e-3

-7e-3

B

STR_03_26_07

Ext

inct

ion

Coe

ffici

ent (

cm-1

/M)

HbHbO

LED 530nmLED 605nmLED 630nm

0

1

2

3

4

5

6 x 104

500 550 600 650Wavelength (nm)

LED

Intensity (AU)

0

1

dR/R

5mm

Fig. 1. Optical imaging wavelengths and corresponding pointspreads. (A) Invitro absorbance spectra for HbR and HbO (molar extinction coefficients) alongwith normalized emission spectra for the three LEDs used. The full LED spectrawere used to compute effective absorptions and path-lengths for the threeimaging wavelengths (see Methods). (B) Imaged point spreads in one represen-tative session. (Upper) Red: 630-nm (oximetric) LED. (Lower) Green: 530-nm(volumetric) LED. Image frames were averaged over the times in parentheses tocapture the peaks of the initial dip, rebound, etc. of the oximetric signal. Notedifferent scale bars to the right of each image.

Sirotin et al. PNAS � October 27, 2009 � vol. 106 � no. 43 � 18391

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

Aug

ust 1

9, 2

020

initial darkening [i.e., increased absorption; Figs. 1B and 2A: initialdip; 605 nm: peak at 1.5 s, dR/R � 0.31 (0.05)%; 630 nm: peak at1.3 s, dR/R � 0.05 (0.03)%¶], a subsequent larger decrease inabsorption [rebound; 605 nm: peak at 3.7 s, dR/R � 0.69 (0.09)%;630 nm: peak at 3.6 s, dR/R � 0.49 (0.05)%], and finally a smallsecondary increase in absorption [undershoot; 605 nm: peak at 8.0 s,dR/R � 0.15 (0.04)%; 630 nm: peak at 8.8 s, dR/R � 0.04 (0.02)%].Volumetric responses at 530 nm, by contrast, were large andmonophasic [peak at 2.9 s, dR/R � 3.3 (0.2) %]. The spatialproperties of these signals are summarized in Table 1. For com-parison, the 75% half-width of the point spread measured in V1using voltage-sensitive dyes was �0.6 mm (estimated from ref. 28).

The interpretation of the initial dip as a focal deoxygenationpreceding a delayed and spatially diffuse active vascular responseleads to specific predictions when comparing oximetric and volu-metric signals. (i) The oximetric point spread should appear first.Any volumetric response should appear considerably later, wellafter the onset of the initial dip. (ii) This volumetric response shouldbe distributed more diffusely over cortex than the initial dip. (iii)During the oximetric rebound (which is presumably dominated bythe blood volume response), the oximetric and volumetric re-sponses should have comparable spatial extents. To test thesepredictions, we compared the volumetric and oximetric pointspreads at the three phases of the oximetry signal. Our results failedto bear out any of the above predictions.

Darkening at the Initial Dip Is Unlikely to Measure a Selective FocalIncrease of [HbR]. Although the initial dip at both 630 and 605 nmwas reliable, it started no earlier than the volumetric (530 nm)response. Quantifying response onsets for each wavelength as thefirst time point where the response showed a statistically significantinflection in slope (Fig. S1) revealed that onset times were essen-tially identical for all three wavelengths (0.4 s at 530 nm; 0.4 s at 605nm; 0.5 s at 630 nm). Thus, there was no delay in the blood volumeincrease relative to the initial dip. We noted that the ratio of theinitial dip to the subsequent rebound was reliably higher for 605than for 630 nm (Fig. 2B), a point that we return to later.

We found no difference in the spatial extent of signals atoximetric and volumetric wavelengths during the initial dip. Evenat its earliest clearly measurable time point (0.8 s), the initial dippoint spread was identical to that of the volumetric point spread[Fig. 2 C and D; 75% half-width � 0.5 mm (SEM � 0.2 mm)�

(volumetric); 0.48 (0.08) mm (oximetric), P � 0.8, n � 11]. Thissurprising result shows that the initial dip is no more focal thanblood volume and suggests that the two may share a commonunderlying mechanism.

Volumetric Signal Remains Spatially Focused, Unlike the Diffuse andDelayed Oximetric Rebound. Both the volumetric and oximetricresponses broadened during the rebound. However, the spatialspread at the center of the volumetric response remained largely

unchanged, maintaining a sharp profile. At its peak (3 s afterstimulus onset), the 75% half-width of the volumetric response grewby only 13% [half-width: from 0.5 (0.2) to 0.60 (0.05) mm] despitea 13-fold increase in amplitude [from 0.26 (0.09) % to 3.36 (0.16)%] (Table 1 and Fig. 2 A, C, and E). By contrast, the spatial spreadat the center of the oximetric response broadened significantlyduring the rebound, taking on a distinctly rounded profile [75%half-width grew from 0.48 (0.08) to 1.15 (0.08) mm, an increase ofa factor of 2.4; P � 9e-6, n � 11]. It is noteworthy that thecross-sectional shape and spatial extent of the volumetric pointspread closely matches those observed with voltage sensitive dyes(28, 29). This finding suggests that the volumetric response, is notonly distinct from the oximetric rebound, but is likely a betterneuroimaging signal tightly coupled to underlying neural activity.

Late Undershoot at the Oximetric Wavelengths Matches the SpatialExtent of the Volumetric Signal. At oximetric wavelengths we oftendetected a small increase in absorption after the rebound that wasspatially localized [half-width at 75% � 0.6 (0.2) mm; Table 1 andFig. 2F]. Other investigators have shown this ‘‘late undershoot’’emerging many seconds after the cessation of neural activity andhave suggested that it may indicate a secondary increase in [HbR](30). This late oximetric signal matched spatially the long-livedvolumetric signal [volumetric half-width � 0.57 (0.05) mm], sug-gesting that both could reflect lingering blood volume.

Spectral Decomposition. The spectrally decomposed [HbO], [HbR],and [HbT] signals showed similar spatially localized responses, eachmonophasic in time (Table 1 and Fig. 3). To relate absorptionchanges at oximetric wavelengths during the initial dip, rebound,and undershoot to these analyzed hemodynamic components weexamined the pointspreads in HbR and HbT separately duringthese phases.

Initial Dip Is Dominated by HbT. Surprisingly, we saw no significantchange in [HbR] during the initial dip (0.8 s after stimulus, P � 0.9,n � 11; Fig. 3 A and B and Fig. S2). Unlike the darkening atoximetric wavelengths, [HbR] at this time point had no reliablespatial structure. Instead, the response was entirely dominated byan increase in [HbT] (P � 0.02; Fig. 3 B–D), whereas [HbR] showedonly a delayed monophasic decrease. The [HbT] point spread wasidentical to the volumetric point spread (Figs. 1 and 2 and Table 1).Our data suggest that the initial dip at oximetric wavelengths isindeed a result of an early and spatially localized increase in [HbT] andnot an increase in [HbR]. These results were independent of thewavelength pairs used for imaging (Figs. S2 and S3) or the modelparameters used for decomposition (see Methods and Fig. S4).

Observed Oximetric Rebound Is Likely Shaped by an InteractionBetween HbT and HbR. We found that the [HbT] signal during therebound had a spatial spread that was consistently, albeit slightlynarrower than the concomitant decrease in [HbR] {Table 1 and Fig.3E arrowhead: [HbR] width � 0.80 (0.03); [HbT]: 0.61 (0.05); P �0.0017} (31, 32). Notably, both the [HbR] and [HbT] profiles were

¶All data cited are averaged across 11 experiments in two animals.

�Half-widths measured at the 75% assess the focal spatial resolution of the point spread.

Table 1. Half-widths of the hemodynamic response point-spreads over V1

Time Oximetric Volumetric HbR HbO HbT

75% half-width (mm)0.8 s (dip) 0.48 (0.08) 0.5 (0.2) – 0.5 (0.2) 0.5 (0.1)3.0 s (rebound) 1.15 (0.08) 0.60 (0.05) 0.80 (0.03) 0.66 (0.04) 0.61 (0.05)9.0 s (undershoot) 0.6 (0.2) 0.57 (0.05) 0.59 (0.03) 0.58 (0.04) 0.58 (0.04)

50% half-width (mm)0.8 s (dip) 1.0 (0.4) 0.73 (0.08) – 0.7 (0.1) 0.74 (0.09)3.0 s (rebound) 1.86 (0.09) 1.35 (0.08) 1.53 (0.07) 1.41 (0.07) 1.37 (0.07)9.0 s (undershoot) 0.7 (0.2) 0.88 (0.05) 0.94 (0.05) 0.91 (0.04) 0.90 (0.04)

18392 � www.pnas.org�cgi�doi�10.1073�pnas.0905509106 Sirotin et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

9, 2

020

sharply peaked at the center and much narrower than the rawoximetric rebound [half-width � 1.15 (0.08) mm; P � 0.0007; Table1 and Fig. 2E]. Thus, neither the [HbT] nor the [HbR] signal byitself can account for the width and the diffuse peak of the oximetricrebound.

We hypothesize that the oximetric rebound reflects the differ-ence between the absorption increase from increasing [HbT] andthe somewhat broader absorption decrease from decreasing [HbR].A simple model of such a difference image, using computed profilesfor [HbR] and [HbT], plausibly reproduces the shape of therebound (Fig. S5).

Late Undershoot Is a Result of Residual HbT. Our analysis alsoconfirmed that the late darkening is dominated by slowly decaying[HbT], persisting for 15 s, rather than a late increase in [HbR].

Predicting the Imaging Signal Time Course Across Wavelengths. Ourcalculated hemodynamic time courses, although obtained withthree imaging wavelengths, correctly predict ISOI time coursesobserved across the full spectrum (Fig. 4). These predictions weremade by using the calculated early increase in [HbT] (increaseddarkening) and delayed decrease in [HbR] (decreased darkening).Signals at isosbestic wavelengths result from a rapid, monophasicdarkening matching the [HbT]. For wavelengths away from theisosbestic point, with progressively stronger absorption by HbR vs.HbO, the delayed [HbR] decrease (Fig. 3B) becomes progressivelymore prominent. This delayed [HbR] decrease leads to a delayedreduction (e.g., at 592 nm) or reversal of the initial HbT-linkeddarkening (e.g., at �600–720 nm where the predicted signaldisplays the familiar initial dip and rebound). These predicted timecourses match published results (e.g., figure 2 of ref. 19). In

5mm

'Rebound'(2.0-4.0 sec)

35e-7

-35e-7

'Undershoot'(7.3-8.7 sec)

[dHbR

]

3e-7

-3e-7

8e-7

-8e-7

HbT

8e-7

-8e-7

HbR

'Initial Dip'(0.7-1.3 sec)

5e-7

-5e-7

15e-7

-15e-7

[dHbT]

A

STR_03_26_07

Nor

mal

ized

Res

pons

e

B

Nor

mal

ized

Res

pons

e

0 2.5 5

0

1

0 2.5 5

0.75

0

1

0 2.5 5

0.75

Radius (mm)

D E F

C

-5e-6

0

5e-6

10e-6

15e-6

Avg.

Con

c. C

hang

e

-0.2

0

0.2

0.4

0.6

0.8

1 HbRHbOHbT

-1-10123

0 1 2

x 10

0

2e-7

4e-7

Avg.

Con

c. C

hang

e

-6

'Initial dip' 'Rebound' 'Undershoot'

Tick spacing: 2 sec

Fig. 3. Spectral decomposition of imaging signal shows no increase in HbR during the initial dip period as defined for the oximetric signal. (A) Spectrally decomposedpoint spreads: same experiment as in Fig. 1B. Conventions are as in Figs. 1 and 2. Only the images for HbR and HbT are shown. (B) The average temporal profile ofconcentration changes at response center after visual stimulation. Spectral decomposition was done separately for each experiment and then averaged. Conventionsare as in Fig. 2A. Note the lack of any detectable increase in HbR despite clear initial dip signals in Fig. 2. (Inset) Magnified view of concentration changes during earlyportion of response, 1–2 s after stimulus. Note the early increase in HbT and the lack of any increase (only a late decrease) in HbR. (C) Same as B, but normalized tomaximal response (HbR signal is thus flipped in sign). Note HbR signal lagging behind HbT at onset but decaying faster toward baseline. (D–F) The radial profiles ofconcentration changes for HbR, HbO, and HbT during the three oximetric response phases (as in Fig. 2). (D) Concentration changes early into the initial dip (t � 0.8 s).Unnormalized because there was no reliable change in HbR at this time point (see A and B). Compare with Fig. 2D. (E) Same as D, but for the rebound phase and scaledto maximum (t � 3 s). Note HbR signal is slightly broader than HbT. (F) Same as E, but for the undershoot phase (t � 9 s). Note similar point spreads in all components.

Sirotin et al. PNAS � October 27, 2009 � vol. 106 � no. 43 � 18393

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

Aug

ust 1

9, 2

020

particular, wavelengths with stronger absorption in HbR vs. HbOshow stronger and more prolonged rebounds after weaker andmore transient initial dips (Fig. 4 B and C) matching our own (Fig.2B) and previous observations (30) of stronger dips closer to theisosbestic point. By contrast, a model of the initial dip as deoxy-genation failed to account for these observations, yielding similarlysized dips across wavelengths, and delayed darkening near isos-bestic points (Fig. S6).

Our findings suggest that the multiphasic temporal profile of theISOI signals at typical oximetric wavelengths, including the mag-nitude and duration of the initial dip, rebound, and undershoot,reflect the interplay of absorptions by HbT and HbR and areproblematic as estimates of neural oxygen metabolism.

DiscussionWith ISOI in alert macaque we observed the expected, robustinitial dip. Our interpretation of this dip is very different,however, from earlier interpretations in macaque or cat. Thelikely reason for this difference is that we separated our imagingsignal into [HbR] and [HbO] by using a model that incorporateschanges in optical path length caused by the large changes inabsorptivity across wavelengths (27). The earlier results (7, 23)were obtained with fixed path lengths, which, as now widelyacknowledged (4, 10), can lead to significant errors in spectro-scopic analysis (17, 27). Our results are robust over a broad rangeof model parameters when using wavelength-dependent pathlengths (Fig. S4). However, control analyses of our imaging dataassuming fixed path lengths resulted in transient initial increasesin [HbR] similar to the earlier published results.

Although we applied these analysis techniques to ISOI in themacaque, our results are consistent with a large body of workpioneering similar analysis in rodents [including alert rodents (22)]and alert humans (33). Those studies arrived at conclusions similarto ours: that the primary response after stimulus onset is a rapid androbust increase in [HbT]. Those groups either failed to see anyinitial increase in [HbR] (e.g., refs. 18–20) or found only a modestincrease accompanied by a large rise in [HbO], consistent with a netincrease in [HbT] (e.g., refs. 21, 22, and 34). Discrepancies betweenrodent and earlier monkey results had been attributed to species

differences (35, 36). By contrast, our findings add to a growing bodyof evidence suggesting that neurovascular coupling mechanismsmay be conserved across species.

Independent of modeling, we see that the volumetric (530 nm)signal is as fast as the oximetric (605 or 630 nm) initial dip,indicating no delay in the rise of [HbT]. We identified signal onsettimes by using a quantitative statistical measure. Our findings areconsistent with those from fMRI (35) and rat ISOI (18, 19). Priorwork in cats and monkeys suggested that changes in HbT aredelayed, but that delay could be an artifact of signal normalizationand appears to be contradicted by other data from the same group(e.g., figure 3 a and b in ref. 7). Our results argue against anysignificant delays between the start of oxygen extraction from bloodand the initiation of the active vascular response (4, 7).

It is important to note that the lack of increased [HbR] does notpreclude increased oxygen consumption or decreased tissue pO2caused by neural activity (19). Stimulation leads to a rapid (�400ms) increase in local cerebral blood flow (18, 19, 37), increasing[HbO] in local blood vessels (38). This increase in flow couldsupport increased oxygen consumption and a decrease of tissuepO2 even as the local blood supply gets enriched in [HbO] anddepleted of [HbR] (19). Our findings highlight the complexity of therelationship between measures of blood volume, flow, and oxygen-ation with local metabolic demand or neural activity.

Our observations could shed light on the elusive nature of thefMRI initial dip compared with the ISOI. Unlike in ISOI, a dip inBOLD equals increased [HbR]. However, a large number of carefulfMRI investigations have either failed to see any initial dip (e.g.,refs. 24 and 35), found it unreliable (39), or obtained it only afterunnaturally intense stimulation (40). Our work together with thelarge body of work in rodents suggests an explanation: the hemo-dynamic response during the ISOI dip (unlike in fMRI) is driven bya rapid active vascular response, dominated by increases in [HbT],and only occasionally accompanied by small increases in [HbR].

Our results suggest that blood volume ([HbT]) is a more reliablesignal to use for neuroimaging than the initial dip. The initial diplikely carries no privileged information about neural activity ormetabolic demand. The [HbT] signal is as rapid and spatiallyfocused and more than an order of magnitude stronger and

10-2

10-1

100

101

102

Dip

Size

/ R

ebou

ndS

ize

Wavelength (nm)400 450 500 550 600 650 700 750 800 850

BA

DC

-6 -4 -2 0 2 4 60

1

Radial Distance (mm)

-0.03

-0.02

-0.01

0

0.01

Pre

dict

ed d

R/R

-0.04

584 nm592 nm600 nm608 nm

Pre

dict

ed d

R/R

0.5

.01

-.04

dR/R

102

103

104

105

106

Ext

inct

ion

Coe

ffici

ent

(cm

-1/M

)

HbRHbO

400 450 500 550 600 650 700 750 800 850

584nm

600nm

592nm

608nm

Wavelength (nm) Tick spacing: 2 sec

Fig. 4. Changes in blood volume predict size of initial dip. (A) Molar extinction coefficient for HbR and HbO. Vertical dashed lines indicate isosbestic points. Shadedregions indicate: ranges where HbR absorbs more strongly than HbO. (B) Predicted response time courses at wavelengths moving from isosbestic to oxymetric markedin A, showing progression from a monophasic signal at the isosbestic 584 nm to a biphasic signal at 600 nm. (Note: The predicted signal time courses here are for thepure spectral wavelengths indicated; while qualitatively similar to our measured imaging signals they are quantitatively different because our LED sources had finitebandwidths giving the corresponding admixture of responses. All quantitative calculations using our LED sources accounted for this finite bandwidth; see Methods).(C) The predicted ratio of the size of the initial dip relative to the rebound across wavelength. Note an explosive increase in this ratio toward isosbestic points. (D)Predicted spatial profiles at 3 s poststimulus onset normalized to maximum response amplitude. Wavelengths same as in B. Note broader spreads at wavelengthsshowing a rebound. (Inset) Not normalized for amplitude.

18394 � www.pnas.org�cgi�doi�10.1073�pnas.0905509106 Sirotin et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

9, 2

020

longer-lived. For ISOI maps, e.g., of orientation, the initial dip hasbeen reported to give cleaner maps than the [HbT] signal, i.e., withfewer vascular artifacts (1, 10). However, these cleaner maps mayresult from spatial filtering and the fortuitous cancellation ofcommon vascular patterns by the difference signal that comprisesthe initial dip (increasing [HbT] with decreasing [HbR]).

An important open question is the relation of these varioushemodynamic signals to underlying neural activity. The [HbT]signal was earlier believed to be diffuse, nonspecific, and dominatedby vascular artifacts (10). Our point spreads imaged using thevolumetric 530 nm (or the computed [HbT], [HbO], or [HbR]) aresharply peaked and remarkably similar in shape and extent to thoseobtained with voltage sensitive dyes (28, 29). The [HbT] responseis thus likely a tight neurovascular signal that faithfully reports thespatial pattern of spiking plus subthreshold neural activity. Thislikely close link between [HbT] signals and neural activity is alsoconsistent with work in anesthetized rodent barrel cortex showingthat neural activity correlates well with [HbT] or [HbO] signals butnot with the initial dip (5).

The correct interpretation of stimulus-driven hemodynamic signals iscritical to modern neuroimaging techniques such as fMRI. The properseparation of the imaging signal into its hemodynamic components is aprerequisite for linking it to underlying neural activity.

MethodsResults were obtained by using continuous, multiwavelength ISOI in alertmacaque V1. Two animals (V and S) were trained on a periodic visual fixationtask. Visual stimuli, consisting of 0.25° light bars against a gray background,were flashed on passively for 1 s while the animal held fixation. Standard

alert-monkey optical imaging techniques (23) were used to record the intrinsiccortical signal, continuously, through a clear artificial dura and glass-frontedrecording chamber mounted over the animals’ V1. The brain surface wasilluminated by using two LED arrays chosen from three sets with wavelengthscentered at 530, 605, and 630 nm. The LED arrays were switched on and offalternately in synchrony with the camera (15 Hz) giving, in effect, simulta-neous optical imaging at 7.5 Hz at the separate LED wavelengths.

Importantly, the imaging signals were spectrally decomposed by using mod-ified Beer-Lambert equations with wavelength-dependent optical path lengths:thecortical tissuewasmodeledasascatteringmediumwithembeddedabsorbers(HbO, HbR). The path lengths of light traveling into and out of cortex wereestimated separately for each LED wavelength by using Monte Carlo simulations.These wavelength-dependent path lengths were then used to decompose themeasured imaged signal into the underlying �[HbR] and �[HbO].

Details of animal training, visual stimuli, optical imaging, and data analysis arein SI Appendix. All experimental procedures were performed in accordance withthe National Institutes of Health Guide for the Care and Use of LaboratoryAnimals and approved by the Institutional Animal Care and Use Committees ofColumbia University and the New York State Psychiatric Institute.

ACKNOWLEDGMENTS. We thank M. E. Goldberg for supporting a dedicatedmachine shop at the Columbia University Medical Center; P. P. Mitra (Cold SpringHarborLabs,ColdSpring,NY)forprovidingChronuxanalysis software;K.Korinekfor designing and fabricating the optical imaging hardware (with Y.B.S.); and G.CantoneandC.Maforhelpingwithrecordings.MonteCarlocodewasdevelopedby C.B. and E.M.C.H., based on the previous works of Drs. Lihong Wang (Wash-ington University, St. Louis) and Andrew Dunn (University of Texas, Austin). Thiswork was supported by National Institutes of Health Grants F31 NS056834 andT32 MH015174 (to Y.B.S.), R01 EY013759 and R01 EY019500 (to A.D.), R21NS053684 and R01 NS063226 (to E.M.C.H.), and R24 EY015634 (to Dr. M. E.Goldberg), the Human Frontier Science Program (E.H.), the Columbia ResearchInitiatives in Science and Engineering, the Gatsby Initiative in Brain Circuitry, TheDana Foundation Program in Brain and Immuno Imaging, and the Esther andJoseph Klingenstein Foundation Fellowship in the Neurosciences (to A.D.).

1. Frostig RD, Lieke EE, Ts’o DY, Grinvald A (1990) Cortical functional architecture and localcoupling between neuronal activity and the microcirculation revealed by in vivo high-resolution optical imaging of intrinsic signals. Proc Natl Acad Sci USA 87:6082–6086.

2. Ogawa S, et al. (1992) Intrinsic signal changes accompanying sensory stimulation: Func-tional brain mapping with magnetic resonance imaging. Proc Natl Acad Sci USA 89:5951–5955.

3. Buxton RB (2001) The elusive initial dip. NeuroImage 13:953–958.4. Vanzetta I, Grinvald A (2001) Evidence and lack of evidence for the initial dip in the

anesthetized rat: Implications for human functional brain imaging. NeuroImage 13:959–967.

5. Nemoto M, et al. (2004) Functional signal- and paradigm-dependent linear relationshipsbetween synaptic activity and hemodynamic responses in rat somatosensory cortex.J Neurosci 24:3850–3861.

6. Grinvald A, Lieke E, Frostig RD, Gilbert CD, Wiesel TN (1986) Functional architecture ofcortex revealed by optical imaging of intrinsic signals. Nature 324:361–364.

7. Malonek D, Grinvald A (1996) Interactions between electrical activity and cortical micro-circulation revealed by imaging spectroscopy: Implications for functional brain mapping.Science 272:551–554.

8. Menon RS, et al. (1995) BOLD-based functional MRI at 4 Tesla includes a capillary bedcontribution: Echo-planar imaging correlates with previous optical imaging using intrinsicsignals. Magn Reson Med 33:453–459.

9. Ances BM (2004) Coupling of changes in cerebral blood flow with neural activity: Whatmust initially dip must come back up. J Cereb Blood Flow Metab 24:1–6.

10. Vanzetta I, Hildesheim R, Grinvald A (2005) Compartment-resolved imaging of activity-dependent dynamics of cortical blood volume and oximetry. J Neurosci 25:2233–2244.

11. Chisum HJ, Mooser F, Fitzpatrick D (2003) Emergent properties of layer 2/3 neurons reflectthe collinear arrangement of horizontal connections in tree shrew visual cortex. J Neurosci23:2947–2960.

12. Das A, Gilbert CD (1995) Long-range horizontal connections and their role in corticalreorganization revealed by optical recording of cat primary visual cortex. Nature 375:780–784.

13. Basole A, White LE, Fitzpatrick D (2003) Mapping multiple features in the populationresponse of visual cortex. Nature 423:986–990.

14. Vnek N, Ramsden BM, Hung CP, Goldman-Rakic PS, Roe AW (1999) Optical imaging offunctional domains in the cortex of the awake and behaving monkey. Proc Natl Acad SciUSA 96:4057–4060.

15. Siegel RM, Raffi M, Phinney RE, Turner JA, Jando G (2003) Functional architecture of eyeposition gain fields in visual association cortex of behaving monkey. J Neurophysiol90:1279–1294.

16. Kim DS, Duong TQ, Kim SG (2000) High-resolution mapping of iso-orientation columns byfMRI. Nat Neurosci 3:164–169.

17. LindauerU,etal. (2001)Noevidenceforearlydecrease inbloodoxygenation inratwhiskercortex in response to functional activation. NeuroImage 13:988–1001.

18. Sheth SA, et al. (2004) Linear and nonlinear relationships between neuronal activity,oxygen metabolism, and hemodynamic responses. Neuron 42:347–355.

19. Dunn AK, Devor A, Dale AM, Boas DA (2005) Spatial extent of oxygen metabolism andhemodynamic changes during functional activation of the rat somatosensory cortex.NeuroImage 27:279–290.

20. Hillman EM, et al. (2007) Depth-resolved optical imaging and microscopy of vascularcompartment dynamics during somatosensory stimulation. NeuroImage 35:89–104.

21. Mayhew J, et al. (2000) Spectroscopic analysis of neural activity in brain: Increased oxygenconsumption following activation of barrel cortex. NeuroImage 12:664–675.

22. Martin C, Martindale J, Berwick J, Mayhew J (2006) Investigating neural-hemodynamiccoupling and the hemodynamic response function in the awake rat. NeuroImage 32:33–48.

23. Shtoyerman E, Arieli A, Slovin H, Vanzetta I, Grinvald A (2000) Long-term optical imagingand spectroscopy reveal mechanisms underlying the intrinsic signal and stability of corticalmaps in V1 of behaving monkeys. J Neurosci 20:8111–8121.

24. Moon CH, Fukuda M, Park SH, Kim SG (2007) Neural interpretation of blood oxygenationlevel-dependent fMRI maps at submillimeter columnar resolution. J Neurosci 27:6892–6902.

25. Vovenko E (1999) Distribution of oxygen tension on the surface of arterioles, capillaries,and venules of brain cortex and in tissue in normoxia: an experimental study on rats.Pflugers Arch 437:617–623.

26. Sirotin YB, Das A (2009) Anticipatory hemodynamic signals in sensory cortex not predictedby local neuronal activity. Nature 457:475–479.

27. KohlM,etal. (2000)Physicalmodel for thespectroscopicanalysisofcortical intrinsicopticalsignals. Phys Med Biol 45:3749–3764.

28. Grinvald A, Lieke EE, Frostig RD, Hildesheim R (1994) Cortical point-spread function andlong-range lateral interactions revealed by real-time optical imaging of macaque monkeyprimary visual cortex. J Neurosci 14:2545–2568.

29. ChenY,GeislerWS,SeidemannE(2006)Optimaldecodingofcorrelatedneuralpopulationresponses in the primate visual cortex. Nat Neurosci 9:1412–1420.

30. Chen-Bee CH, Agoncillo T, Xiong Y, Frostig RD (2007) The triphasic intrinsic signal: Impli-cations for functional imaging. J Neurosci 27:4572–4586.

31. Culver JP, Siegel AM, Franceschini MA, Mandeville JB, Boas DA (2005) Evidence thatcerebral blood volume can provide brain activation maps with better spatial resolutionthan deoxygenated hemoglobin. NeuroImage 27:947–959.

32. Berwick J, et al. (2008) Fine detail of neurovascular coupling revealed by spatiotemporalanalysis of the hemodynamic response to single whisker stimulation in rat barrel cortex.J Neurophysiol 99:787–798.

33. Keller CJ, et al. (2009) Intracranial microprobe for evaluating neuro-hemodynamic cou-pling in unanesthetized human neocortex. J Neurosci Methods 179:208–218.

34. Jones M, Berwick J, Johnston D, Mayhew J (2001) Concurrent optical imaging spectroscopyand laser-Doppler flowmetry: The relationship between blood flow, oxygenation, andvolume in rodent barrel cortex. NeuroImage 13:1002–1015.

35. Silva AC, Lee SP, Iadecola C, Kim SG (2000) Early temporal characteristics of cerebral bloodflow and deoxyhemoglobin changes during somatosensory stimulation. J Cereb BloodFlow Metab 20:201–206.

36. Vanzetta I, Grinvald A (2008) Coupling between neuronal activity and microcirculation:Implications for functional brain imaging. HFSP J 2:79–98.

37. MasamotoK,VazquezA,WangP,KimSG(2008)Trial-by-trial relationshipbetweenneuralactivity, oxygen consumption, and blood flow responses. NeuroImage 40:442–450.

38. Hoge RD, et al. (1999) Investigation of BOLD signal dependence on cerebral blood flowand oxygen consumption: The deoxyhemoglobin dilution model. Magn Reson Med42:849–863.

39. Logothetis N (2000) Can current fMRI techniques reveal the micro-architecture of cortex?Nat Neurosci 3:413–414.

40. Yesilyurt B, Ugurbil K, Uludag K (2008) Dynamics and nonlinearities of the BOLD responseat very short stimulus durations. Magn Reson Imaging 26:853–862.

Sirotin et al. PNAS � October 27, 2009 � vol. 106 � no. 43 � 18395

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

Aug

ust 1

9, 2

020