Report

Single-Neuron Activity and Tissue Oxygenation in the Cerebral Cortex

See allHide authors and affiliations

Science  14 Feb 2003:
Vol. 299, Issue 5609, pp. 1070-1072
DOI: 10.1126/science.1079220

Abstract

Blood oxygen level–dependent functional magnetic resonance imaging uses alterations in brain hemodynamics to infer changes in neural activity. Are these hemodynamic changes regulated at a spatial scale capable of resolving functional columns within the cerebral cortex? To address this question, we made simultaneous measurements of tissue oxygenation and single-cell neural activity within the visual cortex. Results showed that increases in neuronal spike rate were accompanied by immediate decreases in tissue oxygenation. We used this decrease in tissue oxygenation to predict the orientation selectivity and ocular dominance of neighboring neurons. Our results establish a coupling between neural activity and oxidative metabolism and suggest that high-resolution functional magnetic resonance imaging may be used to localize neural activity at a columnar level.

Functional magnetic resonance imaging (fMRI) is a powerful tool used for the noninvasive mapping of neural activity in the brain (1–3). Blood oxygen level–dependent (BOLD) fMRI infers neural activity by measuring small changes in deoxyhemoglobin within the brain's vasculature. Therefore, the coupling between local cerebral hemodynamics and the underlying neural activity is critically important to the interpretation and design of fMRI studies. Increases in neural activity elicit a delayed decrease in deoxyhemoglobin corresponding to a positive change in the BOLD signal (1–4). Most fMRI investigations use this positive BOLD response to map neural activity. However, because the response is also observed from draining veins that are displaced from the sites of activation, the spatial resolution is limited (5–7). Optical imaging and imaging spectroscopy findings suggest that a fast increase in deoxyhemoglobin, occurring before the delayed decrease, may be a better indicator of neural activity (8, 9). This component of the hemodynamic response is often referred to as the “initial dip.” Recent fMRI studies and direct measurements of oxygen tension within the microcirculation have also identified the initial dip (6, 7, 10–12). In studies of the visual cortex, brain imaging maps that emphasize the initial dip exhibit sharper patches and stripes [characteristic of orientation and ocular dominance columns (13)] relative to functional maps that are based on other signals (6, 8,9). It has been proposed that the initial dip reflects an increase in oxygen consumption by active cells and is consequently better localized to the site of neural activity (6,8, 9).

Despite its potential, the existence and interpretation of the initial dip are controversial because it is not always observed using fMRI (14, 15), and its detection using imaging spectroscopy may be confounded by the difficulty in correcting for the wavelength dependence of light scattering in tissue (16,17). An alternative hypothesis is that changes in blood volume, rather than oxygen consumption, could account for the initial dip (18). Although simultaneous recordings of fMRI signals and neural activity were recently reported in the monkey's primary visual cortex (4), the analysis focused on the positive BOLD response and used stimuli that were designed to activate a relatively large uniform area of the cortex. Therefore, the studies did not address neural and hemodynamic coupling at a submillimeter spatial resolution.

In our study, we made simultaneous colocalized measurements of tissue oxygenation and single-cell neural activity in area 17 of the cat's visual cortex. Tissue oxygenation is linked to deoxyhemoglobin through the local oxygen concentration gradient in tissue and through the oxygen-hemoglobin dissociation curve and is therefore expected to reflect the hemodynamic changes measured with fMRI. A quantitative model of cerebral hemodynamics posits that tissue oxygenation and the BOLD response exhibit similar time courses after brief increases in neural activity (19).

We measured tissue oxygenation with a Clark-style polarographic oxygen microelectrode (20) and measured single-cell neural activity simultaneously with an adjacent platinum microelectrode (Unisense A/S). Both sensors were housed within a double-barrel micropipette (21). The field of sensitivity of the oxygen sensor was a sphere approximately 60 μm in diameter. This is smaller than a typical cortical column (300 to 600 μm) (13) and of the same order of magnitude as the average intercapillary distance in the cat's brain (∼30 μm) (22).

We used this combined sensor to identify changes in tissue oxygenation that accompany variations in spike rate during standard measurements of orientation selectivity and ocular dominance (21). These visual parameters have a well-established columnar organization within the visual cortex (13). Average time courses of both neural and oxygen responses are presented in Fig. 1, A and C, for a representative neuron. Changes in tissue oxygenation (solid lines with dashed lines representing ±1 SE) exhibited clear stimulus-induced changes when averaged over a sufficient number of trials. We measured statistically significant oxygen responses for all 21 cells studied (P < 0.0005, t test). Like hemodynamic responses, the time course of the oxygen response exhibited an initial dip followed by a positive peak (Fig. 1, A and C). The transition from dip to peak was delayed by 1.0 to 2.5 s relative to hemodynamic responses measured with optical imaging and fMRI techniques (6–12). A portion of this delay (∼0.75 s) can be attributed to the response time of our sensor (21). The remaining time can be attributed to the high resolution and extravascular nature of our oxygen measurements (23).

Figure 1

Orientation selectivity, ocular dominance, and correlation analysis for a representative neuron. Spike-time histograms (gray solid area) and oxygen responses (solid lines) in (A) and (C) represent the mean response across 72 trials. All stimuli were drifting sinusoidal gratings presented for 4 s. Gray dotted regions indicate stimulus onset and duration; dashed lines represent 1 SE of the mean. (A) Orientation selectivity. Average responses are shown for six orientation conditions. (B) Comparison between neural and oxygen orientation tuning curves. Spike responses are quantified as the average spike rate during the stimulus, minus the spontaneous rate. Oxygen responses are quantified as the percentage change from the baseline occurring 5.75 s after stimulus onset. Average spike and oxygen responses are fit to independent Gaussian functions. (C) Ocular dominance. Responses are shown for left eye (LE, top) and right eye (RE, bottom) stimulation. (D) The bar plot compares the average neural and oxygen response (5.75 s after stimulus onset) between the two eyes. Error bars represent 1 SE of the mean. (E) A contour plot of tissue oxygenation versus spike rate and time. Oxygen response time courses across all conditions are binned according to their average spike rate during the stimulus (overlapping bins are each 7.5% of the neural response range). For each time point, the percentages of oxygen changes within each response bin are averaged together and included in the contour plot. The line above the contour plot depicts stimulus onset and duration, and the arrow indicates the maximum correlation time (8.75 s). (F) Correlation coefficients derived from correlating average spike rate and oxygen changes as a function of time. Correlation coefficients below the dotted line are significantly less than zero (P < 0.0005).

Neural and oxygen responses were recorded for six orientation conditions in the dominant eye. Optimal orientation conditions, which elicited large neural responses, gave rise to the largest initial dips and smallest peaks (Fig. 1A). The same relationship was observed for ocular dominance measurements (Fig. 1C). The ocular dominance was estimated with optimally oriented drifting gratings, which were presented separately to each eye. The example cell in Fig. 1 was left eye dominant and exhibited a larger initial dip in response to left eye stimulation than to right eye stimulation (Fig. 1C). These results suggest two competing mechanisms. First, a dip in the oxygen response begins early, as a result of increased oxygen consumption by activated cells. Second, a peak in the oxygen response begins later, presumably as a result of increased blood flow to the activated area of the cortex. These two mechanisms partially overlap in time, and as a result, each response reduces the size of the other. Furthermore, the two mechanisms operate over different spatial scales. We observed the difference in spatial scales in the responses to nonoptimal stimuli, which presumably excited only surrounding columns. The relatively distant activation from neighboring columns elicits a robust peak in the oxygen response without a robust dip. Therefore, the effect of oxygen consumption by active cells was localized within a cortical column, whereas the compensatory inflow of oxygen appears to be spread out over multiple columns.

Orientation tuning curves, derived from neural and oxygen responses in Fig. 1A, gave similar estimates of orientation selectivity (Fig. 1B). The estimates for orientation preference and tuning width differed by 10.6° and 34.0°, respectively. The oxygen tuning curve was inverted because negative changes in the oxygen response corresponded to increases in neural spike rate. Oxygen and neural measurements of ocular dominance, derived from Fig. 1C, exhibited the same relationship (Fig. 1D). Oxygen measurements of orientation tuning and ocular dominance were calculated from the average oxygen change occurring 5.75 s after stimulus onset. This time gave the best correlation between spike rate and tissue oxygenation for the group data (fig. S1). Although the analysis we present here used 5.75 s exclusively, similar results were obtained using times between 4 and 12 s (fig. S2).

A more detailed picture of the predictive relationship between oxygen and neural signals was obtained through correlation analysis (Fig. 1, E and F). The contour plot (Fig. 1E) shows oxygen response time courses as a function of spike rate for the representative neuron. Red and blue shading represent positive and negative changes, respectively, in tissue oxygenation. Trials with high average spike rates have a large dip and a small peak, whereas those with low average spike rates have a small dip and a large peak. Oxygen responses are significantly correlated with spike rate between 3.25 and 14.0 s after stimulus onset (Fig. 1F). During this time period, correlation coefficients are significantly below zero (P < 0.0005), indicating an inverse relationship between spike rate and percent of oxygen change.

We studied 21 cells from four cats using the stimulus protocol and analysis methods described above. Tests on 16 of the 21 cells resulted in oxygen orientation tuning curves well fit by a Gaussian function (r 2 > 0.75). Of the five cells that did not exhibit orientation tuning in their oxygen responses, three had relatively low neural spike rates. We analyzed these 16 cells further by plotting histograms of the error between neural and oxygen tuning parameters (Fig. 2, A and B). For the majority of cells, the predictions were accurate to within 15° of their optimal orientation (Fig. 2A) and 40° of their tuning width (Fig. 2B). Oxygen orientation tuning curves tended to be wider than neural orientation tuning curves (Fig. 2B). This trend was significant (P < 0.05, t test) and suggests that decreases in tissue oxygenation reflect a population of active cells that have a collective tuning width broader than the tuning width of the individually recorded neuron. The difference between the oxygen responses under right and left eye stimulation showed an inverse relationship when plotted as a function of neural measurements of ocular dominance (r = –0.66) (Fig. 2C). This indicates a larger negative percentage of oxygen change in the stronger eye as compared to the weaker eye.

Figure 2

Summary plot of the difference between oxygen and neural measurements of orientation selectivity and ocular dominance. (A and B) Sixteen of 21 cells had oxygen orientation tuning curves well fit by a Gaussian function (r 2 > 0.75). The centers and widths of these 16 fits are compared with the same parameters derived from the neural orientation tuning curves. (A) Histogram of the error in predicting optimal orientation selectivity using oxygen responses. The absolute value of each difference is presented. Mean difference (arrow) = 8.3°. SD = 8.6°. (B) Histogram of the error in predicting orientation tuning width from changes in tissue oxygenation. Mean difference (arrow) = –20.5°. SD = 31.2°. (C) The difference in percent oxygen change between the right and the left eyes is plotted as a function of the ocular dominance index for each neuron.

Previous functional brain imaging studies have proposed that utilization of the initial dip improves the spatial localization of neural activity (6, 8, 9). Our results provide direct evidence for this hypothesis by demonstrating a close relationship between single-unit neural activity and transient decreases in tissue oxygenation. High-resolution fMRI techniques, based on the initial dip, thus have the potential to resolve functional cortical columns that are thought to be fundamental for a wide variety of brain processes.

Supporting Online Material

www.sciencemag.org/cgi/content/full/299/5609/1070/DC1

Materials and Methods

SOM Text

Figs. S1 and S2

References and Notes

  • * To whom correspondence should be addressed. E-mail: freeman{at}neurovision.berkeley.edu

REFERENCES AND NOTES

View Abstract

Navigate This Article