Neural–metabolic coupling in the central visual pathway

1. Background

A primary challenge in the use of non-invasive brain imaging is to enable broader and more fundamental applications than a limited topographical analysis. The identification of brain functions in both normal and abnormal neuronal systems would represent a significant increase in utility of neural imaging. For this to occur, it is necessary to have a detailed understanding of neural implications of the measurements that are made. Currently, functional magnetic resonance imaging (fMRI) is the most developed approach. In this method, blood oxygenation level-dependent (BOLD) signals are used to infer neural activity. However, there remain basic questions regarding the relationships between BOLD signals and neural function. The BOLD signal is sensitive to oxygen extraction fraction (OEF), the percentage of oxygen removed from the blood by tissue during its passage through the capillary network. The BOLD signal is not necessarily linked to neural activity if there are matched fractional changes in cerebral metabolic rate of oxygen (CMRO₂) and cerebral blood flow (CBF), which results in no change in the OEF. However, fMRI is a powerful technique because in an activated brain area, blood flow increases much more than oxygen metabolic rate [1,2]. This leads to a reduction in the OEF, which contributes to a local increase in the magnetic resonance signal during activated neural activity. These processes have been reviewed previously [3,4]. For the studies we report here, it is important to note that the BOLD signal depends mainly on changes in the concentration of deoxygenated haemoglobin. This means that tissue oxygen consumption provides a direct estimate of the metabolic state of the particular brain region under study. Therefore, tissue oxygen concentration changes along with those of blood flow provide fundamental information about the neural state.

Details have emerged concerning the relationships of vascular and oxygen parameters during different stages of development of non-invasive neural imaging. In positron emission tomography (PET), CBF and CMRO₂ were reported to be coupled during inactivity but substantially dissociated during brain stimulation [1,2]. The assumption for the mismatch was anaerobic glycolysis, but subsequent studies found substantial CMRO₂ responses and increases in deoxygenated haemoglobin [5–9]. It should be mentioned that neural activity is accompanied by relatively large changes in CBF compared with those of CMRO_2. This decreases the concentration of deoxyhaemoglobin, which yields a BOLD response. With neural activity, CBF increase reduces deoxyhaemoglobin while CMRO₂ has the opposite effect. Relative changes in CBF and CMRO₂ affect BOLD response levels, and coupling properties are directly relevant. It is also necessary to note that neural activity includes membrane and spiking processes, facilitation and suppression, and various synaptic functions which are presumed to affect both CBF and CMRO₂ [10–13]. Although details of these processes remain to be established, for the purpose of the investigations to be described here, it is sufficient to point out that tissue oxygen is closely related to the BOLD signal. With neural activation, oxygen metabolism is increased along with that of CBF. This increases tissue oxygen concentration and causes a decrease in deoxyhaemoglobin. The BOLD signal is then increased because it is inversely related to the level of deoxyhaemoglobin. This increase is the primary measurement in fMRI.

Although CBF and CMRO₂ are of basic interest in the process of non-invasive neural imaging, their measurements in behaving organisms are not straightforward. However, it is possible to measure local concentrations of tissue oxygenation in activated brain tissue of a behaving animal. This may be done with a polarographic microelectrode to measure oxygen partial pressure in bulk solution or tissue oxygen concentration [14–16]. The device consists of a platinum cathode and a non-polarized anode. Sensor current is proportional to the concentration of tissue oxygen, and response time varies with size of the exposed platinum surface. A small surface yields rapid electrode responses to altered oxygen tensions. Area of sensitivity is approximately twice the diameter of the electrode tip.

In the studies described here [17–25], we have updated and modified the original design to incorporate a system that enables simultaneous co-localized measurements of tissue oxygen concentration changes during extracellular neural responses of activated cells in the central visual pathway. A schematic of the sensing device we have used is given in figure 1. A double-barrel glass pipette contains a neural microelectrode and an oxygen sensor, both made of platinum, as illustrated. The tip of the electrode extends through an oxygen permeable membrane into neural tissue. The oxygen sensor consists of an anode and cathode. The anode is a silver–silver chloride reference electrode. Tip size is around 30 µm and the field of sensitivity is around 60 µm. Oxygen sensor response time is less than 1 s. Impedance of the platinum neural electrode is from 0.5 to 2 MΩ.

• Download figure
• Open in new tab
• Download PowerPoint

We have used an anaesthetized cat preparation for which there is a history of extensive neurophysiological study that has provided detailed information concerning neural organization of the central visual pathway. In general, the neural system of the cat behaves like that of a scaled down primate. The intention of this work is to monitor neural–metabolic coupling in an intact system during visual processing in order to gain insights that are relevant to non-invasive neural imaging.

2. Neural–metabolic response characteristics of cortical cells

As visual processing progresses from early to central stages, two primary response characteristics emerge, orientation selectivity and ocular dominance. Cells in primary visual cortex are generally sensitive to specific limited ranges of orientation of visual stimuli. And these cells are also the first stage of processing where left and right eye information is functionally combined [26,27]. Data in figure 2 are presented to illustrate and compare neural and metabolic responses, which demonstrate orientation selectivity [22]. First, the neural response range for this cell is limited to around a 30° range between 255° and 285° (figure 2a). Within that range, neural response is vigorous. Correspondingly, tissue oxygen level changes occur for all orientations but the curves have different forms. Within the range of neural firing, there are clear biphasic oxygen responses with substantial initial negative dips followed by generally moderate secondary peaks. The biphasic oxygen response is typical for most of our measurements. It is possible that the initial negative dip represents an immediate oxygen utilization and the positive secondary peak could be the result of blood flow that follows the oxygen response. These notions are speculative and there are discussions in the literature in which it is proposed that blood flow is immediate [28,29]. The other feature that is notable in figure 2a is that orientations removed from those that yield neural responses have substantial secondary peaks in oxygen changes. This is somewhat counterintuitive and we believe that the secondary peaks represent activity in multiple orientation columns beyond those that are detectable by the neural electrode. The implication is that the initial dip is a better indicator of local neural response activity and we have found evidence consistent with this view [22–24,30]. In summary, there is a close correspondence between initial dip oxygen change and neural activity within a preferred orientation range. However, the potential use of the initial dip for fMRI is limited because it is not reliable [31,32].

• Download figure
• Open in new tab
• Download PowerPoint

As shown in figure 2a, a strong neural response is indicated in conjunction with the negative oxygen dip. The test results for orientations away from optimal show decreases in both neural response and initial dip up to the orthogonal orientation for which there is no neural response and a relatively large secondary positive oxygen change. Tuning curves for the same cell, shown in figure 2b, exhibit well-matched inverse tuning characteristics. Summaries for a small group of cells (n = 16), given in figure 2c,d, include, respectively, differences between neural and oxygen parameters for optimal orientations and for tuning widths. Respective mean differences between neural and oxygen values (arrows, figure 2c,d) for these parameters are relatively modest.

Analogous data for ocular dominance are presented in figure 3a for the same example cell as shown in figure 2. Ocular dominance refers to the condition in which a given cortical cell is differentially driven by left and right eyes. For some cells, the two eyes are equally effective. For others, one eye dominates a cell by varying degrees. And some cells are driven only monoptically. For the current cell, left eye activation is stronger for both the neural and tissue oxygen responses (figure 3a). Graphical representation (figure 3b) shows response values for spike count and oxygen change. The left eye exhibits a negative oxygen change in correspondence with a relatively strong neural response. A summary, shown in figure 3c for 21 cells, plots right and left eye differences in oxygen tension versus relative neural ocular dominance. The clearly inverse relationship between the two variables is consistent with relatively large negative oxygen changes for the eye, which elicits stronger neural responses. This result is consistent with previous studies [30,33,34] which suggest that oxygen changes may be localized within a cortical column.

• Download figure
• Open in new tab
• Download PowerPoint

3. Characteristics of tissue oxygen and neural responses to visual stimuli

Responses to visual stimuli have been studied in detail at most stages of the central visual pathway. However, the data obtained are mainly limited to neuronal processes. Parallel metabolic characteristics are generally not available. Experiments described in this section are intended to provide some of this information. As described in §2, typical tissue oxygen responses are biphasic. A small initial negative dip is followed by a relatively large positive peak. As noted in §2, the first phase may reflect immediate usage of oxygen and the second could represent increased blood flow, which is slightly delayed because of draining veins and the initiation of hyperaemia. The relative roles of CBF and CMRO₂, in timing or in magnitude, are not clear [35]. To address this area, we have made neural–metabolic recordings in the lateral geniculate nucleus (LGN) [17,19,23,24] in addition to visual cortex. The dorsal LGN receives direct input from retinal ganglion cells and projects to primary visual cortex. What makes it useful for the present purpose is that it exhibits precise retinotopic organization and has small receptive field size. LGN cells also produce strong responses to visual stimuli, which can be accurately measured with spatial precision.

A simple test of spatial parameters for LGN cells may be explored as follows. A sinusoidal grating can be used as a stimulus to find correct receptive field variables. The size and parameters of the stimulus can be adjusted to match receptive field dimensions and requirements. In general, LGN cells have small receptive fields and high firing rates for optimal gratings. For many LGN cells that we have tested, the smallest stimulus elicits vigorous neural activity and strong monophasic negative oxygen responses. As stimulus size increases, neural response remains strong or is slightly reduced while the initial negative oxygen response becomes smaller and a secondary positive phase emerges. Tests with various sized grating stimuli either covering or positioned outside the receptive field indicate that the negative dip in oxygen responses is clearly coordinated with strong neural activity. This indicates that oxygen consumption is highly specific to local changes in neural activity. The monophasic negative oxygen responses may reflect increases in CMRO₂ and the positive changes could be associated with CBF. The oxygen changes are localized with neural activity but at relatively larger spatial scales.

An extension of these tests can be made to explore the spatial selectivity of the negative oxygen response. Once again, LGN cells are recorded using stimuli that are at first coincident with the receptive field dimensions, and then moved spatially so that they are located at positions that are offset from the centre. For these trials, as the grating stimuli are progressively displaced from the centre of the receptive field, the oxygen initial dip response becomes smaller. These results are consistent with the finding that activity induced oxygen consumption is well localized to the site of neuronal responses. Finally, calculated point spread functions of biphasic tissue oxygen responses in LGN show that negative response components are substantially narrower than those for positive phases [24]. These data are consistent with previous findings suggesting that early hemodynamic responses are related to increases in CMRO₂ [30]. The initial negative dip in oxygen tension appears to be spatially highly specific whereas the secondary positive peak, which may represent CBF activity, appears to be based on a relatively large region. If applied to fMRI, the initial dip would be a better estimate of local neural function. However, as noted in §2, it is well known that the initial dip in fMRI is elusive, variable, weak and may be completely missing in standard measurements [31,32]. This is relevant here because initial dip data in our experiments are generally very robust and reliable.

4. Temporal and spatial summation

The results presented above show that negative tissue oxygen response changes are coupled with neural activity in a closely localized manner. To obtain details concerning oxygen response characteristics, we have conducted tests in which stimulus size can be temporally or spatially altered to evaluate the relative linearity of these variables [17,19]. To do this, visual stimuli with different durations and spatial patterns were presented, while activity from LGN cells was monitored. A test of temporal integration is illustrated in figure 4. For a small grating stimulus at a size that matches the receptive field dimension (figure 4a), multiple unit activity (MUA) along with the oxygen tension change is shown for a 2 s duration presentation. A similar response set is given also for a 16 s stimulus. Another set of data are shown below (figure 4b) for the same LGN location. In this case, the grating is large and contains a blank mask centred over the receptive field. Here, neural responses are minimal, as expected, and there are prominent positive oxygen changes at both 2 and 16 s durations. To examine the degree of temporal linearity for both the negative and positive oxygen components, we included other stimuli of 4 and 8 s durations (not included in figure 4). We then used short presentation data to predict summated responses. For example, we summated a 2 s response with that of a 2 s right-shifted response to predict a 4 s result and compared it with the measured 4 s response. Data for one prediction, for a 16 s presentation, are shown in figure 4c. The initial dip data (blue lines) show a close correlation between measured and predicted results suggesting a linear metabolic temporal summation. On the other hand, there is a considerable discrepancy between measured and predicted results for the positive oxygen response (red lines). Predictions yield considerably larger values than measured results. This over-estimation of a positive oxygen response is consistent with nonlinear temporal summation of the BOLD signal in fMRI [36–38]. If the positive oxygen response is due mainly to an increase in CBF, nonlinear neurovascular coupling will affect the BOLD signal.

• Download figure
• Open in new tab
• Download PowerPoint

An analogous set of tests to those described above is used to explore the extent of spatial integration. To do this, we use the observation that the classical receptive field has an additional region surrounding it that can influence responses even though direct stimulation there cannot excite a neuron. We have conducted tests in both LGN and visual cortex using the following stimuli [19]. First, optimal stimulus parameters are determined for a given cell. They are then used for a grating patch consisting of a pattern within a full circle that matches classical receptive field size. Full size, a left and a right half semicircle size are used as depicted in figure 5a–c. Two additional larger stimuli (figure 5d,e) are used which consist of full-field gratings, one of which contains a blank mask covering the classical receptive field. Tests using these five stimulus patterns were made at an LGN recording site and the data obtained are shown in the middle and right columns for oxygen changes and neural activity profiles, respectively. In figure 5c,e, predicted composite data are included with measured results to examine linearity of spatial summation. This test sequence covers characteristics of spatial summation both within the classical receptive field and in the surrounding region. The oxygen data (middle column) show clear initial dips for each half circle grating and a larger negative change for the full circle stimulus. For the larger stimuli extending beyond the classical receptive field there are minimal initial dips and pronounced secondary positive peaks. The neural data (right column) show responses for each half grating stimulus and a larger response for the full grating in accord with the oxygen data. The large grating stimulus outside the classical receptive field yields only a small amount of spontaneous activity when a blank mask covers the centre, as expected. When the mask is removed, the extended grating yields a response that is reduced in comparison with that for the classical receptive field. Predicted composite responses for the full circle grating are greater than the measured values for both oxygen and neural data. For the full-field response, predicted and measured values are similar for positive oxygen peaks but predicted neural responses are higher than those measured.

• Download figure
• Open in new tab
• Download PowerPoint

Population data from LGN also show nonlinearity of spatial summation. Within the classical receptive field, composite predicted oxygen and neural responses are substantially larger than measured values. Impulse response function estimates indicate a neural basis for the nonlinear spatial summation of tissue oxygen levels within the classical receptive field. For the surrounding area, predicted oxygen responses have relatively large initial dips but similar positive peaks in comparison with measured values. Predicted composite spiking activity is also substantially stronger than measured levels. This suggests a nonlinear coupling between oxygen and neural responses when spatial summation is examined. Considered together, these findings suggest that oxygen initial dip values and neural activity are closely coupled in both space and time.

5. Binocular activation

Binocular activation of the visual pathway is relevant to the study of neural–metabolic coupling as it may provide insights regarding interpretation of haemodynamic signals. A primary question regarding interpretation of BOLD signals concerns the neural nature of the measurement that is represented. Specifically, synaptic activity is considered to be the input stage, and spiking discharge the output of neural response. These components are obviously related, but previous work suggests that synaptic activity may be more directly reflected in the BOLD response [39]. Both CBF and CMRO₂ are presumably involved, but measurements of these factors in an in vivo functioning neural system are difficult to make. The oxygen and neural co-localized measurements we describe here provide a way to activate and compare inhibitory and excitatory neural processes in the visual cortex. This allows us to compare metabolic, spiking and local field potential responses during excitatory or inhibitory binocular interaction.

To obtain these data, drifting sinusoidal gratings were presented to both eyes simultaneously on two cathode ray tube monitors. Preferred grating specifications were determined and dichoptic tests were run such that phase was fixed in one eye and variable for the other [20]. This provided response data for a series of randomly interleaved relative interocular phase values. This sequence is equivalent to relative binocular disparities. The dual neural–metabolic sensor described above was used here to monitor tissue oxygen responses in visual cortex while MUA, and local field potentials (LFP) were measured. The LFP frequency range we focused on was between 25 and 115 Hz (gamma range) as in previous work [25,39,40]. Representative results are presented in figure 6 for MUA, LFP (gamma frequencies), and oxygen response data for six relative interocular phases interleaved with a control stimulus presented monoptically to the dominant eye. In the first profile (figure 6a), MUA responses vary with interocular phase and peak at around 240°. The MUA data show a typical phase-specific binocular interactive profile in which spiking activity varies from below to above the monocular control level. There is clear statistically significant facilitation and suppression exhibited in this response as shown in previous work [27,41,42]. By contrast, LFP power (figure 6b) varies and is relatively strong at all phase values. And initial dip oxygen responses (figure 6c) are also slightly variable and, as in the LFP case, are all larger than those of the monocular control. The positive peak oxygen data (figure 6d) exhibit phase tuning that is different than that for the initial dip and response at 180° is equal to that of the monocular control, but all values, as in figure 6a,b, are positive.

• Download figure
• Open in new tab
• Download PowerPoint

Similar data to those of figure 6a–d have been collected for a population of cortical cells. Three main types of MUA responses have been found. Phase tuning which exhibits both suppression and facilitation, as in figure 6a–d, is one type. Second and third types show uniform suppression or facilitation, respectively, for all binocular phase values. These three types of dichoptic neural interaction have been identified in previous studies [27]. In contrast with the MUA patterns, LFP values are uniformly higher than monocular control results. Initial dip oxygen responses are similar to those for LFP and positive peak oxygen data appear uncoupled to both MUA and LFP. For the cell shown in figure 6a–d, and for other examples, MUA and LFP data are generally comparable. These variables are also consistent with initial dip oxygen results, but only during facilitative binocular interaction. During suppressive dichoptic interaction, LFP and MUA data are dissociated. In this case, LFP data and initial oxygen dips are generally associated.

Coupling characteristics between MUA, LFP, initial dip and positive peak oxygen data can be quantified as follows. Of interest is the change in relative values of monocular and binocular responses. We therefore use a normalized change index

As shown above in the example of figure 6a–d, LFP and initial dip oxygen data change with MUA response during facilitative MUA binocular interaction, but not for that of suppression. To analyse our dataset, we compute C_n values for MUA, LFP and initial dip oxygen responses during facilitation and suppression. Results of this analysis are presented in figure 6e–l. For facilitative interaction, the relationships of change indices between MUA–LFP, MUA–oxygen initial dip and LFP–oxygen initial dip (figure 6e–g) are quite similar as the least-squares regression fits demonstrate. The histogram summary (figure 6h) shows similar average C_n values with no significant differences. The same data analysis for suppressive interaction has a different outcome. The relationships between MUA and LFP and MUA and oxygen dip (figure 6i,j) show modest negative correlations which are not significant. However, LFP is positively and significantly correlated with initial oxygen dip data (figure 6k). The associated averaged C_n histogram data show similar positive LFP and oxygen dip values along with a negative averaged MUA (figure 6l). The suppressive data shown here represent conditions in which activation of one eye is suppressed by stimulation of the other. In this case, the suppressed MUA profile is not matched by LFP or initial oxygen dip responses. At this point, it should be noted again that the initial dip is generally not considered in fMRI work because it is weak, variable or non-existent [31,32]. However, for the preparation described here, initial dip data are reliable, consistent and generally robust. It is also well localized to the site of neural activity [30]. The analysis presented above was also made using the positive oxygen peak response instead of that of the initial dip. Positive peak data do not match alterations in LFP or MUA. However, there are similar average changes to LFP and MUA tuning during excitatory binocular response. During suppression, positive peak data are only matched to those of LFP. In summary, it is possible to examine neural–metabolic coupling during excitatory or inhibitory phases of neural activity. We have measured spiking and LFP activity during visual stimulation and associated changes in tissue oxygen at both the initial dip and positive peak levels. During excitatory binocular interaction, we find that spiking and LFP activity correlate with oxygen metabolism. Suppressive binocular interaction yields a different result. LFP and spiking activity are dissociated, but LFP and oxygen metabolism are tightly coupled. These results suggest that a similar binocular interaction paradigm should be explored in an fMRI procedure.

6. Neural activity, glucose and lactate

Neurovascular coupling involves numerous functional processes. With neural activation, there is neurotransmitter release such as glutamate and γ-aminobutyric acid. There is an increase in oxygen, glucose and ATP consumption. These events stimulate vasoactive chemical agents and metabolites that are involved in CBF. To address a part of the coupling of neural activity with energy metabolism, we have investigated the relationship between neural responses and the principal supportive substrates of glucose and lactate. We use sensors with high spatial and temporal resolution to measure simultaneous co-localized changes in glucose, lactate and neural activity in the striate cortex during visual stimulation. Electrochemical microelectrodes are used to measure tissue glucose and lactate concentration levels while neural spiking or LFP are recorded. Visual stimuli consist of gratings with optimal or contrast varying parameters [21].

The traditional view of glucose and lactate is that they are, respectively, the main neural energy substrate and the principal waste product in an associated metabolism [43]. There have been alternative views that lactate can also serve as an energy source for neurons via a glial released substrate [44,45]. This notion has been developed as the astrocyte-neuron lactate shuttle (ANLS) theory in which glucose is taken up by glial cells to produce lactate which serves to fuel neurons [44,45]. Various measurements relevant to this idea have been made [46–52] but they generally have coarse temporal or spatial resolution not suitable for capturing the rapid changes in metabolite concentration that occur during neural activity. However, electrochemical techniques have also been used which can resolve the rapid changes [53]. For the results described here, we use customized electrochemical glucose and lactate sensors combined with a neural microelectrode to determine coordinated changes in the metabolic substrates during visual activation of cortical neurons. This approach, which involves monitoring of an active in vivo region of cerebral cortex, is intended to reveal insights from an intact physiological system.

Representative data are presented in figure 7 for simultaneous measurements of neural activity, glucose and lactate concentration changes. Four levels of grating contrasts (10%, 20%, 40% and 80%) are used in order to explore relationships at different levels of neural activity. In figure 7a, a 10% contrast activation period of 30 s is shown along with a 10 s prior section which illustrates low-level spontaneous spiking. Glucose and lactate levels (figure 7b,c) are given as percentages of changes relative to the spontaneous baseline. These changes in signals are derived by averaging responses across trials. Responses are normalized and presented as percentages of signal alterations. Note that the percentage response of lactate is much larger than that of glucose. This may be accounted for by a few mechanisms. First, the baseline concentration of glucose is higher than lactate (1.1 mM and 0.05 mM, respectively) for this recording site. Second, the process of aerobic glycolysis converts one molecule of glucose to two molecules of lactate. Third, lactate might also be generated by astrocytic glycogenolysis [54,55]. What is striking in figure 7a–c is the consistent synchronously timed increases in MUA and lactate while glucose is simultaneously decreased. Glucose and lactate return to baseline levels immediately as neuronal activity stops. These rapid changes are not consistent with some previous reports of sustained increased lactate and decreased glucose during stimulation [53,56]. We believe the differences are due to low temporal resolution in previous work. To explore this, we reduce the temporal resolution of the analysis shown above by averaging responses of every five data points and re-plot the data of figure 7a–c into those given in figure 7d–f. These latter data now appear to show sustained changes during visual stimulation consistent with previous reports. To complete the results for all four stimulus contrast levels, mean values are presented in figure 7g which again shows tight coupling of neural, glucose and lactate responses. This coupling is clearly directly connected to increased neural activity as shown in figure 7h. We have extended these findings by use of short-duration stimuli (2 s), varied orientations including preferred and orthogonal and different visual stimulus size. Results for all the variables studied are consistent with those presented in figure 7. In summary, it is clear in an in vivo visual system, that there is a close temporal coordination of neural, glucose and lactate activity.

• Download figure
• Open in new tab
• Download PowerPoint

A final point of this study concerns the observation that glucose consumption is proportional to changes in CBF [1,2]. As our current results show that response profiles of glucose and lactate are tightly coordinated, we have extended this finding with simultaneous recordings of neural activity and CBF [21]. A fine laser Doppler flow needle probe with attached tungsten microelectrode is advanced into visual cortex to obtain recordings. Representative data, shown in figure 8a,b, include MUA and LFP results, respectively, which in this case are modulated in synchrony with the temporal frequency of stimulation. A very different pattern is shown for the CBF response (figure 8c) which consists of a sustained change during the neural response. There is also a significant temporal difference in peak changes of spiking and LFP compared with CBF signals (0.96, 0.9 and 5.34 s, respectively). In summary, the CBF signal is not synchronized with the modulation of neural activity, and there is a clear delay in the CBF signal compared with that of the neural response. Consistent with our finding here, CBF signals were also reported to lag behind CMRO₂ and neural activity although these three measurements are linearly coupled in amplitude [40,57].

• Download figure
• Open in new tab
• Download PowerPoint

Taken together, the dynamic changes of our co-localized measurements of CBF, tissue oxygen, glucose and lactate signals are consistent with the hypothesis that lactate may serve as a mediator of metabolic information rather than as a metabolic substrate, per se [58–60]. As neurons are activated by visual stimulation, tissue oxygen and glucose levels decrease while lactate levels increase. This is followed by blood flow rises after changes in oxygen, glucose and lactate signals. Lactate is vasoactive and dynamically alters microvasculature diameters [61]. It is likely that the lactate generated with glucose consumption may serve to maintain brain energy turnover and possibly regulate CBF. The role of lactate as an informant of metabolic states rather than as an energy substrate may help resolve the controversy over the understanding of astrocytic metabolism and its contribution to neuronal function [44,45,47,62–66]. Therefore, our results appear to be in line with a feedback mechanism in which energy supply is controlled by energy demand [30,67], rather than by a feed-forward model involving neuronal signalling via neurotransmitters [28,29,68,69].

7. Summary

The enormous potential of non-invasive neural imaging for basic and applied research and for clinical application requires clarification of acquired data. Specifically, neuronal processes must be deduced from haemodynamic measurements. Although a considerable effort has been made to clarify this transformation, fundamental questions remain. The work we describe here is an attempt to provide direct parallels between neural and haemodynamic signals. To do this, we use assessments of tissue oxygen concentration changes in cerebral cortex or the LGN while stimuli are used to activate the central visual pathway. The addition of a neural sensor provides co-localized measurements of neuronal activity. Tissue oxygen is directly related to the BOLD signal, which is the primary measurement in fMRI. It can be readily assessed by a micro-sensor in an in vivo preparation. This arrangement provides neural–metabolic comparisons within a functioning visual system in an anaesthetized preparation. We have utilized a variety of visual stimuli with known neural response characteristics and have also characterized co-localized changes in oxygen concentration. The typical oxygen response is bimodal with an initial negative dip followed by a large positive peak. Experimental results demonstrate that the initial dip is more reliable and a better local indicator of neural and metabolic activity than the positive peak. This finding is at odds with assessments of the initial dip in fMRI, which is considered to be unreliable. Beyond characterization, we have considered functional questions. One approach is to use a visual stimulation procedure that provides separate analysis of neural–metabolic events during suppression or facilitation. In this case, we find that LFP and spike activity are coordinated during facilitation but not in a suppression mode. In addition to the dual sensor approach, we have used an analogous system to investigate changes in primary metabolic substances, glucose and lactate, during neural activation. Results show simultaneous changes in these substances that are tightly coupled to neural activity. However, co-localized measures of CBF and neural activity show clear delays in blood flow relative to neuronal response. Although these results are significant, the relatively high temporal resolution of glucose and lactate sensors is not comparable with millisecond action potential data. We also cannot determine if generated lactate is released and consumed or transformed subsequently. And it is possible that astrocytes consume glucose for further function. These and related areas remain to be determined.

Authors' contributions

Both authors conceived the projects, performed data analysis, made the illustrations and wrote the manuscript.

Competing interests

We have no competing interests.

Funding

This work was supported by National Eye Institute Research grant no. EY01175 and CORE grant no. EY03176.