These results suggest that the hemodynamic response is not solely driven by thalamic afferent inputs (P1) but it is largely controlled by secondary and late cortical transmission and influenced by baseline blood flow. In fact, when using a linear regression model, the coupling between the thalamic afferent component P1 and the hemodynamic responses changes across anesthetics. In order to maintain the same neurovascular coupling relationship across different anesthetics we need to add secondary and late SEP components and their interaction with baseline CBF. In particular, the late SEP components P2-N2 alone (plus CBF interaction) are sufficient to predict the hemodynamic responses of all anesthetics simultaneously.
For any single anesthetic used, by changing stimulus train duration, predictions of the hemodynamic responses are good using any SEP component (). This is because all SEP components and hemodynamic responses increase linearly when varying stimulus train duration and there is insufficient variation to reach statistically significant differences, as we showed in (Franceschini et al., 2008
). It is important to notice that by limiting the analysis to single SEP components, the regression coefficients change considerably across anesthetics (differences from 45% for N1, to 60% for P1). This result is consistent with previous studies using LFP and invasive microscopic imaging techniques. Huttunen et al. (Huttunen et al., 2008
), during electrical forepaw stimulation at different frequencies, found linear relationships (R2
~ 0.85-0.87) between LFP and BOLD under alpha-chloralose and urethane anesthesia, but very different regression coefficients. Martin et al. (Martin et al., 2006
) compared LFP and hemoglobin responses in awake and urethane-anesthetized rats over a large range of forepaw stimulation frequencies. While limiting the discussion to the principal LFP component, they found a very different coupling relationship between awake and anesthetize animals.
For the four GABAergic anesthetics, hemoglobin predictions using P2 have the same coupling relationship (P2 regression coefficients =0.42±0.04×10-7
for HbO and =0.29±0.05×10-7
for HbR) while P1 regression coefficients are very different (>60%). Ketamine-xylazine and fentanyl-droperidol have an additional SEP component N2 which is missing, or negligible, with GABAergic drugs within the concentrations used in this study. By combining N2 and P2 in the regression analysis, we obtain the same P2 regression coefficients (within a 35% range) for each of the 6 anesthetics. N1, P2 and N2 are generated by excitatory and inhibitory cortical events (Cauller and Kulics, 1991
; Steriade, 1984
) and the couple P2-N2 in particular describe cortico-cortical transmissions and thus it makes sense to consider them together. Several electrophysiology studies have shown that cortical transmissions are modulated by anesthetic agents more than the thalamic input (Adrian, 1941
; Arezzo et al., 1981
; Cauller and Kulics, 1988
). The fact that P2 for different GABAergic agents has the same regression coefficient, and again the same coefficient when we add N2 as a regressor to predict hemoglobin responses under fentanyl-droperidol, suggests to us that its contribution, and the contribution of cortico-cortical transmission in general, should be considered when studying neurovascular coupling.
A more rigorous analysis is performed by combining all anesthetic agents together and testing for all possible combinations of SEP components with or without baseline blood flow interaction. We used the F-test statistic to take into account the different degrees of freedom when different numbers of regressors are used in the linear regression analysis to predict hemoglobin responses. Using all regressors (P1, N1, P2 and N2 with baseline CBF interaction) did not give statistically significant better predictions of hemoglobin responses than using P2-N2 (either with or without CBF interaction). The best F test statistic and R were obtained using P2-N2 and CBF interactions, confirming the above observation of the predominant involvement of P2-N2 in the hemodynamic response.
While alpha-chloralose and isoflurane produce the largest hemodynamic responses, we verified that they do not solely drive our results. In fact by excluding either one or both of those anesthetics from the combined analysis, P2-N2 still produces better hemoglobin predictions than P1 and N1.
These results suggest that the hemodynamic response is primarily driven by cortico-cortical transmissions and not by thalamic inputs in layer IV. This finding is in agreement with our previous results using parametric stimulation (Franceschini et al., 2008
), but in contrast with the common belief that functional hyperemia is driven by the thalamic afferents' activity in layer IV. A retrograde vasodilation mechanism (Iadecola et al., 1997
) controlled by layer IV thalamic afferent synaptic activity is necessary to support this common belief, since pial arteries regulate the local increase of blood flow into downstream branches (Iadecola, 2004
). If the hemodynamic response is driven by synaptic activity in general, as our data suggest, and not just by the primary synaptic activity in layer IV, then synaptic activity in more superficial layers will initiate superficial vascular responses before any retrograde contribution from layer IV (~100-400 ms delay vs. ~1s delay to propagate ~600 micron (Iadecola et al., 1997
)). With diffuse optical imaging, while we detect hemodynamic changes in both superficial and deeper cortical layers with good temporal resolution, we do not have the spatial resolution to differentiate between cortical layers and cannot determine the origin of the hemodynamic responses. Several groups have tried to resolve laminar differences in the onset of the hemodynamic responses. Using high-resolution fMRI in rats, Silva and Koretsky (Silva and Koretsky, 2002
) found BOLD signal onset starting in layer IV 0.5 sec before starting in layers II and III. Jin and Kim (Jin and Kim, 2008
), in a recent fMRI study of the cat visual cortex, found that cerebral blood volume response (CBV) in superficial cortical layers has a faster time to peak than CBV in middle cortical layers, suggesting that arterial volume increase in the surface of the cortex precedes dilation of microvessels in deeper cortical layers. Using Laser Doppler and electrical stimulation of the trigeminal nerve of rats, Norup Nielsen and Lauritzen (Norup Nielsen and Lauritzen, 2001
) found earlier CBF onset times in superficial layers than in layer IV. Recently, using optical coherence tomography (OCT) to measure blood volume changes in the rat forepaw cortex, Chen et al.
(Chen et al., 2008
) found constant signal onsets for cortical depths between 200-600 micron (layers II-IV), but significantly earlier with respect to the brain surface. The observation of no onset delays from depths of 200 to 600 micron supports the role of secondary and late synaptic activity in initiating vasodilation. We believe that technological advances of techniques such as OCT and two-photon microscopy may soon allow researchers to resolve the issue.
A few other papers indirectly support our finding that the hemodynamic response is driven by cortico-cortical transmissions and not by thalamic inputs. Anna Devor (Devor et al., 2005
) has shown that the hemodynamic response in one cortical column cannot be explained solely by the neuronal activity in that column; rather, neuronal activity in the neighboring columns needs to be included. In the cerebellum, Mathiesen et al.
(Mathiesen et al., 1998
) have shown that the hemodynamic response is mostly sensitive to the level of synaptic activity. If functional hyperemia is controlled by the amount of synaptic activity, the larger number of activated synapses occurs not during the layer IV volley, but from the later cortico-cortical transmissions (Szentagothai, 1978
Still, we do not rule out a smaller contribution of P1 to the hemodynamic response. In our measurements we observe that, during pentobarbital or propofol anesthesia, there is a large P1 response and smaller N1 and P2 responses. Yet we do observe a small change in the hemodynamic signals. This small hemodynamic response under pentobarbital and propofol may be driven by P1. Correlation coefficients between ΣSEP and ΣHb under these two anesthetics are higher for P1 than for N1 and P2 when we average all animals in the same anesthetic group together (). By carrying the analysis to single rats and single anesthetics under pentobarbital or propofol, the correlation coefficients of P1 are not statistically better than the ones of N1 or P2 (p-values>0.05). To isolate and evaluate the P1 contribution to the hemodynamic response, either more invasive methods with better SNR than DOI or EEG or perturbations that exclusively target one SEP component at a time need to be used. For example, by using topical drug application like MK801 (Hoffmeyer et al., 2007
), one may be able to reduce secondary activity without affecting thalamic afferent inputs and estimate individual contributions to the hemodynamic response.
While our results, when fitting for all anesthetics combined, show that P1 does not correlate as well as the subsequent SEP components with the hemodynamic response, we are cautious in disentangling the roles of N1 and P2-N2 in the hemodynamic response. In general, P2-N2 better predicted the hemoglobin responses than N1 (note that using all regressors is statistically significantly better than using only N1 while using all regressors is not significantly better than using only P2-N2). However, the fact that P2 strongly covaries with N1 (Kulics, 1982
; Wrobel et al., 1998
) suggests a close relationship between N1 and P2-N2. This issue needs to be further investigated.
In these experiments, we modulated the stimulation by changing only the stimulus train duration and examined the variation in six groups of animals using six different anesthetics. Previously (Franceschini et al., 2008
) using alpha-chloralose anesthesia and changing stimulus frequency or amplitude, we found that N1 and P2 better predicted the hemodynamic response than P1, by changing stimulus train duration we did not find statistically significant differences between the three SEP components. In part, we wanted to test if this negative result held across six individual anesthetics. While confirming this negative result, we did observe that the variation across anesthetics provided predictive power. It is possible that our results are specific to the duration stimulation paradigm. Experiments across anesthetics, by changing frequency, or amplitude of forepaw electrical stimulation, or stimulation of whisker or visual cortex need to be carried out before making our results more general.
We found that baseline cerebral blood flow has a significant interaction with neurovascular coupling. Specifically, adding baseline BFi
to the SEP regressors substantially improves hemodynamic predictions (increased F and R values). The positive interaction between SEP and baseline cerebral blood flow indicates that, under a condition of constant neural activity, a higher baseline blood flow correlates with a higher hemodynamic evoked response (ΔHb). Apparently in contrast, Sicard et al.
(Sicard and Duong, 2005
) have shown that modulating baseline blood flow by changing inspired O2
does not change the absolute magnitude of evoked hemoglobin changes (ΔBOLD and ΔCBF), but does decrease the relative changes. In Sicard et al. (Sicard and Duong, 2005
), neural activity under different gas concentration was not measured, but assumed constant. This assumption may not be valid, since as shown by Jones et al.
(Jones et al., 2005
) neural activity is affected by hypercapnia. In their experiments they found that while increasing inspired CO2
increased baseline blood flow, it decreased neural activity and evoked hemoglobin responses. Reduction of neural activity with moderate hypercapnia has also been shown in nonhuman primates (Zappe et al., 2008
). Sicard et al.'s (Sicard and Duong, 2005
) finding of constant evoked hemodynamic changes for different baseline blood flow may be masked by changes of neural activity with CO2
As expected, of the six anesthetics used, alpha-chloralose produced the largest hemodynamic responses. Hemodynamic responses under Isoflurane, at the concentration used here (~1.2%), were only slightly smaller than responses under alpha-chloralose. This is because, at these concentrations of isoflurane, baseline CBF and CBV are not strongly affected. At higher concentrations (>1.6%), isoflurane becomes a strong vasodilator (Eger, 1984
) and by either preventing further vasodilation or by saturating cortical tissue with oxygen it may reduce hemodynamic changes in response to stimulation.
We measured only small functional hemoglobin changes in animals under pentobarbital and propofol. Pentobarbital was previously used in a study by Ueki et al. (Ueki et al., 1992
) and its effect on neural and metabolic activity was compared to that of alpha-chloralose and two other anesthetics. Similar to our findings, while they measured principal evoked responses under both pentobarbital and alpha-chloralose, they found an increase in the metabolic rate of glucose under alpha-chloralose anesthesia, but not under pentobarbital anesthesia. Our hemoglobin responses with pentobarbital are small but not zero. The difference may be due to the fact that they used a higher dose of pentobarbital than we did. In general, differences in dose of anesthetics used makes it difficult to compare hemodynamic responses reported in the literature, since hemodynamic responses (as well as electrical responses) depend on the amount of anesthesia used (Dueck et al., 2005
; Purdon et al., 2009
). The advantage of our approach is that we compare the electrical and vascular responses simultaneously and measure their correlation, which, within limits, should not be affected by anesthetic dosages.
The N2 SEP component was measured only with ketamine-xylazine and fentanyl-droperidol, and was not present with GABAergic anesthetics. In the regression model, the N2 contribution to the hemodynamic response was always negative (see last column), suggesting a vasoconstrictive role for this component. More measurements are needed to link N2 activity (likely related to inhibitory interneurons since it disappears with GABAergic anesthetics) with vasoconstriction.
We used scalp electroencephalography (EEG) and diffuse optical imaging (DOI) as functional imaging techniques. These two modalities have poor spatial resolution and are sensitive to large tissue volumes and multiple cortical layers. The inferior spatial resolution with respect to invasive microscopic studies may be a disadvantage of our method, and parallel invasive studies need to be performed. But there are important advantages that justify our study. NIRS and scalp EEG allow for non-invasive measurements and can be directly translated to human studies (Mackert et al., 2008
; Obrig and Villringer, 2003
; Ou et al., 2009
; Shibasaki, 2008
). We believe that controlled animal studies with the same methodologies that are applicable in humans are necessary precursors to neurovascular studies in humans. Also, several invasive neurovascular coupling studies have suggested that the uncoupling between hemodynamic responses and electrophysiology may be an artifact arising from the limited field of view of microelectrodes with respect to the more extensive field of view of standard microscopic hemodynamic measures (Devor et al., 2005
; Ureshi et al., 2005
). We believe that experiments with both macroscopic and microscopic methodologies should be carried out to better understand neurovascular coupling.
Neural input affects secondary and late activity and different anesthetics differently affect neural input and secondary-late activity. With these experiments we found that the hemodynamic response better correlates with secondary-late activity than with the neural input. In conclusion, these results indicate that the magnitude of the hemodynamic response is proportional to the secondary and late SEP components, that baseline blood flow positively affects hemodynamic evoked responses, and that neurovascular coupling is constant across anesthetics. The cause of the effect of different anesthetics on the late SEP requires further investigation.