Previous studies using MEG and NIRS in combination (Mackert et al., 2004
; Sander et al., 2007
; Mackert et al., 2008
) have employed direct current (DC) MEG measurement techniques to study DC or near-DC changes in the magnetic fields in a block design finger-movement paradigm with 30-s blocks of finger movements alternating with rest periods. Limited to one stimulus type (stimulus frequency and duration), these studies cannot explore the linear/nonlinear neurovascular coupling relationship. In contrast, here we investigate the relationship of the NIRS/DOI vascular signals with the detailed time course of MEG current sources. The use of an event-related rather than a block design allows us to employ shorter stimulation periods (1–4 s versus 30 s) and to evaluate the contribution of single neural component to the hemodynamic responses. Our approach enables us to directly compare the results with the vast number of neurovascular coupling findings in animals using invasive electrical and vascular recordings.
The choice of limiting the study to four conditions is dictated by the fact that in order to examine more duration conditions, it is necessary to extend the scanning time to ensure a sufficient SNR in the average signals for the neurovascular coupling analysis. The choice of 1 s as the shortest condition is due to the fact that hemoglobin response for shorter stimuli is too noisy; the choice of 4 s as the longest duration is due to the fact that MEG signal for longer stimuli become noisier because of eye blinks and muscle contractions. In any case, the 1 to 4 s duration chosen is within the range of non-linearity previously observed in the hemodynamic response with BOLD fMRI (Bandettini et al., 2002
We observed a good spatial agreement between the activation areas determined by the two imaging modalities. The neural source location determined using MNE in five subjects is in good proximity with the maximum hemoglobin evoked changes measured with NIRS. This result reveals the benefit of integrating other vascular measurements such as fMRI, which has a good spatial resolution, to help solve the ill-posed MEG inverse problem. Our result also indicates that a combined MEG-DOI inverse formulation may provide additional information for the two ill-posed inverse problems.
We verified that the hemodynamic response is non-linear with respect to stimulus train duration (R2
= 0.96 and 0.22 for zero intersection). Our result shows that the hemodynamic responses to short stimulus trains (<2 s) measured with NIRS are larger than those predicted by a linear model, and it is in agreement with fMRI results in humans (Bandettini and Ungerleider, 2001
; Birn et al., 2001
). With the simultaneous MEG measurements, we tested whether the hemodynamic non-linearity can be justified by a habituation effect in the neural responses. For the tested stimulus train durations, we found that N20 and P60 show a linear relationship with train duration, achieving R2
value above 0.99 with 0.26 zero intersection. On the other hand, due the strong habituation effect in P35, the R2
value for P35 is 0.93; however, this habituation effect is too strong to explain the habituation effect in the hemodynamic response.
The initial hemoglobin overshoot observed in human is not supported by the findings in small animals, for which the habituation effect in hemoglobin response appears for longer stimulus trains (8–20 s) (Martindale et al., 2005
; Franceschini et al., 2008
). This discrepancy may be due to differences between species (humans versus rats) or anticipation effects, which are suppressed by anesthesia in rats.
In addition to linearity/non-linearity of the responses, we performed a convolution analysis to test whether individual components of the current amplitude can predict the hemodynamic response time courses better than input stimulus. To do so, we used either a current amplitude component or the input stimulus as an input in the linear convolution model to predict the hemodynamic responses and determined statistically which factor shows better predictive power. With the stimulus tested, we found that there is no significant difference between the predictions of HbR using N20, P35 or P60 versus using the input stimulus. Instead, the sum of the peak amplitudes and the RMS value can predict HbR response consistently better than the input stimulus. Furthermore, subtracting N20 helps remove possible hyperpolarising DC shift of the membrane potential (Hellweg et al., 1977
), both P35 − N20 and P60 − N20 achieve statistically better prediction of HbR than the stimulus.
The fact that the sum of three peaks and the RMS value of the current dipole amplitude are the best and N20 is the worst in predicting the hemodynamic responses is in agreement with our recent results where we investigated neurovascular coupling using EEG and DOI in rats (Franceschini et al., 2008
). Similar to the human study, the animal study shows that the later peaks (N1 and P2) of the somatosensory evoked potentials are able to predict hemodynamic responses better than the first peak, P1. The EEG evoked potential deflection P1 in rats is equivalent to MEG N20 deflection in humans. This initial peak constitutes the primary cortical response which is generated by synaptic excitation of middle layers by thalamocortical inputs (Li et al., 1956
; Mitzdorf, 1985
; Di et al., 1990
). In most invasive animal studies, this earliest peak is typically used to predict the hemodynamic response (Caesar et al., 2003
; Jones et al., 2004
; Sheth et al., 2004
; Iadecola, 2004
). Historically, the greater focus on P1 may be partially due to the fact that this peak is very stable and persists with deep anesthesia, while subsequent cortical activity is abolished with deep anesthesia. The later cortical activity is more spread both spatially and temporally, and it is less stable with changes in stimulus parameters (Cauller and Kulics, 1991
). Our findings that the hemodynamic responses correlate better with later peaks than with the primary cortical response suggest that the cortical hemodynamic response is largely controlled by synaptic activity related to intracortical processing rather than direct thalamic input. This is probably because the majority of the synapses are triggered with column inputs from other neurons in the cortex, and the hemodynamic response is mostly sensitive to the level of synaptic activity (Mathiesen et al., 1998
). Thus, our results indicate that later components need to be considered to correctly understand the neurovascular coupling.
Despite the better predictions of hemoglobin response by the sum of two or three peaks and the RMS value of the current dipole amplitude, the habituation effects exhibited in the hemodynamic response are not totally accounted for with these parameters. This suggests that the hemodynamic response cannot be described by a single component, but by a weighted combination of multiple components or some later components in the neural signals. A more complex experimental design which provides more variability in the neuronal responses will be necessary to obtain sufficient power in the statistical analysis of the different neurovascular coupling models.