Report

Transcranial Magnetic Stimulation Elicits Coupled Neural and Hemodynamic Consequences

See allHide authors and affiliations

Science  28 Sep 2007:
Vol. 317, Issue 5846, pp. 1918-1921
DOI: 10.1126/science.1146426

Abstract

Transcranial magnetic stimulation (TMS) is an increasingly common technique used to selectively modify neural processing. However, application of TMS is limited by uncertainty concerning its physiological effects. We applied TMS to the cat visual cortex and evaluated the neural and hemodynamic consequences. Short TMS pulse trains elicited initial activation (∼1 minute) and prolonged suppression (5 to 10 minutes) of neural responses. Furthermore, TMS disrupted the temporal structure of activity by altering phase relationships between neural signals. Despite the complexity of this response, neural changes were faithfully reflected in hemodynamic signals; quantitative coupling was present over a range of stimulation parameters. These results demonstrate long-lasting neural responses to TMS and support the use of hemodynamic-based neuroimaging to effectively monitor these changes over time.

The study of brain function makes use of various techniques to modify neural processing. These include neurophysiological, surgical, and pharmacological approaches (1). In general, these techniques may be invasive, irreversible, and not confined to specific brain areas. In contrast, transcranial magnetic stimulation (TMS) (2) provides a noninvasive, reversible, and relatively localized approach that has substantial promise for basic neuroscience and clinical applications (3, 4). In this technique, a magnetic coil placed above the scalp generates electric currents in the underlying cortex. As yet, the manner in which these currents affect neuronal processing is largely undetermined (3, 5).

The full potential of TMS depends not only on a basic understanding of its neural effects, but also on the ability to make direct measurements of these changes in the human brain. This has recently been attempted by combining TMS with noninvasive brain-imaging techniques such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) (6). These methods measure hemodynamics and metabolism to infer changes in neural activity based on known coupling between these variables (7). However, in certain conditions, neural activity may be uncoupled from local hemodynamics. For example, altered brain states such as seizures (8) and cortical spreading depression (9) result in complex and atypical physiological responses that do not fit standard models of neurovascular coupling. It is essential, therefore, to investigate both the direct neural effects of TMS and the relationships among neural, vascular, and metabolic parameters.

To provide an integrated view of the basic effects of TMS, we used several complementary techniques in a controlled physiological preparation. We applied short TMS pulse trains to the visual cortex of the anesthetized cat (n =8) while simultaneously measuring tissue oxygen and neural activity (1012). In separate experiments, we used 570-nm optical imaging of intrinsic signals to measure changes in total hemoglobin (Hbt) within the cortical vasculature (1214). Each trial in our experimental paradigm (Fig. 1) included a pre-TMS baseline (40 s), a short TMS pulse train (1 to 4s, 1 to 8 Hz), and along recovery period (5 to 15 min). Throughout the trial, we alternated visual stimulation with a blank screen to assess the effects of TMS on both evoked and ongoing spontaneous activity (12). Elevation of spike rates during visual stimulation also permitted detection of signal decreases (12, 15).

Fig. 1.

TMS and visual stimulation paradigm. (A) Timeline of a sample trial showing stimulus presentations (green) and inter-stimulus intervals (ISIs) (purple). The visual stimulus was a high-contrast grating displayed for 2 s at intervals of 8 s. TMS (gray box) was applied during an ISI. TMS pulse trains were varied in frequency and duration on separate trials. Single-unit spikes (black ticks), LFP (not shown), and tissue oxygen (not shown) were recorded continuously; activity during TMS was not analyzed because of artifact contamination (fig. S3A). (B) The full TMS trial. Evoked activity represents neural responses during stimulus presentations, and spontaneous activity represents responses that occurred during ISIs.

The neural effects of TMS application are shown in Fig. 2, A and B. Aninitial repeated-measures analysis of variance (ANOVA) on firing rate indicated significant main effects for activity state (spontaneous versus evoked, F241,484 = 65.073, P <10–13) and time (1 to 20 s after TMS, 30 to 90 s, or 180 to 210 s, F241,484 = 3.473, P < 0.05), as well as a significant interaction between these factors (F241,484 = 9.931, P < 0.0001) (12). Accordingly, post hoc tests (Wilcoxon signed-rank) revealed differential response time courses between activity states. Across the population, the spontaneous spike rate increased substantially (∼200%) immediately after TMS (Fig. 2A, left) and remained elevated for ∼60 s (P < 0.001; fig. S4A; fig. S5A, left). In contrast, the evoked firing rate (Fig. 2A, right) showed an immediate decrease (∼50%) and remained significantly suppressed for more than 5 min (P < 0.0001; fig. S4B; fig. S5A, middle). Analogous changes occurred in the power of local field potentials (LFPs) (Fig. 2B), although a distinction was evident with regard to frequency band (fig. S5B). Spontaneous LFP power at higher frequencies (>40 Hz) showed immediate enhancement, whereas lower frequencies (<40 Hz) exhibited a prolonged reduction, similar to evoked activity. This distinction is likely related to the different physiological processes reflected by these frequency ranges (12, 16).

Fig. 2.

Effects of TMS on neural, oxygen, and optical imaging signals. Shown are average time courses of (A) spiking activity, (B) LFP power, (C) tissue oxygen, and (D) total hemoglobin (Hbt) before and after TMS (gray box). All signals are expressed as a percent change from their pre-TMS baselines. Shaded areas represent ±1 SEM. (A) Spontaneous (left) and evoked (right) spiking activity (n = 47 cells). (B) Spontaneous (left) and evoked (right) LFP power (n = 42 sites). (C) Tissue oxygen (n = 21 sites). (D) Hbt (n =3 animals). Insets in (C) and (D) show initial increases. Time periods containing TMS artifacts were removed (fig. S3B). In (D), Hbt was measured by recording the change in 570-nm light reflectance (ΔR/R) from the cortical surface (upper right); scale bar, 1 mm.

To determine how neural changes are reflected in metabolic and vascular signals, we examined measurements of tissue oxygen colocalized with the neural recordings. A repeated-measures ANOVA showed a significant main effect for time (F111,224 = 56.609, P <10–16) but no effect for activity state (F111,224 = 0.0001, P > 0.98; fig. S6B). Therefore, oxygen was further analyzed as a single continuous variable (12, 17). Post hoc Wilcoxon signed-rank tests revealed a biphasic response pattern for oxygen (Fig. 2C). An immediate increase peaked at 10 to 15 s after TMS (Fig. 2C, inset; P < 0.001) and was followed by an extended reduction lasting over 2 min (P < 0.01). Separate measurements of Hbt (Fig. 2D) revealed a similar response: a peak at 10 s (Fig. 2D, inset; P <10–7) and a subsequent prolonged decrease (over 1 min, P < 0.001). This independent data set confirms that changes in blood flow underlie a substantial component of the oxygen response.

The above ANOVAs also revealed significant main effects of pulse frequency (1, 4, or 8 Hz) on neural (F241,484 = 3.522, P < 0.05) and oxygen (F111,224 = 5.739, P < 0.005) data. This raises the possibility that neural and hemodynamic response components covary with stimulation parameters. During the initial response component (<20 s), an increase in TMS pulse frequency caused a monotonic increase in the amplitude of the early oxygen peak and the level of spontaneous neural firing (Fig. 3A, upper right quadrant). At later time points (30 to 90 s), reductions in both tissue oxygen and evoked spiking were larger with higher pulse frequencies (Fig. 3A, lower left quadrant). A more limited data set for pulse train duration showed an analogous trend (fig. S7B). These data suggest that the physiological effects of TMS increase in a dose-dependent manner within this regime of TMS application.

Fig. 3.

Covariation between neural and oxygen data. (A) Changes in spiking activity and oxygen as a function of TMS pulse frequency. Neural activity was indexed by spontaneous spiking during the initial phase (0 to 20 s after TMS) and by evoked spiking during the later phase (30 to 90 s) (15). Error bars in this and subsequent panels represent ±1 SEM; where error bars are not visible, the error was smaller than the plot symbol. (B) Time-lag correlation between oxygen and neural signals (left: spiking activity, n = 117 trials; right: LFP power, 1 to 150 Hz, n = 77 trials). Positive time lags indicate a shift of the neural signal forward in time relative to the oxygen signal. Neural-oxygen correlations were performed for evoked spiking and LFP activity (green) and for spontaneous LFP signals (purple); a similar analysis with spontaneous spiking could not be performed because of low baseline firing rates. Correlation coefficients above the dashed lines are significant over the population (P <0.05, t test). Asterisks denote correlations at positive time lags that are significantly greater than those at negative delays (P < 0.05, paired t test). (C) Neural-oxygen correlation magnitude across bands. LFP bands are defined as follows: δ (delta; 1 to 4 Hz), θ (theta; 4 to 8 Hz), α (alpha; 8 to 12 Hz), β (beta; 12 to 20 Hz), γ (gamma; 20 to 80 Hz), hγ (high-gamma; 80 to 150 Hz), all (1 to 150 Hz). (D) Neural-oxygen correlation latencies across bands.

The relationship between decreases in oxygen and neural activity is consistent with recent studies of negative hemodynamic responses (18, 19). However, reductions in oxygen may be a cause of neural suppression rather than a consequence. In this scenario, normal neural function would be limited by hypoxic conditions (12). To investigate this possibility, we performed a time-lag correlation analysis of simultaneously acquired neural and oxygen data (fig. S8). Both spike rate (Fig. 3B, left) and LFP power (Fig. 3B, right) showed significant correlations with oxygen across a broad range of time lags (P < 0.05, t test). Notably, correlation coefficients were significantly greater at time lags in which the neural signal was shifted forward in time (P <0.05, paired t test). Furthermore, LFP-oxygen correlations were band-specific with regard to magnitude and latency. Gamma and high-gamma bands exhibited the strongest correlations (Fig. 3C), as reported in previous studies of hemodynamic coupling (20). Higher-frequency bands also exhibited peak correlations at shorter latencies (Fig. 3D). This trend, which was most pronounced for spontaneous activity, resulted from initial response increases present in higher-frequency but not lower-frequency bands (Fig. 2B, left). These analyses, along with additional experiments (fig. S9) (12), suggest that oxygen responses follow neural activity in a manner consistent with neurovascular coupling (2124).

A striking aspect of our data is the long duration of neural and hemodynamic changes given the short application of TMS. Although most human studies using similar stimulation paradigms have demonstrated short-term effects, several studies have noted changes in cortical excitability on the order of minutes (25, 26). Human studies using longer-duration stimulation have shown effects lasting hours or even days (27, 28). Such long-term changes in neural function are thought to develop via spike timing–dependent plasticity (27, 29, 30). Notably, alterations in synaptic efficacy have been linked to changes in the temporal relationship between spikes and LFP oscillations (31, 32). To examine our data for a link between spike timing and long-term neural changes, we performed an additional analysis of phase relationships between single-unit spikes and LFP oscillations (12, 33). For pre- and post-TMS time windows, we quantified the degree of phase locking from the distribution of LFP phases at which spikes occurred (Fig. 4A). A striking example of TMS-induced changes in phase distributions is shown for spontaneous activity in Fig. 4B. Compared to the pre-TMS baseline (blue), spike timing relative to the theta oscillation was strongly desynchronized, as evidenced by the increased spread of the distribution after TMS (red). Across all frequency bands, spontaneous activity showed significant reductions in phase locking within the first 30 s after TMS (Fig. 4C, left, P < 0.05, randomization test). By 90 s, this index approached baseline values, and in the gamma band it actually exhibited a significant increase (P < 0.05). Somewhat similar effects were present in evoked activity (Fig. 4C, right). Phase locking to oscillations in the delta band were strongly reduced, whereas increases were present in both the gamma and high-gamma bands (P < 0.05). The capacity of TMS to disrupt precise timing of signals between interconnected neurons advocates its ability to alter brain plasticity (27, 29, 30) in a number of neuropsychiatric contexts (4).

Fig. 4.

Effects of TMS on spike timing relative to LFP oscillations. (A) Illustration of phase locking between spikes (red) and LFP (black). During periods of high phase locking (top), spikes occur at consistent phases in the LFP (left), and the resulting phase distribution is narrow (right). (B) Example of a TMS-induced change in phase locking. Before TMS (blue), spontaneous spikes occur more frequently at preferred phases of theta-band oscillation. In the first 30 s after TMS (red), the phase distribution broadens, indicating a decrease in phase locking. (C) Changes in phase locking across LFP frequency bands for spontaneous (left) and evoked (right) activity. Change in phase locking was determined by comparing the vector strengths (one minus the circular variance) of phase distributions before and after TMS. Light bars show changes in the first 30 s after TMS; dark bars show changes at 60 to 90 s. Asterisks indicate significance (P < 0.05, randomization test).

Consistent with previous work (29, 34, 35), our results reveal long-lasting neural and hemodynamic consequences of TMS that covary with stimulation duration and frequency. In contrast, other studies have reported a distinction where- by low-frequency stimulation (≤1 Hz) causes suppression and high-frequency stimulation (≥8 Hz) leads to facilitation (5). However, this division appears to be oversimplified (5, 36). The precise effects of brain stimulation are fundamentally dependent on many factors (37). For example, several groups have found that identical TMS paradigms elicit opposite physiological effects when applied to neighboring cortical regions (34, 38) or different subjects (36). Within a single site, TMS can produce differential effects depending on the activity state to which stimulation is paired (36, 39). Such reports of variability and state-dependence reveal the complex action of TMS, yet also hint at its potential flexibility as an interventional technique.

Harnessing this potential requires the ability to measure the precise neural effects of TMS over different brain regions and time intervals. Our findings show that TMS-induced modifications of neural activity are readily observed in cerebral hemodynamics, which can be detected by standard neuroimaging techniques. This result confirms recent combined TMS-fMRI studies in which correlations were reported between TMS-induced behavioral changes and hemodynamic signals in functionally related brain regions (39, 40). The capacity of brain imaging to monitor the temporal progression of physiological changes induced by TMS may prove highly beneficial for the development and optimization of both basic neuroscience and clinical applications.

Supporting Online Material

www.sciencemag.org/cgi/content/full/317/5846/1918/DC1

Materials and Methods

Figs. S1 to S9

References

References and Notes

View Abstract

Navigate This Article