Graded recruitment of pupil-linked neuromodulation by parametric stimulation of the vagus nerve

Abstract

Vagus nerve stimulation (VNS) is thought to affect neural activity by recruiting brain-wide release of neuromodulators. VNS is used in treatment-resistant epilepsy, and is increasingly being explored for other disorders, such as depression, and as a cognitive enhancer. However, the promise of VNS is only partially fulfilled due to a lack of mechanistic understanding of the transfer function between stimulation parameters and neuromodulatory response, together with a lack of biosensors for assaying stimulation efficacy in real time. We here develop an approach to VNS in head-fixed mice on a treadmill and show that pupil dilation is a reliable and convenient biosensor for VNS-evoked cortical neuromodulation. In an 'optimal' zone of stimulation parameters, current leakage and off-target effects are minimized and the extent of pupil dilation tracks VNS-evoked basal-forebrain cholinergic axon activity in neocortex. Thus, pupil dilation is a sensitive readout of the moment-by-moment, titratable effects of VNS on brain state.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1. Vagus nerve stimulation (VNS), pupillometry, and two-photon axonal GCaMP imaging in the auditory cortex of awake head-fixed mice.

A Bipolar VNS cuff design (top) and implantation on the left cervical vagus nerve (bottom). B VNS stimulation circuit diagram. Current is applied between the two cuff electrodes using an improved Howland current pump, and return current is measured across a 1-kOhm sensing resistor. The animal and current pump circuit were isolated from the ground (indicated at the bottom), except where noted. C VNS pulse waveforms and parameters. Biphasic pulses, separated by a gap of 50% interpulse interval, were delivered in trains lasting 10 s (top). Five pulse amplitudes, four pulse widths, and three pulse rates were combined to yield 60 unique VNS parameter combinations (bottom). D Photograph of a mouse undergoing pupillometry and VNS in the two-photon imaging setup. E Left: example video frame of the mouse’s eye, with fitted ellipses, overlaid on the pupil (red) and exposed eye area (purple). Right: 60 pupil-size time series (all parameter combinations mentioned in panel C) during an example session time-locked to each VNS train (thick line, session mean; thin lines, individual trains). Gray rectangles indicate 10-s window for each train of VNS pulses. F Left: example mean fluorescence intensity image from in vivo two-photon GCaMP6 imaging of cholinergic axons in the auditory cortex. Right: example axonal signal time series, aligned as in panel E. Same session as in panel E. G Walking speed time series from the same session as in panels E and F. H Full-session time series of pupil size (red), fluorescence (green), and walking speed (blue) from the same session as in E–G. Narrow vertical gray rectangles indicate the time of each VNS train.

Fig. 2. Pupil dilation has a graded dependence on VNS parameters.

A VNS-evoked pupil response time courses separately for each pulse amplitude (left), width (middle), and rate (right), collapsed across two other stimulation parameters. Gray window, 10 s VNS train; red bars, the interval for VNS-evoked pupil response scalar measures (see “Methods”); data are presented as mean values ± s.e.m. (across VNS events). B VNS-evoked pupil response measure for all 60 unique parameter combinations. Response magnitude is indicated by circle size and color. C VNS-evoked pupil response scalars plotted separately for pulse amplitudes and rates (binned across the width, left) and separately for pulse widths and rates (binned across amplitude, right). Colored lines, fitted log-logistic functions (“Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events; see Supplementary Fig. 2 for boxplot representations of the same data separately per animal). All panels: N = 45 repetitions for each unique parameter combination (before artifact rejection; “Methods”). Source data are provided as a Source Data file.

Fig. 3. VNS-evoked pupil dilation requires an intact nerve and localized current.

A VNS-evoked pupil response measures separately per pulse-charge (amplitude × width) and width, for 5 Hz trains (left), 10 Hz trains (middle), and 20 Hz trains (right). Colored lines, fitted log-logistic function (see “Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events). B Grid showing the width and amplitude parameter combinations that were collapsed into our 5 charge/pulse bins. C VNS-evoked pupil time courses, each trace is a charge/pulse bin and train rate. Gray window, VNS train; red bar, the interval for VNS-evoked pupil response scalar measures (see “Methods”); data are presented as mean values ± s.e.m. (across VNS events). D VNS-evoked pupil response measures separately per charge/pulse bin and train rate. Colored lines, fitted log-logistic function (“Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events); stats, two-sided one-sample t test (tested against 0; ***P < 0.001, **P < 0.01; *P < 0.05, corrected for false discovery rate); blue, green, and brown backgrounds indicate no effects, optimal, and off-target pulse charges, respectively (“Methods”). E, F As C, D, but for double-cut nerve, ungrounded sessions. A–DN = 45 repetitions for each unique parameter combination (before artifact rejection; “Methods”). E, FN = 24 repetitions for each unique parameter combination (before artifact rejection, “Methods”). Source data are provided as a Source Data file.

Fig. 4. Graded VNS causes graded neuromodulation recruitment.

A Left: example session mean fluorescence of cholinergic axons in layer 1 of the auditory cortex. Right: as left, but with an overlaid color map of VNS-evoked calcium responses (across all VNS stimulations, see “Methods”). This map was used as a session-wise region of interest for the analysis of calcium signal parameter dependence. This is a representative example of 12 independent repetitions (sessions). B VNS-evoked calcium response measures (“Methods”) for all 30 unique parameter combinations. Response magnitude is indicated by circle size and color. C VNS-evoked calcium responses separately for pulse amplitudes (left), widths (middle), and rates (right) collapsed across two other stimulation parameters. Gray window, 10 s VNS train; green bar at the bottom, the interval for VNS-evoked axonal calcium response measures (“Methods”); data are presented as mean values ± s.e.m. (across VNS events; see Supplementary Fig. 6J, K for boxplot representations of the same data separately per animal). D VNS-evoked calcium response measures separately for pulse amplitudes and rates (left, collapsed across widths) and separately for pulse widths and rates (right, collapsed across amplitudes). Colored lines, fitted log-logistic function (“Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events). E VNS-evoked calcium response measures separately per pulse-charge (amplitude × width) and width, 10 Hz trains (left), and 20 Hz trains (right). Colored lines, fitted log-logistic function (see “Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events). F As in D, but for charge/pulse bins. Stats, two-sided paired-samples t test (tested against the response magnitudes in the lowest charge bin; ***P < 0.001, **P < 0.01; *P < 0.05 false discovery rate corrected); blue, green, and brown backgrounds indicate no effects, optimal, and off-target pulse charges, respectively, defined by the pupil response data (as in Fig. 3D, F; “Methods”). B–DN = 12 repetitions for each unique parameter combination (before artifact rejection; “Methods”). Source data are provided as a Source Data file.

Fig. 5. VNS probabilistically causes locomotor activity.

A VNS-evoked walk probability (see “Methods”) separately for pulse amplitudes and rates (collapsed across widths, left) and separately for pulse widths and rates (collapsed across amplitudes, right). Colored lines, fitted log-logistic function (“Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events). B As A, but for charge bins. Stats, two-sided one-sample t test (tested against 0; ***P < 0.001, **P < 0.01; *P < 0.05 false discovery rate corrected); blue, green, and brown backgrounds indicate no effects, optimal, and off-target pulse charges, respectively, defined by the pupil response data (as in Fig. 3D, F; “Methods”). C Walked distance (walk trials only) after VNS train start, separately for pulse amplitudes (left), widths (middle), and rates (right) collapsed across two other stimulation parameters. Gray window, pulse duration; blue bar, the interval for VNS-evoked walking speed measures (“Methods”); data are presented as mean values ± s.e.m. (across VNS events). D, E As A and B, but for walking speed on walk trials only. All panels: N = 27 repetitions for each unique parameter combination (before artefact rejection; “Methods”). Source data are provided as a Source Data file.

Fig. 6. VNS triggered cortical acetylcholine release mediates a large fraction of pupil dilation.

A VNS-evoked calcium response measures, in absence of walking, separately per charge/pulse bin and train rate. Colored lines, fitted log-logistic function (“Methods”); data are presented as mean values ± 1.96 × s.e.m. (across VNS events); stats, two-sided one-sample t test (tested against 0; ***P < 0.001, **P < 0.01; *P < 0.05 false discovery rate corrected); blue, green, and brown backgrounds indicate no effects, optimal and off-target pulse charges, respectively, defined by the pupil response data (as in Fig. 3D, F; “Methods”). B As A, but for VNS-evoked pupil responses during the imaging sessions. C Scatterplot of the relationship between VNS-evoked pupil responses and VNS-evoked calcium responses, in absence of walking. Blue data points are individual VNS trains; gray data points are calcium response defined bins (equal size); red line, linear fit; Pearson correlation, P < 0.001. A first-order (linear) fit was superior to a constant fit (F1,178 = 116.19, P < 0.001), and a second-order (quadratic) fit was not superior to the first-order fit (F1,178 = 1.2, P = 0.28) (sequential polynomial regression; “Methods”). D As C, but after removing the effect of VNS (partial correlation). Pearson correlation, P < 0.001. A first-order (linear) fit was superior to a constant fit (F1,178 = 26.13, P < 0.001), and a second-order (quadratic) fit was not superior to the first-order fit (F1,178 = 0.08, P = 0.78). E Left, schematic of relationship between VNS and pupil responses. Arrow, regressions; coefficient c quantifies the “total effect”. Right, schematic of mediation analysis of VNS to pupil responses, via cholinergic axons responses (“Methods”). Arrows, regressions; coefficient a × b quantifies the “indirect” (mediation) effect; coefficient c’ quantifies the “direct effect”. F Fitted regression coefficients (kernel density estimate of 5 K bootstrapped replicates) of the total effect (brown), indirect path (mediation; pink), and the direct path (gray). Stats, the fraction of bootstrapped coefficients smaller than 0. All panels: N = 12 repetitions for each unique parameter combination (before artefact rejection; “Methods”). Source data are provided as a Source Data file.