Experimental Study Methodology

Shown is a high-level overview of the study used to validate the frequency-dependent effects of TTNS predicted by the computational model. Participants were healthy adults (18+) asked to abstain from any nicotine/caffeine for 12 hours, and any fluids for 2 hours before the study. Upon arrival they were asked to ingest 750 ml water after emptying their bladder. A 30 min digestion period was employed before stimulation. Participants were pseudo-randomly allocated between groups, and were blind to the condition. Created with BioRender.com/p01p196.

Overview of the Computational Bladder Control Model

A. Shown is a block diagram of the simulated neuronal circuit and bladder model. The model used a modified biophysical representation of the bladder produced previously (dashed box, [23]) which used the firing rates of the pelvic (blue), hypogastric (orange), pudendal (green) efferents to calculate bladder state. Tibial input used to modulate the circuit shown in purple. PMC: Pontine Micturition Centre, PAG: Periaquaductal Grey, PGN: Preganglionic Bladder Neurons, ASC: Ascending Interneurons, PUD: Pudendal Afferent, Onuf: Onuf’s Nucleus, Pel: Pelvic Efferent, Hyp: Hypogastric Efferent. Scissors represent possible severed projections. Labels 1 - 5 represent key regions of tibial modulation using opioidergic (1 - 4) or classical (5) inhibitory mechanisms. Created with BioRender.com/z61j143.

Methodological approach used to obtain animal data

A. Experimental setup and recording site used to obtain bladder data and associated afferent neural signal. B. Bandpass-filtered raw neural signal (top trace), corresponding bladder pressure (middle trace), and smoothed continuous neural firing rate (1st order Butterworth filter with a 1 Hz cutoff frequency, bottom trace) obtained from the experimental setup. Created with BioRender.com/g49j208.

Experimental Study Results

A: effects of low (1 Hz), high (20 Hz), and placebo (0Hz) TTNS stimulation on the time elapsed to first sensation of urge. P-values obtaine from post-hoc Mann-Whitney U tests. B: effects of the runoff period on self-reported urge intensity. Error-bars = 95% Confidence Interval. Urge intensity self-reported on a scale of 0–4 (see methods), runoff period was 10 minutes in duration. C: bivariate and univariate kernel-density estimate plot displaying the relationship between the time elapsed before participants reported the urge to urinate, and the self-reported intensity of the urge at this stage (i.e., before the additional 10-minute runoff period).

Behavior of the Simulated Bladder Model.

Shown is the bladder behavior (top trace) and associated efferent neuronal activity (middle trace) recorded over a 1000 second simulation run. Blowout boxes contain 5 second windows of the full recorded activity (highlighted in red) during filling (left box) and voiding (right box). Each raster trace was obtained from a randomly selected neuron within the Pudendal, Hypogastric, and Pelvic units (Nneurons = 100 in each case).

Computationally Modeling TTNS.

A. Effects of low (1 Hz) and high (20 Hz) frequency TTNS on simulated bladder function compared to baseline (unmodulated) conditions. Total elapsed sim-cycles equate to 1000s total simulation time. B. Frequency dependent effects of TTNS on bladder contraction. Shown is the average total bladder contraction duration for a 500s simulation period under 21 different stimulation frequencies (0-20 Hz, in 1 Hz increments, Nrepeats = 10). Errorbar= 95%CI. C. Effects of disconnecting specific tibial-nerve projections on total contraction duration (for a 500s simulation period) under low-frequency (1 Hz, i) and high-frequency (10 Hz, ii) conditions. In both cases, baseline behavior represents the unmodulated behavior of the system. = p < 0.05, = p < 0.01, = p < 0.001, = p < 0.0001 Mann-Whitney-U test, Nrepeats = 10 in each condition.

Example urge-intensity survey.

Shown is an example of the survey which was given to participants in printed form.

Model Performance During Fitting Process.

Shown is the overlap between ground truth (blue) and simulated (orange) afferent neural activity for a subset of bladder pressure data. Model performance is shown in a random unoptimized state (left), after 200 rounds of Bayesian optimization (center), and after final manual optimization of model weights (right). The magnitude of the overlap between simulated and ground truth data at each stage shown as Normalized Root Mean Square Error (NRMSE). Smoothed firing rate (1st-order Butterworth filter, 1 Hz cutoff frequency) was normalized against the maximum recorded value in each dataset.

Bayesian Optimization Convergence Plot.

Shown is a convergence plot detailing the Normalized Root Mean Square Error (NRMSE) of the ground truth vs. simulated afferent activity data at each phase of the optimization process.

Final Neuronal Model Parameters.

Where possible parameters were matched to the original specifications of the neuronal/synaptic model. Parameters that were altered by the model fitting process are marked as *.