Mechanosensory encoding dysfunction emerges from cancer-chemotherapy interaction

Abstract

Persistent sensory, motor and cognitive disabilities comprise chemotherapy-induced neural disorders (CIND) that limit quality of life with little therapeutic relief for cancer survivors. Our recent preclinical study provides new insight into a condition impacting the severity of chronic CIND. We find that sensorimotor disability observed following cancer treatment exceeds that attributable to chemotherapy alone. A possible explanation for intensified disability emerged from evidence that codependent effects of cancer and chemotherapy amplify defective firing in primary sensory neurons supplying one type of low threshold mechanosensory receptor (LTMR). Here we test whether cancer's modification of chemotherapy-induced sensory defects generalizes across eight LTMR submodalities that collectively generate the signals of origin for proprioceptive and tactile perception and guidance of body movement. Preclinical study enabled controlled comparison of the independent contributions of chemotherapy and cancer to their clinically relevant combined effects. We compared data sampled from rats that were otherwise healthy or bearing colon cancer and treated, or not, with human-scaled, standard-of-care chemotherapy with oxaliplatin. Action potential firing patterns encoding naturalistic mechanical perturbations of skeletal muscle and skin were measured electrophysiologically in vivo from multiple types of LTMR neurons. All expressed aberrant encoding of dynamic and/or static features of mechanical stimuli in healthy rats treated with chemotherapy, and surprisingly also by some LTMRs in cancer-bearing rats that were not treated. By comparison, chemotherapy and cancer in combination worsened encoding aberrations, especially in slowly adapting LTMRs supplying both muscle and glabrous skin. Probabilistic modeling best predicted observed encoding defects when incorporating interaction effects of cancer and chemotherapy. We conclude that for multiple mechanosensory submodalities, the severity of encoding defects is modulated by a codependence of chemotherapy side effects and cancer's systemic processes. We propose that the severity of CIND might be reduced by therapeutically targeting the mechanisms, yet to be determined, by which cancer magnifies chemotherapy's neural side effects as an alternative to reducing chemotherapy and its life-saving benefits.

Keywords: cancer treatment; chemotherapy; cutaneous; neurotoxicity; proprioception; sensorimotor abnormalities; sensory encoding.

FIGURE 1

Impaired Muscle Mechanosensory Neuron Population Code After Cancer Treatment. (A–D) Representative cases of spiking activity in control (grey), OX (blue), cancer (red), and cOIN (purple) as a measure of sensory encoding in spindle group Ia (A),unclassified spindle (B), group Ib (C), group II (D). Blackcircles plot instantaneous firing rates (IFR) measures as pulses per second (pps) of corresponding spike (action potential: vertical lines) intervals. Dashed vertical line marks onset of muscle stretch [3 mm from resting length (Lo)] shown in bottom trace divided into dynamic and static phases by dark grey (150 ms duration after stretch command onset) and light grey (1 s duration after the dynamic phase) bars.

FIGURE 2

Co-Impairment of Key Muscle Mechanosensory Features. Quantification of clusters of encoding parameters: detection threshold: during fast (A) and slow (B) stimuli, static average (C), number of spikes during static (D) and dynamic stimuli (E), and peak dynamic firing rate (F), averaged from four trials, from each neuron in each of the neuronal classes. * indicates statistically significant differences as empirically derived from hierarchical Bayesian model (stan_glm): 95% highest density intervals do not overlap.

FIGURE 3

Latent encoding space reveals unique effects of cancer, chemotherapy and their combination across divergent classes of propriosensors. Principal component (PC) analysis applied to all (n = 31) encoding parameters measured in response to natural stimuli for each neuron class emphasizing: spindle group Ib (A), group II (B), group Ia (C), unclassified spindle (D) with greater opacity. Neuron class means (average of PC1-2 coordinates across all neurons in a given class and treatment group: dark colored circles) are visualized in the new latent encoding space created by PC1–2. Color-coded least-squares elliptical fitting (95% confidence) was computed to emphasize differences between neuron classes and experimental groups. Labelled arrows in (A) indicate the magnitude and direction of cancer, chemotherapy, or their combined effects. (E) Scree plot indicating percentage of explained variance by each PC (grey bar, eigenvalue in %). (F) Corrplot indicates the contribution of variables on the factor map for PC1-5. Larger values indicate the components contribute a larger relative portion, indicating components are of greater importance. (G,H) Summarize the effects of shifting along principle axes in latent encoding space.

FIGURE 4

Hierarchical Bayesian modeling. Tests for significant group differences in LD1 scores reconfigured to operate in a predictive fashion in spindle group Ia (A), group II (B), unclassified spindle (C), group Ib (D). Predictors included in each model are listed to the right of each subplot (1–4). Generative models in utilizing one (1 and 2) or both independent 3) predictor(s) for posterior prediction. The generative model in 4) utilizes both independent predictors and an interaction term for posterior prediction. Grey lines in represent 500 novel (generative) samples drawn from the posterior distributions. Black lines illustrate experimentally observed mean LD1 score. Predictive accuracy was measured by calculating expected log predictive density (ELPD) for each model and benchmarked off of the highest performing model. Delta ELPD (Δ) indicates difference from optimal model. Negative models represent worse predictive performance.

FIGURE 5

Impaired Cutaneous Mechanosensory Neuron Code After Cancer Treatment. Representative cases of spiking activity in control (grey), OX (blue), cancer (red), and cOIN (purple) as a measure of sensory encoding in slowly adapting type I SA1: Merkel corpuscles; (A) and type II SA2: Ruffini endings; (B) and rapidly adapting Meissner corpuscles RA1; (C) and Pacinian corpuscles RA2; (D). Black circles plot instantaneous firing rates (pps) of corresponding spike (action potential) intervals. Solid line below voltage recordings indicates the dynamic and static components of the natural stimulation paradigm, i.e. displacement of the plantar skin in vivo utilized to study the cutaneous neurons (1 mm from resting length: Lo).

FIGURE 6

Co-Impairment of Key Cutaneous Mechanosensory Features. Quantification of clusters of encoding parameters: detection threshold (A), dynamic spike encoding (B) last spike time (C), number of spikes during static (D) and spike variability during static stimulus presentation (E) averaged from four trials, from each neuron in each of the neuronal classes. * indicates statistically significant differences as empirically derived from hierarchical Bayesian model (stan_glm): 95% highest density intervals do not overlap.

FIGURE 7

Chemotherapy for cancer impairs cutaneous population encoding. Population code of 20 neurons recorded from control (grey, (A), OX (blue, (B), cancer (red, (C), and cOIN (purple, (D). Raster plots bottom panels: (A–D) stack representative firing responses of 20 different afferents (1 afferent per row) aligned on identical ramp-hold-release displacement applied to each afferent’s glabrous cutaneous receptive field. Top traces show average firing rate (solid colored lines) and standard error (shaded region) computed per experimental group from the corresponding raster. Population codes for each group were constructed from the same distribution of afferent types: seven slowly adapting type I (SAI: Merkel corpuscles), 3 type II (SAII: Ruffini endings), seven rapidly adapting Meissner corpuscles (RA RA1) and three Pacinian corpuscles (RA2 PC).