Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy Stephen N

Published OnlineFirst April 28, 2020; DOI: 10.1158/0008-5472.CAN-19-2331

CANCER RESEARCH | TRANSLATIONAL SCIENCE

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy Stephen N. Housley1, Paul Nardelli1, Dario I. Carrasco1, Travis M. Rotterman1, Emily Pfahl1, Lilya V. Matyunina1,3, John F. McDonald1,3, and Timothy C. Cope1,2,3

ABSTRACT ◥ For the constellation of neurologic disorders known as chemo- neuron dysfunction. Validated measures of sensorimotor behavior therapy-induced peripheral neuropathy, mechanistic understand- in awake, behaving animals revealed dysfunction after chronic ing and treatment remain deficient. Here, we present the first chemotherapy treatment was exacerbated by cancer. Notably, errors evidence that chronic sensory neuropathy depends on nonlinear in precise forelimb placement emerged as a novel behavioral deficit interactions between cancer and chemotherapy. Global transcrip- unpredicted by our previous study of chemotherapy alone. These tional profiling of dorsal root ganglia revealed differential expres- original findings identify novel contributors to peripheral neurop- sion, notably in regulators of neuronal excitability, metabolism, and athy and emphasize the fundamental dependence of neuropathy on inflammatory responses, all of which were unpredictable from the systemic interaction between chemotherapy and cancer. effects observed with either chemotherapy or cancer alone. Systemic interactions between cancer and chemotherapy also determined the Significance: These findings highlight the need to account for extent of deficits in sensory encoding and ion channel protein pathobiological interactions between cancer and chemotherapy as a expression by single mechanosensory neurons, with the potassium major contributor to neuropathy and will have significant and ion channel Kv3.3 emerging as one potential contributor to sensory immediate impact on future investigations in this field.

Introduction published meta-analysis identified 341 preclinical studies of chemotherapy induced neuropathy, and none assessed interaction between Chemotherapy can achieve high rates of survival in patients with chemotherapy and cancer (10). Omitting cancer from preclinical study common cancers (1) but often causes severe side effects, including produces a fundamental gap in knowledge that may explain why neuropathy, that can limit its use (2, 3). Debilitating sensory disorders, treatments for neuropathic side effects of chemotherapy have been including pain, paraesthesia, and somatosensory loss, reduce quality of unsuccessful in patients with cancer. Cancerous tumors located out- life for many patients and can persist for months or years after side the nervous system can induce cognitive disability and peripheral discontinuing chemotherapy (4, 5). These neurologic disorders occur nervous system dysfunction independent of chemotherapy, surgery, or for up to 80% of patients receiving commonly used antineoplastic comorbidities (11, 12). Moreover, cancer provokes dysregulation in agents, notably antitubulins and proteasome inhibitors, as well as immune, metabolic, oxidative, and neuronal excitability (13, 14), all of platinum-based compounds (5, 6), which are the prescribed adjuvant which are identified as targets through which chemotherapy produces treatment in 50% of cancer cases worldwide (7, 8). There are few neuropathy (2). Convergence of cancer and chemotherapy on the same options available for the prevention or treatment of sensory disorders biological processes seems likely to yield nonlinear interactions, and those that exist largely focus on symptomatic management not leading us to hypothesize that clinically relevant neuropathy emerges prevention (9). Thus, there is an urgent need to better understand the from codependent actions of cancer and chemotherapy. pathogenesis of sensory dysfunction to aid development of mecha- We tested our hypothesis by uncoupling the independent and nism-based therapies. combinatorial effects of cancer and chemotherapy on sensory neurons Preclinical studies of chronic neuropathy in experimental models of in studies of rats that are not feasible in human patients. Here we cancer are unavailable, possibly because of presumption that chemo- present original findings to show that sensory neuropathy depends on therapy alone is sufficient to explain the neuropathology. A recently complex systemic interactions between cancer and chemotherapy. By combining cancer and chemotherapy, we reproduce the relevant clinical condition in preclinical study of rats to produce the first 1School of Biological Sciences, Georgia Institute of Technology, Atlanta, Georgia. transcriptional profile of nervous tissue, to discover recurrent meta- 2 W.H. Coulter Department of Biomedical Engineering, Emory University and bolic reprogramming and perturbed inflammatory and ion channel Georgia Institute of Technology, Georgia Institute of Technology, Atlanta, processes distinct from the independent effects of cancer or chemo- Georgia. 3Integrated Cancer Research Center, Parker H. Petit Institute for Bioengineering and Bioscience, Georgia Institute of Technology, Atlanta, therapy. Parallel codependency is found in novel neuropathic Georgia. responses of mechanosensory neurons, both in their IHC for selected voltage-gated potassium ion channels and in electrophysiologic mea- Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/). sures of encoding naturalistic stimuli in vivo. Collectively, our data show for the first time that sensory neuropathy emerges from code- Corresponding Author: Timothy C. Cope, Georgia Institute of Technology, 555 14th St. NW, Atlanta, GA 30318. Phone: 404-385-4293; Fax: 404-894-9982; pendencies between the systemic effects of cancer and chemotherapy. E-mail: [email protected] Apart from identifying new molecular targets for restoring function to mechanosensory neurons, our findings provide the fundamental Cancer Res 2020;80:2940–55 insight that accounting for chemotherapy–cancer combinatorial doi: 10.1158/0008-5472.CAN-19-2331 effects on the somatosensory nervous system is prerequisite for 2020 American Association for Cancer Research. developing meaningful treatment or prevention of neuropathy.

AACRJournals.org | 2940

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

Materials and Methods roots (lumbar L4–6), muscles, and nerves in the left hindlimb were fi Animal care prepared for stimulation and recording with the rat xed in a rigid frame at the snout, vertebral bodies, distal tibia, and distal femur (knee All procedures and experiments were approved by the Georgia angle 120). Triceps surae muscles (lateral and medial gastrocnemii Institute of Technology Institutional Animal Care and Use Commit- and soleus) were partially freed of surrounding connective tissue and tee. Adult male rats (250–350 g) were studied in terminal experiments marked for their resting length (L ) at ankle angle 90 before their only and were not subject to any other experimental procedures. All o common Achilles tendon was severed at the calcaneus and tied directly animals were housed in clean cages and provided food and water ad to the lever arm of a force- and length-sensing servomotor (model libitum in a temperature- and light-controlled environment in Georgia 305B-LR; Aurora Scientific). Triceps surae nerves were loosely posi- Institute of Technology's Animal facility. tioned in continuity on a unipolar silver stimulating electrode, and all other hindlimb nerves were crushed to reduce extraneous neuronal Rat model of colorectal cancer activity. Dorsal rootlets were carefully freed in continuity from over- We studied the rat model of colorectal cancer developed in Fisher Pirc/þ lying connective tissue and supported on bipolar hook electrodes 344 (F344) rats and known as polyposis in rat colon (Apc ; ref. 15). positioned close to the rootlet's entry into the dorsal spinal cord. Pirc rats generated from a germline mutation in the Apc gene 1137/þ Exposed tissues were covered with warm mineral oil in pools formed (Apcam ) approximate human colorectal cancer. We studied by attaching the edges of severed skin to the recording frame. age-matched male rats (4 months). This age has been previously validated to result in 100% incidence of animals (16) developing colon In vivo intracellular recording cancer and was consistent with our population (Supplementary Pirc/þ Dorsal rootlets positioned in continuity on bipolar recording Fig. S1). Apc rats exhibit regional distribution of tumors that is electrodes were selected for sampling sensory neurons when they best seen in human colorectal cancer (16), histopathology and mor- produced robust action potential activity in response to stretch of phology closely resembling human tumors (adenocarcinomas; ref. 17), triceps surae muscles. Individual axons penetrated in these rootlets by upregulation of well-established progrowth factors, for example, glass micropipettes (30 MW filled with 2 mol/L Kþ acetate) were b -catenin and EGFR, and proliferative markers (Ki-67) seen in human selected for study when electrical stimulation of triceps surae nerves carcinogenesis (Supplementary Fig. S1; ref. 15), dramatic increases in produced orthodromic action potentials that were readily resolvable Wnt signaling believed to be a key factor in human colorectal and had conduction delay of <2 ms. Continuous intracellular record- carcinogenesis, presence of apoptotic-resistant cells unique to carci- ings from sensory neurons were acquired with Spike2 software (ver- nogenesis (15, 16), oxaliplatin induction of apoptosis in colon tumors, sion 8.02). Sensory neurons were classified as Ia by their perfect – survivability (12 15 months) enabling long-term investigation of entrainment to 1-second bouts of high-frequency, small-amplitude oxaliplatin's chronic effects (15). vibration (100 Hz, 80 mm), pause in firing during rising twitch force response (muscle shortening), and by responding with an initial burst Chemotherapy treatment of high-frequency firing [>100 pulses per second (pps)] at the onset of Oxaliplatin was injected intraperitoneally once a week (10 mg/kg, muscle stretch (24). Two stretch paradigms were used to characterize 1 mL 5% dextrose in DMSO) in age-matched rats (4 months) to the firing responses of afferents to physiologically relevant mechanical achieve a cumulative dose of 70 mg/kg over 7 weeks (Fig. 1A), which 2 stimuli. In both paradigms, triceps surae muscles were stretched by scales to a human dose of 420 mg/m (conversion based on rat body 3 mm from L (7% strain). Ramp–hold–release stretches tested afferent ¼ o surface area and Km 6; ref. 18). This dose minimizes nerve encoding of both fast dynamic (20 mm/second, 47% strain rate) and degeneration in patients (19, 20). Throughout the entirety of the static stimuli; successive triplets of triangular stretch tested slow (4 experimental timeline, rats were visually and physically (palpation) mm/second, 9% strain rate) dynamic and activity-dependent encoding inspected for pain, distress, or signs of local infection. No individual rat in dynamic stretch known to be influenced by recent signaling reached set criteria established for early removal from the study: history (24). Strains and strain rates fall within values expected for greater than 10% weight loss, pain, or distress assessed by vocalization, animals engaged in normal activities, for example, locomotion, and absence of grooming, and rats show normal exploratory and feeding have been previously used in our laboratory (24). behaviors (Supplementary Fig. S2). In contrast with previous fl reports (21), no animals exhibited signs of local in ammation, bloat- Tissue collection fi fi ing, or infection, nor did we nd any evidence of brino-purulent At the conclusion of data collection, rats were overdosed with peritonitis during necropsy to recover tumors and other tissues. isoflurane inhalation (5%) then immediately transcardially perfused with cold vascular rinse (0.01 mol/L PBS pH 7.4) followed by room Surgical procedures temperature fixative (2% paraformaldehyde in 0.1 mol/L PBS, pH 7.4). Terminal in vivo experiments were performed 5 weeks after achiev- Muscles and dorsal root ganglia were quickly dissected and postfixed ing clinically relevant chemotherapy doses. They were designed to for 1 hour in the same fixative. After a brief wash in a 0.1 mol/L PBS, measure the firing of individual intact sensory neurons in response to tissues were incubated in 0.1 mol/L PBS containing 20% sucrose at 4C physiologically relevant muscle contraction and stretch with electro- overnight. Three to four animals from each of the four groups were physiologic techniques. All in vivo procedures are well established in selected for whole-transcriptome analysis alone and did not undergo our laboratory and have been extensively described in previous further analysis. Dorsal root ganglia (DRG) from the lumbosacral (L5- publications (22–24). Briefly, rats were deeply anesthetized by inha- S2) spinal cord were surgically removed (4–6 per animal), washed with fl lation of iso urane (5% in 100% O2), intubated via a tracheal cannula, sterile saline and immediately flash frozen (fresh) in 2-methylbutane and then maintained for the remainder of the experiment (up to (isopentane) and prechilled in liquid nitrogen. DRG were then stored – 12 hours by 1.5% 2.5% in 100% O2). Respiratory rate, pCO2, core at 80 C for further analysis (see below). After DRG removal, colonic Pirc/þ temperature, pulse rate and pO2 were continuously monitored and tumors from Apc rats were dissected and counted prior to maintained by adjusting anesthesia and adjusting heat sources. Dorsal postfixation for IHC analysis.

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2941

Housley et al.

IHC and imaging cal clustering. Hierarchical clustering and heatmaps were visualization Details of our IHC procedures have been described previously (25). with the gplots package (heatmap.2; ref. 32) in the R environment Briefly, 50-mm–thick sections of skeletal muscles, DRGs, and tumors (3.5.0). Hierarchical clustering, using the ward.D2 method (33), was were cut using a Cryostat (Leica). Tissues sections from all treatment applied to the eBayes-filtered gene set to obtain unsupervised visual- groups were processed simultaneously. All tissue sections were incu- ization of the gene clusters coordinately expressed among the different bated overnight in primary antibodies diluted in blocking buffer (5% experimental groups. Prior to hierarchical clustering, genes were normal goat serum, 0.3% Triton100 in PBS). The following primary standardized across the samples to a mean of zero and an SD of 1. antibodies used were: rabbit polyclonal anti-NaV1.6 (ASC-009, Alo- Subclusters were derived using the R function cutree (34). mone Laboratories, 1:200), rabbit polyclonal anti-NaV1.1 (ASC-001, Alomone Laboratories, 1:300), rabbit polyclonal anti-NaV1.7 (ASC- Pathway and gene-set enrichment analysis 008, Alomone Laboratories,1:100), rabbit polyclonal anti-Kv3.3 (APC- Hypergeometric testing was conducted on subclusters of named 102, Alomone Laboratories, 1:300), mouse monoclonal anti-Kv1.1 genes using DAVID (Database for Annotation, Visualization, and (K36/, NeuroMab, University of California Davis, Davis, CA; 1:300), Integrated Discovery) and g:Profiler to pinpoint significantly guinea pig polyclonal anti-VGLUT1 (135–104 Synaptic Systems, enriched pathways or gene ontology terms (35). We increase 1:300), chicken polyclonal anti-neurofilament protein (NF-H, Aves statistical power by hypergeometric testing on subclusters, which Laboratories, 1:200), mouse monoclonal anti-ASIC2, (E-20, Santa separates coregulated (up or down) probe sets (36). All biological Cruz Biotechnology,1:200), mouse monoclonal anti-EGRF (SC-120, processes with adjusted P < 0.05 were considered significantly Santa Cruz Biotechnology, 1:100), mouse monoclonal anti-KI67 (SC- enriched. g:Profiler 3–500 (min–max) Benjamini–Hochberg FDR 23900, Santa Cruz Biotechnology, 1:100), mouse monoclonal antiIL6 correction intersection terms (2) only annotated genes were (SC-57315, Santa Cruz Biotechnology, 1:100). After washing with PBS, searched. Parallel identification of pathways enrichment was per- tissue sections were incubated with appropriate fluorescent- formed by gene-set enrichment analysis (GSEA). GSEA uncovers conjugated secondary antibodies (Jackson ImmunoResearch Labora- novel sets of functionally related genes by focusing on all detected tories), diluted in blocking buffer, for 1 hour at room temperature. (above-background) genes. GSEA is unbiased and more sensitive, Following washes with PBS, slides were mounted using Vectashield thus allowing for the detection of even subtle enrichment signals. containing DAPI (Vector Laboratories) to label cell nuclei. The Molecular Signature Database v 6.2 (C2-CP, canonical pathways; C3- MIR, miRNA targets; C3- TFT, transcription factor RNA extraction and amplification targets; C5-BP, GO biological process; C5- CC, GO cellular com- RNA extraction and amplification were performed according to our ponent; C5- MF, GO molecular function C7: immunologic signa- previously described methods (26). Briefly, RNAs were isolated and tures gene sets) with the permutation type set to “gene set,” 15–500 purified from sensory neurons of DRG (L5-S2) using the mRNeasy (min–max), to calculate statistical significance, as suggested for Micro KIT (Qiagen). Total RNA concentration and integrity was fewer than seven replicates; default settings were applied to all the assessed on the Bioanalyzer RNA Pico Chip (Agilent Technologies). other options. For GSEA, a false discovery rate (FDR) of <0.2 was Labeling of RNAs was performed with the FlashTag Biotin HSR RNA considered statistically significant. Labeling Kit (Affymetrix, Thermo Fisher Scientific). Gene expression was determined by microarray analysis on the Affymetrix platform Targeted transcriptional analysis (Rat Transcriptome Array 2.0). Complementary supervised analysis of DEGs leveraged knowledge of proteins known to be expressed at sensory neuron receptor endings Microarray data analysis that directly (25, 37, 38) or indirectly (39, 40) regulate excitability. In In total, 13 [four groups with biologic triplicates per group (4 addition, due to incomplete knowledge of the full suite of proteins that þ biologic replicates from ApcPirc/ þ oxaliplatin group)] global tran- are necessary and sufficient for normal function (40), we expand scriptional expression datasets were generated in this study. Raw targeted transcriptional analysis to incorporate gene families, includ- mRNA expression data were processed using Affymetrix Expression ing ion channels, GPCRs, and channel regulators. Console (EC) software Version 1.4. Raw data probes were normalized using SST-RMA algorithm. Differentially expressed genes (DEG) were 3D digital reconstructions and anatomic quantifications determined through a linear fixed effects model (limma; ref. 27) with a Z-axis stacks of images of muscle spindle receptor endings, robust empirical Bayes (eBayes) framework to moderate the residual DRGs, and tumors were constructedbysequentiallyimagingusing variances (27). This has the effect of increasing the effective degrees of a confocal microscope (LSM 700A, Zeiss). Muscle spindles term- freedom by which gene-wise variances are estimated, thus reducing the inals z-stacks (1-mmsteps)werecapturedwitha40 oil immersion number of false positives for genes with small variances and improving objective (N.A 0.6) at 0.6 digital zoom. DRG and tumors z-stacks power to detect DEG with larger variances (28). (1-mm steps) were captured with a 20 objective (N.A 0.4) at 0.6 digital zoom. Stacks of images were processed and analyzed using Unsupervised multivariate analysis Amaris (Bitplane) or ImageJ (NIH, Bethesda, MD) imaging soft- DEGs were subjected to paired multivariate analysis to discover ware. Figures present images as flat maximal projections of the sum high-level data structure. By serially linking two analytic techniques, of the z-axis optical slices. Analysis of neuron receptor endings principal components analysis (PCA)-driven Hierarchical clustering, (spindles) and soma (DRG) were performed by an investigator we stabilized clustering performance and reduced dimensionality to blinded to treatment group. Immunoreactivity in nerve terminals tractably few continuous variables containing the most important and DRG was determined by measuring the mean pixel intensity of information on which clustering algorithms are focused, which leads to on each en-face portions of the terminal (visualized by the VGLUT1 a better clustering solution (29, 30). PCA and visualization was applied staining) observed on every 1-mm single image or on the medium to DEGs with the factoextra (31) and FactoMineR (29) packages. PCs and large size cell somas observed on the flat maximal projections of capturing high-level structure were then employed to drive hierarchi- the z-stack images, respectively.

2942 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

Intracellular recording preprocessing tistic for all reported parameters was <1.05 (and Monte Carlo SE of Intra-axonal recordings of action potentials together with records of the parameter means was <0.002; ref. 44). We report the expected muscle length and force were digitized (20 kHz) and were monitored mean parameter values alongside 95% credible intervals using the online and stored on computer for later analysis with Spike2 and HDI (Supplementary Fig. S5). custom-written MATLAB scripts. From raw intracellular data of both paradigms, we extracted 31 measured and derived features (Supple- Bayesian model validation mentary Fig. S3) that provide a comprehensive quantification of Models performance was validated (Supplementary Fig. S5) neuronal signaling characteristics. We then categorized features into by computing out-of-sample predictive accuracy using Pareto- four broad clusters that represent functional features encoded by these smoothed importance sampling (PSIS) to perform leave-one-out cross neurons, containing sensitivity (Thr), dynamic (Dyn), static (Stat), and (LOO) validation. Although alternative validation strategies (WAIC) history-dependent (Hx) signaling information. Supplementary are asymptotically equal to LOO, PSIS-LOO provides a more accurate Figure S3 identifies and describes the measured and computed features and stable estimate of model performance and is more robust in the and functional clustering used for inference. finite cases with diffuse priors (weakly informative). We obtained SEs for predictions and for comparison of predictive errors between Statistical analysis of intracellular recordings models to definitively test and rank models based on predictive Linear discriminant analysis (LDA) provided supervised accuracy (elpd ¼ expected log pointwise predictive density). dimensionality reduction for the multiple features of neuron signaling, in an attempt to find a linear combination of features that separated Behavioral analysis and characterized independent and combinatorial treatment effects. We used the ladder rung walking task as a validated outcome to Data were first demeaned, normalized to unit SD, and tested by detect and describe sensorimotor deficits (45). The ladder rung Bayesian one-way ANOVA (stan_glm; ref. 41). The derived covariance apparatus was built as described previously (45). Secure hind foot matrix was then normalized by within-group, pooled covariance. The placement was assessed using the seven-category scale scoring system eigenvectors of that modified covariance matrix defined three canon- introduced by Metz and Whishaw (45) with modifications that did not ical variables that characterized and separated the four treatment distinguish whether or not an error was associated with a deep fall. The groups identified a priori as ApcWTþcontrol, ApcWTþoxaliplatin, total number of errors in forelimb and hindlimb placement was þ þ ApcPirc/ þcontrol and ApcPirc/ þoxaliplatin. LDA and 10-fold cross calculated and the mean error/step ratio was calculated per animal validation of model performance (repeated holdout method) was and represented as percent of error where an increase in the number of performed with the MASS (7.3–51.1) library in the R environment errors/steps corresponds to a decrease in secure foot placement of the (3.5.0). Individual features characterizing the firing responses of hind and/or fore foot. neurons sampled from multiple rats were tested for statistically significant differences with Bayesian one-way ANOVA (stan_glm). Data and material availability Bayesian parameters estimation derives the entire joint posterior Microarray datasets supporting the conclusions of this article are distribution of all parameters simultaneously thus circumventing the available in the Gene Expression Omnibus (GEO) repository under need to correct for multiple tests on the data. accession code GSE126773. Data supporting the conclusions of this Bayesian hypothesis testing was performed by analyzing the prob- article are included within the article and its Supplementary Files. ability mass of the parameter region in question. This leads to a direct Other datasets, code, and models that support the findings of this probability measure that defines values inside the 95% of the highest study are available from the corresponding author upon reasonable posterior density (HDI) more credible than outside values. HDI was request. used to make unbiased inferences by directly comparing the posterior probability distributions between two (or more) contrasts of interests, for example, mean comparisons testing. For example, an HDI of a Results credible difference distribution that does not span zero indicates that Transcriptional profiling of sensory neurons the model predictions for the two conditions of interest are different We first profiled the transcriptomes of lumbosacral (L4-S1) from each other, resulting in evidential support that parameters for DRGs, where sensory neurons (Fig. 1B) are vulnerable to chemo- both populations are unequal. therapy(2).Westudiedwild-type(ApcWT) rats and rats carrying þ an Apc gene mutation (ApcPirc/ ; ref. 15) associated with develop- Bayesian models ment of colon cancer (Supplementary Fig. S1; ref. 46), not only in All models were developed in fully hierarchal Bayesian frame- these rats, but also in 80% of patients with colorectal cancer (46). work (Supplementary Fig. S4) with the rstanarm package (2.18.1; Presence of malignant colonic tumor burden was validated in all þ ref. 41) in the R environment (3.5.0). Rstanarm implements regres- ApcPirc/ rats by IHC evaluation of key indicators of malignancy, sion models in stan,whicharefitusingHamiltonianMarkov for example, Ki67 and EGFR (Supplementary Fig. S1) from tumors Chain Monte Carlo sampling to compute credible parameter collected during terminal experiment dissection. Rats in each values (42, 43). For intercepts and predictors, we use Student t group were randomized to receive injections of oxaliplatin, a distribution with mean zero and four degrees of freedom as the platinum-based compound known to induce neuropathy in prior distribution. The scale of the prior distribution is 10 for the patients with cancer and shown in our previous studies to induce intercept and 2.5 for the predictors. Each model was run with four chronic movement disorders (23). Oxaliplatin was administered independent chains for 400 warm-up, 4,000 sampling steps, and by intraperitoneal injection on a human-scaled dose schedule every second sample (thinning) was discarded. For all parameters, (Fig. 1A;ref.23).Theoxaliplatintreatment schedule resulted in thenumberofeffective(n_eff)sampleswas>500. Convergence was four experimental groups used for all studies: ApcWTþcontrol, þ þ assessed and assumed to have reached the stationary distribution by ApcWTþoxaliplatin, ApcPirc/ þcontrol, and ApcPirc/ þ oxaliplatin ^ ensuring that the Gelman–Rubin shrinkage statistic (rhat, R)sta- (Fig. 1A).

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2943

Housley et al.

ApcWT ApcWT ApcWT A OX or VC 10 mg/kg +control +OX B Dorsal root Coasting ganglia

Receptor ApcPirc/+ ApcPirc/+ ApcPirc/+ 12 31987654 10 11 2+control +OX Week Spinal cord Triceps surae muscles

C D Fosb ApcWT +control ApcWT+control vs. Gad1 Cldn1 WT Pirc/+ vs. Apc +OX Apc +control Igf1 Sox6 425 total 290 total Col1a1 Cybb Cybb Postn Col3a1 31 Igf1 114 73 Fam69c B3gnt5 Plpp3 81 Tril Pilra 199 105 Ccl6 GTL2 Igf1 PeBayesAdj<0.005 Slco1a2 Col1a1 1,704 FC ≥1.2 & ≤-1.2 Lpar1 Inmt Kng1 Bcl3 Angptl4 WT Pirc/+ Apc +control vs. Apc +OX Ptgs2 Mt2A 2089 total Inhba Hmox1 Gfap E Il6 −10 010203040 Fold change ApcPirc/+ * +control F 30 ApcWT +control 15 15 * *

ApcPirc/+

PC3: 7.24% +OX 0 15 PC mean value ApcWT -15 +OX

PC2: 11.14% PC1 PC2 PC3 (65.6%) (11.14%) (7.24%) - -30 15 -15 -30 PC1: 65.6% 30

Figure 1. Cancer–chemotherapy codependence amplifies transcriptional dysregulation. A, Schematic of treatment schedule of oxaliplatin (OX) and vehicle control (VC) generating four independent experimental groups in wild-type (ApcWT) and autosomal-dominant mutation of the adenomatous polyposis coli gene (ApcPirc/þ): ApcWTþcontrol, ApcWTþOX, ApcPirc/þþcontrol, and ApcPirc/þþoxaliplatin rats. B, Diagram of experimental targets of lumbosacral DRG and muscle for assaying transcriptomes and/or cell-specific protein expression in rat models. C, Venn diagram distribution of significantly differentially expressed genes from the lumbosacral DRG among all groupwise contrasts. Significance determined by both P < 0.005 and FC ≥ 1.2 and ≤ 1.2 using a Bayesian moderated linear fixed effects model Pirc/þ (eBayesAdj). D, Bar graphs outline the top differentially expressed genes in Apc þoxaliplatin (purple bars) animals (FC ≥ 4and≤ -4) as compared with the effects observed in ApcWTþ oxaliplatin (blue bars) or ApcPirc/þþcontrol (red bars). E, Group correlation determined by PC analysis of significantly DEGs. Genes were visualized in the new 3D composite space created by PC1-3. Effects of independent chemotherapy (ApcWTþOX) or cancer (ApcPirc/þþcontrol) and their combinatorial (ApcPirc/þþOX) treatment are indicated by curved arrows; dots represent individual animals. F, Bar graphs of the mean values of PC1, PC2, and PC3. Error bars, SEM. , statistically significant differences between experimental groups as empirically derived from hierarchical Bayesian model (stan_glm).

2944 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

We employed an empirical Bayesian (eBayes) linear model (27) Next, we performed two independent downstream analyses (Sup- to modify residual variances on a total of 31,042 (22,875 named) plementary Fig. S6) to test for interaction and to determine which genes and identified 3,426 DEGs across all groupwise contrasts biologic processes might be involved in pathogenesis of sensory < fi fi (PeBayesAdj 0.005). Subsequent post hoc analysis identi ed DEG disorders. First, we subjected the eBayes- ltered database to un- from all possible contrasts [fold change (FC) ≥1.2 and ≤1.2]. supervised PC analysis. DEGs were visualized in the new 3D Overlap between DEGs retrieved from individual comparisons is composite space created by PC1–3(Fig. 1E) that explained 83.98% illustrated in Fig. 1C. In keeping with the notion that chemotherapy of the variance and clearly segregated the four experimental groups. and cancer act through similar molecular mechanisms, we found Inspection of individual PCs revealed significant evidence of nonlinear overlap of 112 DEG affected independently by cancer and chemo- interactions between cancer and chemotherapy captured in PC1 therapy (Fig. 1C). Of note, we found that 73.8% (n ¼ 1,704) of all (65.6% of explained variance), whereas PC2 (11.14%) and PC3 DEGs (n ¼ 2,307) were uniquely affected when cancer and chemo- (7.24%) separated the effects of cancer and chemotherapy, respectively þ therapy interacted (ApcPirc/ þoxaliplatin; Fig. 1C). Figure 1D lists (Fig. 1F). Second, we drove unsupervised hierarchical clustering onto þ the top DEGs (FC ≥ 4and≤ 4) identified in ApcPirc/ þoxaliplatin the PC1 solution to identify groups of coregulated genes that express in comparison with the effects observed in ApcWTþoxaliplatin or interaction (Fig. 2A). Analysis identified two clusters of genes dis- þ ApcPirc/ þcontrol. Collectively, these findings demonstrate the first playing distinct, but internally consistent, expression profiles clearly evidence that cancer exacerbates genetic dysregulation induced by representing patterns of interaction (Fig. 2A, C1 and C2; Supplemen- chemotherapy, unmasking a more than 4-fold increase in DEGs. tary Table S1).

A BCC2c Mean gene expression Response to steroid hormones IL6 IL6

Transcription factor binding 12 543 67 +OX

c C2b Mean gene expression +control Pirc/+

Anterograde transport WT Reterograde transport RNA polymerase Apc Prox. promoter DNA-binding Apc Enshealthment of neurons Axon ensheathment b Myelination 40 μm 40 μm C2 12 543 67

C2a Mean gene expression

Mechanical stimulus Locomotion Regulation of locomotion Gene expression Protein expression Response to lipids D E a ECM * 12 543 67 ** 150 C1b Mean gene expression 10

Voltage-gate K+ Ion channel activity b Regulation of K+ 8 100 K+ transport C1 12 543 67 Mean gene expression C1a 6 MFI: IL6 expression: IL6

Gluconeogenesis 2 50 a Negative reg apoptosis

ROS Log Carbohydrate metabolism Inflammatory response 4 Cytokines z-score Phosphorylation n = 3334 n = 45 49 44 77 MAPK 3334 0 -20 2 12 543 67 − +OX +OX log10(FDR) +OX +OX +OX +OX WT WT WT Pirc/+ +control +control +control +controlPirc/+ +control +controlPirc/+ WT Apc WT Apc WT Pirc/+ Pirc/+ Apc Pirc/+ Apc Apc Apc Apc Apc Apc Apc Apc Apc

Figure 2. Cancer–chemotherapy interaction disrupts distinct biological processes. A, Heatmap of the top differentially expressed genes in the sensory neurons identified by PC1-driven hierarchical-clustering (biological triplicates, n ¼ 3forApcWTþcontrol, ApcWTþOX, ApcPirc/þþcontrol, and four biologic replicates for ApcPirc/þþOX). Data were mean-centered and SD normalized before clustering; upregulated and downregulated genes (row) are shown in red and blue, respectively. Gray and white bars (C1 and C2, respectively) display the main clusters from gene-level hierarchical clustering and were chosen for clarity, while subclusters (A–C) highlight unique biologic characteristics within the main clusters. B, Clusters showing distinct up- or downregulation of key biological processes (Gene Ontology terms) and pathways (Kyoto Encyclopedia of Genes and Genomes). Selected pathway and gene ontology terms significantly associated (FDR ≤ 0.25) with misregulated transcripts in individual clusters are listed on the left of each gene cluster, with significance indicated by horizontal bars [-log10(FDR)]. The vertical bar graphs on the right compare mean SD of gene expression levels in each subcluster (Supplementary Table S1). C, ApcWTþcontrol and ApcPirc/þ OX immunostained against IL6. Scale bar, 40 mm. Gene expression (D) and receptor protein expression [E, mean fluorescence intensity (MFI)] of IL6 as determined by transcriptional profiling and cell-specificIHCof large diameter sensory neurons (> 30 mm). Numbers (n) identify sample size for each group. , statistically significant differences between experimental groups as empirically derived from hierarchical Bayesian model (stan_glm). , eBayesAdj< 0.005 and FC ≥ 1.2 and ≤ 1.2 using a Bayesian moderated linear fixed effects model. Data presented as mean SD.

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2945

Housley et al.

We then subjected clusters of named genes (Supplementary studies restricted investigation to acute neurotoxicity, which limits Table S1) to pathway enrichment analysis (35) and identified unique generalizability to symptoms that persist long after treatment. Our subclusters (Fig. 2A) expressing dysregulation in distinct biological data present the first evidence that Il6 (IL6) expression levels remain processes (Fig. 2B). In keeping with the notion that combinatorial elevated long after treatment cessation without observable morpho- effects of cancer and chemotherapy (oxaliplatin) reflect convergence logic changes. We show that codependent neuropathy significantly onto shared signaling pathways, genes mediating inflammatory increases Il6 (IL6) when compared with chemotherapy or cancer alone. ¼ 6 ’ response, for example, Il6 (PeBayesAdj 7.82 10 ), Ptgs2 (PeBayesAdj In light of IL6 s capacity to decrease excitability following chronic 7 0.0028¼ ), and Cxcr4 (PeBayesAdj ¼ 3.74 10 ), and reactive oxygen exposure (51), our data suggest that neuronal dysfunction (see below) ¼ species processes, for example, Hmbox1 (PeBayesAdj 0.001), were as a result of codependent neuropathy may be expressed as hypoexcit- found to be among the most upregulated in cluster 1a (C1a; Fig. 2B). ability, in contrast with previous work (5). Notably, we observed substantial induction of genes encoding glycolytic and carbohydrate metabolic processes (FDR ≤ 1.83 10 3, Global dysregulation of neuron transcriptomes þ C1a; Fig. 2B) in the ApcPirc/ þ oxaliplatin neurons. Coincident with We then used GSEA as a second independent tool to investigate the these changes was a statistically significant downregulation of genes cellular processes underlying codependent neuropathy. We initially þ mediating lipid metabolism (C2a), suggestive of pervasive metabolic focused on the comparison of ApcWTþ oxaliplatin and ApcPirc/ þ þ þ dysfunction beyond the small subset of dysregulated mitochondrial- oxaliplatin (Fig. 3A) and ApcPirc/ þcontrol and ApcPirc/ þ oxaliplatin related processes that we found in response to ApcWTþ oxaliplatin or (Fig. 3B) and interrogated entire datasets against the Molecular þ ApcPirc/ þcontrol alone (Fig. 2B; Supplementary Fig. S7A and S7B) Signatures Database (Supplementary Table S2). Two novel conclu- and similar to dysfunctional effects previously suggested (2). Among sions can be drawn from the GSEA findings. First, GSEA indepen- þ additional unique transcripts downregulated in ApcPirc/ þ oxaliplatin dently validated codependencies of cancer and chemotherapy in the neurons, Fig. 2B highlights multiple deficits in processes related to DEGs identified by transcriptional profiling using paired unsupervised extracellular matrix control, mechanical stimulus transduction, (C2a), analysis (Fig. 3A and B). GSEA corroborated downregulation of lipid DNA-binding, RNA polymerization, (C2b), transcription factor activ- metabolic pathways [normalized enrichment score (NES) ¼ 1.73] and þ ity, hormone signal transduction (C2c; Supplementary Table S1). upregulation of glycolytic pathways (NES ¼1.94) in ApcPirc/ þ Furthermore, we observed genes in C1b and 2a that represent specific oxaliplatin sensory neurons when compared with either chemotherapy dysregulation of neuronal and sensorimotor function, for example, (Fig. 3A) or cancer alone (Fig. 3B). Focusing analysis on pathways ion-channel genes determining neuronal excitability along with coin- directly involved in neuronal signaling confirmed dysfunction in cident downregulation of locomotion. These data provide the first peripheral neuron ensheathment in the absence of evidence support- evidence that codependent neuropathy induces broad metabolic ing neuron-specific apoptotic response (FDR > 0.83; Fig. 3A). reprogramming, fro example, increased glycolytic and decreased lipid Second, GSEA exposed dysregulation of genes relevant to neuronal metabolism, consistent with cancers capacity to corrupt metabolic excitability particularly potassium ion channels and transporters, processes (47, 48), which extends knowledge beyond the limited and others participating in synaptic communication, for example, impairment observed in mitochondria (Supplementary Fig. S8). SNARE and glutamatergic transmission (Fig. 3A and B). Dysre- Markers for nerve degeneration, inconsistently observed in patients gulation of these and other gene sets (e.g., proinflammatory che- treated with antineoplastic drugs (49), exhibited mixed results. While mokines) was not predicted from the independent effects of cancer we observed few significantly downregulated genes associated or chemotherapy alone (Supplementary Table S2). Unlike previous with myelination processes (C2b; P ≤ 7.12 10 3, Fig. 2B), we also studies that consistently report dysregulated voltage-gated sodium found significant induction of processes involved in the preservation channels (52), we find little evidence of differentially expressed of neurons (e.g., enrichment of negative regulation of apoptosis sodium channels or disrupted regulatory pathways (52). Instead, we P ≤ 5.96 10 3) in C1a. Mixed results from these data suggests find targeted DEGs related to potassium ion channels unlike those heterogeneous neuronal populations in the DRG may contain discrete previously identified. We draw two conclusions from these data. subpopulations with selective vulnerability, whereas others are First, ion channel clusters are differentially vulnerable to codepen- resistant to degeneration, which is consistent with mixed results seen dent neuropathy. Second, chronic neuropathy may be mechanisti- in human studies (49) and highlights the need for future cell-type cally linked to dysregulation of ion channels distinct from those specific investigations that are currently unavailable. identified in acute preparations. Taken together, our analyses indicate significant dysregulation of þ Dysregulation is conserved at protein level genes and proteins mediating core cellular processes of ApcPirc/ þ Because inflammatory pathways play multiple roles in regulating oxaliplatin sensory neurons. Our integrated transcriptomics and neuron function (50), we focused attention on their significant protein-level findings confirmed that cancer and chemotherapy affect response to codependent neuropathy. We selected the most dysregu- common processes (13, 53, 54). Moreover, our findings support our lated gene in the inflammatory pathways, Il6, for molecular validation central hypothesis that clinically relevant neuropathy depends on by assessing gene—protein correspondence. Sensory neurons in dorsal systemic interaction between cancer and chemotherapy by showing root ganglia were immunolabeled with mAbs targeting IL6 (Fig. 2C). for the first time that codependence exacerbates neuropathy and As predicted from the transcriptional profiling (Fig. 2D), the IL6 revealing it operates through distinct mechanistic pathways, unpre- protein was constitutively expressed in ApcWTþcontrol rats at lower dicted from cancer or chemotherapy alone. þ levels than in ApcPirc/ control animals or ApcWTþ oxaliplatin animals, þ and at significantly higher levels in ApcPirc/ þ oxaliplatin neurons Novel voltage-gated ion channel dysfunction (Fig. 2E), demonstrating that codependent neuropathy was conserved Sensory deficits, for example, somatosensory and hearing loss, are in protein expression (Fig. 2E; Supplementary Fig. S9). Previous commonly reported by patients long after cessation of chemotherapy, studies implicate IL6's capacity to mediate hyperexcitability and but knowledge of the combined effects of cancer and chemotherapy is morphologic changes associated with neuron death (51). However, missing. Mechanosensory systems, for example, cochlear hair cells,

2946 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

ApcWT ApcPirc/+ Downregulated Upregulated Cellular response to IL6 +OX +OX A 0.0 − −0.1 NES = 2.06 Response to IL6 (24) FDR = 0.01 FGA −0.2 SMAD4 Cellular response to IL6 (20) −0.3 PTGIS −0.4 RELA − STAT3 Cytokine-cytokine receptor interaction (187) 0.5 IL6R −0.6 IL6ST −0.7 CCL2

Positive regulation of ROS metabolic process (75) Enrichment score FGF23 −0.8 SBNO2 JAK/STAT signaling pathway (117) IL6 WT Pirc/+ Lipid biosynthesis (17) Apc +OX Apc +OX Regulation of gluconeogenesis Regulation of gluconeogenesis (32) CLK2 0.0 DGAT2 − NES = -1.94 GCG Cellular response to glucose starvation (30) 0.1 FDR = 0.02 SESN2 −0.2 LEP −0.3 WDR5 Negative regulation of potassium ion transport (29) MST1 −0.4 GCK −0.5 HNF4A Regulation of neurotransmitter uptake (15) FBP1 −0.6 LEPR Regulation of calcium ion dependent exocytosis (70) Enrichment score ADIPOQ SERPINA12 PTPN2 Voltage gated calcium channel complex (32) ApcWT+OX ApcPirc/++OX IL6 Calcium ion regulated neurotransmitter (28) Neuron enshealthment 0.6 NES = 2.04 CXCR4 Regulation of synaptic transmission glutamatergic (47) 0.5 FDR = 0.02 POU3F1 0.4 LPAR1 UGT8 Positive regulation of synaptic transmission glutamatergic (17) 0.3 CLDN11 0.2 GAL3ST1 DHH Ensheathment of neurons (83) 0.1 SH3TC2 QKI PNS system axon ensheathment (21) Enrichment score 0.0 KCNJ10 MPP5 DAG1 Positive regulation of neuron apoptotic process (47) PARD3 ApcWT+OX ApcPirc/++OX 3012 −1 −2 −3 ApcPirc/+ ApcPirc/+ Normalized enrichment score Glutamatergic synapse transmission +control +OX 0.0 ApcWT+OX vs. ApcPirc/++OX , FDR < 0.25 NES = −1.96 B −0.1 GRIK3 FDR = 0.11 SYT1 − score 0.2 RAB3GAP1 t − GLUL 0.3 GRM3 men − DRD2

h 0.4

c ADCYAP1 Downregulated Upregulated i −0.5 GRM8 nr GRM7 E − 0.6 PTK2B Response to IL6 (24) UCN PTGS2 Cellular response to IL6 (20) ApcPirc/++control ApcPirc/++OX

Positive regulation of ROS metabolic process (75) CACNB1 KCNK9 KCNK3 Regulation of gluconeogenesis (32) KCNJ13 Voltage gated cation channels KCNE1 HCN2 Synaptic transmission cholinergic (31) 0.1 KCNS2 NES = −1.81 0.0 CACNA1H Regulation of synaptic transmission glutamatergic (47) FDR = 0.11 KCND3 −0.1 RYR1 CACNB2 −0.2 KCNH6 STAT phosphorylation (38) CATSPER2 −0.3 CACNA1B SNARE complex (110) −0.4 KCNAB2 CACNG5

Enrichment score −0.5 KCNQ4 Voltage gated cation channel (114) CACNA1C KCNG2 KCNAB3 Proteinaceous ECM (292) Pirc/+ ApcPirc/++OX HTR3A Apc +control KCNC4 ABCC8 − − − KCNJ1 012 1 2 3 KCNG3 Normalized enrichment score GRM7 KCNJ5 ApcPirc/++control vs. ApcPirc/++OX , FDR < 0.25 CACNG3

Figure 3. GSEA and pathway analyses test codependence of cancer and chemotherapy. A, Enriched gene sets in sensory neurons identiﬁed by GSEA (MSigDB, C2-CP, canonical pathways; C3-MIR, miRNA targets; C3-TFT, transcription factor targets; C5-BP, GO biological process; C5-CC, GO cellular component; C5- MF, GO molecular function C7: immunologic signatures gene sets from the comparison of ApcWTþoxaliplatin and ApcPirc/þþoxaliplatin. B, ApcPirc/þþcontrol and ApcPirc/þþoxaliplatin groups. Individual frames (horizontal gray bars) identify functionally related (internally consistent) groups of gene sets expressing distinct up- or downregulation as compared with ApcWTþoxaliplatin. Number of overlapping genes are displayed in parentheses to the right of the corresponding GO term. Representative enrichment plots are shown in the middle column. Heatmap representation of leading-edge genes is shown on right. Statistical signiﬁcance determined by permutation testing with normalized enrichment score and Benjamini–Hochberg FDR < 0.25. ROS, reactive oxygen species; ECM, extracellular matrix.

þ and various somatosensory neurons in skin and muscle are responsible rying DEGs identified in ApcPirc/ þ oxaliplatin rats compared with all for encoding this information and exhibit deficits following chemo- other groups against genes encoding proteins known to be constitu- therapy (6). We therefore next focused on identifying the combined tively expressed in mechanosensory neurons that mediate unique effects of cancer and chemotherapy on a representative member of contributions to neuronal signaling (Supplementary Fig. S10; mechanosensory neurons, called muscle spindle neurons. Muscle refs. 25, 37, 40). We found no evidence for independent nor combi- spindles share many of the molecular mechanisms (40) underlying natorial effects of cancer or chemotherapy at the genetic level (Fig. 4A) mechanotransduction and encoding found in this broad class of for channels mediating mechanotransduction (e.g., Asic2, Piezo2, sensory neurons [e.g., hair cells (55), Merkel (56)]. Furthermore, ENaCs), signal amplification, and spike encoding [e.g., Scn1a (Nav1.1), signaling by these neurons encodes sensory features of muscle Scn8a (Nav1.6), Scn9a (Nav1.7), Cacna1s,-c,-d, -f (Cav1.1-4), Kcnn2 mechanics necessary for perceiving (proprioception) and coordinating (SK2), Kcna1 (Kv1.1)]. body position and movement, functions, which, when impaired, have Next, we performed cell-specific IHC analyses to test gene–protein the potential to explain persistent disorders in patients (57). We correspondence. We found no evidence to suggest dysregulated pro- performed targeted analyses of our transcriptome database by que- tein expression nor did evidence emerge to suggest altered protein

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2947

Housley et al.

Kv3.3 A BCGene expression DProtein expression 12.5 Upregulated Downregulated ** * 200 KCNC3 +control 6.9 WT KCNN2 150 Apc 10 6.6 100 CACNA1C 40 μm MFI Kv 3.3 CACNA1F 6.3 Kv3.3 50 expression KCNC3 2

CACNA1S 3334 12 10 8 9 7.5 Log +OX P value

+OX +OX +OX

Pirc/+ +OX WT WT +control +controlPirc/+ +control +controlPirc/+ WT WT Apc Pirc/+ Apc Apc Pirc/+ Apc Apc eBayes Apc Apc

10 Apc Apc 40 μm 5

- Log CACNA1D E vGluT1 Kv3.3 Merge SCN8A

SCN1A 2.5 +control WT SCN9A ASIC2 SCNN1A Apc 20 μm 20 μm 20 μm 0 PIEZO2 SCNN1B SCNN1G vGluT1 Kv3.3 Merge -4 -2 0246 ApcPirc/+ ApcWT Log2(fold change): +OX vs. +control +OX Pirc/+ Apc

20 μm 20 μm 20 μm

Figure 4. Targeted transcriptional analysis reveals mechanisms for cancer–chemotherapy interaction. A, Volcano plot comparing up- and downregulated genes in ApcPirc/þþoxaliplatin and ApcWTþcontrol sensory neuron transcriptomes. Annotation of genes involved in mechanotransduction, signal ampliﬁcation, action

potential generation and maintenance is shown. Significance was determined as Benjamini–Hochberg FDR < 0.01 and log2-fold change ≥ 1. B, Gene expression of Kcnc3 encoding Kv3.3, as determined through transcriptional profiling (Kcnc3: 1369133_a_at). C, Confocal images of dorsal root ganglia immunolabeled against Kv3.3 illustrate expression of ApcPirc/þþoxaliplatin relative to ApcWTþcontrol neurons. D, Quantification of receptor protein expression [mean fluorescence intensity (MFI)] of Kv3.3 determined through averaging over all annulospiral endings (617 from 12 ApcWTþcontrol neurons, 434 from 10 ApcWTþoxaliplatin neurons, 459 from 8 ApcPirc/þþcontrol neurons, and 362 from 9 ApcPirc/þþoxaliplatin neurons) for each neuron per experimental group. E, Confocal images of mechanosensory nerve terminals in muscle spindles immunolabeled against VGluT1 (green), Kv3.3 (red), and merged (yellow) illustrate protein expression levels and distribution in ApcPirc/ þþOX relative to ApcWTþcontrol. , statistically significant differences between experimental groups as empirically derived from hierarchical Bayesian model (stan_glm). , PeBayesAdj < 0.005 and FC ≥ 1.2 and ≤ 1.2 using a Bayesian moderated linear fixed effects model. Data presented as mean SEM.

distributions (Supplementary Fig. S11–S13) for most proteins, includ- groups indicate that Kv3.3 expression was significantly lower in þ ing those mediating mechanotransduction (ASIC2), signal amplifica- ApcPirc/ þ oxaliplatin compared with ApcWTþcontrol (Fig. 4D), þ tion (Nav1.1, Nav1.6, Nav1.7), spike encoding (Kv1.1), and mechan- ApcPirc/ þcontrol, or ApcWTþ oxaliplatin (Supplementary Fig. S15), osensory sensitivity (VGLUT1). These findings demonstrate that demonstrating that codependent dysfunction is conserved in many of the molecular mechanisms responsible for mechanotrans- cell-type–specific protein expression. Moreover, conserved Kv3.3 duction and determining excitability of mechanosensory neurons were expression in neuromuscular junctions suggests that Kv3.3 þ preserved in the ApcPirc/ þ oxaliplatin rats. These findings narrow the dysfunction was constrained to sensory neurons (Supplementary field of candidate mechanisms for which treatment might reasonably Fig.S16).Thisisthefirst evidence implicating Kv3.3 in the restore normal function. development of neuropathy and the first data, to our knowledge, Having exhausted examination of proteins known to be expressed in demonstrating a channelopathy persists long after treatment ces- mechanosensory neurons, we extended our search to include ion sation. The capacity of Kv3.3 to drive fast spiking and to enhance channels yet to be identified, but known to regulate neuronal signaling transmitter release–processes that are required for mechanotrans- in other classes of neurons (58–60). As one of the most dysregulated duction and spike encoding–suggests that this ion channel might voltage-gated ion channels, Kcnc3 (encoding protein Kv3.3) emerged provide both a potential novel mechanism for the dysfunction of as a novel candidate (Fig. 4A and B; Supplementary Fig. S14). mechanosensory neurons and a potential target for therapy. Pharmacologic and genetic perturbation of Kv3.3 (Kcnc3) impairs neuronal signaling (60) and is causally linked to ataxias (61) that are Neuronal signaling is impaired consistent with functional deficits observed in patients with cancer Next, we examined whether the codependent neuropathy we dis- long after cessation of chemotherapy. We then immunolabeled covered at gene and protein levels had negative consequences on sensory cell bodies in DRG and their receptor endings taken from mechanosensory function in living animals (Supplementary Fig. S17). ApcWTþcontrol with mAbs targeting Kv3.3, which revealed the first Applying electrophysiologic methods to rats in vivo, we recorded evidence of this ion channel in all large diameter cell bodies in the spiking activity from single mechanosensory neurons responding to DRG that supply mechanosensors (Fig. 4C; Supplementary Fig. S15) naturalistic mechanical stimuli (Fig. 5A). From these spiking and endings (Fig. 4E). Results of comparison across experimental responses, we collected 31 measured and derived parameters (average

2948 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

Intra axonal A microelectrode C Spine clamp * * * 20 100 0.50

Bipolar electrodes Nerve Spinal cord cross section electrodes ): Fast Triceps -1

Achilles 10 50 0.25 In vivo physiological sensory stimulus

(20 mm/s) (4 mm/s) Sensitivity

Fast Slow Static firing Spikes (#): Fast Dynamic firing velocity velocity+repeated Spikes (#): Fast Muscle

length Servomotor 1/Threshold (ms 3 mm 20 mm/s 750 ms

B ApcWT+control 400 0 0 0

+OX +OX +OX +OX +OX +OX 200 WT WT WT +control +control Pirc/+ +control +control Pirc/+ +control +control Pirc/+ WT WT WT Apc Pirc/+ Apc Pirc/+ Apc Pirc/+ Apc Apc Apc 0 Apc Apc Apc Apc Apc Apc ApcPirc/++control 400 D 200

NEFH NEFH 0

WT 400 Apc +OX +OX IFR (Hz) WT +control 200 WT Apc

0 Apc 15 μm 15 μm

Pirc/+ NEFH NEFH 400 Apc +OX

200 +OX +control Pirc/+ 0 Pirc/+ Apc L 15 μm 15 μm

3 mm o 200 ms Apc IL6

Figure 5. Mechanosensory signal degradation exacerbated by the cancer–chemotherapy interaction. A, Diagram of anesthetized rat stabilized in spine clamps for physiological recordings of triceps surae (medial and lateral gastrocnemius and soleus) muscles and nerves. Records taken intra-axonally from single mechanosensory neurons in lumbosacral dorsal roots (L4-S1) while applying fast (20 mm/second) and slow (4 mm/second) naturalistic mechanical stimuli delivered in the periphery. B, Representative cases of spiking activity in ApcWTþcontrol, ApcWTþoxaliplatin, ApcPirc/þþcontrol, and ApcPirc/þþoxaliplatin as a measure of sensory encoding. Black circles plot instantaneous ﬁring rates (pps) of corresponding spike (action potential) 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 gray (150 ms duration after stretch command onset) and light gray (1-second duration after the dynamic phase) bars. C, Neuronal spiking parameters (n ¼ 31 averaged from four trials in each neuron, four shown from fast ramp stimulus) representing different features of sensory stimulus [ApcWTþcontrol (n ¼ 11), ApcWTþoxaliplatin (n ¼ 19), ApcPirc/þþcontrol (n ¼ 20), and ApcPirc/þþ oxaliplatin (n ¼ 10)]: mean initial burst frequency (pps) signaling stretch onset; the number of spikesduring dynamic sensory stimulation (shown asa dark gray bar inB); the number of spikes during static sensory stimulation (shown as a light gray bar in B); sensitivity assessed as the inverse of latency to stimulus detection (ms-1), that is, lower sensitivity corresponds to longer latency and higher threshold. D, Confocal image of a neuroﬁlament heavy chain (NEFH, in green) with immunolabeling of the terminal axon and receptor structure in ApcWTþcontrol, ApcWTþOX, ApcPirc/þþcontrol, and ApcPirc/þþ oxaliplatin rats. Scale bar, 15 mm. , statistically signiﬁcant differences between experimental groups as empirically derived from hierarchical Bayesian model (stan_glm); 95% highest density intervals do not overlap between groupwise contrasts. Data presented as mean SEM.

of four trials; n ¼ 11 ApcWTþcontrol, n ¼ 19 ApcWTþOX, n ¼ 20 history-dependent reduction in spike number (Supplementary þ þ ApcPirc/ þcontrol, n ¼ 10 ApcPirc/ þ oxaliplatin; Supplementary Fig. S18). þ Fig. S3) from which we extract four functional features encoded by We next tested whether ApcPirc/ or oxaliplatin treatment alone these neurons—sensitivity, dynamic, static, and history-dependent induce signaling dysfunction. Although we find transcriptional dys- þ signaling information (Supplementary Fig. S3). In ApcWTþcontrol regulation in ApcPirc/ þcontrol (Fig. 1C; Supplementary Fig. S7A), we rats, we found that muscle stretch elicited the spiking expected from found a remarkable degree of concordance with signaling in þ sensory neurons in normal animals (24) and humans (62), specifically ApcWTþcontrol (Fig. 5B). ApcPirc/ þcontrol neuronal signaling dis- high frequency initial bursting at stimulus onset, increasing dynamic played comparable sensitivity to stimuli and showed similar static firing during increasing stimulus, sustained static firing with slow behavior across all neurons (n ¼ 20; Fig. 5B and C) as the accommodation during the hold phase of stretch (Fig. 5B), and ApcWTþcontrol. We then found that oxaliplatin treatment of ApcWT

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2949

Housley et al.

3 LD3 (11%) LD3 * * ApcWT +OX *ApcPirc/+ ApcPirc/+ +OX +control 4 -2 ApcWT -4 +control LD2 (34%) LD1 (54%) 6 -3 B Predictive modelsmodels D Predictive models

Predictor(s) Predictor(s) Chemotherapy ELPD Cancer ELPD Cancer = −54.1 = −48.8 Probability density Probability density

C E

Chemotherapy Cancer ELPD ELPD Interaction Chemotherapy = −50.2 = 0 Probability density Probability density

-10 -5105 0-10 -5105 0 LD1 LD1 Predicted (PPD) Observed

Synergystic: Opposing: Emergent: F Sensitivity cluster G Dynamic cluster H History-dependent cluster Static cluster * * * *

20 250 40 120

Predicted Predicted Predicted Predicted 15 200

10 20 60 Sensitivity

150 Spikes (#): Slow Static firing Dynamic firing Mean firing rate (pps): Fast 1 / Length threshold (mm): Fast 5 History dependence Peak firing rate ramp (pps): Fast Observed Observed Observed Observed 100 0 0 0

+OX +OX +OX +OX +OX +OX +OX +OX WT WT WT WT +control +control Pirc/+ +control Pirc/+ +control +control +control Pirc/+ +control +control Pirc/+ WT Apc WT WT Pirc/+ Apc Pirc/+ Apc WT Apc Pirc/+ Apc Pirc/+ Apc Apc Apc Apc Apc Apc Apc Apc Apc Apc Apc

Figure 6. Cancer–chemotherapy codependence exacerbates sensory dysfunction beyond that predicted by cancer or chemotherapy alone. A, Neuronal spiking parameters [n ¼ 31, averaged from four trials, from each neuron (n ¼ 60)] representing different features of sensory stimuli were subjected to linear discriminant (LD) analysis. Neuronal signaling was visualized in the new 3D composite space created by LD1-3. 3D ellipsoids enclosing 68% of data were computed with least-squares elliptical fitting to emphasize differences between control and ApcPirc/þþoxaliplatin neurons. Effects of independent (oxaliplatin or ApcPirc/þ) and combinatorial (ApcPirc/þþ oxaliplatin) treatment are indicated by curved arrows. B–E, the hierarchical Bayesian model used to test for significant group differences in LD1 scores reconfigured to operate in a predictive fashion. Predictors included in each model are listed on the right of each plot. Generative models in B–D utilizing one (B and C) or both independent (D) predictor(s), that is, cancer (ApcPirc/þ) or chemotherapy (oxaliplatin) for posterior prediction. The generative model in E utilizes both independent predictors and an interaction term for posterior prediction. (Continued on the following page.)

2950 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

rats induced mild signaling deficits that were less pronounced, liplatin rats, they are in line with the inconsistency of physical evidence occurred in a small proportion of neurons (5/19), and were primarily for nerve degeneration reported in clinical studies (49). Furthermore, restricted to sensitivity functional clusters (Fig. 5B and C; Supple- the ability of muscle vibration to entrain firing of these neurons in all mentary Fig. S19D), validating our previous findings of restricted treatment groups as they do in ApcWTþcontrol rats further refutes the deficits observed in a different strain of healthy rats treated with existence of degeneration's necessity in signaling disorders (Supple- oxaliplatin alone (23). mentary Fig. S21). Overall, our findings suggest that the structure and In contrast to the observations in ApcWTþcontrol, ApcWTþ oxa- core ability of mechanosensory neurons to produce action potentials þ liplatin, or ApcPirc/ þcontrol rats, we found drastically impaired remain unimpaired by codependent neuropathy and does not depend þ neuronal signaling in ApcPirc/ þ oxaliplatin rats (Fig. 5B). High on dying-back degeneration. frequency initial bursting was attenuated 4-fold and sensitivity to stimuli was reduced by 3-fold at both high (Supplementary Fig. S19D) Dysfunction depends on cancer–chemotherapy interaction and low velocity stretch (Supplementary Fig. S18 and S19D). During Data presented in Fig. 5 suggests damaging cancer–chemotherapy dynamic stimuli, we found marked reduction in the number of spikes interactions are conserved in vivo and have consequences in sensory þ over both stretch conditions (Fig. 5C). Neurons in ApcPirc/ þ oxa- function. To assess the extent of interaction quantitatively, we took an liplatin rats also failed to sustain firing (Fig. 5B). In addition, we unbiased statistical approach by subjecting all neurons (n ¼ 60, 240 observed fewer spikes (Fig. 5C) during rapid accommodation imme- total trials) from all experimental groups to a machine learning diately preceding signal deletion, despite constant stimulus (Fig. 5B). algorithm [linear discriminant (LD) analysis]. Our approach reduced þ We then simulated naturalistic compensation by subjecting ApcPirc/ þ complex feature space into canonical variables giving us a high-level oxaliplatin neurons to repeated trials at 1, 2, and 3 background understanding of where interaction emerges, without biased feature stimulus (Lo strain) intensity, which failed to completely rescue selection a priori. Our analysis yielded three canonical variables signaling back to ApcWTþcontrol levels (Supplementary Fig. S20B– (Fig. 6A) that achieved overall 94.7% classification accuracy (Supple- S20D). These data are the first to directly demonstrate a damaging mentary Fig. S19A). We then visualized neuronal signaling in the functional interaction between the systemic effects of cancer and new 3D composite space created by LD1-3 (Fig. 6A). By projecting chemotherapy in sensory neurons in a living animal. While vali- high-dimensional parameter and feature data onto a simplified dating our previous findings of signaling deficits in chemotherapy 3D canonical space, statistically significant nonlinear interaction alone(23),thesedataindicatethatrestrictivesignalingdeficits between cancer and chemotherapy clearly emerged in the first dimen- induced by chemotherapy or cancer alone were insufficient to sion (LD1; 54.23% proportion of variance; Fig. 6A; Supplementary reproduce the magnitude, direction, and number of dysfunctional Fig. S19B). Notably, dysfunction in LD1 induced by the cancer— parameters induced by codependent neuropathy. chemotherapy interaction occurred in the opposite direction to that predicted by independent effects, and its magnitude was amplified Dysfunction does not depend on degeneration greater than their sum (Fig. 6A; Supplementary Fig. S19B and S19C). We next tested whether the impaired neuronal signaling of To test the statistical significance of the cancer—chemotherapy inter- þ ApcPirc/ þ oxaliplatin neurons might be explained by dying-back action, we conducted Bayesian model comparison with full factorial degeneration of sensory nerve terminals. While the underlying and all restricted models using leave-one-out cross validation. We then mechanisms for chemotherapy induced neuropathy are not under- quantified and validated each model's predictive performance by stood, current opinion identifies “dying-back” axon degeneration as a computing the expected log predictive densities (ELPD; measure of major pathology in this disorder (19). Blinded reviewers assessed the a model's out-of-sample predictive accuracy in Fig. 6B–E). We found structural integrity of sensory afferents and receptor endings immu- decisive evidence in favor of a model including a cancer– nolabeled against neurofilament protein (NEFH; ref. 25). We then chemotherapy interaction predictor (ELPD diff ≥ 48 SE <8.1; tested for functional evidence of degeneration by comparing differ- Fig. 6E). Moreover, our data and generative modeling conclude that ences in conduction delays, a physical measure of nerve degeneration codependent interaction (Fig. 6B–D) is necessary to accurately and that is used clinically (49). Neither histologic observation of sensory reliably reproduce clinically relevant neuronal signaling deficits þ afferents and receptors (Fig. 5D) nor axon conduction tests (Supple- observed in ApcPirc/ þ oxaliplatin rats (Fig. 6E) mentary Fig. S21) revealed evidence of dying-back degeneration Having determined the significance of an interactive effect, we þ of sensory nerve terminals (ApcPirc/ þ oxaliplatin: 1.41 0.13 ms, then asked which functional clusters are highly enriched in LD1. By n ¼ 10. ApcWTþcontrol: 1.52 0.14 ms, n ¼ 11; Supplementary examining the discriminant function coefficients, we found that Fig. S21). Our findings demonstrate that transcriptional changes we static signaling and sensitivity functional clusters represent a large observed for some markers of nerve degeneration (Fig. 3)were portion of the explained variance in LD1. These findings suggest the insufficient to yield nerve degeneration. While our results indicate sensitivity and static signaling clusters are most susceptible to þ selective resistance of mechanosensory neurons in ApcPirc/ þ oxa- codependent neuropathy (Supplementary Fig. S19D).

(Continued.) Gray lines in B–E 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 (D symbol) indicates difference from optimal model. Negative models represent worse predictive performance. F, Sensitivity assessed as the inverse of latency to stimulus detection (mm-1), that is, lower sensitivity corresponds to longer latency and higher threshold. G, Peak firing rate (pps) achieved during the dynamic sensory stimulus (shown as dark gray bar in Fig. 5B). H, Change in dynamic spike number in successive, slow stretch dynamic sensory stimulus and mean firing rate (pps) achieved during the static sensory stimulus (shown as light gray bar in Fig. 5B). Recordings were analyzed from ApcWTþcontrol (n ¼ 11), ApcWTþ oxaliplatin (n ¼ 19), ApcPirc/þþcontrol (n ¼ 20), and ApcPirc/þþOX (n ¼ 10) in B–G, , statistically significant differences between experimental groups as empirically derived from hierarchical Bayesian model (stan_glm). Gray arrows in F–H indicate direction and significant differences from ApcWTþcontrol predicted from the linear sum of the independent effects of ApcWTþ oxaliplatin and ApcPirc/þþcontrol groups. Black arrows indicate experimentally observed effects of their combination. Circles with slashes indicate no difference predicted for ApcWTþ oxaliplatin group and ApcPirc/þþcontrol groups. Data presented as mean SEM.

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2951

Housley et al.

þ While dynamic signaling provided relatively modest contributions to observed in either ApcWTþ oxaliplatin nor ApcPirc/ þcontrol rats. LD1 (lower coefficient weights in Supplementary Fig. S19D), examining Our findings demonstrate that complex systemic perturbations of the parameter-level data allowed us to discover the expression of distinct chemotherapy and cancer result in synergistic interactions for sensi- þ classes of interactions. Independently, oxaliplatin and ApcPirc/ induce tivity signaling characteristics (Fig. 6F), opposing interactions for peak highly conserved (87.5%) opposing effects exclusively within the dynam- firing rate during dynamic signaling (Fig. 6G), and emergent inter- þ ic functional cluster. We consistently found that ApcPirc/ mutation actions for history-dependent and static signaling characteristics þ increased and oxaliplatin treatment decreased dynamic firing (Fig. 6G). (Fig. 6H). In summary, ApcPirc/ þ oxaliplatin rats express widespread This led us to predict that their combination would nullify dysfunction neuronal signaling dysfunction that targets all functional clusters, was and approximate ApcWTþcontrol signaling. Instead, we observed independent of structural or physiologically detected degeneration and opposing interactions that resulted in drastic reduction in dynamic were not fully accounted for by a global decrease in sensitivity. firing properties. In contrast, for history-dependent functional clusters, we found that cancer–chemotherapy interaction emerged exclusively in Behavioral dysfunction exacerbated by cancer–chemotherapy þ ApcPirc/ þ oxaliplatin neurons, since neither ApcWTþ oxaliplatin nor interaction þ þ ApcPirc/ þcontrol neurons were disrupted (Fig. 6H). ApcWTþcontrol and ApcPirc/ þcontrol rats achieved a high inci- Taken together, we found that 58% of the neuronal signaling dence of rear and fore foot placement, when walking on the rungs of an þ parameters in ApcPirc/ þ oxaliplatin rats showed high-confidence uneven horizontal ladder, with a foot-drop error rate of only 2.4% interaction effects, in that those deficits were greater than those (2.7% SD, 6 animals; Fig. 7A; Supplementary Movie S1) and 2.5%,

ABD 40 Gait errors total Forelimb errors per step cycle 20 *

30 * 10 Error rate (%)

0 20

Error rate (%) * CE Hindlimb errors

20 * 10 *

10 Error rate (%) 1, reproduced from Vincent et al 2016 0 with permission

+OX +OX

+control +control Pirc/+ WT 0 Pirc/+ Wistar 1 Apc Apc Apc

Figure 7. Behavioral dysfunction exacerbated by cancer–chemotherapy interaction. A, Total gait errors as measured by secure fore- and hindfoot placement was scored by errors/step cycle during ladder rung walking. Recordings were analyzed from ApcWTþcontrol (n ¼ 6), Wistarþ oxaliplatin (n ¼ 7), ApcPirc/þþcontrol (n ¼ 4), and ApcPirc/þþ oxaliplatin (n ¼ 7) animals. B and C indicate the contributions of fore- (B) and hindlimb (C) errors to total error rates. D, Photograph shows left forelimb error in double image (simultaneous side and underneath views) of ApcPirc/þþ oxaliplatin rat walking on ladder rungs. E, Photograph shows left hindlimb error in double image (simultaneous side and underneath views) of ApcPirc/þþ oxaliplatin rat walking on ladder rungs (representative cases of entire ladder rung walking trials can be viewed in Supplementary Movies S1 and Movie S3 for ApcWTþcontrol and ApcPirc/þþ oxaliplatin rats, respectively). , statistically signiﬁcant differences between experimental groups as empirically derived from hierarchical Bayesian model (stan_glm); 95% highest density intervals do not overlap between groupwise contrasts. 1, Wistarþ oxaliplatin data reproduced from Vincent and colleagues 2016 (23) with permission.

2952 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

respectively (2.9% SD, 4 animals; Fig. 7A; Supplementary Movie S2). large increases in both gene and protein expression. As a result of its þ In contrast, ApcPirc/ þ oxaliplatin exhibited a 19.2% foot-drop error effect in suppressing neuronal firing behavior (51), IL6 has the rate [5.6% SD, 7 animals; hierarchical Bayesian model (stan_glm) vs. potential to explain decreased signaling by mechanosensory neurons. þ ApcWTþcontrol and ApcPirc/ þcontrol; Fig. 7A; Supplementary Mov- We identify both Kv3.3 and IL6, therefore, as targets worthy of testing þ ie S3]. ApcPirc/ þ oxaliplatin error rates significantly exceeded our for their potential value in treating movement disorders. Additional previous findings following oxaliplatin treatment alone [Wistarþ points of interaction likely exist, which remain either undiscussed (e.g., oxaliplatin) of 8.4% (3.1% SD, 7 animals; hierarchical Bayesian FosB) or potentially masked due to transcriptional profiling of DRG þ model (stan_glm) vs. ApcPirc/ þ oxaliplatin; Fig. 7]. Independent that may result in variable differential expression across heterogenous analysis of fore- (Fig. 7B and D) and hind-foot (Fig. 7C and E) error cell types (Fig. 1D). rates revealed that skilled motor behavior degradation was signifi- We observed abnormal skilled sensorimotor behavior in þ cantly exacerbated by the emergence of fore-limb errors (10.4% 3.9 ApcPirc/ þ oxaliplatin rats. Deficits were significantly greater than þ SD) not present in ApcWTþcontrol (0%), ApcPirc/ þcontrol those previously observed in our lab from animals treated with (2.3%2.8SD), or Wistarþ oxaliplatin (0%; hierarchical Bayesian oxaliplatin alone (23). The emergence of novel fore-limb deficits model (stan_glm)). typically not observed in this challenge to hind-limb capacities provides further evidence that exacerbation effects are unpredicted from studying chemotherapy alone. Furthermore, our data indicates Discussion that intact proprioceptive sensory input from muscle spindles is Here we present original evidence that cancer transforms the necessary to ensure limb placement accuracy in a skilled behavioral nature and magnitude of neuropathy induced by chemotherapy task and may be sufficient to explain sensorimotor deficits expe- alone. Our preclinical study of rats is the first to compare the rienced by cancer survivors. neuropathic effects of chemotherapy and cancer, both indepen- Our study, being restricted to a single time point following che- dently and in combination, impractical in human study. These motherapy, does not assess the stability of potential treatment targets comparisons for global transcriptional analysis of sensory neurons that emerge from codependent neuropathy. It is reasonable to expect, in DRG revealed dysregulation of genes uniquely induced, ampli- however, that the codependence undergoes dynamic change. Cancer, fied, or suppressed by the combination of cancer and chemotherapy. presumably also its associated systemic effects, undergoes complex Codependence was conserved as impaired spike encoding of progression as subpopulations of cancer cells differentially resist, mechanosensory stimuli and novel ion channelopathy. While the adapt, or succumb to chemotherapy. Similarly, surgical resection of current report is limited to rats, extensive conservation of core tumors or surgical associated analgesics might impose additional molecular and cellular processes across mammalian species (63) points of interaction that have the potential to impact DRG neuro- leads reasonable expectation that cancer–chemotherapy codepen- biology. Moreover, biological systems themselves initiate dynamic dency, whether different in detail, extends to humans. All consid- responses to perturbations. In the nervous system, homeostatic reg- ered, we conclude that inattention to codependencies necessarily ulation initiates compensatory mechanisms to offset perturbations in prevents the development of mechanism-based treatments for neuronal excitability (59). It is thought provoking in this regard to sensory neuropathy, which remains unexplained and unabated in consider the possibility that the chronic hypoexcitability, which patients receiving chemotherapy for cancer. follows the acute hyperexcitability (2, 19), that develops during Our transcription analyses expose two potential targets for treating chemotherapy reflects the actions of a dysregulated compensatory sensorimotor disorders among the debilitating patient symptoms that mechanism. Understanding these nonlinear interactions and their persist following treatment of various cancers with platinum-based effects on potential treatment are both challenging and necessary for compounds and other antineoplastic agents, for example, taxanes. developing effective cancer treatment. Patients display deficits in the spatiotemporal parameters (speed and stride) of walking gait, in control of posture, and in balance relying on Disclosure of Potential Conflicts of Interest proprioception (64). Acknowledging that these disabilities may arise No potential conflicts of interest were disclosed. from a variety of lesions in the nervous system, the deficits we observe here at the receptor origin of detection and encoding of muscle mechanics would necessarily impair movements and postures and Authors’ Contributions cannot be fully compensated by other senses, for example, vision. Conception and design: S.N. Housley, P. Nardelli, T.C. Cope Furthermore, signaling by mechanosensory neurons in rats closely Development of methodology: S.N. Housley, P. Nardelli, J.F. McDonald, T.C. Cope resembles that in human (62). For these reasons, we assign special Acquisition of data (provided animals, acquired and managed patients, provided attention to depressed signaling by muscle spindles and to the asso- facilities, etc.): S.N. Housley, P. Nardelli, D.I. Carrasco, T.M. Rotterman, E. Pfahl, ciated decrease in expression of the voltage-gated ion channel Kv3.3 L.V. Matyunina, J.F. McDonald, T.C. Cope and its gene Kcnc3. Reports that Kcnc3 gene knockout impairs firing Analysis and interpretation of data (e.g., statistical analysis, biostatistics, responses of neurons and results in ataxia (61) promote this gene and computational analysis): S.N. Housley, P. Nardelli, J.F. McDonald, T.C. Cope its ion channel as potential contributors to signaling deficits observed Writing, review, and/or revision of the manuscript: S.N. Housley, P. Nardelli, fi J.F. McDonald, T.C. Cope in this study. At this time, we are uncertain about the suf ciency of Administrative, technical, or material support (i.e., reporting or organizing data, Kv3.3 channelopathy to explain deficient mechanosensory signaling in constructing databases): S.N. Housley, D.I. Carrasco, T.M. Rotterman, T.C. Cope cancer treated by chemotherapy. Uncertainty arises, in part, because Study supervision: S.N. Housley, T.C. Cope comprehensive understanding of the molecular mechanisms underlying the function of these sensory neurons is lacking, although Acknowledgments advanced by our discovery of Kv3.3 in mechanoreceptors of healthy This work is supported by NIH grant R01CA221363 and R01HD090642. We animals. Another candidate target for treating sensorimotor disorders thank the staff of the Physiological Research Laboratory at Georgia Institute of is the inflammatory signaling molecule IL6, shown here to express Technology. We are grateful to Dr. Thomas Burkholder for technical assistance,

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2953

Housley et al.

discussions, and comments. We thank Dr. Mengnan Zhang and Dr. Evan Clayton for advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate discussions and comments. this fact.

The costs of publication of this article were defrayed in part by the Received August 1, 2019; revised March 11, 2020; accepted April 23, 2020; payment of page charges. This article must therefore be hereby marked published ﬁrst April 28, 2020.

References 1. Andre T, Boni C, Mounedji-Boudiaf L, Navarro M, Tabernero J, Hickish T, et al. 22. Nardelli P, Powers R, Cope TC, Rich MM. Increasing motor neuron excitability Oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment for colon cancer. to treat weakness in sepsis. Ann Neurol 2017;82:961–71. N Engl J Med 2004;350:2343–51. 23. Vincent JA, Wieczerzak KB, Gabriel HM, Nardelli P, Rich MM, Cope TC. A 2. Cavaletti G, Marmiroli P. Chemotherapy-induced peripheral neurotoxicity. novel path to chronic proprioceptive disability with oxaliplatin: distortion of Nat Rev Neurol 2010;6:657. sensory encoding. Neurobiol Dis 2016;95:54–65. 3. Alcindor T, Beauger N. Oxaliplatin: a review in the era of molecularly targeted 24. Vincent JA, Gabriel HM, Deardorff AS, Nardelli P, Fyffe RE, Burkholder T, et al. therapy. Curr Oncol 2011;18:18. Muscle proprioceptors in adult rat: mechanosensory signaling and synapse 4. Seretny M, Currie GL, Sena ES, Ramnarine S, Grant R, MacLeod MR, et al. distribution in spinal cord. J Neurophysiol 2017;118:2687–701. Incidence, prevalence, and predictors of chemotherapy-induced peripheral 25. Carrasco DI, Vincent JA, Cope TC. Distribution of TTX-sensitive voltage-gated neuropathy: a systematic review and meta-analysis. Pain 2014;155:2461–70. sodium channels in primary sensory endings of mammalian muscle spindles. 5. Sisignano M, Baron R, Scholich K, Geisslinger G. Mechanism-based treatment J Neurophysiol 2017;117:1690–701. for chemotherapy-induced peripheral neuropathic pain. Nat Rev Neurol 2014; 26. Lili LN, Matyunina LV, Walker L, Benigno BB, McDonald JF. Molecular 10:694–707. profiling predicts the existence of two functionally distinct classes of ovarian 6. Avan A, Postma TJ, Ceresa C, Avan A, Cavaletti G, Giovannetti E, et al. cancer stroma. Biomed Res Int 2013;2013:846387. Platinum-induced neurotoxicity and preventive strategies: past, present, and 27. Smyth GK. Linear models and empirical bayes methods for assessing differential future. Oncologist 2015;20:411–32. expression in microarray experiments. Stat Appl Genet Mol Biol 2004;3:1–25. 7. Galanski M, Jakupec MA, Keppler BK. Update of the preclinical situation of 28. Ritchie ME, Silver J, Oshlack A, Holmes M, Diyagama D, Holloway A, et al. A anticancer platinum complexes: novel design strategies and innovative analytical comparison of background correction methods for two-colour microarrays. approaches. Curr Med Chem 2005;12:2075–94. Bioinformatics 2007;23:2700–7. 8. Johnstone TC, Park GY, Lippard SJ. Understanding and improving platinum 29. Husson F, Josse J, Pages J. Principal component methods-hierarchical clustering- anticancer drugs–phenanthriplatin. Anticancer Res 2014;34:471–6. partitional clustering: why would we need to choose for visualizing data. Applied 9. Hershman DL, Lacchetti C, Dworkin RH, Lavoie Smith EM, Bleeker J, Cavaletti Mathematics Department; 2010. Available at: http://factominer.free.fr/more/ G, et al. Prevention and management of chemotherapy-induced peripheral HCPC_husson_josse.pdf. neuropathy in survivors of adult cancers: American Society of Clinical Oncology 30. Arguelles€ M, Benavides C, Fernandez I. A new approach to the identification of clinical practice guideline. J Clin Oncol 2014;32:1941–67. regional clusters: hierarchical clustering on principal components. Appl Econ 10. Currie GL, Angel-Scott H, Colvin L, Cramond F, Hair K, Khandoker L, et al. 2014;46:2511–9. Animal models of chemotherapy-induced peripheral neuropathy: a machine- 31. Kassambara A, Mundt F. Factoextra: extract and visualize the results of mul- assisted systematic review and meta-analysis: a comprehensive summary of the tivariate data analyses. R package version. 2016. Available at: https://cran.r- field to inform robust experimental design. bioRxiv 2018:293480. DOI: https:// project.org/web/packages/factoextra/index.html. doi.org/10.1101/293480. 32. Warnes MGR, Bolker B, Bonebakker L, Gentleman R. Package ‘gplots’. Various R 11. Boyette-Davis JA, Eng C, Wang XS, Cleeland CS, Wendelschafer-Crabb G, Programming Tools for Plotting Data 2016. Available at: https://cran.r-project. Kennedy WR, et al. Subclinical peripheral neuropathy is a common finding in org/web/packages/gplots/index.html. colorectal cancer patients prior to chemotherapy. Clin Cancer Res 2012;18: 33. Murtagh F, Legendre P. Ward's hierarchical agglomerative clustering method: 3180–7. which algorithms implement Ward's criterion? J Classification 2014;31:274–95. 12. Ahles TA. Brain vulnerability to chemotherapy toxicities. PsychoOncology 2012; 34. Becker R, Chambers J, Wilks A. The New S Language. Pacific Grove, CA: 21:1141–8. Wadsworth & Brooks/Cole; 1988. 13. Reuter S, Gupta SC, Chaturvedi MM, Aggarwal BB. Oxidative stress, inflam- 35. Huang DW, Sherman BT, Lempicki RA. Systematic and integrative analysis of mation, and cancer: how are they linked? Free Radic Biol Med 2010;49:1603–16. large gene lists using DAVID bioinformatics resources. Nat Protoc 2008;4:44. 14. Zheng Q, Fang D, Cai J, Wan Y, Han J-S, Xing G-G. Enhanced excitability of 36. Hong G, Zhang W, Li H, Shen X, Guo Z. Separate enrichment analysis of small dorsal root ganglion neurons in rats with bone cancer pain. Molecular pain pathways for up-and downregulated genes. J Roy Soc Interface 2014;11: 2012;8:24. 20130950. 15. Amos-Landgraf JM, Kwong LN, Kendziorski CM, Reichelderfer M, Torrealba J, 37. Woo S-H, Lukacs V, De Nooij JC, Zaytseva D, Criddle CR, Francisco A, et al. Weichert J, et al. A target-selected Apc-mutant rat kindred enhances the Piezo2 is the principal mechanotransduction channel for proprioception. modeling of familial human colon cancer. Proc Natl Acad Sci U S A 2007; Nat Neurosci 2015;18:1756. 104:4036–41. 38. Simon A, Shenton F, Hunter I, Banks RW, Bewick GS. Amiloridesensitive 16. Irving AA, Yoshimi K, Hart ML, Parker T, Clipson L, Ford MR, et al. The utility of channels are a major contributor to mechanotransduction in mammalian muscle Apc-mutant rats in modeling human colon cancer. Dis Model Mech 2014;7: spindles. J Physiol 2010;588:171–85. 1215–25. 39. de Nooij JC, Simon CM, Simon A, Doobar S, Steel KP, Banks RW, et al. The PDZ- 17. Washington MK, Powell AE, Sullivan R, Sundberg JP, Wright N, Coffey RJ, et al. domain protein Whirlin facilitates mechanosensory signaling in mammalian Pathology of rodent models of intestinal cancer: progress report and recom- proprioceptors. J Neurosci 2015;35:3073–84. mendations. Gastroenterology 2013;144:705–17. 40. Bewick GS, Banks RW. Mechanotransduction in the muscle spindle. Pflug€ Arch 18. Reagan-Shaw S, Nihal M, Ahmad N. Dose translation from animal to human Eur J Physiol 2015;467:175–90. studies revisited. FASEB J 2008;22:659–61. 41. Gabry J, Goodrich B. rstanarm: Bayesian applied regression modeling via Stan. R 19. Park SB, Goldstein D, Krishnan AV, Lin CSY, Friedlander ML, Cassidy J, et al. package version (2.18.1); 2018. Available at: https://mc-stan.org/rstanarm. Chemotherapyinduced peripheral neurotoxicity: A critical analysis. CA Cancer J 42. Brooks S, Gelman A, Jones G, Meng X-L. Handbook of Markov Chain Monte Clin 2013;63:419–37. Carlo. Boca, Raton, FL: CRC press; 2011. 20. Grisold W, Cavaletti G, Windebank AJ. Peripheral neuropathies from che- 43. Hoffman MD, Gelman A. The No-U-turn sampler: adaptively setting path motherapeutics and targeted agents: diagnosis, treatment, and prevention. lengths in hamiltonian monte carlo. J Mach Learn Res 2014;15:1593–623. Neuro Oncol 2012;14(suppl_4):iv45–i4. 44. Gelman A, Rubin DB. Inference from iterative simulation using multiple 21. Cavaletti G, Tredici G, Petruccioli M, Donde E, Tredici P, Marmiroli P, et al. sequences. Stat Sci 1992;7:457–72. Effects of different schedules of oxaliplatin treatment on the peripheral nervous 45. Metz GA, Whishaw IQ. The ladder rung walking task: a scoring system and its system of the rat. Eur J Cancer 2001;37:2457–63. practical application. J Vis Exp 2009:e1204. DOI: 10.3791/1204.

2954 Cancer Res; 80(13) July 1, 2020 CANCER RESEARCH

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

46. Kwong LN, Dove WF. APC and its modifiers in colon cancer. Adv Exp Med Biol 55. Chalfie M. Neurosensory mechanotransduction. Nat Rev Mol Cell Biol 2009; 2009:656;85–106. 10:44. 47. DeBerardinis RJ, Lum JJ, Hatzivassiliou G, Thompson CB. The biology of cancer: 56. Maksimovic S, Nakatani M, Baba Y, Nelson AM, Marshall KL, Wellnitz SA, et al. metabolic reprogramming fuels cell growth and proliferation. Cell Metab 2008;7: Epidermal Merkel cells are mechanosensory cells that tune mammalian touch 11–20. receptors. Nature 2014;509:617. 48. Hay N. Reprogramming glucose metabolism in cancer: can it be exploited for 57. Proske U, Gandevia SC. The kinaesthetic senses. J Physiol 2009;587: cancer therapy? Nat Rev Cancer 2016;16:635. 4139–46. 49. Burakgazi A, Messersmith W, Vaidya D, Hauer P, Hoke A, Polydefkis M. 58.RudyB,ChowA,LauD,AmarilloY,OzaitaA,SaganichM,etal.Con- Longitudinal assessment of oxaliplatin-induced neuropathy. Neurology 2011; tributions of Kv3 channels to neuronal excitability. Ann N Y Acad Sci 1999; 77:980–6. 868:304–43. 50. Vezzani A, Viviani B. Neuromodulatory properties of inflammatory cyto- 59. Waxman SG, Zamponi GW. Regulating excitability of peripheral afferents: kines and their impact on neuronal excitability. Neuropharmacology 2015; emerging ion channel targets. Nat Neurosci 2014;17:153. 96:70–82. 60. Akemann W, Knopfel€ T. Interaction of Kv3 potassium channels and resurgent 51. Nelson TE, Ur CL, Gruol DL. Chronic interleukin-6 exposure alters electro- sodium current influences the rate of spontaneous firing of Purkinje neurons. physiological properties and calcium signaling in developing cerebellar purkinje J Neurosci 2006;26:4602–12. neurons in culture. J Neurophysiol 2002;88:475–86. 61. Figueroa KP, Minassian NA, Stevanin G, Waters M, Garibyan V, Forlani S, et al. 52. Sittl R, Lampert A, Huth T, Schuy ET, Link AS, Fleckenstein J, et al. Anticancer KCNC3: phenotype, mutations, channel biophysics—a study of 260 familial drug oxaliplatin induces acute cooling-aggravated neuropathy via sodium ataxia patients. Hum Mutat 2010;31:191–6. channel subtype Na(V)1.6-resurgent and persistent current. Proc Natl Acad 62. Burke D, Hagbarth K-E, Lofstedt€ L. Muscle spindle activity in man during Sci U S A 2012;109:6704–9. shortening and lengthening contractions. J Physiol 1978;277:131–42. 53. Ahles TA, Saykin AJ. Candidate mechanisms for chemotherapy-induced cog- 63. Breschi A, Gingeras TR, Guigo R. Comparative transcriptomics in human and nitive changes. Nat Rev Cancer 2007;7:192. mouse. Nat Rev Genet 2017;18:425. 54. Argyriou AA, Bruna J, Genazzani AA, Cavaletti G. Chemotherapy-induced 64. Marshall TF, Zipp GP, Battaglia F, Moss R, Bryan S. Chemotherapy-induced- peripheral neurotoxicity: management informed by pharmacogenetics. Nat Rev peripheral neuropathy, gait and fall risk in older adults following cancer Neurol 2017;13:492. treatment. J Cancer Res Pract 2017;4:134–8.

AACRJournals.org Cancer Res; 80(13) July 1, 2020 2955

Cancer Exacerbates Chemotherapy-Induced Sensory Neuropathy

Stephen N. Housley, Paul Nardelli, Dario I. Carrasco, et al.

Cancer Res 2020;80:2940-2955. Published OnlineFirst April 28, 2020.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-19-2331

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2020/04/28/0008-5472.CAN-19-2331.DC1

Cited articles This article cites 56 articles, 9 of which you can access for free at: http://cancerres.aacrjournals.org/content/80/13/2940.full#ref-list-1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at Subscriptions [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/80/13/2940. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.