Early prodromal sensory neuropathy in Pink1-SNCA double mutant Parkinson mice

BACKGROUND: Parkinson's Disease (PD) is frequently associated with a prodromal sensory neuropathy, which manifests as sensory loss and eventually chronic pain. METHODS: The present study assessed somatosensory functions and the pathology of somatosensory neurons in Pink1-/-SNCAA53T double mutant mice (Pink1SNCA) who develop a PD-like disease > 15 months of age. RESULTS: Pink1SNCA mice showed the rst manifestations of sensory dysfunctions at 6 months of age, which were characterized by a loss of thermal sensitivity but sparing of mechanical sensation, and decline of olfactory perception of pleasant odors. Mechanistically, the loss of thermal sensitivity was attributed to a strong reduction of transient receptor potential TRPV and TRPA channels at mRNA, protein and functional levels in the dorsal root ganglia (DRGs). Primary sensory neurons responded much less to stimulations with capsaicin, formalin or depolarization as assessed by calcium imaging. RNA sequencing pointed to additional deregulations of voltage dependent potassium channel subtypes, lowering of acidic broblast growth factor driven support, and alterations of sphingolipid metabolism. Analyses of sphingolipids of multiple classes revealed an accumulation of glucosylceramides mainly in the DRGs and loss of sphingosines and sphinganines in the sciatic nerve. Transmission electron microscopy of the DRGs revealed a higher frequency of neurons with swollen mitochondria, which was associated with a decrease of mitochondrial respiration in sensory ganglia, but not in the brain. CONCLUSION: The data support the view of a prodromal sensory disease in PD resulting from a combination of lipid allostasis and mitochondrial damage. Hence, Pink1SNCA mice phenocopy the PD-associated prodromal sensory neuropathy and reveal novel mechanistic insight. part inconclusive. the present study, we used a previously described double mutant mouse model of PD to study premotor sensory phenomena and biological and morphological correlates in primary somatosensory neurons of these mice to understand better the underlying pathophysiological mechanisms and the progressive nature of the sensory loss. The mice carry a loss of function knock-in mutation of PTEN induced kinase (Pink1) and they are heterozygous for the human A53T mutation of alpha synuclein (SNCA-A53T). These Pink1SNCA mice develop spontaneous motor function decits at advanced ages with an incidence of about 30% above 15 months of age. We show that sensory decits occur much and are associated with a prominent loss of transient receptor potential channels, mitochondrial damage and alterations of bioactive lipid patterns in somatosensory ganglia and peripheral nerves strengthening the pathological role of glucosylceramides.

The predominant morphological features of PD are intraneuronal alpha-synuclein-rich deposits, the major components of Lewy bodies [2,23,24], but the pathophysiology is only partly understood. Mitochondrial damage, dysfunctions of protein degradation via the proteasome and via autophagolysosomes contribute to the progressive loss of dopaminergic and other neurons. Defective mitophagy can lead to leakage of mitochondrial DNA to the cytosol that triggers innate immune system responses. In addition, it has been recognized in recent years that metabolic deregulations of bioactive lipids including ceramides and their metabolites increase the toxicity of alpha-synuclein [25][26][27]. Patients who are heterozygous carriers of glucocerebrosidase (GBA1) mutations tend to develop a rapidly progressive disease [28], and such mutations propagate alpha synuclein deposition in PD model organisms [29][30][31][32]. GBA1 is a lysosomal enzyme that catalyzes the degradation of ceramides to glucosylceramides (GlcCer) and subsequent generation of lactosylceramides (LacCer) and gangliosides. Low GBA1 activity and accumulations of GlcCer even occur in PD patients without known mutations of GBA1 [33,34] and point to a link between PD and lysosomal storage disorders [35].
Indeed we have recently observed in patients with sporadic PD that high concentrations of plasma GlcCer are associated with pain ratings and with sensory losses as assessed via quantitative sensory testing, suggesting that GlcCer may contribute to the progression of the disease from peripheral sensory neurons to the brain and to clinical PD-associated pain (accompanying manuscript). There are no apparent remedies to speci cally alleviate PD-associated pain or prevent sensory losses [36,37] in part owing to limited knowledge of PD-associated somatosensory pathology in model organisms.
Motor functions and morphological characteristics of PD have been studied in several rodent-, primateand y models of the disease, but sensory functions and underlying pathological features of sensory neurons were rarely addressed [38] and the few studies of non-motor manifestations such as olfaction, anxiety-like behavior and gastrointestinal functions are fragmented and in part inconclusive.
In the present study, we used a previously described double mutant mouse model of PD to study premotor sensory phenomena and biological and morphological correlates in primary somatosensory neurons of these mice to understand better the underlying pathophysiological mechanisms and the progressive nature of the sensory loss. The mice carry a loss of function knock-in mutation of PTEN induced kinase (Pink1) and they are heterozygous for the human A53T mutation of alpha synuclein (SNCA-A53T). These Pink1SNCA mice develop spontaneous motor function de cits at advanced ages with an incidence of about 30% above 15 months of age. We show that sensory de cits occur much earlier and are associated with a prominent loss of transient receptor potential channels, mitochondrial damage and alterations of bioactive lipid patterns in somatosensory ganglia and peripheral nerves strengthening the pathological role of glucosylceramides.
Mice had free access to food and water, and they were maintained in climate-controlled rooms with a 12 h light-dark cycle. Behavioral experiments were performed between 10 am and 3 pm. The experiments were approved by the local Ethics Committee for animal research (Darmstadt, Germany), adhered to the guidelines for pain research in conscious animals of the International Association for the Study of PAIN (IASP) and those of the Society of Laboratory Animals (GV-SOLAS) and were in line with the European and German regulations for animal research.

Analysis of nociception and somatosensory functions
Nociceptive tests were performed at 2, 4, 10 and 12 months of age with 8 mice per group. Mice were habituated to the test room and the test chambers for three consecutive days before the baseline measurement.
The latency of paw withdrawal on pointy mechanical stimulation was assessed using a Dynamic Plantar Aesthesiometer (Ugo Basile, Comerio, Italy). The steel rod was pushed against the plantar paw with ascending force (0-5 g, over 10 s, 0.2 g/s) and then maintained at 5 g until the paw was withdrawn. The paw withdrawal latency was the mean of three consecutive trials with at least 30 s intervals.
The sensitivity to painful heat stimuli was assessed by recording the paw withdrawal latency with a Hot Plate (52 °C or 30-55 °C surface, Föhr Medical Instruments, Germany) or with the Hargreaves test (IITC Life Science). In the latter, an infrared lamp was placed with a mirror system underneath the respective hind paw. By pressing the start button the lamp starts to emit a heat-beam until the paw is withdrawn, which stops the lamp. The mean paw withdrawal latency of three tests with at least 10 min intervals was used for statistical analysis.
The Orofacial Pain Assessment Device (OPAD, Stoelting, Ireland) allows for evaluation of facial thermal nociception by using a reward-con ict paradigm. The OPAD cage consists of a plexiglas chamber with metal grid oor and an adjustable slit giving access to the nipple of the reward bottle, which is anked with two PC-controlled thermal peltier elements. To receive the reward (diluted milk in water), the mouse has to touch the thermodes with its cheeks. Hence, the mouse has to decide between receiving a reward or escaping the aversive heat or cold stimulus. Mice were fasted overnight to increase the appetite. Mice were trained three times a week for two weeks at innocuous temperatures (36.5-38 °C) to get a stable baseline of at least 600 licks in 18 min. During test periods, temperature circles starting from neutral (37 °C, 3 min) to aversive cold (15 °C, 10 °C each 3 min) or aversive heat (45 °C, 54 °C each 3 min) were applied using a ramping protocol. Circles were repeated up to 20 min. The ANY-maze software (Version 4.99, Stoelting) registered licks and contacts with the thermodes. The average and total numbers of lickings and contacts at the de ned temperatures were used for analysis. Ramping times were excluded.

Analysis of olfactory functions
The Novel Odor Recognition Test (NOdorR) was used to assess whether mice were able to detect and differentiate between odors. Mice were habituated to a 3-chambered test box (sizes as for social tests) for 3 days, each for 10 min. The test sessions consisted of three different stages. During (1) habituation, mice were again adapted to the box for 10 min. During (2) "Learning", two petri dishes with perforated lids containing 2 × 2 cm lter papers soaked with identical odors (1% vanilla) were placed in both side chambers. The mouse was placed in the middle and was allowed to explore odors for 10 min. In (3) "Novelty", the settings were identical to the learning stage except that one of the familiar odors (1% vanilla) was replaced with a novel one, which was undiluted rose water. After each test session, the open eld was carefully cleaned. Filter papers and petri dishes were freshly prepared for each mouse. Visits of chambers, visits of odors and the time spent with odors were recorded with the VideoMod2 Software. In analogy, odor preferences were tested for urine mixtures of stranger mice.

Analysis of social cognition and memory
The social test apparatus consisted of a rectangular, three-chambered box (42.5 cm wide, 22.2 cm high) with a central chamber 17.8 cm in length and two side chambers each 19.1 cm in length, according to recommended speci cations [78]. The central chamber was separated from the side chambers by removable partitions with doors allowing the animal to move freely between chambers. The "stranger" stimulus mouse was positioned in a grid enclosure (Ugo Basile, Italy) allowing close interactions and nose contact but preventing the stranger mouse from initiating social contact. The enclosures had an internal diameter of 7 cm, and height of 15 cm. The top and bottom were constructed from grey PVC.
We used two stimulus mice in each experiment with the same gender and strain as the test mouse, but housed in another room. Mice were habituated to the test box for 3 days, each for 10 min. The sessions were similar as for the NOdorR: (1) Habituation: the animal was placed in the middle chamber with the dividers closed to allow exploration of the middle chamber for 5 min. (2) Sociability test: after the 5-min habituation period, an unfamiliar adult gender-matched mouse (stranger 1) was placed inside the grid enclosure in one of the side chambers. An identical empty grid enclosure was placed in the opposite chamber. The dividers were then raised, allowing the test mouse to move freely among the chambers over a 10-min test session. (3) Novelty: The original stranger mouse (stranger 1) remained in its grid enclosure on one side of the apparatus. A new unfamiliar mouse (stranger 2) was placed in the grid enclosure on the opposite side. The behavior was observed for 10 min. Visits of strangers and the time spent with strangers were recorded. A "visit of stranger" required close proximity of the nose with the grid enclosure with a maximum distance of one cm.

Phenomaster
The TSE Phenomaster offers an automated metabolic and behavioral monitoring in home cage environments. Drinking and feeding behavior were monitored with high-precision weight sensors for liquid and food dispensers, which are integrated into the lid of the cage. Mice were adapted to the drinking bottles for one week in their home cage and to the Phenomaster® cage for 2 consecutive days before starting the experiment. Drinking, feeding and voluntary wheel running were recorded for 24 hours.

Analysis of motor functions
Motor coordination and endurance were assessed with the accelerating RotaRod test (16-32 rpm, ramp 3 rpm/min, cut-off 5 min; Ugo Basile, Italy) or a RotaRod at constant speed. Mice were trained with 2-3 training runs. The running time in test trials was averaged from three trials. The cut-off time was 300 s.
Motor functions of the front limbs were assessed with the mouse staircase test (Campden Instruments Ltd., UK). The test apparatus comprises two compartments, a start compartment with a hinged lid and a test compartment containing a small central platform with two staircases (8 steps) on each side. Each step has a cup for a reward, which was lled with sweet pellets. At the beginning, mice were set on a restriction diet to increase the appetite. Mice were adapted to the sweet rewards for three days in the home cage (d 1-3) and in two training trials (d 5-6). Test sessions were performed at day 7 to 9 and lasted 30 min. Mice were placed into the test compartment, and the behavior to grasp, lift and collect food was monitored with a video camera, and analyzed with VideoMod2. The numbers of pellets remaining on the steps or oor were counted.
Grip strength was assessed with a computerized Grip-Strength Meter (Model 47200, Ugo Basile, Italy). The apparatus consists of a grasping grid connected to a force sensor. To measure the grip strength of front limbs and/or hind limbs, mice were held by the tail and allowed to grasp the grid. As soon as the mouse grasped the grid, the mouse was pulled backwards by the tail until the grip was lost. The peak force of each measurement was automatically recorded in gram-force (gf) by the software. The grip strength was the mean of three trials.
Culture and staining of primary DRG neurons Primary adult dissociated DRG neuron-enriched cultures were prepared by dissecting mouse dorsal root ganglia (DRGs) into 1x PBS (Phosphate Buffered Saline, Gibco, Germany), followed by digestion with 5 mg/ml collagenase A and 1 mg/ml dispase II (Roche Diagnostics, Germany). Triturated cells were centrifuged through a 15% BSA (bovine serum albumin) solution and plated on poly-L-lysine and laminin coated cover slips in Neurobasal medium (Gibco) containing 2% (vol/vol) B27 supplement (Gibco), 50 µg/ml Pen-Strep, 100 ng/ml NGF and 200 mM L-glutamine. After incubation for 2 h, 2 ml Neurobasal medium was added and neurons were cultured for up to 48 h depending on the experimental requirements. Cells were kept at 37 °C, 5% CO2, 95% humidity. Primary DRG neurons were used for calcium imaging and immunohistochemistry stainings.

Calcium Imaging
Calcium uxes were measured uorometrically as the ratio of the absorbances at 340 and 380 nm (F 340/380) in cultured adult DRG neurons (see DRG primary cultures). Calcium-imaging experiments were performed with a Leica calcium-imaging setup, consisting of a Leica DMI 4000 b inverted microscope equipped with a DFC360 FX (CCD) camera, Fura-2 lters, and an N-Plan 10x/0.25 Ph1 objective lens (all from Leica). Images were captured every two seconds and were processed with the LAS AF-software (Leica). Cells were loaded with 5 µM of the Ca 2+ -sensitive uorescent dye Fura-2-AM-ester (Biotium), incubated for 40 min at 37 °C and washed three times with ringer solution (Fresenius). Coverslips were then transferred to a perfusion chamber with a ow rate of 1-2 ml/min at room temperature. Baseline ratios were recorded with ringer solution for 100-180 s, before application of either 0.01% formalin to activate TRPA1 or 50 nM capsaicin (Sigma) to activate TRPV1 ion channels for 100 s or 26 s, respectively. After wash-out with ringer solution, cells were perfused with 100 mM KCl (high K+) to assess depolarization-evoked calcium currents and the viability of the neurons. Data are presented as changes in uorescence ratios (F340/380) normalized to baseline ratios. The analysis encompassed 300-350 neurons per condition of 10-12 independent DRG cultures of each three mice per group per stimulus, which were 18 months old. The maximum, the time of maximum and area of the fold increase versus time curve was calculated by integration (Origin Pro 2020 software) with the baseline xed at Y = 1. The time courses and areas were used for statistical comparison.

Quantitative real-time PCR (QRT-PCR)
Mice were sacri ced via carbon dioxide followed by rapid blood withdrawal. The dorsal root ganglia were rapidly dissected and frozen in liquid nitrogen. Total RNA was extracted from mouse tissue and reversely transcribed with random primers. Twenty nanograms of cDNA equivalent were subjected to quantitative real-time polymerase chain reaction (qRT-PCR). The PCR reaction and amplicon detection were done on a TaqMan (AB 7500 Applied Biosystems; Life Technologies Corporation, Carlsbad, CA) using a FastStart Universal Master Mix (Roche Diagnostics, Mannheim, Germany) with SYBR Green uorescence staining. TRP channel mRNA were ampli ed with speci c primers and were normalized to the housekeeping gene protein phosphatase 1 (PPP1CA). The comparative threshold cycle (CT) method was used for quanti cation of relative mRNA expression. The primer pairs are summarized in Suppl. Table 1a.

Immunohistochemistry
Mice were terminally anesthetized with iso urane and cardially perfused with cold 0.9% NaCl followed by 2% paraformaldehyde (PFA) for xation. Tissues were excised, post xed in 2% PFA for 2 h, cryoprotected overnight in 20% sucrose at 4 °C, embedded in small tissue molds in cryo-medium and cut on a cryotome (10 µm for DRGs and ScN, 12 µm for SC). Slides were air-dried and stored at − 80 °C. After thawing, slides were immersed and permeabilized in 1xPBS with 0.1% Triton-X-100 (PBST), then blocked with 3% BSA/PBST, subsequently incubated overnight with the rst primary antibodies in 1% BSA/PBST at 4 °C.
After washing three times with PBS, slides were incubated with the secondary antibodies for 2 h at room temperature, followed by 10 min incubation with DAPI and embedding in Aqua-Poly/Mount. The general settings were optimized for the respective antibodies and tissues. Primary antibodies included STING, IFI16, Tuji, Ceramides, SNCA (Suppl . Table 1b). Secondary antibodies were labeled with uorochromes (Invitrogen, Sigma, Life Technologies). Slides were analyzed on an inverted uorescence microscope (BZ-9000, KEYENCE, Germany and Axio Imager Z1, Zeiss, Germany). Tiled images were obtained to reconstruct whole DRG or spinal cord sections to assess different regions of DRG neurons and bers.
For analysis of neurite outgrowth and morphology, primary neuron cultures were washed in PBS, xed in 4% PFA and immunostained with antibodies directed again neuro lament of 200 kDa and subsequent Alexa-488 or Cy3 labeled secondary antibodies, and with phalloidin-Alexa-594. Images were captured on an inverted Axio Imager Z1 uorescence microscope (Zeiss, Jena, Germany). Tiled images were obtained to reconstruct the whole cultures and assess density of neurons and dendrites.
For quanti cation, RGB images were converted to binary images using threshold setting implemented in ImageJ with minor adjustments. The particle counter of FIJI ImageJ was used to assess the area covered with neuronal immunoreactive structures.

Transmission electron microscopy
Mice were terminally anaesthetized with carbon dioxide and perfused transcardially with cold 0.9% sodium chloride (NaCl) followed by perfusion with the xation solution containing 4% PFA and 4% glutaraldehyde in 0.1 M cacodylate buffer, pH 7.4. DRGs were excised and post xed in the xation solution for 90 min. After washing 3 times with 0.1 M cacodylate buffer, the tissue was osmicated (1% OsO4 in 0.1 M cacodylate buffer) for 60 min, washed 3 times with 0.1 M cacodylate buffer and dehydrated in ethanol, 1% uranyl acetate and propylene oxide. The tissue was then incubated in Durcupan resin (Fluka, Durcupan ACM-Kit) overnight and embedded in fresh Durcupan resin at 56 °C for 48 h. Ultrathin sections (54 nm) were cut with a diamond blade, collected on single slot Formvar-coated copper grids and contrasted with lead citrate. Sections were analyzed using a Zeiss electron microscope (Zeiss EM900) and imaged with a slow-scan CCD-Camera.
Oxygraph analysis of mitochondrial OXPHOS activity Brain sections (prefrontal cortex, hippocampus) and DRG plus trigeminal nerve plus ganglion were immediately transferred into ice-cold Respiration medium MiR06Cr containing 280 U/ml catalase (Pesta&Gnainer, 2012). Tissue was homogenized using a motor-driven tightly tting glass/Te on Potter-Elvehjem homogenizer. Mitochondrial respiration was measured using high-resolution respirometry (Oxygraph-2 k, Oroboros Instruments, Innsbruck, Austria) with DatLab software 6.1.0.7 (Oroboros Instruments, Innsbruck, Austria). LEAK-respiration was induced by the addition of complex I -linked substrates pyruvate (5 mM), malate (0.5 mM) and glutamate (10 mM). Complex I -linked respiration was measured after adding ADP (2.5 mM) in a saturating concentration. To measure complex II-linked respiration, rotenone (0.5 µM) was added to block complex I followed by the addition of succinate (10 mM). Maximum uncoupled respiration (ETS, electron transfer system capacity) was measured after stepwise titration of FCCP ((carbonyl cyanide-p-tri uormethoxyphenylhydrazone). Residual oxygen consumption (ROX) was determined after sequential inhibition of complex III with Antimycin A and complex IV with azide. Absolute respiration rates were corrected for ROX and normalized for the protein content.
In brief, tissue samples were homogenized by adding ethanol:water (1:3, v/v, tissue concentration 0.02 µg/ml) using a ball mill mixer (MM400, Retsch, Haan, Germany) with 4 zirconium oxide grinding balls (25 Hz for 2.5 min). Subsequently, 50 µl of the homogenate were extracted using a liquid-liquidextraction method. Plasma sample volumes were 10 µl. The quanti cation of all analytes was performed using a hybrid triple quadrupole-ion trap mass spectrometer QTRAP 5500 (Sciex, Darmstadt, Germany) equipped with a Turbo-V-source operating in positive ESI mode. Sphingolipids were separated using an Agilent 1290 HPLC system equipped with a Zorbax C18 Eclipse Plus UHPLC column (50 Å~ 2.1 mm, 1.8 µm, Agilent technologies, Waldbronn, Germany). Quality control samples of three different concentration levels (low, middle, high) were run as initial and nal samples of each run. For all analytes, the concentrations of the calibration standards, quality controls and samples were evaluated by Analyst software 1.6 and MultiQuant Software 3.0 (Sciex) using the internal standard method (isotope-dilution mass spectrometry). Calibration curves were calculated by linear or quadratic regression with 1/x weighting or 1/x2 weighting. Variations in accuracy of the calibration standards were less than 15% over the range of calibration, except for the lower limit of quanti cation (LLOQ), where a variation in accuracy of 20% was accepted. Lipid concentrations in tissue are expressed as pg/mg of the tissue or ng/ml of plasma. For multivariate statistical analyses, lipid data were normalized on the mean value of the respective tissue, and they are expressed as percentages of the mean because the concentrations of different lipids differ by several orders of magnitude.

Statistics
Group data are presented as mean ± SD or mean ± sem, the latter for behavioral time course data, speci ed in the respective gure legends. Data were analyzed with SPSS 24 and Graphpad Prism 8.0 and Origin Pro 2020. Data were mostly normally distributed, or log-normally distributed. For testing the nullhypothesis that groups were identical, the means of two groups were compared with 2-sided, unpaired Student's t-tests. The Mann Whitney U test was used as a non-parametric alternative in case of violations of t-test requirements. Time course data or multifactorial data were submitted to 2-way analysis of variance (ANOVA) using e.g. the factors 'time' and 'genotype'. In case of signi cant differences, groups were mutually compared at individual time points using post hoc t-tests according to Dunnett, i.e. versus the control group, or according to Šidák. In case of violations of sphericity, degrees of freedom were adjusted according to Huynh Feldt. Asterisks in gures show multiplicity-adjusted P-values. The areas under the curve of calcium imaging data were log-normally distributed and were Log2-transformed for statistical comparisons. The areas were calculated by integration and the distribution was obtained by Kernel density estimation (Origin Pro).
Principal component analysis was used to de ne the lipid species, which accounted most for the variance between genotypes. Further multivariate analyses included canonical discriminant analysis to assess the predictability of group membership and separation of genotypes and tissues. Partial least square analysis was used if analytes exceeded the number of samples per group. The statistical analysis of RNAseq data is explained below.

RNA sequencing and analysis
DRGs were removed and ash-frozen on dry ice. RNA was harvested using Trizol reagent. Illumina TruSeq RNA Sample Prep Kit (TruSeq Total RNA with Ribo-Zero rRNA depletion) was used with 1 µg of total RNA for the construction of sequencing libraries. Libraries were prepared according to Illumina's instructions.
Sequencing was performed with an Illumina Next Generation sequencing system with a sequencing depth of 50 Mio reads per sample.
Sample quality was assessed with demultiplexed fastq.gz les and subsequently the alignment was performed with SeqMan NGen 16 (Lasergene) using the reference genome mm10 provided from UCSC (GRCm38) [80] as template, a minimum read length of 35 bp and automatic adapter trimming. Results were displayed with ArrayStar 16 (Lasergene) including the amount of mapped reads, target length, source length and position, strand, genes and gene IDs, annotated according to the mm10 assembly, and reads were normalized according to RPKM. Alternative DEseq2 normalization yielded equivalent results.
Normalized reads were analyzed with ArrayStar, which uses general linear models to assess differential expression. Genes were ltered for at least 10 valid values out of 17 samples with normalized reads > 0.05 to exclude low expression genes. Data were log2 transformed, single missing values were imputed from the normal distribution, and results were displayed as scatter plots, MA-plots and Volcano plots, the latter showing the log2 difference i.e. fold change (positive for upregulated genes and negative for downregulated genes) versus the -log10 of the t-test P value. The P value was set at 0.05 and adjusted according to Benjamini Hochberg. Hierarchical clustering was employed to assess gene expression patterns using Euclidean distance metrics. Results were displayed as heat maps with dendrograms.
Key regulated genes (based on P-value, fold change and abundance) were further analyzed for gene ontology annotation enrichments for 'cellular component', 'biological process' and 'molecular function', KEGG, Biocarta and Reactome pathways, SMART domains and SP-PIR-keywords to assess common localizations and functions of signi cantly regulated genes. GO analyses were done with the "term enrichment" and "functional gene clustering" tools of The Database for Annotation, Visualization and Integrated Discovery (DAVID, version 6.8) (http://david.abcc.ncifcrf.gov/home.jsp) [81]. In addition, gene set enrichment analyses (GSEA) (http://www.gsea-msigdb.org) [82] were used to further assess functional implications of up-or downregulated genes and to obtain a gene ranking and heat map of the leading edge 50 up-and downregulated genes. GSEA generates ranked gene lists based on fold difference and P value. The RNAseq data have been deposited as GEO dataset with the provisional accession number GSE146091.

Results
Loss of thermal nociception and of nutritive olfaction in middle-aged Pink1SNCA mice General health and adequate motor functions are prerequisites to interpret sensory tests, which rely on paw withdrawal or locomotion. During adult life up to middle age (9-10 months of age), Pink1SNCA mice had normal body weight (Fig. 1A), and they behaved normally in most motor function tests including voluntary wheel running, grip strength (Fig. 1A) and distances travelled in mazes. Only in the accelerating Rota Rod, half of the Pink1SNCA mice did not reach the upper target running time of 5 min, whereas all wildtype controls succeeded (Fig. 1A). Hence, voluntary locomotion, which is required for paw withdrawal, is unimpaired at this age.
Pink1SNCA mice had normal paw withdrawal latencies upon mechanical stimulation, but showed a dramatic loss of thermal heat sensitivity in Hargreaves and Hot Plate tests (Fig. 1B), and they had lost the normal warm preference in a Thermal Gradient Ring test, in which mice can freely choose their temperature of well-being (Fig. 1C). The behavior suggests a loss of thermal sensation, which was con rmed for trigeminal hot/cold stimulation in the Orofacial Pain Assessment Device (OPAD), in which mice have to touch heated or cooled metal bars to get access the a rewarding milk bottle. Pink1SNCA had more contacts and more licks at unpleasant hot or cold temperatures suggesting that they did not feel the unpleasantness (Fig. 1D).
In the Novel Odor Recognition and memory test, Pink1SNCA mice were less interested in the pleasant odors vanilla or rose than wildtype control mice, and they did not reach out to the sweet pellets presented on the steps of a staircase test, suggesting a loss of olfactory sensitivity (Fig. 1E). However, they maintained a stronger appeal to vanilla than rose like the controls suggesting that they were able to discriminate between odors (Fig. 1E).
Repetition of the odor recognition test with social odors (urine mixtures) yielded different results. Pink1SNCA responded equally to the social odor and were more interested in the novel odor relative to the familiar one in the second phase of the test (Fig. 1F). Sensation and processing of pleasant odors in mice requires dopaminergic neurons of the olfactory bulb, which are a primary site PD pathology, whereas social odors depend on neurons of the vomeronasal organs and accessory olfactory bulb [39], which are likely less affected in PD mice. Social behavior is therefore preserved. In con rmation, social contacts with familiar or stranger mice in the sociability test were normal in Pink1SNCA mice (Fig. 1G). Analyses of the total distances travelled in the maze tests revealed that Pink1SNCA mice were overactive in social tests or social environments (Fig. 1H).

Onset of premotor sensory loss
Nociceptive tests were repeated in younger mice to assess the onset of thermal sensory losses. There were no differences at 2 months of age but the rst signi cant protraction of paw withdrawal upon heat stimulation occurred at 4 months of age ( Fig. 2A), hence 11 months before the earliest occurrence of spontaneous motor de cits. Motor function Rota Rod tests at 4 months were not impaired, but some Pink1SNCA mice needed more trials to achieve the running goal of 300 s, whereas all wildtype controls succeed on the rst or second training run ( Fig. 2A).
There was no sensory loss in single mutant mice at 9-10 months of age but rather nociceptive hypersensitivity for mechanical stimuli in Pink1 −/− mice, or heat stimuli in SNCA A53T mice, suggesting a progression from hypersensitivity to sensory loss with the number of molecular defects.

RNA sequencing reveals loss of acidic broblast growth factor in the DRGs
The behavior of Pink1SNCA mice shows a prodromal loss of thermal somatosensation, which may be caused by a predominant loss of sensory neurons with unmyelinated C-bers, or a speci c downregulation of ion channels, which contribute to the sensation of heat and cold. In search of putative mechanisms, we performed gene expression analyses by RNA sequencing in DRG tissue of prodromal middle-aged to old mice (Fig. 3). Clustering of mice for the 1000 top regulated genes (at 90% con dence level with 1.5fold change) perfectly separated Pink1SNCA from wildtype mice. Apart from the knockout of Pink1, a prominent down-regulated gene was acidic broblast growth factor, Fgf1 (Fig. 3B, C), which is known to restore the survival of dopaminergic neurons in PD models [40,41] and contributes to neurogenesis [42]. Up-regulated genes included DNase1l3, CCL27 and CCL25, IFI44 and IL31ra, and pointed to increased DNAse activity and activation of the immune system.
Gene ontology analyses agreed with the top candidates. The major terms, which were enriched in downregulated genes (top 200) were "proteolysis", "cell differentiation", and the cellular compartment, "transmembrane". Major terms in upregulated genes were "nuclease", "lipid metabolism", "ion channel/ion transport" and "chemotaxis/immune response". GSEA gene set enrichment analysis agrees with these top GO terms, in particular a negative enrichment of "Notch signaling" (Hallmark gene sets), "axon guidance" (Reactome gene sets) or "focal adhesion" (KEGG gene sets). Leading edge genes according to GSEA are shown in Suppl. Figure 1 and selected pathways in Suppl. Figure 2 Hence, from the behavioral studies with loss of thermal sensation and from the RNAseq studies with prominent deregulations of Fgf1, ion channels, innate immune genes and DNase we decided to follow up on key aspects that may ultimately lead to thermal sensory loss, i.e. TRP channel dysfunction, sphingolipid metabolism, mitochondrial dysfunction and DNA triggered innate immune responses. The RNAseq data have been deposited as GEO dataset with the provisional accession number GSE146091.
Loss of TRP-channel dependent calcium uxes in sensory neurons of the DRGs The behavioral results pointed to a speci c early-onset loss of thermal sensation, agreeing well with results of Quantitative Sensory Tests in sporadic PD patients, where the predominant QST phenotype is thermal sensory loss plus mechanical hypersensitivity. Thermal sensitivity relies on the expression and functioning of TRP channels, mainly TRPVs for heat, and TRPA1 and TRPMs for cold.
To assess functions of these TRP channels, primary DRG neurons of 18 months old Pink1SNCA and wildtype mice were stimulated with capsaicin to activate TRPV1 and formalin to activate TRPA1. Calcium in ux was determined as Fura2 absorbance ratios from 300-350 neurons per group and stimulus. High-K + responsive neurons (80-90% in each culture) were used for group comparisons (Fig. 4). Time courses of [Ca 2+ ]i and violin plots of the areas under the curve (AUC) revealed lower peak calcium in ux and reduced net increases upon stimulation irrespective of the stimulus (Fig. 4A-D). DRG neurons of Pink1SNCA mice were more heterogeneous than WT neurons and showed a triple-phasic distribution upon stimulation with capsaicin or formalin representing non-responding, low-responding and highresponding neurons. Similarly, a double peak distribution upon High-K + stimulation contrasted the monophasic distribution of the wildtype neurons, and revealed that about half of the Pink1SNCA neurons had lower depolarization-evoked calcium raises. The morphology of primary DRG neuron cultures, neurite outgrowth and immuno uorescence analyses of markers of autophagy and protein aggregates did not differ between genotypes (Suppl. Figure 3). Hence, the changes of excitability were not caused by neuronal death.
Immuno uorescence analyses of TRPV1 in DRGs showed a reduction of TRPV1-immunoreactive neurons in DRGs of Pink1SNCA mice (Fig. 4E, quanti cation 4F), which was con rmed at RNA level via rt-PCR for TRPV1 and TRPA1 (Fig. 4G). The RNA pro le of various TRP channels obtained via RNAseq showed a reduction of TRPVs and TRPA1, and increase of TRPMs, whereas TRPCs were mostly unaffected (Fig. 4H). The gene expression analyses agree with the calcium imaging results.
Immuno uorescence studies of DRGs with neuronal markers and markers of autophagy and apoptosis did not reveal overt differences of autophagy or neuronal death (Suppl. Figure 4), suggesting that the observed alterations of speci c ion channels were not the consequence of neuronal loss. Indeed, there were no differences of gene expression of voltage-gated calcium or sodium channels, but a number of subunits of voltage-gated potassium channels were increased (Suppl. Figure 2A), which regulate the excitability of somatosensory neurons [43] and responsiveness of TRP channels [44]. Biogenic amine receptors and transporters were in part deregulated (Suppl. Figure 2B), whereas mitophagy/autophagy marker were mostly unaffected (Suppl. Figure 2C), except Pink1 itself.

Accumulation of glucosylceramides in DRGs and loss of sphingolipids in the sciatic nerve
We have previously shown that ceramides accumulate progressively in the olfactory bulb of Pink1 −/− single mutant mice [45], and glucosylceramides altered the responsiveness of TRP channels [46], suggesting that alterations of sphingolipids contribute to PD-associated sensory dysfunctions.
To address this aspect, sphingolipids of different classes were analyzed in DRGs, sciatic nerve (ScN), spinal cord (SC) and plasma from 12 months old Pink1SNCA and wildtype control mice (n = 8 per group).
The analysis encompassed dehydro-ceramides, ceramides, glucosyl-and lactosylceramides, as well as sphinganine, sphingosine and the respective phospho-SPH's. Plasma levels did not differ between groups (not shown), but there were substantial differences in tissue levels, which mostly originated from DRGs and ScN. Discriminant Principal Component Analysis using the lipids in DRGs and ScN as input revealed that GluCer16:0 (increased), the SPH's (decreased) and LacCer24:1 (decreased) accounted for most of the variance between groups. Based on PC1 and PC2, groups were clearly separated (Fig. 5A). Scatter 3D plots of key lipids with all tissues (i.e. also including spinal cord) show some overlap of 95% CI ellipsoids (mainly caused by SC) but still separate groups (Fig. 5B). Further pairs were compared in a scatter matrix (Suppl. Figure 5A) and reveal a clear group separation based on GluCer16:0 versus either sphinganine or sphinganine-1-phosphate. Canonical Discriminant analysis shows the tissue speci c group separations (Suppl. Figure 5B). Polar plots showing the mean of normalized lipids as percentages (relative to the mean of all mice set to 100%) reveal the tissue speci c patterns of sphingolipid alterations (Fig. 5C, Suppl. Figure 5C). In the DRGs, the accumulation of ceramides is prominent whereas loss of sphingosines and sphinganines predominates in the sciatic nerve. The spinal cord is not affected. The increase of ceramides was also detected in DRG neurons of aged mice (Fig. 4D). Tissue speci c concentrations and ANOVA statistics of lipids are shown in Fig. 6.
Accumulation of GlcCer and lower levels of LacCer in the DRGs suggest that the lysosomal metabolism of GlcCer via glucocerebrosides (GBA1) to LacCer is impaired, which would agree with associations of GBA1 dysfunctions with PD [28], and with increased alpha synuclein toxicity in the presence of GBA1 mutations [27,29,47]. Gene expression analysis (RNAseq) in the DRGs of our mice did not reveal lower GBA1 transcription (Suppl. Figure 2D). Instead, low ceramidase (Asah2) and increases of Cers4 (ceramide synthase), Degs1 (sphingolipid desaturase) and Smpd5 (sphingomyelin phosphodiesterase) suggest a combination of lower ceramide breakdown and increased formation.
Reduction of oxygen consumption of sensory neurons pointing to mitochondrial defects Intracellular accumulation of GlcCer owing to GBA1 mutations are supposed to interfere with mitochondrial functions and autophagolysosomal removal of damaged mitochondria via mitophagy [48,49]. Using oxygraph respirometry, we measured cellular oxygen consumption and OXPHOS activity with 'substrate-uncoupler-inhibitor' titration protocols in freshly prepared tissue of the prefrontal cortex, hippocampus and of sensory ganglia (DRGs and trigeminal ganglia) in 18 months old Pink1SNCA versus wildtype control mice (Fig. 7A exemplary respirograms and 7B quanti cation). There were no differences between genotypes in the brain, but oxygen consumption and Complex I and Complex II activity were signi cantly reduced in sensory ganglia of Pink1SNCA mice. There were no differences in citrate synthase activity at any site (Suppl. Figure 5D).
The ultrastructural morphology of dark neuronal mitochondria (crista type) in the DRGs of 16 months old mice was alike in Pink1SNCA and wildtype control mice (Fig. 8). There were also no other overt differences in terms of cellularity, neuronal morphology, myelin sheaths or vascular cells (Fig. 8). The DRG neurons displayed a high number of autophagolysosomes, which were present in similar numbers in both genotypes, and also occur in DRGs of younger mice [50].
However, there was a higher proportion of neurons highly packed with white very prominent mitochondria in Pink1SNCA DRGs (overview upper panel in Fig. 8), which do not normally occur in the DRGs of younger mice [50]. The frequency of neurons with such mitochondria was 0-3.5% in sections of WT DRGs and 6-15% in Pink1SNCA DRGs. These white mitochondria in part gave a swollen impression and had lost or disorganized cisternae. Similar changes of mitochondrial morphology have been previously described in neurons of the cortex and substantia nigra of Parkin-SNCA double mutant mice [51].

Increase of innate immune markers in the DRGs
It has been suggested that mitochondrial damage in combination with defective mitophagy results in deposits of mitochondrial DNA (mtDNA) in the cytosol which triggers a cGAS-cGAMP-STING (cGAMP = cyclic guanosine monophosphate adenosine monophosphate, cGAS = cGAMP synthase) mediated innate immune response leading to increases of interferon regulated genes [52][53][54][55]. It was shown that mtDNAevoked in ammation in Pink1 de cient mice could be rescued by concurrent loss of STING [56]. Hence, considering the observed mitochondrial dysfunctions and deregulations of innate immune pathways we assessed the key candidates, which showed differential gene expression in RNAseq experiments, IFI16/IFI204 and STING (Fig. 9). The immuno uorescence of IFI16/IFI204 suggests a stronger distribution of IFI16 in the cytosol in Pink1SNCA mice (Fig. 9A), and RNAseq shows a decrease of its RNA (Fig. 9B). The immuno uorescence of STING did not show overt differences (Fig. 9A), but at RNAseq level, STING expression was increased (Fig. 9B). Analysis of multiple interferon-stimulated genes (Fig. 9B) and of chemokines (Suppl. Figure 2E) would agree with a stimulation of innate immune pathways.

Discussion
Sensory neuropathies occur in PD with higher frequency than in age-matched control subjects, and the onset is often years before the occurrence of motor dysfunctions [16]. The high vulnerability of the sensory neurons may result from their anatomy, in particular very long axons requiring organelle transport over long distances and exposure to peripheral environments, but the pathophysiology is not well understood. In contrast to olfactory sensory neurons, which are also affected early in the course of PD, only few neurons of the dorsal root ganglia express marker of dopaminergic neurons [57].
We show in the present study that Pink1SNCA double mutant mice phenocopy the prodromal sensory neuropathy. They develop a progressive loss of thermal sensation far earlier than the earliest occurrence of clinical motor dysfunctions. The sensory phenotype is reminiscent of QST phenotypes of PD patients, which presents with a predominant loss of thermal sensation and preservation of mechanical sensitivity or mechanical hypersensitivity [18,58]. Pink1SNCA mice replicate the thermal versus mechanical dichotomy. Thermal sensation relies on TRPV1 positive DRG neurons [59,60], and heat nociception is conveyed mainly through non-myelinated C-bers [61], suggesting that this population is particularly vulnerable. However, DRG neurons are highly diverse with functional overlap, and the behavioral readouts do not clearly distinguish the most affected subpopulation.
Morphologically, we did not detect overt signs of neuronal death or damage in DRGs of Pink1SNCA mice, but it has been shown that the density of sensory ber terminals in the skin is reduced in PD patients [15,58], which is not a speci c observation for PD associated sensory neuropathies but occurs frequently in sensory neuropathies associated with metabolic diseases such as diabetes mellitus [62]. The similarities suggests that metabolic dysfunctions of sensory neurons, which are not or not yet visible at the morphological level are important determinants of the sensory decline in PD. In addition, subsets of neurons displayed high numbers of white mitochondria that do not normally occur at younger age and appeared to be swollen with partially disrupted cristae. Such morphological abnormalities bear some resemblance to those described in human PD patients [63][64][65]. Mitochondria are susceptible to morphological artefacts such as swollen appearance that are caused by inadequate xation, but occur to some extent with any xation technique. However, we did not observe swollen mitochondria in DRGs in younger mice in a previous technically identical study [50], and swollen mitochondria were consistently more abundant in Pink1SNCA mice than in controls although the tissue was prepared in parallel (three per genotype). In addition, the morphology agreed with the functional loss of mitochondrial respiration and support the view of a mitochondria-dependent damage in PD neurons.
We have observed in PD patients that sensory loss as assessed by QST measurements and pain intensity ratings are associated with high levels of plasma glucosylceramides. We show in the present study that these glucosylceramides are increased in the DRGs of Pink1SNCA mice strengthening the idea that they contribute to the pathology of sensory neurons. Mutations of GBA1, the key lysosomal enzyme catalyzing GlcCer degradation, have been associated with autophagolysosmal dysfunctions [66,67], increased alpha-synuclein toxicity [27,68] and disruptions of mitochondrial energy production [49]. GBA1 expression was unaltered in the DRGs of Pink1SNCA mice, and the ultrastructural morphology of lysosomes and lipid droplets in DRG neurons appeared to be normal for the age. Instead, gene expression analysis suggested that sphingomyelin degradation converging on GlcCer generation was increased, which would agree with the loss of sphinganines and sphingosines in the sciatic nerve and in DRGs, and points to alterations or myelination. The data suggest that GlcCer may accumulate even in the absence of GBA1 defects, which agrees with previous observations in humans [33,34]. Hence, sphingolipid allostasis in PD may have a broader clinical relevance for PD pathophysiology, not restricted to GBA1 mutation carriers.
We showed previously that GlcCer has dual effects on sensory neurons. It increased the fraction of nonresponding neurons on stimulation, but augmented the capsaicin-evoked calcium in ux in responsive neurons (accompanying manuscript). The duality re ected the clinical phenotype where sensory loss is often associated with burning pain. It is not clear, if and how GlcCer affect ion channel properties. So far, most studies addressing the functional consequences of GBA1 deletion assumed that the observed pathology was caused by glucosylceramides without directly measuring these lipids. GBA1 deletion or mutation increased alpha-synuclein aggregates [26,27], led to mitochondrial calcium overload [49], ER stress [69] and disruption of autophagolysosomal pathways [67]. TRP channels are regulated by phosphorylation, redox modi cation, calmodulin and membrane insertion [70][71][72][73][74][75]. The ceramide nature of GlcCer suggests that membrane insertion or lipid raft function may be affected [46], ultimately leading to degradation and downregulation of TRPV1, which would agree with the immuno uorescence and gene expression analyses of the DRGs in Pink1SNCA mice. The very early behavioral onset of declining heat sensitivity in Pink1SNCA mice suggest that assessment of warm detection thresholds may qualify as early diagnostic tool.
To stop progression of neurodegeneration one would like to know the causal chain of pathophysiologic events. From the genetics of Pink1SNCA mice, one would assume that defective mitochondria trigger subsequent dysfunctions. Indeed, mitochondrial respiration was reduced, and there was a higher proportion of cells in the DRGs with white mitochondria, which gave a swollen impression. Defective mitophagy owing to Pink1 de ciency is supposed to lead to incomplete removal of damaged mitochondria and may result in a cytosolic mtDNA driven activation of innate immune responses via the cGAS-STING pathway [56,76,77]. Previous brain microarray studies from aging Pink1SNCA mice had suggested progressive innate immune pathway dysregulations [45]. However, in the DRGs, sciatic nerve and spinal cord we did not observe invasion with in ammatory cells or satellite cell activation, although STING itself, IFI44l, Irf6 and Irf7 and the chemokines CCL27 and 25 were upregulated at RNA level. Hence, the immune aspect was subtle, although it was in agreement with previous observations, and rather does not explain the sensory loss.

Conclusion
In summary, we show that Pink1SNCA double mutant mice phenocopy the prodromal PD-associated sensory neuropathy, which often precedes motor manifestations for years and poses substantial strain to patients, particularly if it causes chronic pain and is mis-diagnosed owing to the lack of motor impairments. The mice develop a progressive loss of thermal sensation far earlier than the earliest occurrence of clinical motor dysfunctions. We have evaluated multiple mechanistic aspects, which theoretically contribute to the prodromal sensory loss in Pink1SNCA mice. There is no single gene or pathway acting as single originator. However overall, the deregulations of sphingolipid metabolism and (likely consequent) TRP channel disappearances and mitochondrial damage were most convincing to explain the behavioral phenomena of early prodromal somatosensory loss and suggest that sphingolipid metabolic pathways pose novel therapeutic options. analysis lab. TD provided EM knowledge and discussed data. IT initiated the study, supervised the studies, analyzed calcium imaging, immuno uorescence and omics data, drafted the manuscript and made the gures. All authors contributed to drafting or editing of parts of the manuscript and approved the nal version of the manuscript.       Oxygraph respirometry in cortex, hippocampus and trigeminal ganglia A: Exemplary respirograms of each one Pink1SNCA and wildtype control mice in freshly prepared homogenates of the prefrontal cortex, hippocampus, trigeminal ganglia and trigeminal nerve. B: Quantitative OXPHOS analysis of each 6 mice per genotype for each site. Mitochondrial respiration was quanti ed as oxygen ux. LEAK-respiration was induced by adding pyruvate, malate and glutamate (PMG, CI substrates). CI-respiration was initiated by adding ADP in a saturating concentration. To measure CII respiration, rotenone was added to block CI, followed by adding succinate. Maximum uncoupled respiration was measured after stepwise titration with the uncoupler, FCCP. Residual oxygen consumption (ROX) was determined after sequential inhibition of complex III with antimycin A and complex IV with azide. Absolute respiration rates were corrected for ROX and normalized for the protein content. Data were analyzed per 2-way ANOVA and subsequent