Dysregulation of sphingolipid metabolism contributes to bortezomib-induced neuropathic pain

Bortezomib (Velcade) is a dipeptide boronic acid inhibitor of the 26S proteasome widely used in the treatment of multiple myeloma and mantle cell non-Hodgkin’s lymphoma (Chen et al., 2011). More than 40% of patients receiving bortezomib develop chronic, distal, and symmetrical sensory peripheral neuropathy often accompanied by a neuropathic pain syndrome (chemotherapy-induced painful peripheral neuropathy [CIPN]) that may last for weeks, months, or even years after drug termination (Farquhar-Smith, 2011). CIPN greatly compromises the quality of life of cancer survivors and patients under treatment and can interfere with effective anticancer treatment (Farquhar-Smith, 2011). Dose reduction or treatment discontinuation is often the sole recourse for patients who develop neuropathic pain, because traditional analgesics are ineffective or have their own serious unwanted side effects (Farquhar-Smith, 2011). Advances in strategies to address this growing problem are urgently needed with increased efficacy of cancer therapy that has resulted in nearly 14 million cancer survivors in the United States (de Moor et al., 2013), many suffering from these long-term side effects (Farquhar-Smith, 2011). The mechanisms underlying CIPN remain poorly understood and differ among the chemotherapeutic agents (Han and Smith, 2013). However, neuropathological changes in the dorsal horn of the spinal cord (DHSC) have been documented (Han and Smith, 2013).

As a desirable anticancer mechanism, bortezomib activates the sphingolipid biosynthesis pathway to produce the potent proapoptotic sphingolipid ceramide (Gong et al., 2014). Ceramide is hydrolyzed by ceramidases to sphingosine that is phosphorylated by sphingosine kinases (SphK1 and 2) to produce sphingosine-1-phosphate (S1P; Spiegel and Milstien, 2011). S1P initiates signaling through a family of five cognate G protein–coupled receptors (S1PR1–5; Spiegel and Milstien, 2011) and, in contrast to ceramide, is a potent antiapoptotic sphingolipid (Ogretmen and Hannun, 2004). It is hypothesized that the ceramide/S1P rheostat plays a critical role in regulating cancer cell fate with elevated levels of ceramide inducing cell death and elevated levels of S1P leading to survival and proliferation (Ogretmen and Hannun, 2004). Thus, therapeutic strategies aimed at reducing S1P bioavailability or signaling are attracting attention as novel anticancer agents (Ogretmen and Hannun, 2004). Numerous studies have demonstrated that the S1PR1 functional antagonist and orally available multiple sclerosis drug FTY720 (fingolimod; Gilenya; Brinkmann et al., 2010) has potent antitumor activity (Hait et al., 2015; White et al., 2016) and can synergize with bortezomib to target multiple myeloma cells in vitro and multiple myeloma in vivo xenografts (Beider et al., 2017).

Despite having opposing effects on cancer cells, compelling evidence is emerging that ceramide and S1P share potent inflammatory and nociceptive actions (Salvemini et al., 2013). It is noteworthy that altered sphingolipid metabolism caused by mutations in serine palmitoyltransferase (SPT) contributes to neuropathic pain in humans (Dawkins et al., 2001). Whether sphingolipid dysregulation drives bortezomib-induced neuropathic pain is not known. Here, we demonstrate that S1PR1-dependent neuroinflammatory signaling pathways contribute to the development of bortezomib-induced neuropathic pain identifying S1PR1 as a molecular target for therapeutic intervention.

Bioactive sphingolipid metabolites are powerful signaling molecules, and minor changes are known to have significant impact in signal transduction (Hannun and Obeid, 2008). Liquid chromatography–electrospray ionization–tandem mass spectrometry (LC-ESI-MS/MS) analysis of multiple sphingolipid species (Fig. 1 A) revealed that bortezomib treatment (Zheng et al., 2012; Janes et al., 2013) caused dysregulation of sphingolipid metabolism in the DHSC. Indeed, bortezomib induced significant increases in the levels of ceramide (Figs. 1 B and S1 A) and its de novo biosynthetic pathway precursors, dihydrosphingosine and dihydroceramide (Figs. 1 B and S1 B). Bortezomib also increased the levels of sphingosine, S1P, and dihydro-S1P (DH-S1P), which are metabolites of ceramide and dihydroceramide (Fig. 1 B). In contrast, there were no observable changes in sphingomyelin (Fig. 1 B). These results suggest that dysregulation of sphingolipid metabolism from the de novo pathway within the DHSC is among the early events of bortezomib-induced neuropathic pain. Dysregulation of sphingolipid metabolism was associated with time-dependent development of neuropathic pain (mechano-allodynia and mechano-hyperalgesia; Figs. 2 and S2). Our sphingolipidomic data supporting the contribution of the de novo pathway in bortezomib-induced neuropathic pain were substantiated with studies performed with myriocin, a well-characterized inhibitor of SPT (Delgado et al., 2006), the first and rate-limiting enzyme of de novo pathway. Myriocin, when used at doses reported to block SPT (Delgado et al., 2006) and S1P levels in DHSC (Muscoli et al., 2010) significantly attenuated the development of neuropathic pain (Fig. 1, C and D). Collectively, the data strongly suggest that bortezomib alters S1P signaling by triggering SPT and de novo sphingolipid metabolism.

Because S1P and DH-S1P are potent bioactive mediators and ligands with similar affinities for S1PR1 (Strub et al., 2010), we explored the functional contribution of S1PR1 using NIBR14, a methyl ester prodrug that is rapidly hydrolyzed in vivo to the carboxylic acid, NIBR15, a potent and selective S1PR1 antagonist (Angst et al., 2012). i.th. administration of NIBR14 attenuated the development of mechano-allodynia and mechano-hyperalgesia in rats (Figs. 2 A and S2 A). In support of these pharmacological effects, i.th. administration of S1pr1-targeting dicer-substrate short interfering RNA (DsiRNA) during bortezomib treatment recapitulated the pharmacological blockade (Figs. 2 B and S2 B) by significantly reducing S1PR1 protein levels in the DHSC (Fig. 2 C). These data provide convergent evidence that blocking S1PR1 signaling mitigates bortezomib-induced neuropathic pain.

The pharmacological efficacy and potency of oral administration of S1PR1-targeted agents, which is the preferred clinical approach, was examined using S1PR1 competitive (NIBR14) and functional (FTY720, ponesimod) antagonists (Bigaud et al., 2014). FTY720 (once phosphorylated by SphK2 to the biologically active FTY720-P) and ponesimod are S1PR1 agonists that act as “functional antagonists” by potently and irreversibly down-regulating S1PR1 (Bigaud et al., 2014). Oral administration of NIBR14 or FTY720 during bortezomib treatment dose-dependently prevented neuropathic pain from developing for the 20-d observation period after the completion of bortezomib treatment (Figs. 2, D and E; and S2, C and D; ED₅₀ values in Table 1). Oral administration of ponesimod exhibited similar beneficial effects (Fig. S2, E and F). The beneficial effects observed with FTY720 were independently verified by another laboratory in our group using another method of behavioral assessment (Fig. 2, F and G).

Numerous studies have convincingly demonstrated the antitumor effects of FTY720 on many types of cancer cells in vitro and in tumor-bearing animals (Yasui et al., 2005; Alinari et al., 2011; Neviani et al., 2013; Hait et al., 2015). Therefore, S1PR1 agents are not expected to reduce the chemotherapeutic effectiveness of bortezomib. We confirmed this with in vitro tumor cell–killing studies using FTY720. Based on pharmacokinetic studies in rats (Brinkmann et al., 2010), the anticipated plasma levels of FTY720 at doses providing maximal blockade of bortezomib-induced neuropathic pain are ∼0.5–0.7 nM. When tested at a much higher dose, FTY720 (5 µM) did not significantly diminish the in vitro tumor cell toxicity of bortezomib in human multiple myeloma (RPMI 8226) cells (LC₅₀, 34 nM; 95% confidence interval [CI], 27–42 nM; n = 5) compared with bortezomib and vehicle (LC₅₀, 31 nM; 95% CI, 23–42 nM; n = 5).

In the central nervous system (CNS), astrocytes express much higher levels of S1PR1 than microglia and neurons (Zhang et al., 2014), marking these cells as a prime target for S1PR1 signaling in the development of bortezomib-induced neuropathic pain. To test this possibility, we used mice with floxed S1pr1 recombined with a cre gene under the control of a glial fibrillary acidic protein (Gfap) promoter that results in the deletion of S1pr1 in GFAP⁺ astrocytes in the CNS (Choi et al., 2011). We confirmed the deletion of S1pr1 in the spinal cord (Fig. 3, A and B) and intact S1pr1 in dorsal root ganglion (DRG) tissues (Fig. 3 A). When compared with control littermates, male mice with astrocyte-specific deletions of S1pr1 did not develop neuropathic pain (Fig. 3 C). In animal models of multiple sclerosis, Choi et al. (2011) demonstrated that these mice develop a lesser form of the disease while losing the ability to respond to FTY720. Because in our pain model the complete deletion of S1pr1 from astrocytes resulted in no development of pain, we used mice with astrocyte-specific reductions of S1pr1 to examine the role of S1pr1 in astrocytes on the pharmacological effects of S1PR1 antagonists. Mice with astrocyte-specific reductions of S1pr1 had delayed development of neuropathic pain compared with controls and, because of reduced levels of S1PR1, they also lost their responsiveness to the beneficial effects of FTY720 (Fig. 3 D). These data, which mimic those obtained with S1PR1 antagonists and siRNA, strongly suggest that activation S1PR1 signaling in astrocytes is necessary for both bortezomib-induced neuropathic pain and the analgesic effects of FTY720, thus identifying astrocytes as a primary cellular substrate of S1PR1 activity.

Bortezomib-induced dysregulation of sphingolipid metabolism in the spinal cord was associated with increased levels of the astrocyte marker GFAP (Fig. 4, A–D and G) in the superficial dorsal horn (laminae I and II), confirming a previous study (Robinson et al., 2014). The increase in GFAP suggests an increase in the number of astrocytes, their activation, or both in the DHSC in response to bortezomib. S1P triggers astrocytes to become reactive (Sorensen et al., 2003), and activation of S1PR1 can induce astrocyte migration in culture (Mullershausen et al., 2007). In other neurodegenerative diseases with a prominent astrocyte-driven inflammatory component (e.g., multiple sclerosis, Alzheimer’s disease, or Huntington’s disease), blocking the S1PR1 axis with S1PR1 antagonists attenuates astrocyte activation and inhibits increased production of inflammatory cytokines in the spinal cord; in contrast, levels of IL-10 increase (Brinkmann et al., 2010; O’Sullivan and Dev, 2017). These antiinflammatory effects of S1PR1 antagonism are further supported by a recent study showing that FTY720 can prevent endotoxin activation, migration, and ensuing release of inflammatory mediators in primary astrocyte cultures, but not microglia (Rothhammer et al., 2017). In our study, bortezomib-induced dysregulation of sphingolipid metabolism was also associated with concomitant increases in TNF and IL-1β (Fig. 4, H and I). In contrast, antiinflammatory cytokines (IL-10 and IL-4) decreased compared with those in vehicle-treated rats (Fig. 4, J and K). Concurrent administration of FTY720 attenuated increased GFAP immunolabeling (Fig. 4, E–G) and production of TNF and IL-1β (Fig. 4, H and I) associated with bortezomib treatment while preventing reductions in IL10 and IL4 (Fig. 4, J and K). Similar effects on cytokine expression were obtained with ponesimod (Fig. 4, H–K). This S1PR1-driven neuroinflammation in the spinal cord may, in turn, establish “feed-forward” mechanisms that can sustain dysregulation of sphingolipid metabolism. For example, TNF and IL-1β can activate the major enzymes involved in the biosynthesis of ceramide and S1P pathways: SphK1, sphingomyelinases, SPT, and ceramidases (Snider et al., 2010). Interruption of this cross talk early in the bortezomib regimen may explain how neuropathic pain is prevented with S1PR1-targeted agents from developing for many days after bortezomib discontinuation (Figs. 2 and S2).

Spinal synaptic mechanisms underlying the development of bortezomib-induced neuropathic pain have not been identified. We investigated whether glutamatergic synaptic activities in the DHSC of rats with bortezomib-induced neuropathic pain were altered using whole-cell patch-clamp recording techniques. Miniature excitatory postsynaptic currents (mEPSCs) were recorded from neurons in the spinal outer lamina II receiving monosynaptic input from the primary afferents (Yoshimura and Jessell, 1989; Weng et al., 2006). Alterations in mEPSC frequency indicate changes in the presynaptic transmitter release probability, whereas alteration in mEPSC amplitude indicates changes in the postsynaptic glutamate receptor function (Lissin et al., 1999; Yan et al., 2013; Yan and Weng, 2013). We found that both the frequency (6.10 ± 0.44 Hz) and amplitude (26.37 ± 0.52 pA) of mEPSCs in neurons from rats with bortezomib-induced neuropathic pain were significantly higher (P < 0.0001, Welch’s uncorrected t test) than in vehicle-treated rats (frequency, 3.14 ± 0.33 Hz; amplitude, 22.87 ± 0.77 pA), suggesting that glutamate release from presynaptic terminals and ligand-gated glutamate receptor function in postsynaptic neurons are enhanced in bortezomib-treated rats (Fig. 4, L–O).

Previous studies by others and us have shown that spinal glutamatergic synaptic activities are enhanced by inflammatory mediators (e.g., TNF, IL-1β, and MCP1) under pathological pain conditions (Kawasaki et al., 2008; Yan and Weng, 2013). However, their regulation by the S1PR1 axis is not known. NIBR15 treatment significantly reduced the mEPSC frequency, but not amplitude, from 6.10 ± 0.44 Hz to 4.09 ± 0.23 Hz (Fig. 4, L and N) in DHSC neurons from rats with bortezomib-induced neuropathic pain. Furthermore, perfusion of the selective S1PR1 agonist SEW2871 in normal rats significantly increased mEPSC frequency with no significant change in amplitude (Fig. 4, M and O). Collectively, these data suggest that S1P in the DHSC of bortezomib-treated rats enhances glutamate release from presynaptic terminals, which is blocked by S1PR1-targeted agents.

These findings demonstrate the dysregulation of excitatory glutamatergic signaling as a second factor contributing to neuropathic pain induced by chemotherapeutics (Boyette-Davis et al., 2015). More importantly, our study demonstrates for the first time that S1PR1 signaling is responsible for increased presynaptic glutamate release and the genesis of bortezomib-induced neuropathic pain. Because we found that the development of neuropathic pain is lost when S1PR1 is deleted in astrocytes (Fig. 3 C) and FTY720 lost its effectiveness when S1PR1 is reduced in astrocytes (Fig. 3 D), it is likely that S1PR1-targeted agents suppress astrocyte activity and ensuing neuroinflammation (Rothhammer et al., 2017), attenuating the ability of proinflammatory cytokines to enhance glutamatergic synaptic activity (Clark et al., 2013).

The increased efficacy of cancer therapy has led to nearly 14 million cancer survivors in the United States (de Moor et al., 2013). Many of these survivors continue to suffer from long-term neuropathic pain states that are poorly managed. To address this major unmet medical need, we examined the effectiveness of targeting S1PR1 in models of established bortezomib-induced neuropathic pain. At the time of peak pain (day 25), subcutaneous minipump infusions of FYT720, ponesimod, or NIBR14 over 6 d (Janes et al., 2014) completely reversed neuropathic pain (Fig. S3) throughout the entire period of drug infusion.

The mechanisms of bortezomib dysregulation of sphingolipid metabolism are not known, but several possibilities are envisioned (Fig. 4 P). The limited ability of bortezomib to cross the blood–brain barrier (Adams et al., 1999) suggests that sphingolipid metabolism in the CNS may be triggered by peripheral inflammatory changes (Seruga et al., 2008). Accordingly, bortezomib increases levels of various proinflammatory cytokines, such as TNF and IL-1β (Liang et al., 2014), in plasma that gain access to the CNS (Yarlagadda et al., 2009) and activate de novo biosynthesis to increase DH-S1P and S1P (Snider et al., 2010; Maceyka and Spiegel, 2014). Alternatively, bortezomib-induced events in the DRG (Carozzi et al., 2013) may provoke production of reactive oxygen and nitrogen species (Duggett and Flatters, 2017) and release of glutamate (Ghelardini et al., 2014) from presynaptic nerve endings that could activate glial sphingolipid metabolism (Kim et al., 2012). The resulting increased S1PR1 signaling in astrocytes promotes their activation, migration, and ensuing release of inflammatory mediators (Rothhammer et al., 2017) to reinforce neuroinflammation and drive the establishment and maintenance of the neuronal sensitization in the CNS. In turn, enhanced reactive oxygen and nitrogen species or glutamatergic signaling may perpetuate the cycle and transition the DHSC toward a chronic hypersensitized state.

In conclusion, by using a multidisciplinary approach, our studies provide new insight into the underlying S1PR1-driven mechanisms engaged during bortezomib-induced neuropathic pain and establish the foundation for proof-of-concept studies with S1PR1-targeted agents that are already in clinical use (FTY720; Brinkmann et al., 2010) or advanced clinical trials for other indications (ponesimod; Bigaud et al., 2014). Our studies provide a compelling case for the consideration of repurposing FTY720 as an adjuvant to bortezomib for the prevention and treatment of chemotherapy-related neurotoxicity to address an immense unmet medical need.

Male Sprague Dawley rats (200–220 g starting weight) were from Harlan Laboratories. For S1pr1 knockout and knockdown mice, transgenic mouse colonies were descendants of original homozygous S1pr1^fl/fl;GFAP-Cre breeder mice gifted to us by J. Chun (The Scripps Research Institute, La Jolla, CA; Choi et al., 2011) and interbred to maintain their homozygosity for the floxed S1pr1 gene with regular backcrosses with C57BL/6J wild-type mice (Harlan Laboratories) to reduce inbred genetic defects. Prior characterizations of these conditional mutants showed the removal of S1pr1 from GFAP⁺ astrocytes in the CNS (Choi et al., 2011). Mice with astrocyte-specific deletion of S1pr1 (S1pr1^fl/fl;GFAP-Cre) and their control littermates (S1pr1^fl/fl) were generated from mating male S1pr1^fl/fl mice with female S1pr1^fl/fl;GFAP-Cre mice. Mice with astrocyte-specific reductions in S1pr1 (S1pr1^fl/+;GFAP-Cre) and their control littermates (S1pr1^fl/+) were generated from mating male C57BL/6J wild-type mice with female S1pr1^fl/fl;GFAP-Cre mice. All mice were genotyped before breeding or use in experiments by endpoint PCR of ear-punch DNA. The primers used to detect floxed and deleted S1pr1 were spb1 (Invitrogen; 5′-TTCTAGCCAAGGCATTGCTCCA-3′), Edg1M/S1pr1 (IDT; 5′-CCTCAGAGTCTGGGTAAATTGT-3′), and b1t1c (Invitrogen; 5′-CACAACCTCCTTCTGATGAA-3′) primers. The primers used to detect Gfap-cre were gfap-cre forward (5′-ATGGTCTGGCTCCAGGTACC-3′) and cre reverse (5′-CAGCATTGCTGTCACTTGGTC-3′). All mice were ear-tagged for identification, and sex- and age-matched littermate controls were used in all experiments.

Animals were housed two to four per cage (for rats), five per cage (for mice), or single housed after surgery in a controlled environment (12-h light/dark cycle) with food and water available ad libitum. All experiments were conducted with the experimenters blinded to treatment conditions. All experiments were performed in accordance with the guidelines of the International Association for the Study of Pain and the National Institutes of Health and approvals from the Saint Louis University, University of Georgia, and the University of California, San Diego, Institutional Animal Care and Use Committees.

i.th. compounds (10 µl) were administered via cannulas made from PE10 tubing (427401; Becton Dickinson) that were chronically implanted in rats by using the L5/L6 lumbar approach as previously described (Little et al., 2012).

Rodents were lightly anesthetized with 3% isoflurane and maintained on 2% isoflurane in 100% O₂. An incision was made in the interscapular region for subcutaneous implantation of primed 7-d osmotic minipumps (Alzet 2001; Alza) that infused 1 µl/h as previously described (Janes et al., 2014). Minipumps were filled according to the manufacturer's specifications with test compounds or their vehicle. Immediately after surgery, i.p. loading doses of test compounds or their vehicle were administered, and minipumps were allowed to deliver test agents for 6 d. After implantation, the animals were singly housed for the remainder of the experiment.

In rats, bortezomib (S1013; Selleck Chemicals) or its vehicle (5% Tween 80, P1754 [Sigma-Aldrich]; 5% ethanol, 493546 [Sigma-Aldrich]; and saline) was administered by i.p. injection in rats on 5 consecutive days (day 0 to day 4; 0.2 mg/kg) at an injection volume of 0.2 ml for a final cumulative dose of 1 mg/kg (Janes et al., 2013). In mice, bortezomib (0.4 mg/kg i.p.) was administered every other day three times per week for 4 wk (day 0 to day 25).

All compounds in the prevention models were administered concurrent with bortezomib at the doses and by the routes described in the text after behavior measurements. For the reversal studies, compounds were administered via osmotic minipumps implanted on D24 or D25. FTY720 (fingolimod) was purchased from Cayman Chemical (162359-56-8). Ponesimod (ACT-128800, (Z,Z)-5-[3-chloro-4-((2R)-2,3-dihydroxy-propoxy)-benzylidene]-2-propylimino-3-O-tolylthiazolidin-4-one) was prepared by Shanghai ChemPartner Co. according to the method of Bolli et al. (2010).

For synthesis of NIBR14, the methyl ester pro-drug, the Novartis competitive antagonist NIBR14 (mol wt = 600.17) was prepared as described in the literature (Angst et al., 2012). Molecular identity was confirmed by LCMS (50–95% acetonitrile in 0.05% TFA over 6 min) performed by using a Waters Alliance SQ 3100 system (positive ion electrospray mode) with an Agilent Eclipse (XDB-C18; 4.6 × 150 mm, 5 μm) column (retention time = 7.15 min; C₃₁H₃₈ClN₃O₅S; m/z = 600 [M + H⁺]).

For synthesis of NIBR15, the methyl ester prodrug NIBR14 (90 mg, 0.15 µmol) was dissolved in 6 ml acetic acid and treated with 1.5 ml of concentrated HCl. This mixture was stirred at 50°C for 12 h and concentrated. Recrystallization from acetonitrile afforded 45 mg (48% yield) NIBR15 HCl salt (mol wt = 622.60), confirmed by LCMS (15–95% acetonitrile in 0.05% TFA over 6 min; retention time = 6.97 min; C₃₀H₃₇Cl₂N₃O₅S; m/z = 586 ([M + H⁺-Cl]⁺); Angst et al., 2012).

For myriocin, [(2S,3R,4R,6E)-2-Aamino-3,4-dihydroxy-2-(hydroxymethyl)-14-oxo-6-eicosenoic acid] from Mycelia sterilia was purchased from Sigma-Aldrich (35891-70-4).

Commercially available 27-mer duplex DsiRNA targeting common splice forms of S1pr1/Edg1 transcript (RNC.RNAI.N017301.12.1, GenBank accession no. NM_017301; Integrated DNA Technologies; 5′-rCrCrUrGrUrArCrArArArGrCrArGrArGrUrArCrUrUrCrCTG-3′ and 5′-rCrArGrGrArArGrUrArCrUrCrUrGrCrUrUrUrGrUrArCrArGrGrArU-3′) or its nontargeting control sequence (siNT; Integrated DNA Technologies; DS NC1: 5′-rCrGrUrUrArArUrCrGrCrGrUrArUrArArUrArCrGrCrGrUAT-3′ and 3′-rArUrArCrGrCrGrUrArUrUrArUrArCrGrCrGrArUrUrArArCrGrArC-5′; Walk et al., 2012) were diluted in RNase-free water to 0.2 µg/µl. S1pr1-targeting (siS1PR1) or siNT sequence (10 µl) was administered after bortezomib injections via i.th. catheters on days 0, 2, and 4.

Baseline behavioral measurements to assess mechano-hypersensitivity were always taken before the daily administration of test substances or their vehicles. Pilot tests to examine the effects of S1PR1 antagonists alone in vehicle-treated rats excluded a potential impact of S1PR1 antagonism on baseline paw withdrawal thresholds (PWTs). Thus, systemic or i.th. injections of FTY720 and NIBR14 did not alter baseline PWTs when given alone at the highest dose used in our manuscript in rats that received the vehicle of bortezomib only. Accordingly, we did not include a vehicle plus S1PR1 antagonist group in our experimental design.

Mechano-allodynia was measured as previously described (Janes et al., 2014). Mechanical PWT in grams [PWT, (g)] was measured using an electronic von Frey test (dynamic plantar aesthesiometer, model 37450; Ugo Basile) with a cutoff set at 50 g in rats or calibrated von Frey filaments (Stoelting; range in mice: 0.07–2.00 g; in rats: 2–26 g) using the “up-and-down” method (Dixon, 1980). Mechano-allodynia was defined as a significant (P < 0.05) reduction in mechanical mean absolute PWT (g) at forces that failed to elicit withdrawal responses before chemotherapy treatment (day 0). Mechano-hyperalgesia was measured in the rat hind paw using the Randall and Sellitto paw pressure test as previously described (Janes et al., 2014) by using an analgesiometer (model 37215; Ugo Basile). The nociceptive PWT (g) was measured as the force that caused the rat to withdraw its paw (cutoff set at 250 g). Mechano-hyperalgesia was defined as a significant (P < 0.05) reduction in nociceptive PWT (g) compared with withdrawal responses before chemotherapy treatment (day 0). The values from both paws were averaged because chemotherapy-induced neuropathy results in bilateral allodynia and hyperalgesia with no differences between left and right hind PWT (g). The ED₅₀ and the corresponding 95% CI were determined from dose responses by curve fitting the percentage of prevention of mechano-allodynia or mechano-hyperalgesia on day 25 using the least sum of square method by a normalized three-parameter, nonlinear analysis (Hill slope = 1) with the use of GraphPad Prism (Release 5.03; GraphPad Software, Inc.). The percentage of prevention = (PWT (g)_treatment − mean PWT (g)_bortezomib)/(mean PWT (g)_vehicle − mean PWT (g)_bortezomib) × 100.

In studies performed by an independent laboratory from our group, behavior was measured by using a previously reported von Frey method (Siau et al., 2006). In brief, 4 and 15 g of von Frey hairs were applied bilaterally to the plantar hind paw five times per side with 1–2 min between successive stimuli, and the percentage of stimuli that elicited a withdrawal reflex was recorded. A significant increase in the number of responses to the 4 g of hair is indicative of mechano-allodynia, and an increase to the 15 g of hair indicates mechano-hyperalgesia.

After lipid extraction, sphingolipids were quantified by LC-ESI-MS/MS (4000 QTRAP; AB Sciex) as described previously (Hait et al., 2009).

Total protein lysates (rat, 20 µg; mouse, 5 µg) from L4–6 were denatured in Laemmli’s buffer, resolved on 4–20% SDS-PAGE, and electrotransferred to nitrocellulose membrane. The membranes were blocked for 1 h in 5% milk in 1× PBS, pH 7.4, with 0.05% Tween 20, probed overnight with rabbit anti–S1PR1 antibody (rat, 1:5,000; mouse, 1:30,000; ab11424; Abcam; and MA/PA1-1040; Thermo Fisher Scientific) and visualized with goat anti–rabbit IgG-horseradish peroxidase (1 h, room temperature [RT], 1:10,000–15,000, 7074; Cell Signaling Technology) and Clarity chemiluminescent reagents (1705061; BioRad). The signal was documented using a ChemiDoc imager (BioRad) and analyzed using ImageLab Software v.5.0 (BioRad). Horseradish peroxidase bound to the membrane was inactivated by two incubations in 30% H₂O₂ at RT for 15 min (Sennepin et al., 2009) and reprobed for β-actin by using mouse anti–β-actin antibody, clone AC-15 (1 h, RT; rat, 1:2,000; mouse, 1:10,000; A5441; Sigma-Aldrich) and goat anti–mouse IgG–horseradish peroxidase (1 h, RT; 1:3,000; 04-18-15; KPL). The density of the S1PR1 band was normalized to the corresponding β-actin band and expressed as percent S1PR1/β-actin.

Ear snip, spinal cord, or DRG DNA was isolated by using a QIAmp DNA mini isolation kit (Qiagen) and quantitated by NanoDrop (NanoDrop Products) for PCR as previously described (Choi et al., 2011). In brief, DNA (20–25 ng) was subjected to PCR (25 µl) in duplicate by using two separate PCR reactions: (1) Edg1M (S1pr1) (5′-CCTCAGAGTCTGGGTAAATTGT-3′; Integrated DNA Technologies), SPB1 (5′-TTCTAGCCAAGGCATTGCTCCA-3′; Invitrogen), and B1T1C (5′-CACAACCTCCTTCTGATGAA-3′; Invitrogen) sites by using the following amplification cycle conditions: 94°C, 5 min; 35 cycles (94°C, 30 s; 59°C, 45 s; 72°C, 1 min), and 72°C, 7 min and (2) GFAP-cre by using forward GFAP-cre primers (5′-ATGGTCTGGCTCCAGGTACC-3′; Integrated DNA Technologies) and reverse cre primers (5′-CAGCATTGCTGTCACTTGGTC-3′; Integrated DNA Technologies) by using the following amplification cycle conditions: 94°C, 3 min; 35 cycles (94°C, 25 s; 59°C, 35 s; 72°C, 45 s); and 72°C, 7 min. Blue/orange 6× loading dye (2 µl; Promega) was added, and 10 ng PCR product was loaded for 1.8% agarose gel electrophoresis in 1× Tris-borate-EDTA buffer. When measured against a 100-bp DNA ladder (Promega), PCR yielded the following bands: 177 bp (wild-type S1pr1), 335 bp (floxed S1pr1), 450 bp (deletion product), and 625 bp (Gfap-cre). Animals with a 450-bp band in ear snip DNA samples indicated germline deletions of S1PR1 and were excluded from breeding and experiments. For each experiment, the heterozygous S1pr1^fl/+;Gfap-Cre mice were paired with their heterozygous S1pr1^fl/+ littermates, and their genotypes were blinded to the experimenters.

Immunofluorescence and image analysis was performed by using modifications to previously reported methods (Little et al., 2012; Janes et al., 2014). The spinal cord (L4–6) was harvested, transferred to optimum cutting temperature compound, and rapidly frozen in an isopropanol/dry ice bath. Transverse sections (20 µm) were cut by using a cryostat, collected on gelatin-coated glass microscope slides, and stored at −20°C. Spinal cord sections were fixed in 10% buffered neutral formalin for 10 min (SF100; Thermo Fisher Scientific), blocked (10% normal goat serum 10000C; Thermo Fisher Scientific; 2% BSA, BP9706; Thermo Fisher Scientific; and 0.2% Triton X-100, T8787; Sigma-Aldrich) in PBS (BP39920; Thermo Fisher Scientific) for 1 h and then immunolabeled by using an 18-h incubation (4°C) with mouse monoclonal anti-GFAP antibody, clone G-A-5 (1:200, G3893; Sigma-Aldrich). After a series of PBS rinses, sections were incubated for 2 h with a goat anti–mouse Alexa Fluor 488 antibody (1:250, A-11001; Invitrogen). The coverslips were mounted with Fluorogel II containing DAPI (17985-51; Electron Microscopy Sciences) and photographed with an Olympus FV1000 MPE confocal microscope (multiline argon lasers with excitation at 405 nm and 488 nm) using a 10× objective (UPLSAPO; 0.4 numerical aperture) for regional fluorescence intensity image analysis and with a 60× oil-immersion objective (PLAPONl, 1.42 numerical aperture) and 2.4× optical zoom (0.1-µm pixel dimensions in the X–Y plane and the pinhole set at 1 Airy unit) for higher magnification images. Images were acquired within the dynamic range of the microscope (i.e., no pixel intensity values of 0 or 255 in an 8-bit image). Sections treated with isotype control (X0931; Agilent-Dako) at concentrations equivalent to the primary antibody yielded only nonspecific background fluorescence. Image analysis of GFAP immunolabeling was performed by using the National Institutes of Health freeware program ImageJ (version 1.43; Schneider et al., 2012). The bilateral superficial DHSC (laminae I and II) at the L4, L5, and L6 levels was outlined on images using the ImageJ region of interest tool. The superficial DHSC was determined and confirmed by using adjacent cresyl violet–stained sections and an atlas (Paxinos and Watson, 1998). Images received automated background threshold corrections (histogram-based threshold) before analysis. The mean fluorescence intensity (MFI) of the superficial DHSC was determined as previously reported (Little et al., 2012; Janes et al., 2014). MFI was calculated using the following equation: MFI = i(pp² ÷ p²), where i is the mean gray value, pp² is the positive pixel area, and p² is total pixel area. There were no significant differences bilaterally within each group, so MFI was calculated as a combined value for each animal and reported as fold change compared with the vehicle group.

Cytokine levels were assessed in DHSC (L4–6) homogenates using a commercially available multiplex cytokine kit (Bio-Rad Laboratories) as previously described (Janes et al., 2014).

Male Sprague Dawley rats (170–180 g) were administered bortezomib (0.2 mg/kg) or its vehicle (5% Tween 80, 5% ethanol in saline) on days 0–4. This resulted in significantly reduced (40.47 ± 8.13%; P < 0.001) mechanical PWTs by day 7 with bortezomib treatment. On day 7, traverse L4–5 spinal cord slices were prepared as previously described (Weng et al., 2007; Nie and Weng, 2009). Surgery was performed in deeply isoflurane-anesthetized rats to expose and remove the spinal lumbar enlargement segments. The section was placed in ice-cold sucrose artificial cerebrospinal fluid (234 mM sucrose, 3.6 mM KCl, 1.2 mM MgCl₂, 2.5 mM CaCl₂, 1.2 mM NaH₂PO₄, 12.0 mM glucose, and 25.0 mM NaHCO₃) presaturated with 95% O₂ and 5% CO₂. The pia-arachnoid membrane was removed from the section before it was attached with cyanoacrylate glue to a cutting support and then glued onto the stage of a vibratome (Series 1000; Technical Products International). Transverse spinal cord slices (400 µm) were cut in the ice-cold sucrose artificial cerebrospinal fluid and preincubated in Krebs solution (117.0 mM NaCl, 3.6 mM KCl, 1.2 mM MgCl₂, 2.5 mM CaCl₂, 1.2 mM NaH₂PO₄, 11.0 mM glucose, and 25.0 mM NaHCO₃ at 35°C) oxygenated with 95% O₂ and 5% CO₂ at 35°C. Each single slice was placed in the recording chamber (volume, 1.5 ml), perfused with Krebs solution at 35°C, and saturated with 95% O₂ and 5% CO₂. Borosilicate glass recording electrodes (resistance, 3–5 MΩ) were pulled and filled with an internal solution containing 135 mM potassium-gluconate, 5.0 mM KCl, 2.0 mM MgCl₂, 0.5 mM CaCl₂, 5.0 mM Hepes, 5.0 mM EGTA, 5.0 mM ATP-Mg, 0.5 mM Na-GTP, and 10 mM QX-314.

Live dorsal horn neurons in the spinal outer lamina II were visualized by microscopy and approached by using a three-dimensional motorized manipulator (Sutter Instrument), and whole-cell configurations were established by applying moderate negative pressure after electrode contact (Nakatsuka et al., 2003). Recordings of mEPSCs were made from neurons receiving monosynaptic input from the primary afferents by using the criteria established previously (Yoshimura and Jessell, 1989; Weng et al., 2006). mEPSCs were recorded at a membrane potential at −70 mV in the presence of 1 µM tetrodotoxin (T8024; Sigma-Aldrich), 10 µM bicuculline (14340; Sigma-Aldrich), and 5 µM strychnine (S0532; Sigma-Aldrich) in the external solution to block voltage-gated sodium channels, GABA_A, and glycine receptors. Data were recorded using Axopatch 700B amplifiers, digitized at 10 kHz, and analyzed off-line. The frequency and amplitude of mEPSCs in 3 min before and during perfusion of tested drugs were analyzed and averaged using a peak detection program (MiniAnalysis; Synaptosoft Inc.).

The effects of FTY720 on the tumor cell toxicity of bortezomib on human multiple myeloma cells (RPMI 8226) was assessed by using an MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide, M5655; Sigma-Aldrich) assay as previously described (Janes et al., 2013).

The data are expressed as mean ± SD or mean ± SEM for n animals and analyzed by paired or unpaired t test or one-way or two-way repeated-measures ANOVA with Holm-Sidak corrected comparisons. Significant differences were defined as P < 0.05, and the false discovery rate for sphingolipid analysis was controlled by Benjamini-Hochberg procedure at q < 0.05 using SPSS software (v21; IBM). All other statistical analyses were performed using GraphPad Prism.

Fig. S1 shows bortezomib-induced ceramide and dihydroceramide acyl chain species in the spinal cord. Fig. S2 shows the effects of i.th. or oral administration S1PR1-targeted agents on bortezomib-induced mechano-hyperalgesia. Fig. S3 shows the effects of S1PR1-targeted agents on the reversal of established bortezomib-induced mechano-hyperalgesia.

We thank Leesa Bryant for her technical assistance in some of the animal experiments.

This work was supported by the Mayday Fund Clinical Evaluation of Novel Biomarkers to Select and Treat Chronic Pain grant, the Leukemia and Lymphoma Society (6241-13 to D. Salvemini and G.J. Bennett), National Institutes of Health (NIH) T32 Training Grant GM008306 (to K. Stockstill, K. Braden, and C.M. Harada), NIH grants RO1NS064289 (to H.-R. Weng) and R01GM043880 (to S. Spiegel), and NIH-National Cancer Institute Cancer Center Support Grant P30 CA016059 (to the Virginia Commonwealth University Lipidomics Core).

D. Salvemini and G.J. Bennett are founders of BioIntervene, Inc., which has licensed related intellectual property from Saint Louis University. All other authors declare no competing financial interests.

Author contributions: K. Stockstill, T.M. Doyle, J.W. Little, K. Janes, Z. Chen, K. Braden, F. Lauro, L.A. Giancotti, C.M. Harada, R. Yadav, and W.H. Xiao performed the experiments. K. Stockstill, T.M. Doyle, and J.W. Little analyzed the data. X. Yan and H.-R. Weng performed the patch-clamp experiments. W.L. Neumann performed the NIBR14 synthesis. S. Spiegel performed the sphinglipidomics. K. Stockstill, T.M. Doyle, J.W. Little, W.L. Neumann, S. Spiegel, J.M. Lionberger, and G.J. Bennett assisted with the manuscript. H.-R. Weng and D. Salvemini wrote the manuscript. D. Salvemini conceived, designed, and led the project.