Caspase-Dependent HMGB1 Release from Macrophages Participates in Peripheral Neuropathy Caused by Bortezomib, a Proteasome-Inhibiting Chemotherapeutic Agent, in Mice

Given the role of macrophage-derived high mobility group box 1 (HMGB1) in chemotherapy-induced peripheral neuropathy (CIPN) caused by paclitaxel, we analyzed the role of HMGB1 and macrophages in the CIPN caused by bortezomib, a proteasome-inhibiting chemotherapeutic agent used for the treatment of multiple myeloma. Repeated administration of bortezomib caused CIPN accompanied by early-stage macrophage accumulation in the dorsal root ganglion. This CIPN was prevented by an anti-HMGB1-neutralizing antibody, thrombomodulin alfa capable of accelerating thrombin-dependent degradation of HMGB1, antagonists of the receptor for advanced glycation end-products (RAGE) and C-X-C motif chemokine receptor 4 (CXCR4), known as HMGB1-targeted membrane receptors, or macrophage depletion with liposomal clodronate, as reported in a CIPN model caused by paclitaxel. In macrophage-like RAW264.7 cells, bortezomib as well as MG132, a well-known proteasome inhibitor, caused HMGB1 release, an effect inhibited by caspase inhibitors but not inhibitors of NF-κB and p38 MAP kinase, known to mediate paclitaxel-induced HMGB1 release from macrophages. Bortezomib increased cleaved products of caspase-8 and caused nuclear fragmentation or condensation in macrophages. Repeated treatment with the caspase inhibitor prevented CIPN caused by bortezomib in mice. Our findings suggest that bortezomib causes caspase-dependent release of HMGB1 from macrophages, leading to the development of CIPN via activation of RAGE and CXCR4.


Introduction
Chemotherapy-induced peripheral neuropathy (CIPN) frequently occurs in cancer patients undergoing chemotherapy with platinum agents, taxanes, vinca alkaloids, proteasome inhibitors, etc. CIPN impairs the patients' quality of life and the dose reduction or discontinuation of the chemotherapy [1,2]. The pathogenesis of CIPN largely remains unclear, and there are presently no effective pharmacological interventions approved for the prevention or treatment of CIPN, although the recent systemic literature reviews have ascertained a limited benefit from the use of duloxetine, a selective serotonin and noradrenalin reuptake inhibitor, on the established CIPN [3].
We have demonstrated that high mobility group box 1 (HMGB1), one of the damageassociated molecular patterns (DAMPs), plays a role in the development and maintenance of CIPN in rodents treated with paclitaxel, vincristine, or oxaliplatin [2,[4][5][6][7]. HMGB1, a The mouse was placed in a transparent plastic box (10 × 10 × 10 cm) on a risen wire mesh floor. After acclimatization to the environment for 1 h, the surface of mid-planter in hindpaw was stimulated with von Frey filaments of distinct strengths (0.008, 0.02, 0.04, 0.07, 0.16, 0.4, 0.6, or 1.0 g). The 50% nociceptive threshold was calculated using an up-down method [20]. Indeed, the stimulation with filaments was started with a middle intensity of 0.07 g. In the absence of a paw withdrawal, the stimulation was performed by one stronger filament and repeated until the observation of positive response. If a paw withdrawal was observed, one weaker filament was used for next stimulation. The stimulation was performed 5 times, including the first positive response. The pattern for the positive or negative responses was recorded and the 50% response threshold was calculated using the following formula: 50% g threshold = [10 (Xf + kδ) ]/10,000 Xf = value in log units of the final von Frey filament; k = tabular value for the pattern of positive/negative responses; δ = mean difference in log units between stimuli. Measurements of nociceptive threshold in critical experiments were performed in a blinded fashion. For the induction of CIPN, mice received i.p. bortezomib at 0.1-0.4 mg/kg on days 0, 2, 5, 7, 9, and 12, as described previously [18,21].

Determination of Protein Levels of HMGB1 and Its Targeted Receptors
On day 14 after the onset of bortezomib treatment, mice were anesthetized with i.p. pentobarbital at 10 mg/kg following i.p. administration of a mixture of midazolam at 4 mg/kg and medetomidine at 0.3 mg/kg. After blood was collected from the abdominal aorta, the mouse was sacrificed by exsanguination. As described previously [22], after the skin and muscles on the back side were incised to expose the spine (chest to lumbar region), the spinal canal was cut open, and the bilateral dorsal root ganglia (DRG) at L4-L6 spinal levels were excised under microscope observation. Thereafter, the bilateral sciatic nerves were excised from the left and right thighs. Plasma HMGB1 levels were analyzed using an ELISA kit (Shino-Test Corporation, Tokyo, Japan). Protein levels of HMGB1, RAGE, TLR4, and CXCR4 in the DRG and sciatic nerve were measured by Western blotting, as reported previously [23]. Primary antibodies were: an anti-HMGB1 rabbit polyclonal antibody (Abcam, Cambridge, UK) (1: 5000 dilution), anti-RAGE rabbit polyclonal antibody (Abcam) (1:1000 dilution), anti-TLR4 rabbit polyclonal antibody (Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA) (1:10000 dilution), anti-CXCR4 rabbit polyclonal antibody (Novus Biologicals, Littleton, CO, USA) (1:1000 dilution), and anti-glyceraldehyde 3-phosphate dehydrogenase (GAPDH) rabbit polyclonal antibody (Santa Cruz Biotechnology, Inc.) (1:5000 dilution). A horseradish peroxidase (HRP)-conjugated anti-rabbit IgG goat antibody (Cell Signaling Technology, Beverly, MA, USA) (1:5000 dilution) was used as a secondary antibody. The immunoreactive proteins were visualized with a chemiluminescence detection reagent (Chemi-Lumi One Super, Nacalai Tesque, Inc., Kyoto, Japan), and the bands were detected by Image Quant 400 (GE Healthcare, Little Chalfont, Buckinghamshire, UK) and quantified using densitometric software (ImageJ downloaded from https://imagej.nih.gov/ij/download.html (accessed on 1 August 2019).

Depletion of Macrophages in Mice
Liposomal clodronate at 1.05 mg/mouse was administered i.p. on days 5 and 7 of bortezomib treatment, and nociceptive threshold was measured daily until day 8. To confirm the successful macrophage depletion, the spleen was isolated from the mice that were sacrificed by cervical dislocation following the measurement of nociceptive threshold on day 8 after the onset of bortezomib treatment and subjected to detection of macrophages by flow cytometry using FITC anti-mouse F4/80 (1:100; Biolegend, Inc., San Diego, CA, USA) and PE anti-mouse/human CD11b (1:200; Biolegend, Inc.), as described previously [23]. To check the effect of macrophage depletion on the established CIPN after bortezomib treatment, the mice received a single i.p. administration of liposomal clodronate at 1.05 mg/mouse on day 13 after the onset of i.p. bortezomib.
2.9. Determination of Expression, Release, and Cytoplasmic Translocation of Nuclear HMGB1 and Protein Levels of Cleaved Caspase-8 in RAW264.7 Cells RAW264.7 cells were seeded and cultured in a 6-well plate (12 × 10 4 cell/well) or 60-mm diameter culture dish (36 × 10 4 cell/dish) filled with 10% FCS-containing culture medium for 24 h. After additional 24-h incubation in 1% FCS-containing medium in a volume of 1 mL, the cells stimulated with bortezomib at 0.1-100 nM for 9-24 h or MG132 at 0.1-0.3 µM for 12 h [24,25]. Then, exactly 0.5 mL of the supernatant medium was collected and centrifugated. After the assessment of protein content by the Bradford method, HMGB1 levels in the supernatant were determined using an ELISA kit (Fuso Pharmaceutical Industries, Ltd., Osaka, Japan) or by Western blotting using the abovementioned antibodies, as described elsewhere [5]. Protein levels of cleaved caspase-8 in the pellets were determined by Western blotting using an anti-mouse cleaved caspase-8 (Asp387) rabbit antibody (1:1000 dilution) that recognizes both p43 (pro-domain with p18) and p18, the cleaved products of caspase-8 when caspase-8 was cleaved at Asp387 (Cell Signaling Technology), together with the above-mentioned secondary antibody.
For the immunostaining experiments, RAW264.7 cells (3 × 10 4 cells/well) were seeded on cover glass in a 24-well plate filled with 10% FCS-containing culture medium and stimulated as described above. The cells, after stimulated with bortezomib at 10 nM for 9 h, were fixed with 4% PFA in PBS. The subcellular localization of HMGB1 was detected by immunofluorescence staining using an anti-HMGB1 rabbit polyclonal antibody (Shino-Test Corporation) (1:500 dilution), TRITC-conjugated anti-rabbit IgG (Sigma-Aldrich) (1:400 dilution), and DAPI (1:200 dilution), as described previously [5]. Fluorescence images were visualized using a confocal laser fluorescence microscopy (FV10C-O, Olympus). Translocation of HMGB1 from the nucleus to cytoplasm were calculated as the ratio of nuclear HMGB1 to cytoplasmic HMGB1 using FV10-ASW software (Olympus).

Detection of Nuclear Condensation or Fragmentation by Hoechst Staining
RAW264.7 cells were seeded and cultured in a 6-well plate (12 × 10 4 cell/well) containing 10% FCS-containing culture medium for 24 h. After additional 24-h incubation in 1% FCS-containing medium, the cellsstimulated for 12 h with bortezomib at 10 nM, which was a concentration capable of releasing HMGB1 in the preliminary experiments. The cells were collected and fixed with 1% glutaraldehyde for 30 min at room temperature. After washing with PBS, the cells were stained with Hoechst 33342 (Sigma-Aldrich) at 1 mM for 15 min. The stained cells were placed on a slide glass and covered with a cover glass. The number of cells with nuclear condensation or fragmentation was counted under UV excitation light with an inverted fluorescence microscope (BX50, Olympus, Tokyo, Japan). An anti-HMGB1-neutralizing antibody, when given i.p. at 1 mg/kg before each dose of bortezomib, six times in total, completely prevented the development of the CIPN following bortezomib treatment ( Figure 1B). Interestingly, the anti-HMGB1-neutraling antibody, when administered on day 14 after the onset of bortezomib treatment, transiently elevated the nociceptive threshold lowered by bortezomib in the mice, an effect quickly disappearing thereafter ( Figure 1C). Thus, HMGB1 inactivation after the establishment of CIPN could not reverse the progression of CIPN in bortezomib-treated mice, as shown in An anti-HMGB1-neutralizing antibody, when given i.p. at 1 mg/kg before each dose of bortezomib, six times in total, completely prevented the development of the CIPN following bortezomib treatment ( Figure 1B). Interestingly, the anti-HMGB1-neutraling antibody, when administered on day 14 after the onset of bortezomib treatment, transiently elevated the nociceptive threshold lowered by bortezomib in the mice, an effect quickly disappearing thereafter ( Figure 1C). Thus, HMGB1 inactivation after the establishment of CIPN could not reverse the progression of CIPN in bortezomib-treated mice, as shown in rodent models for CIPN caused by paclitaxel and oxaliplatin [4,6]. Protein levels of HMGB1 tended to increase in the sciatic nerves and plasma, although it significantly decreased in the DRG ( Figure 1D), in agreement with results in rats with CIPN caused by paclitaxel or vincristine [4].

TM/TMα Prevents and Reverses the CIPN Caused by Bortezomib in a Thrombin-Dependent Manner in Mice
As did the anti-HMGB1-neutralizing antibody (see Figure 1B,C), TMα, capable of promoting thrombin-dependent degradation of HMGB1, when given i.p. repeatedly at 3 mg/kg, completely prevented the development of the CIPN following repeated treatment with bortezomib at 0.4 mg/kg (Figure 2A). rodent models for CIPN caused by paclitaxel and oxaliplatin [4,6]. Protein levels of HMGB1 tended to increase in the sciatic nerves and plasma, although it significantly decreased in the DRG ( Figure 1D), in agreement with results in rats with CIPN caused by paclitaxel or vincristine [4].

TM/TMα Prevents and Reverses the CIPN Caused by Bortezomib in a Thrombin-Dependent Manner in Mice
As did the anti-HMGB1-neutralizing antibody (see Figure 1B,C), TMα, capable of promoting thrombin-dependent degradation of HMGB1, when given i.p. repeatedly at 3 mg/kg, completely prevented the development of the CIPN following repeated treatment with bortezomib at 0.4 mg/kg ( Figure 2A). It is to note that repeated treatment with TMα itself did not alter nociceptive threshold in vehicle-treated mice (Supplementary Figure S1). TMα also reversed the established CIPN when administered i.p. at 3 or 10 mg/kg once on day 14 after the onset of bortezomib treatment ( Figure 2B). Argatroban, a parenteral direct thrombin inhibitor, when administered i.p. at 10 mg/kg, 30 min before each administration of TMα at 3 mg/kg, 6 times in total ( Figure 2C), abolished the anti-CIPN effect of TMα in the mice treated with bortezomib at 0.4 mg/kg ( Figure 2C,D). Most interestingly, in the mice that received repeated i.p. administration of bortezomib at a low dose, 0.1 mg/kg, repeated i.p. administration of argatroban at 10 mg/kg significantly facilitated the development of allodynia ( Figure 2E) and elevated the increased plasma HMGB1 levels ( Figure 2F).

HMGB1-Targeted Receptors Involved in the CIPN Caused by Bortezomib in Mice
We next tested whether the CIPN caused by bortezomib involves the HMGB1-targeted membrane receptors, such as RAGE, TLR4, and CXCR4, known to participate in the CIPN caused by paclitaxel or oxaliplatin [4][5][6]. FPS-ZM1 at 1 mg/kg, a RAGE antagonist, and AMD3100 at 8 mg/kg, a CXCR4 antagonist, when administered i.p. 30 min before each administration of bortezomib at 0.4 mg/kg, 6 times in total, completely prevented the development of CIPN following repeated bortezomib treatment, whereas TAK-242, a TLR4 antagonist, administered at 3 mg/kg in the same schedule, did not exhibit such an anti-CIPN effect ( Figure 3A).
It is to note that repeated treatment with FPS-ZM1 itself did not alter nociceptive threshold in vehicle-treated mice (Supplementary Figure S1). Most interestingly, FPS-ZM1 but not AMD3100 or TAK-242, given i.p. once on day 14 after the onset of bortezomib, transiently reversed the established CIPN in the mice treated repeatedly with bortezomib at 0.4 mg/kg ( Figure 3B). These behavioral results are apparently in agreement with the overexpression of RAGE, but not TLR4 or CXCR4, in the DRG observed on day 14 after the onset of bortezomib treatment ( Figure 3C), although RAGE was not upregulated in the sciatic nerves of the bortezomib-treated mice ( Figure 3D). There is a report showing the involvement of TLR5 in the allodynia induced by exogenously applied HMGB1 in mice [26]. However, TH1020, a TLR5 antagonist, when administered i.p. at 1 mg/kg repeatedly and once in the above-described manner, neither prevented nor reversed the CIPN in bortezomib-treated mice, respectively ( Figure 3E,F).

Involvement of Macrophages in the Developmental Stage of CIPN Following Bortezomib Treatment in Mice
Given evidence that the origins of endogenous HMGB1 responsible for CIPN development appear to differ depending on chemotherapeutics employed [2,[5][6][7], we asked if different inhibitors or a depletor of macrophages could affect the CIPN caused by bortezomib. Repeated i.p. administration of minocycline at 30 mg/kg, a macrophage/microglia inhibitor, or ethyl pyruvate at 80 mg/kg, known to inhibit HMGB1 release from macrophages, blocked the CIPN development following bortezomib treatment ( Figure 4A,B).
Liposomal clodronate, a macrophage depletor, when administered i.p. twice at 1.05 mg/mouse on days 5 and 7, dramatically decreased the number of macrophages in the spleen ( Figure 4C), and significantly reduced the CIPN development following bortezomib treatment ( Figure 4D). A single administration of minocycline at the same dose, known to inhibit not only peripheral macrophages but also spinal microglia, significantly reversed the established CIPN in bortezomib-treated mice ( Figure 4E), although neither ethyl pyruvate nor liposomal clodronate affected the established CIPN ( Figure 4F,G).

Involvement of Macrophages in the Developmental Stage of CIPN following Bortezomib Treatment in Mice
Given evidence that the origins of endogenous HMGB1 responsible for CIPN development appear to differ depending on chemotherapeutics employed [2,[5][6][7], we asked if different inhibitors or a depletor of macrophages could affect the CIPN caused by bortezomib. Repeated i.p. administration of minocycline at 30 mg/kg, a macrophage/microglia inhibitor, or ethyl pyruvate at 80 mg/kg, known to inhibit HMGB1 release from macrophages, blocked the CIPN development following bortezomib treatment ( Figure 4A,B).
Liposomal clodronate, a macrophage depletor, when administered i.p. twice at 1.05 mg/mouse on days 5 and 7, dramatically decreased the number of macrophages in the spleen ( Figure 4C), and significantly reduced the CIPN development following bortezomib treatment ( Figure 4D). A single administration of minocycline at the same dose, known to inhibit not only peripheral macrophages but also spinal microglia, significantly reversed the established CIPN in bortezomib-treated mice ( Figure 4E), although neither ethyl pyruvate nor liposomal clodronate affected the established CIPN ( Figure 4F,G).  Immunofluorescent analyses showed that repeated treatment with bortezomib caused macrophage accumulation in DRG at L4-L6 spinal levels but not in the sciatic nerve on day 3 after the onset of bortezomib treatment ( Figure 5A,B,D,E), while no such increase in the number of macrophages was detected in the sciatic nerve or DRG on day 14 after the onset of bortezomib treatment ( Figure 5A,C,D,F). It is noteworthy that there were significant (p < 0.001) differences in the number of macrophages between days 3 and 14 when analyzed by two-way ANOVA, which might the reflect non-specific effect of repeated i.p. injections or variations on measurement because of unexpected biases. Immunofluorescent analyses showed that repeated treatment with bortezomib caused macrophage accumulation in DRG at L4-L6 spinal levels but not in the sciatic nerve on day 3 after the onset of bortezomib treatment ( Figure 5A,B,D,E), while no such increase in the number of macrophages was detected in the sciatic nerve or DRG on day 14 after the onset of bortezomib treatment ( Figure 5A,C,D,F). It is noteworthy that there were significant (p < 0.001) differences in the number of macrophages between days 3 and 14 when analyzed by two-way ANOVA, which might the reflect non-specific effect of repeated i.p. injections or variations on measurement because of unexpected biases.

Bortezomib and MG132, Proteasome Inhibitors, Evoke HMGB1 Release from Macrophage-Like RAW264.7 Cells
In mouse macrophage-like RAW264.7 cells, stimulation with bortezomib at 10 and/or 100 nM caused the cytoplasmic translocation of nuclear HMGB1 at 9 h ( Figure 6A,B) and extracellular release of HMGB1 at 24 h, as assessed by Western blotting ( Figure 6C).

Bortezomib and MG132, Proteasome Inhibitors, Evoke HMGB1 Release from Macrophage-Like RAW264.7 Cells
In mouse macrophage-like RAW264.7 cells, stimulation with bortezomib at 10 and/or 100 nM caused the cytoplasmic translocation of nuclear HMGB1 at 9 h ( Figure 6A,B) and extracellular release of HMGB1 at 24 h, as assessed by Western blotting ( Figure 6C). The time-course experiments employing an ELISA kit for HMGB1 measurement showed that extracellular HMGB1 concentrations gradually increased during the time period from 12 to18 h after the onset of stimulation with 10 nM bortezomib ( Figure 6D). MG132, a well-known proteasome inhibitor, at 0.3 µM, like bortezomib at 10 nM, caused HMGB1 release in 12 h ( Figure 6E).

Involvement of Caspase in Bortezomib-Induced HMGB1 Release from RAW264.7 Macrophages
We then tested effects of some inhibitors of cell signals on the bortezomib-induced HMGB1 release from RAW264.7 cells. Pyrrolidine dithiocarbamate (PDTC) and SB203580, inhibitors of NF-κB and p38MAPK, respectively, which abolished the paclitaxel-induced HMGB1 release from macrophages in our previous study [5], failed to reduce the bortezomib-induced HMGB1 release from RAW264.7 macrophages (Figure 7A,B). The time-course experiments employing an ELISA kit for HMGB1 measurement showed that extracellular HMGB1 concentrations gradually increased during the time period from 12 to18 h after the onset of stimulation with 10 nM bortezomib ( Figure 6D). MG132, a well-known proteasome inhibitor, at 0.3 µM, like bortezomib at 10 nM, caused HMGB1 release in 12 h ( Figure 6E).

Involvement of Caspase in Bortezomib-Induced HMGB1 Release from RAW264.7 Macrophages
We then tested effects of some inhibitors of cell signals on the bortezomib-induced HMGB1 release from RAW264.7 cells. Pyrrolidine dithiocarbamate (PDTC) and SB203580, inhibitors of NF-κB and p38MAPK, respectively, which abolished the paclitaxel-induced HMGB1 release from macrophages in our previous study [5], failed to reduce the bortezomibinduced HMGB1 release from RAW264.7 macrophages (Figure 7A,B).  STO-609, pyridone 6, and SP600125, inhibitors of the Ca 2+ /calmodulin-dependent protein kinase kinase (CaMKK), and JAK and JNK, respectively, which had been reported to block LPS-induced HMGB1 release from macrophages [13][14][15], did not significantly inhibit the bortezomib-induced HMGB1 release from RAW264.7 cells ( Figure 7C). Given evidence for the relationship between apoptosis and HMGB1 release in macrophages [27,28], we next tested whether caspase inhibitors could inhibit the bortezomib-induced HMGB1 release from RAW264.7 cells. Z-VAD-FMK, a pan caspase inhibitor, at 20 µM completely inhibited the HMGB1 release from macrophages in response to bortezomib at 10 nM ( Figure 7D). In contrast, surprisingly, Z-VAD-FMK rather promoted paclitaxel-induced HMGB1 release from macrophages ( Figure 7D). We then found that Z-IETD-FMK, a caspase-8 inhibitor, at 20 µM significantly reduced the bortezomib-induced HMGB1 release from macrophages ( Figure 7E) and that bortezomib treatment gradually increased protein levels of two cleaved products of caspase-8, p43 (containing p18) and p18, at 3-9 h in a time-dependent manner ( Figure 7F,G). Actually, bortezomib at 10 nM caused apoptosis characterized by nuclear condensation or fragmentation in macrophages at 12 h ( Figure 7H,I).

Systemic Administration of a Caspase Inhibitor Prevents the CIPN Caused by Bortezomib in Mice
Finally, we then asked whether systemic administration of Z-VAD-FMK, the caspase inhibitor, could inhibit the CIPN in mice treated with bortezomib. Repeated i.p. administration of Z-VAD-FMK at 1 mg/kg significantly reduced the development of CIPN following repeated treatment with bortezomib at 0.4 mg/kg in the mice (Figure 8). STO-609, pyridone 6, and SP600125, inhibitors of the Ca 2+ /calmodulin-dependent protein kinase kinase (CaMKK), and JAK and JNK, respectively, which had been reported to block LPS-induced HMGB1 release from macrophages [13][14][15], did not significantly inhibit the bortezomib-induced HMGB1 release from RAW264.7 cells ( Figure 7C). Given evidence for the relationship between apoptosis and HMGB1 release in macrophages [27,28], we next tested whether caspase inhibitors could inhibit the bortezomib-induced HMGB1 release from RAW264.7 cells. Z-VAD-FMK, a pan caspase inhibitor, at 20 µM completely inhibited the HMGB1 release from macrophages in response to bortezomib at 10 nM ( Figure 7D). In contrast, surprisingly, Z-VAD-FMK rather promoted paclitaxel-induced HMGB1 release from macrophages ( Figure 7D). We then found that Z-IETD-FMK, a caspase-8 inhibitor, at 20 µM significantly reduced the bortezomib-induced HMGB1 release from macrophages ( Figure 7E) and that bortezomib treatment gradually increased protein levels of two cleaved products of caspase-8, p43 (containing p18) and p18, at 3-9 h in a time-dependent manner ( Figure 7F,G). Actually, bortezomib at 10 nM caused apoptosis characterized by nuclear condensation or fragmentation in macrophages at 12 h (Figure 7H,I).

Systemic Administration of a Caspase Inhibitor Prevents the CIPN Caused by Bortezomib in Mice
Finally, we then asked whether systemic administration of Z-VAD-FMK, the caspase inhibitor, could inhibit the CIPN in mice treated with bortezomib. Repeated i.p. administration of Z-VAD-FMK at 1 mg/kg significantly reduced the development of CIPN following repeated treatment with bortezomib at 0.4 mg/kg in the mice (Figure 8). It is to note that repeated treatment with Z-VAD-FMK itself did not alter nociceptive threshold in vehicle-treated mice (Supplementary Figure S1). It is to note that repeated treatment with Z-VAD-FMK itself did not alter nociceptive threshold in vehicle-treated mice (Supplementary Figure S1).

Discussion
The present study demonstrates the involvement of macrophage-derived HMGB1 in the development of CIPN in mice treated with bortezomib, in agreement with the mechanisms for the CIPN caused by paclitaxel [5]. Surprisingly, our findings from experiments employing RAW264.7 macrophages indicate that the bortezomib-induced HMGB1 release from macrophages involves caspase-dependent apoptosis but is independent of p38MAPK and NF-κB, which are involved in the paclitaxel-induced HMGB1 release from macrophages [5]. Considering the results from in vivo inhibition experiments, our study strongly suggests that repeated treatment with bortezomib causes caspase-dependent release of HMGB1 from macrophages, which in turn triggers neuronal sensitization by activating RAGE and accelerating CXCL12/CXCR4 signaling, resulting in CIPN (Figure 9).

Discussion
The present study demonstrates the involvement of macrophage-derived HMGB1 in the development of CIPN in mice treated with bortezomib, in agreement with the mechanisms for the CIPN caused by paclitaxel [5]. Surprisingly, our findings from experiments employing RAW264.7 macrophages indicate that the bortezomib-induced HMGB1 release from macrophages involves caspase-dependent apoptosis but is independent of p38MAPK and NF-κB, which are involved in the paclitaxel-induced HMGB1 release from macrophages [5]. Considering the results from in vivo inhibition experiments, our study strongly suggests that repeated treatment with bortezomib causes caspase-dependent release of HMGB1 from macrophages, which in turn triggers neuronal sensitization by activating RAGE and accelerating CXCL12/CXCR4 signaling, resulting in CIPN (Figure 9). . Scheme for bortezomib-induced caspase-dependent HMGB1 release from macrophages and CIPN development, in contrast to caspase-independent mechanisms for paclitaxel. (A) Inhibition of proteasome by bortezomib causes caspasedependent apoptosis of macrophages followed by the release of HMGB1, which in turn causes neuronal sensitization via activation of RAGE and acceleration of CXCL12/CXCR4 signals, leading to CIPN. (B) Paclitaxel causes HMGB1 release from macrophages through activation of the ROS/p38MAPK/NF-κB pathway [5], independently of caspase (see Figure  7D), and the extracellular HMGB1 develops CIPN in a manner dependent on RAGE and CXCR4 [5], as shown in the CIPN caused by bortezomib.
The present study, together with a series of our studies on CIPN [2,[4][5][6]29], ascertains that HMGB1 is a common crucial molecule responsible for the development of CIPN in rodents treated with different chemotherapeutics including bortezomib, paclitaxel, oxaliplatin, and vincristine. However, it is surprising that the origins of HMGB1 and molecular mechanisms for HMGB1 release vary depending on chemotherapeutics employed, e.g., the origins of HMGB1 are macrophages for bortezomib and paclitaxel [5] versus non-macrophage cells for oxaliplatin [6], and the mechanisms for HMGB1 release involve caspase  Figure 9. Scheme for bortezomib-induced caspase-dependent HMGB1 release from macrophages and CIPN development, in contrast to caspase-independent mechanisms for paclitaxel. (A) Inhibition of proteasome by bortezomib causes caspasedependent apoptosis of macrophages followed by the release of HMGB1, which in turn causes neuronal sensitization via activation of RAGE and acceleration of CXCL12/CXCR4 signals, leading to CIPN. (B) Paclitaxel causes HMGB1 release from macrophages through activation of the ROS/p38MAPK/NF-κB pathway [5], independently of caspase (see Figure 7D), and the extracellular HMGB1 develops CIPN in a manner dependent on RAGE and CXCR4 [5], as shown in the CIPN caused by bortezomib.
The present study, together with a series of our studies on CIPN [2,[4][5][6]29], ascertains that HMGB1 is a common crucial molecule responsible for the development of CIPN in rodents treated with different chemotherapeutics including bortezomib, paclitaxel, oxaliplatin, and vincristine. However, it is surprising that the origins of HMGB1 and molecular mechanisms for HMGB1 release vary depending on chemotherapeutics employed, e.g., the origins of HMGB1 are macrophages for bortezomib and paclitaxel [5] versus nonmacrophage cells for oxaliplatin [6], and the mechanisms for HMGB1 release involve caspase for bortezomib versus ROS/p38MAPK/NF-κB signals for paclitaxel [5] (Figure 9). Our findings in inhibition experiments suggest that the extracellular HMGB1 derived from macrophages targets RAGE and CXCR4, but not TLR4 or TLR5, in the production of CIPN following bortezomib treatment in mice, being consistent with our previous study characterizing CIPN caused by paclitaxel in mice [5] (Figure 9). However, it is to be noted that spinal, but not peripheral, TLR4 might play a role in the CIPN caused by paclitaxel in C57BL/6 mice [5,30], and that the CIPN caused by oxaliplatin involves TLR4 in addition to RAGE and CXCR4 [6].
The present findings that repeated treatment with TMα completely prevented the CIPN following bortezomib treatment, as it did in rodents treated with paclitaxel, vincristine, or oxaliplatin [4][5][6], encourage its clinical application for the prevention of CIPN in various cancer patients undergoing different chemotherapeutic agents, in addition to colorectal cancer patients undergoing oxaliplatin-based chemotherapy [12]. The cancellation of the anti-CIPN effect of TMα by argatroban in bortezomib-treated mice (see Figure 2D) suggests its dependence on endogenous thrombin, as reported elsewhere [6,29,31]. Our results that repeated administration of argatroban significantly enhanced plasma HMGB1 levels and aggravated CIPN itself in mice treated with a subeffective dose of bortezomib (see Figure 2E,F), suggest the protective role of the endothelial TM/thrombin system as an HMGB1-inactivating system against CIPN, in agreement with our previous study using a mouse model of CIPN caused by oxaliplatin [6].
The links between macrophages, HMGB1, and CIPN do not necessarily explain the overall picture of molecular mechanisms underlying CIPN caused by bortezomib. In the present study, we found that HMGB1 levels in the sciatic nerve and plasma tended to increase only on day 14 after the onset of bortezomib treatment (see Figures 1D and 2F), while macrophage accumulation in DRG was observed on day 3, but not day 14 (see Figure 5D-F). Thus, there are still some concerns to be addressed by further follow-up studies, such as the investigation of the effects of HMGB1 on CIPN in macrophage-depleted mice, monitoring of the time-course of changes in HMGB1 levels in the tissue and plasma.
Our findings that bortezomib as well as MG132, a well-known another proteasome inhibitor, directly stimulated the release of HMGB1 from mouse macrophage-like RAW264.7 cells (see Figure 6E) indicate that proteasome inhibition itself causes HMGB1 release in macrophages. There is evidence that HMGB1 can be released from macrophages during apoptosis [27,28], and that bortezomib induces apoptosis of cancer cells through stabilization and inhibition of proteasomal degradation of pro-apoptotic proteins such as p53 and IκB [32][33][34]. Therefore, it is understandable that bortezomib induced caspase activation and apoptosis in macrophages, leading to caspase-dependent HMGB1 release (see Figure 7D-I), and that the CIPN caused by bortezomib in mice was prevented by repeated administration of the caspase inhibitor (see Figure 8) in the present study. On the other hand, in our previous study [5], LPS-RS, AMD3100, and FPS-ZM1, antagonists of TLR4, CXCR4, and RAGE, respectively, that prevented the CIPN in paclitaxel-treated mice, did not inhibit the paclitaxel-induced HMGB1 release from macrophages. Nevertheless, considering the differences of molecular mechanisms for the release of HMGB1 from macrophages by bortezomib and paclitaxel, it is interesting to evaluate the involvement of autocrine or paracrine acceleration mechanisms in the bortezomib-induced HMGB1 release from macrophages in future.
Our results that the inactivation of HMGB1 by its neutralizing antibody or TMα reversed the established CIPN in mice treated with bortezomib (see Figures 1C and 2B) suggest the role of HMGB1 in the maintenance of the CIPN, as reported in the CIPN caused by paclitaxel or oxaliplatin in rodents [4,6]. However, neither liposomal clodronate, a macrophage depletor, nor ethyl pyruvate, known to inhibit HMGB1 release from macrophages, had such effect on the established CIPN (see Figure 4F,G), suggesting that macrophages are not involved in the maintenance of CIPN. This is consistent with the detection of macrophage accumulation in the DRG on day 3, but not day 14, after the onset of bortezomib treatment (see Figure 5D-F). Nonetheless, minocycline, a macrophage/microglia inhibitor, reversed the established CIPN (see Figure 4E), as did the anti-HMGB1-neutralizing antibody or TMα (see Figures 1C and 2B), suggesting a role of microglia in the maintenance of CIPN caused by bortezomib, although this has yet to be analyzed in-depth by future studies. Furthermore, the endothelium and nociceptors could be origins for the extracellular HMGB1, because the endothelial cells and nociceptive neurons secrete HMGB1 in response to inflammatory stimuli and/or following surgical injury to the sciatic nerves [35,36]. Our studies to clarify the involvement of HMGB1 derived from non-macrophage/non-microglial cells in CIPN are now in progress. Interestingly, blockers of RAGE, but not CXCR4, TLR4, or TLR5, also transiently reversed the established CIPN (see Figure 3B,F), which might be supported by our results that RAGE, but not TLR4 or CXCR4, in the DRG was upregulated on day 14 after the onset of bortezomib treatment (see Figure 3C), in agreement with the previous report concerning the bortezomib-induced upregulation of RAGE in the spinal cord [37]. However, the inhibition of the HMGB1/RAGE system in the persistent phase of CIPN following bortezomib might be quickly compensated by unknown pronociceptive systems, considering the short duration of the anti-CIPN effects of the anti-HMGB1-neutralizing antibody, TMα and RAGE antagonist (see Figures 1C, 2B and 3B).

Conclusions
In conclusion, the present study indicates that the activation or acceleration of RAGE and CXCR4 signals by macrophage-derived HMGB1 plays a crucial role in the development of CIPN in bortezomib-treated mice, as it did in paclitaxel-treated mice, which is negatively regulated by TM/TMα in a thrombin-dependent manner. However, the molecular mechanisms for the HMGB1 release from macrophages are different between bortezomib and paclitaxel, i.e., caspase-dependent apoptosis is involved in the HMGB1 release from macrophages and the development of CIPN caused by bortezomib, but not paclitaxel [5] (Figure 9). Thus, the present study sheds light on the previously unknown mechanisms for CIPN caused by bortezomib, including caspase-dependent HMGB1 release from macrophages, and justify pharmacological intervention targeting HMGB1 for the prevention of CIPN in cancer patients undergoing chemotherapeutics including bortezomib.