Erythropoietin Attenuates Postoperative Cognitive Dysfunction by Shifting Macrophage Activation toward the M2 Phenotype

Postoperative cognitive dysfunction (POCD) may be driven by transference of the innate immune response to the brain after aseptic surgical damage. Macrophages are key mediators of innate immunity that can display a pro-inflammatory M1 phenotype or an anti-inflammatory M2 phenotype. Erythropoietin (EPO) is a hematopoietic hormone that exerts anti-inflammatory effects by influencing macrophage function. We hypothesized that EPO would prevent POCD by promoting macrophage phenotype switching to the M2 phenotype post-surgery. To evaluate the effects of EPO on POCD and macrophage polarization post-surgery, we administered EPO (5,000 U/kg) with or without an arginase inhibitor (amino-6-boronohexanoic acid, 10 mg/kg) to ICR mice before and after abdominal surgery. Forty-eight hours post-surgery, we assessed memory, synapse function, and macrophage/microglial phenotypes in the spleen and hippocampus. We also investigated M1/M2 phenotypes in RAW264.7 and BV2 cells stimulated with lipopolysaccharide and interferon-γ (M1 inducers) in the presence or absence of EPO. EPO prevented POCD, decreased surgery-related synaptic dysfunction, and attenuated pro-inflammatory cytokine generation in the hippocampus. Moreover, EPO suppressed M1-related genes expression and promoted M2 genes expression in the spleen and hippocampus post-surgery. Furthermore, EPO decreased the proportions of macrophages/microglia expressing an M1 surface marker (CD40) and increased those expressing an M2 surface marker (CD206). Arginase inhibition abolished the beneficial effects of EPO on POCD. In vitro, EPO treatment promoted switching of RAW264.7 and BV2 cells stimulated with M1 inducers to an M2 phenotype. In conclusion, EPO prevents POCD by promoting macrophage phenotype switching toward the M2 phenotype.


INTRODUCTION
Postoperative cognitive dysfunction (POCD) is a major neurological complication after surgery that adversely affects the quality of life, social independence, and mortality of patients (Steinmetz et al., 2009;Mashour et al., 2015). Even acute cognitive decline after surgery can result in persistent cognitive dysfunction (Saczynski et al., 2012;Inouye et al., 2016). Although several mechanisms have been associated with the development of POCD, mounting preclinical evidence suggests that inflammation in the brain (neuroinflammation) induced by systemic inflammation after surgery is a core pathogenic factor (Cibelli et al., 2010;Terrando et al., 2010;Hovens et al., 2014;Vacas et al., 2014). Furthermore, a recent meta-analysis confirmed that the systemic inflammatory response to surgery is associated with POCD in clinical settings (Peng et al., 2013).
Surgical interventions inevitably damage tissue and lead to activation of the innate immune response. Macrophage act as key mediators of innate immunity. Macrophages that are activated after surgery can infiltrate the brain parenchyma to induce microglial activation and neuroinflammation. The resulting increase in pro-inflammatory cytokine expression, especially within hippocampus, eventually leads to cognitive dysfunction (Wan et al., 2007;Dilger and Johnson, 2008;Terrando et al., 2011;Degos et al., 2013). However, macrophages can be differentially activated in response to environmental cues or pathophysiological conditions (Wang et al., 2014). Activated macrophage phenotypes include the classically activated M1 phenotype and the alternatively activated M2 phenotype. The M1 phenotype is induced by microbial products, pro-inflammatory cytokines, and destructive inflammatory mediators, such as interleukin (IL)-12, IL-23, nitric oxide, and reactive oxygen species. In contrast, the M2 phenotype can be induced by IL-4, IL-10, or IL-13, and this phenotype facilitates the clearance of cellular debris through phagocytosis, as well as the production of protective and trophic factors, such as ornithine and polyamines, through the arginase pathway. Therefore, in various disease conditions, the M1 phenotype initiates and maintains inflammation, while the M2 phenotype prevents or resolves chronic inflammation. To date, there is little evidence regarding the association between POCD and patterns of macrophage activation.
Erythropoietin (EPO) is a 30.4-kDa glycoprotein hormone that promotes red blood cell production by inhibiting the apoptosis of erythroid progenitor cells (Lombardero et al., 2011). Accordingly, EPO has been applied in clinical settings for the prevention and management of perioperative anemia (Steuber et al., 2016). The cognate receptor for EPO is expressed in various cell types; thus, EPO has pleiotropic actions in nonhematopoietic as well as hematopoietic cells (Lombardero et al., 2011). Immune cells also express the EPO receptor and are thus subject to modulation by EPO (Brines and Cerami, 2012). Several studies have reported that the treatment of activated macrophages with EPO significantly reduces the production of M1-like pro-inflammatory mediators and enhances M2associated phagocytic activity (Nairz et al., 2011;Liu et al., 2014;Luo et al., 2016). Since EPO is commercially available in the form of recombinant human EPO (rhEPO) and widely used in the treatment of anemia now, it is very important to identify its other therapeutic potentials or mechanisms of action. Therefore, in the present study, we tested whether EPO treatment would promote the activation of macrophage/microglia toward the M2 phenotype, reduce neuroinflammation, and thus limit the extent of post-surgical cognitive dysfunction. We examined cognitive dysfunction after abdominal surgery because abdominal surgery is one of the most commonly performed surgeries and is reported to induce POCD in clinical settings (Monk et al., 2008). The model of abdominal surgery adopted in this study includes physical irritation to the viscera and mesenteric ischemiareperfusion to mimic human exploratory abdominal surgery.

Animals
All experimental procedures and animal care were approved by the Institutional Animal Care and Use Committee of Yonsei University, and were conducted in accordance with the Guide for the Care and Use of Laboratory Animals approved by the Association for the Assessment and Accreditation of Laboratory Animal Care. Adult male ICR mice (Orient, Seongnam, GyeongGi-Do, South Korea; aged 8-10 weeks) were used in the experiment. All mice were housed in a controlled animal facility at Yonsei University. The mice were housed in groups of five per cage on a 12-h light/dark cycle with food and water available ad libitum.

Experimental Design and Animal Grouping
The mice were randomly divided into four groups as follows: (1) control, (2) surgery, (3) EPO-control, and (4) EPO-surgery. Mice in the control group remained naïve to the experimental conditions. Mice in the EPO-control group only received rhEPO treatment, without any anesthetic or surgical procedures. Mice in the surgery group underwent abdominal surgery only. Finally, mice in the EPO-surgery group received rhEPO treatment and underwent abdominal surgery. At 1 h prior to surgery, mice in the EPO-surgery received rhEPO (5,000 U/kg, subcutaneous [s.c].; Epokine 1,000 U/ml, CJ Healthcare, Seoul, South Korea) diluted in 200 µl of phosphate-buffered saline (PBS). They received a second injection 24 h after the first injection. Mice in the EPO-control group received two injections of rhEPO with a 24 h interval between them. For surgery, the animals were anesthetized with 3% sevoflurane and maintained under anesthesia with 2% sevoflurane in a mixture of oxygen and air (inspired oxygen fraction, 0.3). Mice were placed on a heating pad while anesthetized to prevent hypothermia. Abdominal surgery was performed as previously described with some modifications (Barrientos et al., 2012;Hovens et al., 2014). Briefly, a vertical incision was made along the midline, along the linea alba, and the intestines were exteriorized and rubbed with the operator's fingers for 30 s. In addition, the superior mesenteric artery was dissected and clamped for 40 min. After removal of the clamp, the intestine was placed back into the peritoneal cavity and the abdominal muscle and skin were closed using suture. Tramadol (20 mg/kg intraperitoneal [i.p.]) was administered only to the mice undergoing surgery for post-operative analgesia.
Cognitive function was assessed using a behavioral test battery carried out 48 h post-surgery. The levels of inflammatory cytokines (IL-1β and tumor necrosis factor alpha [TNF-α]) and extracellular field potentials in the hippocampus were measured at the same time point in order to evaluate hippocampal synaptic function. To assess the pattern of macrophage/microglial activation, mRNA expression levels of selected M1-related (TNFα and CXCL10) and M2-related (arginase-1 and IL-10) genes were measured in spleen and hippocampus tissues at 6 and 48 h post-surgery. Furthermore, proportions of M1-and M2like macrophages/microglia were also assessed in the spleen and hippocampus tissues using flow cytometry. For the collection of spleen and hippocampal tissues, mice were anesthetized with a mixture of 30 mg/kg zoletil (Virbac Laboratory, Carros, France) and 10 mg/kg xylazine (Bayer Korea Ltd., Seoul, South Korea) and euthanized by transcardial perfusion with 0.9% normal saline at each time point. The hippocampus was isolated under a dissecting microscope as previously described (Spijker, 2011).
Finally, to confirm a beneficial role for M2 macrophage activation in cognitive function post-surgery, we performed an additional study in which a selective arginase inhibitor was co-administered with rhEPO prior to surgery. Arginase activity mediates most M2-associated anti-inflammatory effects (Pesce et al., 2009) and is positively associated with M2 macrophages (Yang and Ming, 2014). Therefore, we assessed cognitive function and levels of inflammatory cytokines in the hippocampus at 48 h post-surgery in mice that were treated with both rhEPO and an arginase inhibitor (EPO-surgery-AI group). Mice in the EPOsurgery-AI groups received rhEPO (5,000 U/kg, s.c.; Epokine 1,000 U/ml, CJ Healthcare, Seoul, South Korea) diluted in 200 µl of phosphate-buffered saline (PBS) at the same time points as the EPO-Surgery group (1 h before surgery and 24 h after the first administration), and the selective arginase inhibitor, amino-6-boronohexanoic acid (10 mg/kg, i.p.; ABH ammonium salt, Enzo Life Sciences, Farmingdale, NY, USA) dissolved in PBS 30 min before each rhEPO injection. The dose of the selective arginase inhibitor was selected based on a previous study in which 10 mg/kg of the ABH salt abolished arginase activity in M2 phenotype macrophages (Amantea et al., 2016).

Neurobehavioral Assessment
Neurobehavioral function was assessed by three different types of tests at 48 h post-surgery. Open field testing was performed to assess exploratory activity. Mice were tested in a square open field arena (40 × 40 × 40 cm) and behavior was recorded for 5 min. Total distance moved was assessed. The passive avoidance and novel object recognition tests were selected to assess hippocampal-dependent function (Antunes and Biala, 2012;Izquierdo et al., 2016). Passive avoidance testing was used to evaluate learning and memory in the mice. The apparatus used was a rectangular chamber divided into dark and lighted compartments. During the acquisition phase, mice were placed into the lighted compartment; when mice entered the dark compartment, they received an electric shock (0.5 mA) for 3 s. In the test phase, mice were place in the lighted compartment and the latency to enter the dark compartment was measured and recorded. The acquisition trial was performed at 24 h after surgery and the test trial was performed at 48 h postsurgery. Novel object recognition test was adopted to evaluate cognition, specifically recognition memory, in the mice. During the habituation phase, each mouse was allowed to explore the black box (40 × 40 × 40 cm) used for testing for 5 min. During the familiarization phase, the mice were placed into the black box, which contained two identical objects (A + A) and were allowed to explore for 5 min. During the test phase, each mouse was returned to the box with the two objects, where one object was changed to a novel object (A + B), and they were allowed to explore for 3 min. During both the familiarization and test phases, the time spent in exploring each object was measured and recorded. At the end of each test, the apparatus and objects were cleaned with 50% ethanol. The habituation phase was performed immediately prior to surgery, the familiarization phase was performed at 24 h post-surgery, and the test trial was performed at 48 h post-surgery.
All the neurobehavioral tests were recorded on video and analyzed with an image analyzing system (SMART v2.5.21 software and SMART video Tracking system, Panlab Harvard Apparatus, Barcelona, Spain) by an assessor blinded to the treatment groups. Some mice performed the open field test and the passive avoidance test sequentially. The remaining mice only performed the novel object recognition test.

Electrophysiology
To investigate the response of the hippocampal CA1 circuit to Schaffer collateral (SC) input, hippocampal brain slices (thickness, 400 µm) were prepared from the mice at 48 h postsurgery. Briefly, the mice were anesthetized with sevoflurane and the brain was rapidly cooled via transcardial perfusion with ice-cold sucrose-artificial cerebrospinal fluid (aCSF). Brains were next removed and placed in the iced-cold sucrose-aCSF. Coronal slices were allowed to recover in aCSF for 30 min at 35 • C, and then incubated in aCSF for 1-4 h at room temperature (23-25 • C) before being placed into the recording chamber. Standard aCSF contained 124 mM NaCl, 2.5 mM KCl, 2.5 mM CaCl 2 , 1.3 mM MgSO 4 , 1.0 mM NaH 2 PO 4 , 26.2 mM NaH 2 CO 3 , 11 mM glucose, 2 mM Na pyruvate, and 1 mM Na ascorbate, saturated with 95% O 2 /5% CO 2 . Sucrose-aCSF contained 195.5 mM sucrose, 2.5 mM KCl, 2.5 mM CaCl 2 , 1.3 mM MgSO 4 , 1 mM NaH 2 PO 4 , 26.2 mM NaHCO 3 , 11 mM glucose, 2 mM Na pyruvate, and 1 mM Na ascorbate, saturated with 95% O 2 /5% CO 2 . All experiments were conducted at 27-29 • C. For electrophysiological experiments, electrodes with 3-6 M pipette resistance were used and whole-cell recordings were obtained from neurons under visual guidance using infrareddifferential interference contrast optical guidance. The CA3 and dentate gyrus regions were cut away immediately prior to longterm potentiation (LTP) experiments to isolate the CA1 area. The stimuli were applied to the SC pathway and field excitatory postsynaptic potentials (fEPSPs) were recorded from the CA1 stratum radiatum using an extracellular glass pipette (3-5 M ) filled with aCSF. SC/commissural fibers in the stratum radiatum were stimulated using a concentric bipolar electrode placed 200-300 µm from the recording pipette. The test stimulation in all fEPSP experiments was 0.1 Hz in frequency and 0.2 ms in duration. An input/output (I/O) curve was plotted by recording fEPSPs in response to stimulation with increasing voltages (0.3-0.9 V). The stimulus for LTP was a voltage that elicited 50% of the maximum fEPSP. Baseline synaptic responses were recorded for 20 min and then LTP was induced with two high-frequency stimulation trains (100 Hz frequency, 1 s duration) separated by a 20-s inter-train interval. The recordings were performed every 30 s for 1 h using an axopatch 1D amplifier (Molecular Devices, Sunnyvale, CA, USA) digitized at 10 KHz and filtered at 2 KHz with Digidata 1322A and pClamp 9.0 software (Molecular Devices).

Cell Cultures and Treatments
Mouse microglial BV2 cells were cultured in RPMI 1640 medium (Hyclone TM , GE Healthcare Life Sciences, Logan, UT, USA) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin solution. Mouse macrophage RAW 264.7 cells (KCIB, Seoul, Korea) were cultured in highglucose Dulbecco's modified eagle medium (Hyclone TM , GE Healthcare Life Sciences) supplemented with 10% FBS and 1% penicillin-streptomycin solution. Cells were incubated at 37 • C in a humidified atmosphere in the presence of 5% CO 2 . For experiments, each cell line was seeded in six well plates at a density of 2 × 10 6 per well 24 h prior to treatment. Cells were next pretreated with 5 U/ml rhEPO or PBS for 30 min and then stimulated with 50 ng/ml lipopolysaccharide (LPS) combined with 20 ng/ml interferon (IFN)-γ or PBS for 24 h.

Enzyme-Linked Immunosorbent Assay (ELISA)
For the in vivo experiment, the hippocampi were obtained at 48 h post-surgery and stored at −80 • C until use. To measure levels of hippocampal IL-1β and TNF-α, the hippocampal tissues were lysed using tissue protein extraction reagent (T-PER R Tissue Protein Extraction Reagent, Thermo Scientific TM , Waltham, MA, USA) containing protease and phosphatase inhibitor cocktail (100X Halt protease and phosphatase inhibitor cocktail, #1861281 Thermo Scientific TM ) added at a volume of 20 µl per 1 mg of tissue. The tissues were then homogenized and centrifuged at 13,000 rpm for 10 min to obtain sample supernatants. Supernatant protein concentrations were measured with a BCA Protein Assay Kit (Thermo Scientific TM ) according to the manufacturer specifications. Levels of mouse IL-1β and TNF-α in hippocampal lysates were assayed using high-sensitivity ELISA kits (Quantikine R ELISA, R&D Systems Inc., Minneapolis, MN, USA) according to the manufacturer specifications. Briefly, samples were added to the assay plates at a volume of 50 µl/well and incubated for 2 h at room temperature. After washing plates with the wash buffer from the kit, mouse IL-1β and TNF-α conjugates were added to each well and incubated for 2 h. The reaction was stopped and well absorbances were read at 450 nm using a microplate reader. For the in vitro experiment, we measured levels of IL-12 and IL-10 in BV2 and RAW 264.7 cell culture supernatants. Briefly, supernatants were collected at 24 h after LPS + IFN-γ stimulation and centrifuged at 1200 rpm for 3 min to remove cell debris. The supernatant protein concentrations were measured using a spectrophotometer (NanoDrop R ND-1000, Thermo Fisher Scientific Inc., Wilmington, DE, USA) at the A280 absorbance level. Levels of IL-12 and IL-10 in supernatants were assayed using high-sensitivity ELISA kits (Quantikine R ELISA, R&D Systems) according to the manufacturer specifications and the well absorbances were read at 450 nm using an ELISA plate reader.

Real-Time Polymerase Chain Reaction
For the in vivo study, total RNA was prepared from spleen and hippocampal tissues using an RNeasy R Mini Kit (QIAGEN, Austin, Texas, USA) according to the manufacturer specifications. Briefly, recommended amounts of tissue (25 mg) were placed into the kit lysis buffer (containing an additional 10 µl β-mercaptoethanol for every 1 ml lysis buffer) and processed as recommended. For the in vitro study, BV2 and RAW 264.7 cells were collected at 24 h after LPS + IFN-γ stimulation and RNA was isolated using the same method as that described above. As per the manufacturer instructions, RNA was eluted with 30-50 µl of RNase-free H 2 O. Samples were immediately aliquoted and stored at −80 • C until use. RNA concentrations were quantified using a spectrophotometer (NanoDrop R ND-1000) at the A260 absorbance level. The purity of the RNA was assessed by calculating the A260/A280 ratio of each sample. cDNA was synthesized from 1 µg of total RNA for each sample using the Maxime RT PreMix kit (Oligo dT Primer, iNtRON Biotechnology, Inc., Seongnam-si, Gyeonggi-do, Korea). Diluted cDNA was amplified with SYBR R Green PCR Master Mix (Applied Biosystems, Foster City, CA, USA) in a final reaction volume of 50 µl. The PCR reaction was performed using an ABI prism 7500 sequence detector (Applied Biosystems). The PCR program included initial denaturation at 95 • C for 10 s followed by 40 cycles of 95 • C for 5 s, 60 • C for 34 s, 95 • C for 15 s, 60 • C for 1 min, and 95 • C for 15 s. The nucleotide sequences of the primers used in this study are shown in Table 1. The cycling threshold values for TNF-α, CXCL-10, IL-1β, arginase-1, IL-10, and Ym-1 were normalized to those of β-actin.

Flow Cytometry
For fluorescent-activated cell sorting (FACS) analyses, fresh spleen and hippocampal tissues were placed on a petri dish on ice and dissociated into small pieces using a clean razor blade. FACS buffer (PBS containing 0.2% bovine serum albumin and 5 mM glucose) supplemented with 0.5 mg/mL type I collagenase was added to the dissociated tissues and the samples were incubated for 10 min. The samples were then centrifuged at 250 × g for 3 min and the resultant pellets were each resuspended in 500 µl FACS buffer. Following additional centrifugation at 250 × g for 3 min, the pellets were re-collected and the samples were blocked in anti-mouse CD16/CD32 (1:50, BD Biosciences, San Jose, CA, USA) for 10 min on ice. After blocking, the samples were centrifuged at 250 × g for 3 min and the resulting pellets were incubated with one of two antibody sets (M1 set: CD11b-PE, F4/80-APC, and CD40-FITC; M2 set: CD11b-FITC, F4/80-APC, CD206-PE) diluted in FACS buffer. Thereafter, the cell suspensions were filtered through a 40-µm nylon mesh cell strainer. Fluorescent cells were immediately identified using a flow cytometer (LSR II, BD bioscience, San Jose, CA, USA) and analyzed using FlowJo software version 10 (FlowJo LLC, Ashland, OR, USA). Primary gating was based on forward scatter (FSC) and side scatter (SSC) to isolate single cell events. Cell debris, dead cells, or doublets were eliminated by excluding low FSC and

Mouse cDNA
Primer sequences high SSC. Macrophage/microglia were identified using CD11b and F4/80. The M1 and M2 polarization states were identified by CD40 and CD206 expression, respectively.

Statistical Analysis
Data are expressed as the mean ± standard error of the mean (SEM). The groups were compared using a one-way analysis of variance followed by Tukey post hoc tests and unpaired t-tests as appropriate. Linear regression was used to estimate the slopes of the fiber volley (FV)/fEPSP ratios in the electrophysiological experiments. Differences with a P < 0.05 were deemed statistically significant. All statistical analyses were performed using Prism version 5.0 software (GraphPad Software Inc., La Jolla, CA, USA).

EPO Minimizes Cognitive Dysfunction and Synaptic Dysfunction Following Surgery
With regard to the open field testing, there were no significant between-group differences in the distances the mice moved at 48 h post-surgery ( Figure 1A, P = 0.774). For evaluations of learning and memory, there were significant between-group differences in the latency times in the passive avoidance test (Figure 1B, P = 0.006), and the duration spent exploring the novel object in the novel object recognition test (Figure 1C, P < 0.001). The latency times in the passive avoidance test were significantly shorter in the Surgery group compared to the control and EPO-surgery groups. Furthermore, mice in the surgery group spent significantly less time exploring the novel object in the novel object recognition task in comparison to all other groups. These data suggest that rhEPO prevented and/or ameliorated POCD in mice.
To investigate the effects of surgery and rhEPO on synaptic function, we measured synaptic strength and LTP in hippocampal slices using fEPSPs at 48 h post-surgery. Synaptic strength was assessed by generating I/O curves and comparing the slopes of the FV/fEPSP ratios in SC-CA1 hippocampal synapses. The synaptic strength was significantly different between the groups (Figure 1D, P = 0.010). The Surgery group had significantly lower fEPSP slopes at similar presynaptic FV amplitudes when compared with the control group, indicating a decrease in synaptic strength. The surgery-induced decrease was significantly ameliorated in the EPO-surgery group. Next, we investigated LTP in SC-CA1 synapses, which plays a key role in memory and cognitive function (Bliss and Collingridge, 1993). The differences in LTP between the groups were also statistically significant ( Figure 1E, P < 0.001). LTP was significantly attenuated in the surgery group compared to control group. Moreover, LTP was almost completely restored in the EPO-surgery group, to a level comparable with that of the control group.
Finally, we measured the protein expression levels of IL-1β and TNF-α in hippocampal tissues (Figure 1F), since these proinflammatory cytokines have been reported to underlie POCD Zhang et al., 2016) and high levels of IL-1β can interfere with synaptic plasticity and LTP (Rachal Pugh et al., 2001). There were a significant between-group differences in levels of IL-1β and TNF-α in the hippocampus (IL-1β, P < 0.001; TNF-α, P < 0.001). The levels of IL-1β and TNF-α in the hippocampus were significantly higher in the surgery group compared to the control and EPO-control groups. Importantly, rhEPO administration partially attenuated the surgery-induced up-regulation of pro-inflammatory cytokines in the EPO-surgery group.

EPO Alters the Pattern of Macrophage/Microglia Phenotype Following Surgery
Flow cytometry was used to quantitatively assess M1 versus M2 phenotype of macrophages/microglia in the post-surgery mice. The cells were labeled with antibodies specific for macrophages/microglia (CD11b and F4/80); of these, CD40 + macrophages/microglia were classed as M1-phenotype and CD206 + macrophages/microglia were classed as M2-phenotype in the spleen (Figure 6) and the hippocampus (Figure 7). The EPO-control group was omitted from this analysis because EPO did not affect the phenotype of the resting statemacrophage/microglia as shown previously (Figures 2-5). The relative proportions of CD40 + cells and CD206 + cells among the CD11b + /F4/80 + cells were compared, and significant differences were found between the groups (Figure 6C: CD40 + cells in the spleen, P < 0.001; Figure 6D: CD206 + cells in the spleen, P < 0.001; Figure 7C: CD40 + cells in the hippocampus, P = 0.005; Figure 7D: CD206 + cells in the hippocampus, P < 0.001). As depicted in Figure 6C, the proportion of CD40 + cells in the spleen was significantly increased in the surgery group compared to the control group, whereas an increased proportion of CD40 + cells was not observed in the EPOsurgery group. Conversely, there were no significant differences . β-actin was used for normalization. *P < 0.05, **P < 0.01, and ***P < 0.001.
in the proportion of CD206 + cells between the surgery and control groups; however, the proportion of CD206 + cells was significantly greater in the EPO-surgery group compared to the control and surgery groups ( Figure 6D). Similar patterns of polarization were observed in the hippocampus compared to the spleen (Figures 7C,D).

Arginase Inhibition Abolishes EPO-Induced Improvements in POCD
Arginase is thought to mediate the anti-inflammatory effects of M2 macrophages (Pesce et al., 2009). Therefore, we assessed post-operative cognitive function and inflammatory cytokine levels in the hippocampus following administration of a specific arginase inhibitor with EPO in order to estimate the influence of macrophage M2 polarization on POCD. There were significant between-group differences in the results from passive avoidance and novel object recognition tests (Figure 8A: passive avoidance test, P = 0.001; Figure 8B: novel object recognition test, P < 0.001). As we previously observed in the passive avoidance test, the latency to enter the dark box of the surgery group was significantly shorter than that of the control group, whereas the latency in the EPO-surgery group was similar to that of the control group. Arginase inhibition prevented the restorative effect of EPO on the latency, such that the latency of the EPOsurgery-AI group was similar to that of the surgery group. In the novel object recognition test, the time spent exploring the novel object was close to 50% in the EPO-surgery-AI group, suggesting that these mice had trouble in distinguishing the novel object, and thus impaired cognitive function. Furthermore, the levels of the pro-inflammatory cytokines IL-1β and TNF-α in the hippocampus were significantly different between the groups (Figure 8C: IL-1β, P < 0.001; Figure 8D: TNF-α, P < 0.001). The levels were higher in the EPO-surgery-AI group than in the control and EPO-surgery groups.

DISCUSSION
The main findings of this study are as follows: (1) EPO treatment suppressed increases in pro-inflammatory cytokines in the hippocampus following surgery, and appeared to prevent or alleviate the post-operative impairments in LTP and synaptic dysfunction, which was associated with decreased POCD. (2) EPO increased the proportion of macrophages/microglia with an M2-phenotype in vivo in spleen and hippocampal tissues post-surgery and in vitro after stimulation of M1 inducers. (3) Arginase inhibition abolished EPO-associated improvements in cognitive function and neuroinflammation post-surgery.
The core pathophysiology of POCD has been reported to involve neuroinflammation that results from systemic inflammation after surgical damage (Degos et al., 2013;Vacas et al., 2014). Elevated levels of pro-inflammatory cytokines such as IL-1β in the hippocampus can impair synaptic function and lead to cognitive dysfunction (Rachal Pugh et al., 2001;Cibelli et al., 2010;Terrando et al., 2010;Zhang et al., 2016). In our study, EPO improved the behavioral indicators of POCD, and prevented the increase in pro-inflammatory cytokine expression in the hippocampus post-surgery. Although previous evidence indicates that EPO has anti-inflammatory effects in the central nervous system and on the periphery (Agnello et al., 2002;Nairz et al., 2011), the present study is the first to demonstrate that the anti-inflammatory effects of EPO may be attributable to M2-phenotype macrophage. Specifically, we found that EPO increased the expression of M2-related genes and increased the proportion of M2-like macrophages in the spleen and hippocampus post-surgery, and our in vitro experiment demonstrated that EPO could directly change the phenotype of activated macrophage. Furthermore, the effects of EPO on FIGURE 5 | In vitro evaluation of the M2 phenotype in RAW 264.7 and BV2 cells after LPS+IFN-γ treatment. The mRNA expression levels of selected M2-related genes in RAW 264.7 (A) and BV2 (C) cells. IL-10 protein expression in RAW 264.7 (B) and BV2 (D) cell supernatants. β-actin was used for normalization. *P < 0.05, **P < 0.01, and ***P < 0.001. cognitive function and neuroinflammation were abolished by the inhibition of arginase, which is thought to mediate M2 antiinflammatory effects. Macrophages can be characterized as M1 or M2 according to the metabolism of arginine (Rath et al., 2014;Yang and Ming, 2014). While M1 macrophages metabolize arginine via inducible nitric oxide synthase to produce the cytotoxic radical nitric oxide, M2 macrophages utilize arginase to metabolize arginine to ornithines and polyamines that promote cell proliferation and tissue repair. Therefore, it may be inferred that the effect of EPO on macrophage phenotype led to the suppression of pro-inflammatory cytokine expression in the hippocampus and thus prevented POCD. Additionally, our in vitro results indicate that EPO did not induce an M2 phenotype, but rather induced M2 switching in the presence of M1 activation cues. We did not observe any changes in M1 or M2-related gene expression when cells were treated with EPO in the absence of M1 inducers. Consistent with this idea, EPO did not alter the expression of any M1-or M2-related genes in the absence of a surgical insult in vivo.
EPO receptors on erythroid cells are comprised of two identical EPO receptor subunits, whereas non-erythroid cells such as immune cells express a heterodimeric receptor that consists of an EPO receptor subunit and a β common receptor subunit (Brines and Cerami, 2012). Thus, EPO receptors in erythroid tissues have a much higher affinity for EPO than those in other tissues. Indeed, the activation of non-hematopoietic EPO receptors may require a higher dose of EPO than that used to stimulate hematopoietic EPO receptors. For example, 600 U/kg EPO is sufficient to reduce the need for perioperative allogenic blood transfusion in patients (Steuber et al., 2016;Zhao et al., 2016). In contrast, Nairz et al. demonstrated that 5,000 U/kg EPO in mice influenced the actions of activated macrophages through the EPO receptor (Nairz et al., 2011). For this reason, we used a dose of 5,000 U/kg EPO in our experiment to test the utility of EPO for altering macrophage polarization in POCD. One limitation of this approach is that high doses of EPO are associated with an increased risk of thrombosis in clinical settings (Bohlius et al., 2006;Corwin et al., 2007;van der Meer and van Veldhuisen, 2011). Alternative compounds such as carbamylated EPO are currently being developed to circumvent such side effects. Carbamylated EPO is an EPO receptor agonist that lacks erythropoietic and hematopoietic activities (Brines and Cerami, 2012), which may make it safer for controlling the postoperative innate immune response in patients. Carbamylated EPO was recently tested in several disease models and showed promising results (van Rijt et al., 2014;Chen et al., 2015).
Although little is known about how macrophage polarization affects postoperative recovery, considerable insight can be derived from studies of infection and cancer (Labonte et al., 2014). M1 macrophage polarization is largely related to protection against bacterial infection and cancer based on the bactericidal and anti-tumoral effects of M1 macrophages. (D) Quantification of the ratio of M2 macrophages/microglia (CD206 + ) to total macrophages/microglia (n = 6 per group). *P < 0.05 and **P < 0.01.
Alternatively, viral infection more commonly leads to M1 polarization and the severity of inflammation is often correlated with disease severity. Therefore, M2 polarization may play a role in alleviating the severity of viral infection. Expanding on this idea, our study suggests that M2 polarization alleviates surgeryinduced aseptic inflammation to prevent POCD.
This study had some limitations. First, although the trajectory of cognitive function should be evaluated over a period that includes complete surgical recovery in order to properly assess POCD (Nadelson et al., 2014), we assessed cognitive function in mice at a single time point 2 days after abdominal surgery. This time point was selected based on the results of a previous study using a similar surgery model that reported that cognitive impairments were most pronounced during the first week postsurgery (Hovens et al., 2014). Second, we were not able to distinguish between the effects of EPO on microglia versus macrophages in our in vivo experiment; blood-brain barrier permeability can be increased with surgery (Bi et al., 2017;Yang et al., 2017), and systemically administered EPO has been reported (observed?) to cross the blood-brain barrier not only in some brain injury models (Brines et al., 2000;Sirén et al., 2001) but also in healthy mice (Ehrenreich et al., 2004). However, a previous study indicated that the infiltration of systemic macrophages into the hippocampus is essential for the development of POCD (Degos et al., 2013), suggesting that macrophages might be largely responsible for the effects of EPO in POCD. Moreover, the direct stimulation of microglia with EPO in vitro produced a different pattern in arginase expression from that observed in the hippocampus. Third, we could not exclude the possibility that EPO affects macrophage/microglia partly through the actions on the other cell types in vivo since various cell types have EPO receptors. Fourth, we did not include a group that only received anesthesia in this study, which may be a weakness of the study since anesthesia has been reported to contribute to POCD (Arora et al., 2014). Finally, narcotic analgesics are known risk factors of postoperative cognitive dysfunction (Bilotta et al., 2013), and therefore the effect of tramadol given to the mice undergoing surgery cannot be completely excluded.

CONCLUSION
Our data support the idea that EPO shifts macrophage activation from an M1 to an M2 phenotype, and this may be associated with its ability to prevent POCD.

AUTHOR CONTRIBUTIONS
JL and B-NK conceived and designed the experiments. JL, EHK, SK, SYC, and J-HJ performed the experiments. JL and B-NK analyzed the data and wrote the paper. EJK and SC provided experimental technical support and contributed to the paper revision. All authors read and approved the final manuscript.