Cerebral inflammation and mobilization of the peripheral immune system following global hypoxia-ischemia in preterm sheep

Background Hypoxic-ischemic encephalopathy (HIE) is one of the most important causes of brain injury in preterm infants. Preterm HIE is predominantly caused by global hypoxia-ischemia (HI). In contrast, focal ischemia is most common in the adult brain and known to result in cerebral inflammation and activation of the peripheral immune system. These inflammatory responses are considered to play an important role in the adverse outcomes following brain ischemia. In this study, we hypothesize that cerebral and peripheral immune activation is also involved in preterm brain injury after global HI. Methods Preterm instrumented fetal sheep were exposed to 25 minutes of umbilical cord occlusion (UCO) (n = 8) at 0.7 gestation. Sham-treated animals (n = 8) were used as a control group. Brain sections were stained for ionized calcium binding adaptor molecule 1 (IBA-1) to investigate microglial proliferation and activation. The peripheral immune system was studied by assessment of circulating white blood cell counts, cellular changes of the spleen and influx of peripheral immune cells (MPO-positive neutrophils) into the brain. Pre-oligodendrocytes (preOLs) and myelin basic protein (MBP) were detected to determine white matter injury. Electro-encephalography (EEG) was recorded to assess functional impairment by interburst interval (IBI) length analysis. Results Global HI resulted in profound activation and proliferation of microglia in the hippocampus, periventricular and subcortical white matter. In addition, non-preferential mobilization of white blood cells into the circulation was observed within 1 day after global HI and a significant influx of neutrophils into the brain was detected 7 days after the global HI insult. Furthermore, global HI resulted in marked involution of the spleen, which could not be explained by increased splenic apoptosis. In concordance with cerebral inflammation, global HI induced severe brain atrophy, region-specific preOL vulnerability, hypomyelination and persistent suppressed brain function. Conclusions Our data provided evidence that global HI in preterm ovine fetuses resulted in profound cerebral inflammation and mobilization of the peripheral innate immune system. These inflammatory responses were paralleled by marked injury and functional loss of the preterm brain. Further understanding of the interplay between preterm brain inflammation and activation of the peripheral immune system following global HI will contribute to the development of future therapeutic interventions in preterm HIE.


Background
Hypoxic-ischemic encephalopathy (HIE) is one of the most important causes of brain injury in preterm infants [1]. Preterm infants suffering from HIE develop cognitive disorders in 25 to 50% of all cases and 5 to10% suffer from severe motor deficits (such as cerebral palsy) [2]. The hippocampus plays a key role in cognition and several studies suggest that hypoxia-ischemia (HI)-induced injury to the hippocampus may predispose to cognitive disorders later in life [3][4][5][6][7]. Motor deficits in preterm HIE are mainly attributable to injury of white matter in the immature brain [1]. Despite the high prevalence of neurological sequelae, no therapeutic interventions are available to treat HIE in preterm infants. Cooling therapy, which has been shown to improve neurodevelopmental outcome in mild cases of HIE in term infants, is associated with adverse outcomes in preterm infants and has therefore not yet been established as standard clinical care for this vulnerable patient group [8][9][10].
White matter injury, the clinical hallmark of preterm HIE, is caused by injury to the highly vulnerable immature oligodendrocytes in the preterm brain. HI-induced damage to immature oligodendrocytes impedes effective differentiation into mature myelinating oligodendrocytes leading to hypomyelination of the preterm brain [11][12][13]. Microglial activation is considered to be involved in the injury to immature oligodendrocytes [13]. Microglia are the resident innate immune cells in the brain and play a central role in the initiation of an inflammatory response aimed at resolving injury caused by HI [14][15][16]. Excessive activation of microglia, however, results in a detrimental cerebral inflammatory response with neurotoxic consequences [15,16]. In addition to cerebral inflammation, experimental data from adult rodent models of focal HI (stroke) suggest a role for the peripheral immune system in the etiology of cerebral HI. More precisely, several studies showed that acute brain injury after focal ischemia is followed by a massive activation of the peripheral immune system with rapid mobilization of immune effector cells from the spleen [17,18]. These mobilized effector cells can invade the brain and aggravate the existing injury [17].
Given the importance of cerebral inflammation and peripheral immune system activation in focal HI of the adult brain, we hypothesized that similar inflammatory responses are involved in the etiology of preterm brain injury following global HI. To test this hypothesis, preterm instrumented sheep were exposed to 25 minutes of umbilical cord occlusion (UCO) at 0.7 gestation. At this time of gestation, neurodevelopment of fetal sheep is equivalent to that of a preterm human infant of 28 to 32 weeks [19][20][21]. During this neurodevelopmental stage of the human and ovine fetus, the preterm brain is highly prone to develop white matter injury following global HI [11][12][13].

Animal experiments
The study was approved by the Animal Ethics Research Committee of Maastricht University, The Netherlands. Fetuses of time-mated Texel ewes were instrumented at 101 ± 1.1 (mean ± SD) days gestation. Before surgery, ewes received i.v. prophylactic antibiotics (1000 mg amoxicillin and 200 mg clavulanic acid). Anesthesia was induced by i.v. thiopenthal (15 mg/kg). After intubation, general anesthesia was maintained with 1 to 2% isoflurane guided by depth of sedation and supplemented by remifentanyl i.v. (0.75 μg/kg/min) for analgesia. Vital parameters and depth of sedation were continuously monitored by certified personnel. A catheter was placed in the maternal long saphenous vein to provide access for a peri-operative saline drip (250 mL/hour) and postoperative blood sampling and administration of the prophylactic antibiotics during four days.
Fetuses were catheterized with 3.5 French polyurethane umbilical vessel catheters (Tyco Healthcare Group, Mansfield, Massachusetts, USA) placed in the femoral artery and the brachial vein. Three custom-made electrocardiogram (ECG) shielded electrodes (Cooner Wire Co., Chatsworth, CA, USA) with silver plates (5 mm) were sewn on the chest for fetal heart rate recordings. Two pairs of custom-made electroencephalogram (EEG) shielded electrodes (Cooner Wire Co.) with silver tips were placed bilaterally on the dura over the parasagittal parietal cortex (5 mm and 15 mm anterior to point bregma and 10 mm lateral), with a subcutaneous silver reference electrode (10 mm) placed in the neck. The EEG electrodes were secured with cyanoacrylate glue and covered with fetal skin. All animals were instrumented with an inflatable vascular occluder (OC16HD, 16 mm, In Vivo Metric, Healdsburg, California, USA) placed around the umbilical cord. A catheter for amniotic pressure recording was placed in the amniotic sac. Before closure of the uterus 80 mg of Gentamycin was administered into the amniotic sac. All fetal catheters and leads were exteriorized through a trocar hole in the flank of the ewe.
After surgery, ewes were housed in a confined space to allow handling and continuous perfusion of the catheters with heparinized saline (25 IU/mL, 0.2 mL/hr). Surgical wounds were inspected daily and treated with chlortetracycline spray to prevent infection. Animals had ad libitum access to water and food. The welfare of the animals was monitored daily by certified personnel.

Experimental design
Fetuses were instrumented at 101 ± 1 (mean ± SD) days of gestation (experimental day −4). After surgery, the ewe and her fetus were allowed to recover for four days. On experimental day 0, fetuses were randomly allocated to either be subjected to 25 minutes of umbilical cord occlusion (HI group, n = 8) or sham occlusion (sham group, n = 8). In the HI group, the occluder was rapidly inflated with sterile saline and complete occlusion was confirmed with a sudden drop in heart rate and subsequent arterial blood gas analysis indicating acidemia, hypoxia and hypercapnia ( Figure 1). Such an insult has been previously shown to result in global HI and subsequent cerebral hypoperfusion [19,22]. After (sham) umbilical cord occlusion, a reperfusion period of 7 days followed. At the end of the experiment (experimental day 7), both ewe and fetus were euthanized by administration of pentobarbital (200 mg/kg).

Data acquisition
Blood pressure, amniotic pressure, EEG and ECG data were acquired and digitized by a custom-made MPAQ unit (Maastricht-Programmable AcQuisition system, Maastricht Instruments BV, Maastricht, The Netherlands) with IDEEQ software (Maastricht Instruments BV). All data were sampled at 1000 Hz and stored on hard-disk for offline analysis. Analog filtering was applied to the ECG data, with a 1 Hz high-pass filter and a 200 Hz low-pass filter. Heart rate (beats per minute) was extracted from the ECG by R-top identification. Blood pressure and amniotic pressure data were not filtered. Fetal mean arterial blood pressure was calculated by online subtraction of the amniotic fluid pressure from the femoral artery pressure.
The EEG data were filtered using a 0.5 to 30 Hz 4 th order Butterworth band-pass filter. EEG signal with an amplitude >1000 μV was considered an artifact and removed from analysis (<1% of data). After filtering, EEG background analysis was performed using an amplitudeand time-threshold based algorithm [23]. Burst activity was defined as an epoch with an amplitude >30 μV and a duration >1 s in both channels. Interburst intervals (IBI) were defined as epochs with an amplitude <30 μV and a duration >3 s in both channels. Segments not meeting sham HI 6  fetal heart rate (HR) in beats per minute (bpm); (C) blood gas: arterial pH; (D) blood gas: arterial partial oxygen pressure (pO 2 ); (E) blood gas: arterial partial carbon dioxide pressure (pCO 2 ). Shaded areas (MABP and HR) and error bars (blood gases) depict standard deviation (SD). Min/' = minutes, d = day.
above criteria were classified as undefined. Using these criteria, mean IBI length per 30 minutes and per 24 hours segments was calculated and used as a surrogate for functional brain suppression for all animals during the period starting two days before UCO (day −2) until the end of the experiment (day 7).

Brain immunohistochemistry analysis
For the analysis of IBA-1 immunoreactivity (IR), digital images of the hippocampus, subcortical white matter (SCWM) and periventricular white matter (PVWM) were acquired at 100x magnification using an Olympus BX51 microscope (Olympus, Tokyo, Japan). In the regions of interest (ROIs) areal fraction of IBA-1 IR was determined with a standard threshold to determine positive staining using Leica Qwin Pro V 3.5.1 software (Leica, Rijswijk, The Netherlands). Within the hippocampus IBA-1 IR was additionally analyzed in the CA1-2, CA3 and dentate gyrus (DG) sub regions. IBA-1 IR areal fraction in the ROIs was assessed in six consecutive coronal sections (posterior hippocampus/mid-thalamus level) per animal (sham, n = 6; HI, n = 6) by an independent observer who was blinded to the experimental conditions.
Analysis of the O4 immunohistochemical staining clearly showed that within the periventricular white matter three sub-regions had region-specific preOL characteristics in sham animals that responded differently to global HI. Therefore O4 staining was assessed in these three different regions of interest in the PVWM. In addition, O4 staining was assessed in the SCWM. Regions of interest are indicated in Figure 2.
To assess O4 immunoreactivity, we adapted the method previously reported by Back et al. [11]. A differential count was performed, discriminating between immature (ring-shaped membrane staining, no processes), mature (ring-shaped membrane staining, extensively branched processes) and degenerative (fragmented membrane staining, fragmentation of processes, signs of cell death; nuclear condensation and apoptotic bodies) phenotype of the O4 positive cells. The sum of the differential count resulted in the total number of O4 positive cells. Differential counts were performed in six consecutive coronal sections (posterior hippocampus/mid-thalamus level) per animal (sham, n = 3; HI, n = 3). The investigator who performed the differential count was blinded to the experimental conditions. In each region of interest, O4 positive cells were counted in eight randomly chosen fields of view with a 40× objective equipped with a counting grid (0.0625 mm 2 ) using a Nikon Eclipse E400 microscope (Nikon, Amsterdam, The Netherlands).
Differential counts of MPO positive cells in the brain were performed to assess the localization of these cells in relation to the cerebral vasculature. Numbers of intravascular, perivascular and interstitial cells were counted in the hippocampus, periventricular white matter and subcortical white matter. Six coronal sections per animal (sham, n = 3; HI, n = 3) were studied at the posterior hippocampus/midthalamus level. In each section, cells were counted in eight fields of view (focused on the cerebral vasculature) per region of interest (hippocampus, periventricular white matter and subcortical white matter) with a 20x objective equipped with a counting grid (0.25 mm 2 ) using a Nikon Eclipse E400 microscope (Nikon, Amsterdam, The Netherlands).
Since in the MPO analysis the fields of view were not randomly chosen, but focused on the cerebral vasculature, the numbers of cells were expressed as cells per field of view (FOV).
All images of immunohistochemical staining in the brain (IBA-1, O4, MBP, MPO) presented here were obtained with an Olympus AX-70 microscope (Olympus, Tokyo, Japan) equipped with a digital camera.
Endogenous peroxidase was inactivated by incubation with 0.3% H 2 O 2 that was dissolved in methanol. Antigen specific binding was prevented by incubating the slides for 30 minutes with 5% bovine serum albumin (BSA). Slides were incubated overnight at 4°C with the diluted primary antibody (cleaved caspase-3 1:200, CD3 1:200, MPO 1:500) followed by incubation with the appropriate secondary biotin labeled antibody. Immunostaining was enhanced with Vectastain ABC peroxidase Elite kit (PK-6200, Vector Laboratories) followed by a NiDAB staining. Sections were counterstained with 0.1% Nuclear Fast Red washed, dehydrated and coverslipped. The number of caspase-3 positive cells in the spleen were counted in twenty (to accommodate heterogenic distribution) fields of view per animal (sham, n = 6; HI, n = 6) with a 20× objective equipped with a counting grid (0.25 mm 2 ) using a Nikon Eclipse E400 microscope. The number of caspase-3 positive cells was expressed in cells/mm 2 .
For the analysis of CD3 and MPO immunoreactivity (IR), digital images of spleen sections were acquired at 100× magnification using a Leica DM200 microscope equipped with a Leica DFC295 digital camera (Leica Microsystems) and Leica Application Suite (LAS) software (Leica LAS V 3.7, Leica Microsystems). Areal fraction of CD3 and MPO IR was determined in five sections per animal (sham, n = 6; HI, n = 6) with a standard threshold to determine positive staining using Leica Qwin software (Leica Qwin Pro V 3.5.1, Leica).

Flow cytometry
At the end of the experiment (day 7), the spleen was immediately harvested after sacrifice. Single-cell splenocyte suspensions were obtained by dissociating freshly sampled spleen tissues in gentleMACS ™ C-tubes (MiltEnyi, Leiden, The Netherlands) filled with Gibco W Iscove's Modified Dulbecco's Medium (IMDM) (Life Technologies, Bleiswijk, The Netherlands) using the gentleMACS ™ Dissociator (MiltEnyi). Subsequently, the cell suspensions were passed through a 70 μm cell strainer (BD Biosciences, Erembodegem-Aalst, Belgium). Splenocytes were stored in nitrogen in freezing medium containing IMDM medium with 10% heat-inactivated fetal calf serum and 10% dimethylsulfoxide (DMSO).
To study the cellular composition of the spleen 7 days after global HI, 200,000 splenocytes per animal (sham, n = 8; HI, n = 8) were stained for detection of lymphocytes (mouse anti sheep CD45-biotin; AbDSerotec, Düsseldorf,

Statistics
Summary statistics of animal characteristics (gestational age at UCO, body weight) are shown as means with 95% confidence intervals (CI). For analysis of O4, MPO and activated caspase-3 parameters, cell counts in each section's region of interest were first averaged per field of view (n = 8 for O4 and MPO; n = 20 for activated caspase-3). Groups' comparisons (sham vs. HI) with respect to all outcome parameters were drawn either with independent t-tests, or with random intercept models in case of repeated measurements per animal (e.g. different sections per brain). Variables, whose distributions were positively skewed, were log-transformed previous to statistical testing. To facilitate interpretation, averages on the log scale were back transformed to the original scale (antilog) and are presented as geometric means and corresponding 95% CIs. Average (additive) differences on log transformed data become 'multiplicative' on the original scale. Thus, the displayed geometric means for the sham and HI groups should be compared in relative terms, not as difference in averages (mean sham minus mean HI), but rather as a ratio of the sham geometric mean with respect to the HI geometric mean (mean sham divided by mean HI). The interpretation of the geometric means ratio is provided for example for areal fraction (%) IBA-1 immunoreactivity in the subcortical white matter (see Results section).
For analysis of the EEG parameter (IBI length), log transformation also preceded parametric inferences regarding groups' comparisons and temporal dynamics of mean IBI length before and after UCO (or sham). To accommodate both the interrupted nature of the experimental follow-up, with UCO happening on day 0 of the experiment, as well as the correlation among longitudinal measurements of individual fetuses, a piecewise mixed regression model was fitted [27]. This mixed model approach allowed additionally for heterogeneity of groups' variances to be accounted for. In the model, time (pre and post UCO, measured in days), group (sham vs. HI), and a dummy for pre and post UCO times (pre-post), were the fixed effects factors. Fetuses (subjects) were the random factor. The addition of random effects was meant to model individual variability relative to the group's average. Variables selection was carried out via the top-down procedure based on likelihood ratio (LR) tests for fixed effects and tests for the covariance structure.

Animal characteristics
Fetal body weight did not differ between the sham and HI group; sham mean: 1,782 g (1571; 1993) versus HI mean: 1,742 g (1482; 2002), P = 0.677. There was no significant difference in gestational age at the time of UCO between the sham and HI groups; sham mean: 105.6 days (104.6; 106.5) versus HI mean: 105.5 days (104.7; 106.3), P = 0.717.

Fetal vital parameters
Fetal vital parameters and blood gases during UCO are depicted in Figure 1. After an initial compensatory rise, mean arterial blood pressure gradually declined from 40 mmHg to 10 mmHg at the end of 25 minutes UCO ( Figure 1A). Mean fetal heart rate rapidly fell after initiation of UCO from around 200 beats per minute (bpm) at baseline to below 100 bpm at the end of UCO ( Figure 1B). All vital parameters normalized within 30 minutes of reperfusion time.
Blood gas data ( Figure 1C-E) indicate that average pH dropped from 7.4 at baseline to 6.8 at the end of UCO. Mean partial oxygen pressure decreased from 20 mmHg at baseline to values below 5 mmHg at the end of UCO. Mean partial carbon dioxide pressure increased from baseline levels of 40 mmHg to values around 90 mmHg at the end of UCO. Upon reperfusion hypoxemia and hypercapnia resolved within minutes. Normalization of pH values occurred after 60 to 90 minutes (data not shown).

Microglial activation and proliferation
Microglia (IBA-1) were studied to determine the local inflammatory response in the brain. Areal fraction (%) of IBA-1 immunoreactivity (IR) was studied in the SCWM, PVWM and hippocampus (Figure 2A). IBA-1 IR was significantly increased in the SCWM of animals exposed to HI compared with sham ( Figure 4D-F); sham geometric mean: 25.6% (95% CI 17.0; 38.4) versus HI geometric mean: 70.0% (95% CI 46.6; 105.1), P = 0.003. Thus, the ratio of the two geometric means, HI with respect to sham is 70.0/25.6 = 2.73. The corresponding interpretation is that the geometric mean of the areal fraction in the HI group is 2.73 higher (173% increase) than the geometric mean of the sham group.
Microglia in sham animals exhibited a quiescent state characterized by extensively branched thin processes (inserts Figure 4E and H, and Figure 5B, E and H). In contrast, in HI-exposed animals, microglia with thick cell bodies and retracted processes were observed indicating an activated state (inserts Figure 4F and I, and Figure 5C, F and I).
Region-specific pre-oligodendrocyte vulnerability Differential counts of O4-positive preOLs were performed in the medial, intermediate and lateral PVWM and in the SCWM ( Figure 2B). Analysis of the O4 staining showed that in sham animals the medial and lateral PVWM were predominantly populated by O4-positive cells with a mature phenotype (Figure 6A-B and G-H). In contrast, O4-positive cells in the SCWM and intermediate PVWM were predominantly of immature phenotype ( Figure 6D-E and J-K). Following global HI all regions showed an increase in preOLs with degenerative morphology (Figure 6A, D, G and J) which reached statistical significance in the regions which were populated by mature preOLs in sham conditions (medial and lateral PVWM). In the medial PVWM, the areal density of O4-positive cells with degenerative phenotype significantly increased in animals exposed to HI compared to sham ( Figure 6A

MBP
Brain sections were stained for myelin basic protein (MBP) to detect white matter injury following HI. In sham animals abundant MBP-positive myelin sheaths and myelin-producing cells (mature oligodendrocytes) were observed in the subcortical white matter (Figure 7). Following global HI a marked reduction of both myelin sheaths and mature oligodendrocytes was observed (Figure 7). In the PVWM (regions 1, 2 and 3 in Figure 2) no MBP immunoreactivity was detected in both sham and HI animals (data not shown) indicating that these white matter regions were not myelinated at this developmental stage of the preterm brain.

Cerebral neutrophil invasion
Brain sections were stained for MPO to detect neutrophils that invaded the brain following HI. Although microglia can also produce MPO, neutrophils and microglia can be easily distinguished by intensity of staining, localization and morphology [29][30][31]. Following global HI the total number of MPO-positive cells inside or adjacent to the cerebral vasculature significantly increased in the hippocampus ( Figure 8A MPO-positive cells in the brain sections studied were large round-shaped cells with lobular intracellular structures situated in and around the cerebral vasculature ( Figure 8B-C, E-F and H-I). These morphological features and their localization indicate that these MPOpositive cells were neutrophils. MPO-positive cells lacking neutrophil morphology were rarely detected (data not shown).
Flow cytometry analysis of whole blood showed that the percentage of living (7-AAD negative) CD45-positive lymphocytes expressing CD11b (neutrophils), CD4 (helper T-cells) or CD8 (cytotoxic T-cells), did not differ between sham and HI groups in the circulation on day 1 (data not shown). This indicated that global HI induced nonpreferential mobilization of immune cells 24 hours following global HI.

Splenic involution
Seven days after global HI, splenic weight was analyzed as an indication of activation of the peripheral immune system. Spleen weight, corrected for fetal BW, was significantly decreased in fetuses exposed to HI ( Figure 10A); sham geometric mean 2.5 g/kg BW (95% CI 1.7; 3.6) versus HI geometric mean 1.7 g/kg BW (95% CI 1.5; 1.9), P = 0.033. Consistently, spleen size was markedly reduced following HI ( Figure 10B To assess whether splenic involution was caused by increased apoptotic cell death, spleen sections were stained for activated caspase-3. The number (cells/mm 2 ) of activated caspase-3 positive cells in the spleen did not differ between sham and HI groups ( Figure 11A
Similarly, no differences in splenic cell populations were found between sham and HI animals using flow

EEG suppression
The fetal EEG was continuously recorded in HI-exposed (n = 8) and sham (n = 8) fetuses during the complete study period. EEG analysis was performed from two days before UCO until the end of the experiment after a reperfusion period of seven days. Interburst interval (IBI) length was assessed to determine suppression of brain function following HI. EEG suppression was indicated by prolonged IBI length. Figure 13A displays the observed IBI length values over time (background; grey) of two animals, one sham fetus and one HI fetus, averaged over 30 minutes. Time point '0' indicates day of UCO. Note the distinct difference in IBI length between the HI versus sham fetuses after occlusion, with more prolonged length and larger fluctuations for the former. Superimposed on the 30 minutes data in Figure 13A are also the IBI length values averaged over 24 hours, for the same two animals (foreground; full circles). For model simplicity, the piecewise regression model was fitted on the 24 hours data. Figure 13B displays the IBI length temporal dynamics as estimated by the final regression model. Its fixed and random effect parameters are displayed in Tables 1 and 2. The predicted lines (according to the model) are superimposed on observed 24 hour values (all animals). Note the greater IBI length variability in the HI fetuses induced by the occlusion. This more variable responsiveness was captured in the model by the significant random slope variance for in the HI group (Tables 1 and 2).
There is a clear upwards shift (increase) in the average IBI length values (log scale) for the HI group compared to sham after UCO (significant interaction between prepost dummy and group variables, capturing the average change in IBI length level for the HI group immediately after UCO). Mean IBI length after UCO remained higher throughout the measured time span (one week) for the HI group. With respect to temporal changes, it is noteworthy to mention that before UCO the averaged values remain stable over time (pre-UCO time variable did not reach statistical significance), contrary to post-UCO time. After UCO, IBI length seemed to change in a curve-linear pattern for both groups (significant quadratic and cubic post-time parameters in the model). In summary, IBI length was significantly increased following UCO and remained higher in the HI group during the seven day reperfusion time, indicating prolonged suppression of preterm brain function following global HI.

Discussion
In this study we showed that global HI caused profound inflammation of the preterm ovine brain which was paralleled by mobilization of the peripheral innate immune system. These inflammatory changes were associated with suppressed brain function, brain atrophy, regionspecific vulnerability of preOLs and hypomyelination, which are known to correlate with white matter disease, the clinical hallmark of preterm HIE [13]. Microglial proliferation following global HI was demonstrated by immunohistochemical staining of IBA-1, a specific marker for microglia under normal and neuroinflammatory conditions [24][25][26]32,33]. Moreover, we showed morphological transformation of IBA-1 positive microglia from ramified into amoeboid state indicating activation of these cells following global HI in preterm sheep [34,35]. These microglial changes, as seen after global ischemia in preterm lambs, are in line with the microglial response after focal ischemia of the adult brain which typically occurs within 24 hours after the insult [36][37][38].
Mobilization of the peripheral innate immune system in our model was demonstrated by a non-preferential recruitment of immune cells into the circulation within 24 hours following global HI as well as a marked influx of neutrophils into the HI-exposed preterm hippocampus and white matter seven days after the global HI insult. This invasion of neutrophils, which is an important histopathological finding in cerebral ischemia, was primarily localized in the perivascular zone [39][40][41]. The influx of neutrophils, which typically occurs as a second hit within 48 to 72 hours after cerebral ischemia, is considered to further aggravate acute inflammation of the brain that was initiated by immediate cell death and microglial activation by enhanced free radical attack [36][37][38].
A permeable blood brain barrier (BBB) is a prerequisite for the influx of neutrophils such as seen following global HI [41,42]. Clinical [43] and experimental [44][45][46] evidence showed disruption of the BBB in perinatal HI. The influx of neutrophils also indicates that the immature immune system is capable of responding to an inflammatory stimulus induced by global HI. This concept is in line with recent literature challenging the dogma that the preterm immune system is naive [47][48][49]. However, the fact that neutrophils were predominantly observed in the perivascular zone following HI, and to a lesser extent in the interstitium, indicated that the capacity of fetal neutrophils to transmigrate was immature [48].  We postulate that the cerebral neutrophils, as seen in our model, were derived from the spleen, since this organ is considered to be the predominant source of invading neutrophils following focal ischemia of the brain [18,50,51]. This concept is further supported by work of Ajmo et al. who showed in a rat model of focal cerebral ischemia that splenectomy reduced neutrophil influx and microglial activation, ultimately diminishing ischemic brain injury [17]. In line with this hypothesis, we observed that neutrophil invasion into the preterm brain following global HI was associated with marked splenic involution. Since splenic apoptosis was not affected by global HI, we suggest that involution of the spleen was caused by mobilization of neutrophils and other immune cells rather than HI-induced splenic cell death. However, 7 days after global HI we did not detect changed immune cell composition in the spleen. The latter is in line with a neonatal mouse study of Winerdal et al. which showed an altered splenic cellular response between 24 and 72 hours after HI, subsequently neutrophil invasion into the brain peaked 7 days after ischemia [52]. Together these findings suggest that splenic cellular changes in our model occurred within hours after the global HI-insult. This shows an important limitation of the current study in which temporal dynamics of the splenic cellular immune response were not studied. Notably, HI-induced involution of the spleen may have clinical postnatal consequences since splenic involution has been associated with an increased risk of postnatal infectious complications such as early onset sepsis [21,53]. Splenectomy in our model of global HI is required to confirm the role of the spleen as a source of neutrophils.
The cerebral and systemic inflammatory changes observed in our study were accompanied by prolonged suppression of preterm brain function. The persistent reduction in brain activity as observed in our study is in line with previous studies in comparable ovine models of global HI [22,54]. Clinical evidence showed that persistent suppression of EEG activity is associated with poor outcome, indicating the severity of the HI-insult applied in this study [55,56]. Furthermore, global HI induced severe brain atrophy in our study which has previously been associated with neuronal injury [57,58].
Our findings showed that HI-induced cerebral inflammation was paralleled by region-specific preOL vulnerability. In the SCWM, loss of morphologically immature O4-positive preOLs was associated with hypomyelination,   [22,54]. In the PVWM, we observed loss of morphologically immature preOLs as well as increased numbers of morphologically mature preOLs following global HI. These findings suggested that the preOL response following global HI varied from degeneration to proliferation and depended on the region and the morphological maturity of preOLs. This concept is supported by recent studies suggesting that myelination failure of the preterm brain following HI may be caused by a combination of preOL degeneration, regeneration and arrested maturation [59][60][61]. The observed white matter injury and functional impairments following global HI are typical findings in preterm infants with HIE underlining the translational character of the preterm sheep model.

Conclusions
This study provides evidence that cerebral inflammation and mobilization of the peripheral innate immune system are paralleled by injury and functional loss of the preterm brain following global HI. We postulate that the spleen plays a key role in preterm HIE by providing immune effector cells to the circulation and subsequently to the injured brain. Further studies should focus on the interplay between preterm brain inflammation and the activation of the peripheral immune system following global HI. Better understanding of the involvement of cerebral and systemic inflammation in the course of global HI will contribute to the development of future therapeutic interventions in preterm HIE.

Competing interests
The authors of this manuscript declare that there are no actual or potential conflicts of interest. The authors affirm that there are no financial, personal or other relationships with other people or organizations that have inappropriately influenced or biased their research.
Authors' contribution RJ designed and performed the animal experiments, analyzed the data and wrote the manuscript under the supervision of TW and BK. VLP performed and supervised the statistical analysis. AZ, WJ and PA performed and interpreted the EEG analysis. DO, JJC and MS performed animal experiments. SM and EK performed immunohistochemistry. JPC designed custom-made software for the analysis of immunohistochemistry. JV, WG, HV, HS, AG, TW and BK significantly contributed to conception, design, data interpretation and editing of the manuscript. All authors read and approved the final manuscript.