Evaluating the effect of post-traumatic hypoxia on the development of axonal injury following traumatic brain injury in sheep

Damage to the axonal white matter tracts within the brain is a key cause of neurological impairment and long-term disability following traumatic brain injury (TBI). Understanding how axonal injury develops following TBI requires gyrencephalic models that undergo shear strain and tissue deformation similar to the clinical situation and investigation of the effects of post-injury insults like hypoxia. The aim of this study was to determine the effect of post-traumatic hypoxia on axonal injury and inflammation in a sheep model of TBI. Fourteen male Merino sheep were allocated to receive a single TBI via a modified humane captive bolt animal stunner, or sham surgery, followed by either a 15 min period of hypoxia or maintenance of normoxia. Head kinematics were measured in injured animals. Brains were assessed for axonal damage, microglia and astrocyte accumulation and inflammatory cytokine expression at 4 hrs following injury. Early axonal injury was characterised by calpain activation, with significantly increased SNTF immunoreactivity, a proteolytic fragment of alpha-II spectrin, but not with impaired axonal transport, as measured by amyloid precursor protein (APP) immunoreactivity. Early axonal injury was associated with an increase in GFAP levels within the CSF, but not with increases in IBA1 or GFAP + ve cells, nor in levels of TNF α , IL1 β or IL6 within the cerebrospinal fluid or white matter. No additive effect of post-injury hypoxia was noted on axonal injury or inflammation. This study provides further support that axonal injury post-TBI is driven by different pathophysiological mechanisms, and detection requires specific markers targeting multiple injury mechanisms. Treatment may also need to be tailored for injury severity and timing post-injury to target the correct injury pathway.


Introduction
Traumatic brain injury (TBI) is the result of a mechanical insult to the head. It contributes to worldwide death and disability more than any other traumatic insult, with over 69 million cases and 10 million deaths occurring per year worldwide (Dewan et al., 2018;Kureshi et al., 2021;Hyder et al., 2007). One of the most significant consequences of a TBI is the development of traumatic axonal injury, also known as diffuse axonal injury (DAI) when widespread. Estimates suggest that DAI occurs in more than 80% of all motor vehicle accident induced TBI cases, and is consistently associated with worse outcome post-injury (Maas et al., 2008;Finfer and Cohen, 2001). Axonal injury in the context of TBI is induced by head rotational accelerations which cause brain tissue to undergo shear and compression that can cause deformation at high strain rates (Margulies and Thibault, 1989). Axonal injury is demonstrated across all severities of TBI (Johnson et al., 2013;Browne et al., 2011;Donovan et al., 2014;Mohamed et al., 2020). Importantly, the majority of axonal damage in TBI is not due to physical tearing of axons at the point of primary injury (Johnson et al., 2013), but is caused by the initiation of secondary injury processes which promote progressive axonal injury and eventual axonal swelling and detachment (Buki and Povlishock, 2006). These secondary injury processes include increased membrane permeability with subsequent calcium influx and activation of calpains and caspases leading to cytoskeletal degradation, alongside mitochondrial dysfunction and oxidative stress (Johnson et al., 2013;Smith et al., 2003).
Although axonal injury is a predictor of poor outcomes following TBI (Sidaros et al., 2008), the pathophysiology of DAI is still poorly understood. This is further hampered by the fact that most pre-clinical research has been limited to the use of lissencephalic species such as rodents (DiLeonardi et al., 2009;Marmarou et al., 1994;Perez-Polo et al., 2015;Laurer and McIntosh, 1999;Lighthall, 1988;Lipshitz and Chen, 2009), the brains of which respond differently to shear strain compared to gyrencephalic brains. Following impact in the gyrencephalic brain, shear strain is maximal at the sulcal depths (Fagan et al., 2020;Hoffe et al., 2021;Mazurkiewicz et al., 2021). The presence of sulci focuses mechanical stress away from the cortex to sulcal depths; however, in lissencephalic brains this mechanical stress is distributed uniformly across the superficial cortical layers due to the absence of sulci Cloots et al., 2008). Gyrencephalic brains also have a higher proportion of subcortical white matter to grey matter, facilitating the modelling of axonal injury (Vink, 2018). Furthermore, larger brains undergo greater tissue deformation and shear strains, with the same acceleration applied to a smaller brain, like the rodent, producing lower strains and less injury (Margulies et al., 1990;Cullen et al., 2016). Various gyrencephalic animal models including those using non-human primates, pigs and sheep have been used to investigate the development of DAI following TBI (Browne et al., 2011;Cullen et al., 2016;Gennarelli et al., 1982;Lewis et al., 1996;Van Den Heuvel et al., 2000;Van den Heuvel et al., 1999;Lafrenaye et al., 2015;Anderson et al., 2003), more closely replicating the pattern and distribution of axonal pathologies seen clinically (Vink, 2018).
Given that the pathology of DAI develops over a delayed time course, it may be exacerbated by subsequent physiological insults, such as hypoxia. Clinically, acute post-injury hypoxia is associated with poor clinical outcomes (Chesnut et al., 1993;Chi et al., 2006;Davis et al., 2009;Manley et al., 2001). In pre-clinical models, immediate posttraumatic hypoxia exacerbates the development of DAI, heightens neuroinflammation and worsens motor and cognitive outcome compared to TBI alone (Hellewell et al., 2010;Yan et al., 2011;Parikh et al., 2016). While hypoxia has been modelled independently of TBI in a gyrencephalic ferret model, where hypoxia caused white matter injury (Wood et al., 2018;Wood et al., 2019), this is yet to be modelled in conjunction with TBI.
Alongside the need to understand how secondary factors may influence axonal injury development, predictors of injury severity are also required. The acute arterial blood pressure (ABP) following TBI has been suggested as a surrogate marker of injury severity, with patients presenting with higher ABP experiencing greater rates of in-hospital mortality (Zafar et al., 2011;Ley et al., 2011) and worse long-term outcomes as measured by the Glasgow Outcome Scale at 6 months post-injury (Butcher et al., 2007). Transient hypertension followed by sustained hypotension is typical of experimental TBI and has been validated in rodent, porcine and ovine models (Marmarou et al., 1994;Lewis et al., 1996;Butcher et al., 2007;McIntosh et al., 1989) and may assist as an acute biomarker of injury severity.
Here we sought to first investigate the angular and linear acceleration produced in an ovine model of TBI. We then determined how this modified ovine injury model affected blood pressure acutely and the development of axonal injury and inflammation at 4 hrs post-injury. Lastly, we determined whether the acute blood pressure response, axonal injury and inflammation in an ovine model of TBI were affected by post-injury hypoxia.

Experimental design
The study was performed within the guidelines established by the National Health and Medical Research Committee of Australia and was approved by the Animal Ethics Committee of the South Australian Health and Medical Research Institute (SAM325). Skeletally mature Merino wethers (N = 14, 55-65 kg, 18-24 months) were randomly allocated to receive either sham surgery or TBI, and then further randomly allocated to either hypoxic or normoxic conditions for 15 mins following surgery. The group allocations were as follows: sham:normoxia (n = 3), sham:hypoxia (n = 3), TBI:normoxia (n = 4) and TBI: hypoxia (n = 4) Animals were group housed in conventional sheep paddocks for at least two weeks acclimatization prior to surgery. Animals were moved to individual indoor pens 48 h prior to surgery and fasted overnight for at least 12 h prior to surgery.

Animal preparation
An external jugular venous catheter was placed for anaesthesia and crystalloid fluid administration (sodium lactate, Baxter Health, Australia). Anaesthesia was then induced with ketamine (100 mg/kg I. V.) and diazepam (5 mg/ml I.V.). Animals were endotracheally intubated and anaesthesia was maintained with 1.5-2.0% isoflurane in a normoxic inhalational mix (30 O 2 /70 N 2 , 4 L/min) and ketamine (4 mg/ kg/h I.V). A pre-injury cerebrospinal fluid (CSF) sample was then acquired via lumbosacral tap. A femoral artery catheter was placed and was fitted with a three way tap to allow insertion of a Codman microsensor pressure probe (Codman & Shurtleff Inc., MA) and for arterial blood gas samples to be taken. Arterial blood pressure was monitored continuously with recording in LabChart Reader (ADInstruments, v7.2). Arterial blood samples were acquired at 5 min intervals during the 15 min hypoxic period until 5 mins after the hypoxic interval in the hypoxia animals, and then hourly after hypoxia in all animals, for monitoring of arterial blood gasses to maintain physiological pO 2 , pCO 2 and pH, with adjustment of ventilation parameters as required.

Head acceleration and velocity
To measure the linear and angular head accelerations and velocities during the injury event, a custom accelerometer array consisting of four triaxial accelerometers (Endevco Model 35B-2, PCB Piezotronics of North Carolina, Inc.) arranged in a 3-2-2-2 configuration with 40 mm spacing, was fixed to the animals' head via a flat 80 × 20 mm polymethylmethacrylate (PMMA) mantle (Hamilton, Australia). The PMMA was mixed to a putty-like consistency and moulded around bilateral stainless steel screws placed 20 mm about the midline cranial suture, approximately 25 mm anterior to the bregma. Four cancellous bone screws secured a stainless steel plate to the PMMA mantle, on which the accelerometer array was rigidly mounted.
The triaxial output from each accelerometer was acquired at 50 kHz via custom LabView code, using four cDAQ-9232 modules in a four-slot cDAQ-9174 chassis (National Instruments, Austin, Texas, USA). The location of the accelerometer array, relative to the animals' head, was defined by digitising notches on the accelerometer array, and anatomical landmarks on the head, using a coordinate measuring device (MicroScribe 3DX, Revware, USA). The anatomical landmarks were: the notches of the zygomatic processes of the malar bones (left and right), which are well-defined and easily palpated; and, the bregma, which was accessed via the skin incision created to mount the accelerometer array (Anderson et al., 2003;Anderson, 2000). Computed tomography (CT) images were not available for these animals, so the landmarks were used in combination with a previously determined relationship to define an anatomical coordinate system (ACS) of the sheep head, for which the origin coincided with the head center of mass and the horizontal plane was aligned with "forward gaze" head posture. (Sharkey et al., 2022). Angular and linear accelerations and velocities of the head were derived using custom MATLAB code (v.9.8, R2020a, Mathworks, UK). Briefly, the unprocessed acceleration data was filtered (2 way, 4-pole, Butterworth with cut-off frequency 1650 Hz) (Society for Automotive Engineers, 2003), the nine linear acceleration components were used to calculate linear and angular accelerations in the array coordinate system using the method described by Padgaonkar et al. (1975) then a transformation matrix was applied to the processed data to transform both linear and angular head accelerations to the ACS head , expressed as both components and resultant magnitude. Angular and linear velocities were calculated by numerical integration of the transformed acceleration data.

Head displacement
Two high speed cameras (iSpeed TR, Olympus, UK) were stereocalibrated, with the field of view approximating the maximum deviation of the head during the injury event. Three quadrant markers were placed on the accelerometer array to track head motion at 1 kHz during the injury event, and on the barrel of the captive bolt stunner (see below) to determine the vector of the bolt trajectory relative to the animal's head immediately prior to injury. The array quadrant marker coordinate system was related to the ACS head via the coordinate measuring device, in a manner similar to that described above. Quadrant marker locations were tracked in both high speed camera views (iSpeed Suite Control Pro, iX Cameras, UK) and 3D coordinates were determined using MATLAB's Stereo Camera Calibrator application. Coordinate data was then filtered (2 way, 4-pole, Butterworth with cutoff frequency 100 Hz). Head rotation and translation about/along each axis of ACS head , relative to the initial head position prior to the injury event, were calculated by solving for Euler angles.

Induction of injury
A modified humane captive bolt animal stunner (KL Model, Karl Schermer, Germany) mounted in a rigid frame secured with an 80 kg counterweight was used to induce injury, modified from a previous study (Anderson et al., 2003). This device used a blank gun powder charged cartridge (#21 charge) to propel a captive bolt (385 g) fitted with a custom-made 4 cm concave silicone tip (RP Prototype Limited, China) to induce a single impact at the midpoint between the right supraorbital process and the right external auditory meatus. The animal's head was suspended in a cervical sling (Model SUSP297, Astir, Australia), secured around the mandible. Following impact, animals were maintained under anaesthesia for four hours to allow development of early axonal injury. At 30 mins and four hours post-injury, CSF was obtained via lumbosacral tap, with failure to retrieve a 30 min sample in four animals, and a four hour sample in two animals, due to technical difficulties.

Hypoxia
Immediately following injury, animals were exposed to either a hypoxic (10% O 2 /90% N 2 ) or a normoxic (30% O 2 /70% N 2 ) gas mixture for 15 min. Hypoxic animals were maintained at peripheral capillary O 2 saturation (SpO 2 ) levels between 65% and 70%, as monitored via pulse oximetry, with the hypoxic gas mixture continuously titrated in order to maintain SpO 2 levels within the target range.

Termination and fixation
At 4 hrs after injury or sham surgery, intravenous heparin (5000 I.U, 5 ml; Pfizer, NY, United States) was injected and animals were then humanely euthanized via exsanguination and common carotid perfusion with cold TRIS-buffered saline under isoflurane anaesthesia. The animals were decapitated and inspected for skull fracture. Brains were then extracted and examined for the presence of contusions or extracranial bleeding. Brains were cut into 5 mm coronal slices using a custom-made sheep brain matrix. Alternating sections were either postfixed in 10% formalin or dissected for regions of interest (corpus callosum, internal capsule) which were flash frozen in liquid nitrogen and stored at − 80 • C.

Arterial blood pressure analysis
To investigate the ABP response to injury, a pre-injury baseline was calculated using three consecutive time points at four second intervals immediately prior to injury. The greatest deviation from this value within the first 10 min post-injury was used to calculate the maximum change in arterial blood pressure. A return to baseline was recorded once there were five consecutive identical values at one minute intervals. If this had not occurred within the first hour following injury, animals were assigned the maximum time of 3600 s.

Immunohistochemistry
Brain tissue sections were processed and embedded with tissue sections with 5 µm sections cut at anteroposterior level +3, +4 and +5 cm. For staining, sections were dewaxed then immersed in methanol/ hydrogen peroxide (0.5%) to block endogenous peroxidase activity for 30 min. After an antigen retrieval (Table 1), sections were blocked in normal horse serum (NHS; Vector laboratories; Cat No. S-2000-20) for 30 min. Primary antibodies were then applied and incubated overnight at room temperature (Table 1). Negative control slides received NHS only. A secondary antibody was applied for 30 min, slides washed with PBS and the tertiary antibody (horseradish peroxidase streptavidin, 1:1000; Vector laboratories; Cat No. SA-5004) applied for 1 h. 3,3′diaminobenzidine (DAB; Vector Laboratories; Cat No. SK-4100) was applied for antigen visualisation, before counterstaining in haematoxylin. Slides were coverslipped and scanned (Hamamatsu NanoZoomer 2.0RS, Hamamatsu, Japan).

Histological analysis
Scanned slides had regions of interest (ROI) outlined as depicted ( Fig. 1), with the ROIs incorporating major white matter tracts and specific grey matter regions of the brain where axonal injury is usually particularly abundant following TBI.
For APP analysis a blinded investigator manually counted APPpositive axons at 40x magnification throughout the entire ROI, with results reported as APP+ve profiles per mm 2 . Given the more extensive SNTF immunoreactivity, analysis was performed through QuPath v0.2.3. The ROIs were outlined, and sampling boxes (2.1 mm 2 ) were placed in each region of interest. A blinded investigator then counted the For IBA1 and GFAP ROIs were similarly outlined and sampling boxes (2.1 mm 2 ) placed in each region of interest 40x images from these sampling boxes were exported and analysed using Fiji Image J (v1.5.2). Color deconvolution was applied to isolate DAB+ve cells, and images were then converted to binary. Automatic thresholding (0-131) filtered out any background processes and the "analyse particle" function (175infinity) was used to count the DAB+ve cell bodies. Any counts that were more than two standard deviations from the mean were manually reviewed and counted to ensure consistency and accuracy. For semiquantitation of the area occupied by microglia and astrocytes, the mean value of these binary images was calculated, with the average taken across the images for each region of interest and expressed as percent area.

Inflammatory cell and cytokine analysis
The CSF and two brain regions, the internal capsule and corpus callosum, were further analyzed for expression of pro-inflammatory cytokines. Increased levels of pro-inflammatory cytokines have previously been reported within the CSF following injury (Clay Goodman et al., 1990;Morganti-Kossman et al., 1997;Shiozaki et al., 2005). The internal capsule and corpus callosum were chosen as regions of interest as they have been shown to be key white matter structures showing microstructural damage following TBI (O'Phelan et al., 2018). Furthermore, in models of diffuse injury microglia specifically converge on injured axons (Lafrenaye et al., 2015), with these two regions showing an increase in SNTF+ve profiles. As such, myBioSource ELISA kits were used for examination of TNF-α (#MBS778330), IL-1β (#MBS778369), and IL-6 (#MBS778334) within the CSF, corpus callosum and internal capsule, whilst GFAP levels were examined in the CSF only with a ProMega ELISA kit (G560A).
For protein extraction within the corpus callosum and internal capsule, radioimmunoprecipitation assay buffer, phosphatase inhibitor (Sigma Aldrich; Cat No. 04906845001) and a Roche mini EDTA-free protease inhibitor tablet (Sigma Aldrich; Cat No. 04693159001) were added to the tissue which was then homogenized with a pellet pestle. Samples were sonicated and centrifuged at 14,000 rpm at 4 ℃ for 15 min. Protein concentrations were estimated using a Pierce BCA protein assay kit (Thermo-scientific; Cat No. 23227) as per the manufacturer's instructions. A pre-injury and post-injury CSF sample were also examined for each animal, with one sham:hypoxia animal excluded due to failure to obtain a post-injury sample.
For the ELISA 100 µl of each sample was loaded in duplicate and the assay completed as per the manufacturer's instructions. Plates were read using a Synergy HTX Multi-Mode Microplate Reader (Biotek Instruments). Absorbance readings (450 nm) were calibrated as standard curves and cytokine concentrations were interpolated. These concentrations were standardised against the total protein in each sample and were expressed as pg/ml.

Statistical analysis
Histological and physiological data was analysed in SPSS (IBM; v27). Bilateral regions of interest were averaged across all three anteroposterior levels. A two-way multivariate analysis of variance (MAN-OVA) on all regions examined was performed, with injury and hypoxia as between group factors to analyse the number of injured axons or inflammatory cells per mm 2 and brain tissue cytokine levels. A three-way repeated ANOVA was used to analyse CSF samples, evaluating the effects of hypoxia, injury, and time. A two-way ANOVA evaluated blood pressure, with hypoxia and injury as the factors. Pairwise comparisons were conducted using Bonferroni's adjustment for multiple comparisons. GraphPad prism (v8.4.3) was used to generate figures and perform simple linear regression models of correlation. Statistical significance was determined as p < 0.05.

Impact location and direction
The location and direction of impact of the bolt, relative to the head ACS, were relatively consistent apart from one outlier (JS007, the first animal injured in this test series; Supplementary Table 1, Supplementary Video 1).

Angular and linear head acceleration, and head displacement
Transient peaks in resultant angular acceleration occurred in the 2 ms immediately following impact, for all animals (Fig. 2). The first peak corresponded to a local maxima in negative vertical (z-axis) angular acceleration, occurring at 0.81 ± 0.10 ms post impact, and the second to peak positive vertical (z-axis) angular acceleration at 1.64 ± 0.36 ms post impact. At these time points, angular acceleration about the x-(lateral) and y-axes (forward) was typically less than one third of that about the z-axis. The second peak in resultant angular acceleration was greater than the first peak for all animals (232.46 ± 87.49 vs 121.68 ± 32.48 krad/s 2 ). Peak resultant linear acceleration (17620 to 60272 km/ s 2 ) occurred at approximately the same time as the second angular acceleration peak (1.61 ± 0.37 ms post-impact) ( Supplementary Fig. 1). Each linear acceleration component was similar in magnitude at the time of peak resultant linear acceleration. Peak resultant angular velocity ranged from 26 to 122 rad/s ( Supplementary Fig. 2) and peak resultant linear velocity ranged from 5.56 to 15.11 m/s ( Supplementary  Fig. 3). Amongst all animals, peak resultant head translation was 193 ± 53 mm and it occurred 366 ± 135 ms after impact ( Supplementary  Fig. 4). Head rotation was variable between animals, but the largest Fig. 1. Regions of interest analysed across three, 1 cm separated anteroposterior brain segments: a) internal capsule, b) grey and white matter of the cerebral cortex, c) cingulum, d) corpus callosum, e) caudate nucleus of the striatum, f) hippocampus and g) thalamus.

Fig. 2.
Resultant and component angular accelerations of the head, with the coordinate system origin at the head centre of mass and the horizontal (x-y) plane aligned with "forward gaze" head posture, for each experiment (Sharkey et al., 2022). rotations occurred about the z-axis for 6/8 animals ( Supplementary  Fig. 5).

Macroscopic injury description
Each of the eight animals subjected to TBI were examined for macroscopic injuries including skull fracture, contusions (coup and contrecoup) and intracranial bleeding ( Table 2). Seven of eight animals demonstrated skull fracture at the impact site, with a lower peak angular and linear acceleration noted in the one animal without skull fracture. Only 4/8 had a focal cortical contusion, 2/8 basal subarachnoid hemorrhage (SAH) and none a contrecoup contusion.

Physiological monitoring
Following hypoxia, physiological variables were measured hourly in all animals, with pCO 2 , pO 2 , and pH within normal limits for all groups throughout the duration of the experiment (Supplementary Table 3). Three-way repeated measures ANOVA found no main effect of injury, hypoxia, or time, for any of the parameters examined.

Axonal injury
Axonal injury was evaluated via APP immunostaining for examination of axonal transport disruption (Fig. 4) and SNTF for examination of activation of calpain mediated axonal injury (Fig. 5). Assessment of APP+ve axons across the regions of interest, found no interaction of injury*hypoxia (F 1,7 = 0.10, p = 0.1), nor main effects of injury (F 1,7 = 0.99, p = 0.14) or hypoxia (F 1,7 = 0.10, p = 0.06). The appearance of the few APP+ve profiles within the tissue are shown in Fig. 4B-E, as small continuous accumulations along the length of the axon, or small punctate staining for axons travelling in a transverse direction.
To determine whether linear and angular acceleration or velocity were associated with axonal injury identified via SNTF immunoreactivity a correlation via simple linear regression was performed. Injured animals were grouped together given that hypoxia had no additional effect on axonal injury development. This revealed no relationship between the presence of SNTF+ve axons and peak linear (R 2 = 0.05., p = 0.61; Fig. 6A) or angular acceleration (R 2 = 0.17, p = 0.31; Fig. 6B). Similarly no relationship was found between SNTF+ve axons and peak linear velocity (R 2 = 0.02, p = 0.77; Fig. 6C) or angular velocity (R 2 = 0.004, p = 0.88; Fig. 6B).

Discussion
This study aimed to evaluate the effect of post-traumatic hypoxia on the development of axonal injury and inflammation in a modified ovine model of TBI (Lewis et al., 1996;Anderson et al., 2003;Van Den Heuvel et al., 1998), with evaluation of resultant angular and linear acceleration during impact. Post-traumatic hypoxia did not affect the development of axonal injury or neuroinflammation following injury, and TBI alone produced minimal axonal injury, that was detected as an increase in SNTF+ve axons but not with APP immunostaining. The level of injury was insufficient to drive early neuroinflammatory changes, with no changes in the number of microglia or astrocytes in either white or grey matter regions, or pro-inflammatory cytokine levels within key white matter areas.
The ranges of peak resultant linear acceleration (3-8 km/s 2 ), angular acceleration (98-376 krads/s 2 ) and angular velocity (26-122 rad/s) were comparable to previous work by  km/s 2 , 81-227 krads/s 2 , and 39-118 rad/s, respectively). However, Anderson et al. observed substantial macroscopic pathologies including coup and contrecoup contusions in 10/11 injured animals, and extensive microscopic APP+ve axonal pathology (Anderson et al., 2003). Here, minimal macroscopic pathology was noted following injury, but there were higher rates of skull fracture (88%, vs 64%) (Anderson et al., 2003), which may relate to the modified captive bolt tip. The silicon cover was intended to mitigate against soft tissue damage and skull fracture, but may have contributed to the dual acceleration peak compared to the single peak reported previously (Anderson et al., 2003). Qualitative analysis of the high speed video footage, along with the shape of the silicone cover post-test, suggested that the first peak resulted from contact between the head and the rim of the silicone cover, that the silicon cover then deformed, and the second peak resulted from subsequent contact between the head and the flat steel plate with a compressed, thin, silicone layer. At low impact velocities skull fracture reduces contusions (Ren et al., 2020;Yavuz et al., 2003), which may explain the differences between this study and the previous work in this model (Anderson et al., 2003).
In pig models of TBI, diffuse brain injury thresholds and outcomes appear to depend on the magnitude of the angular acceleration, or angular velocity, the duration of the event, and the axis about which rotation occurs (Browne et al., 2011;Cullen et al., 2016;Gennarelli et al., 1982;Eucker et al., 2011). For example, axial-plane rotation produces more severe DAI than sagittal-plane rotation in pigs (Browne et al., 2011;Cullen et al., 2016). Despite the current study inducing peak angular accelerations exceeding the range used for non-impact miniature pig TBI studies (15-59 krad/s 2 ) (Browne et al., 2011), and peak angular velocities somewhat similar to those previously reported (100-146 rad/s) (Browne et al., 2011), more widespread APP+ve axonal injury involving the cortex, thalamus, midbrain and brainstem has been reported in that model (Browne et al., 2011;Johnson et al., 2016). The non-impact pig TBI model rotates the head about a defined axis, driven by a HYGE linear actuator and linkage , and the kinematics are typically reported from sensors mounted on the linkage which in some conditions has peak angular kinematics dissimilar to that of the head (Mayer et al., 2021). In the previous ovine impactacceleration TBI model, the head is unconstrained and can undergo motions in six-degrees-of-freedom (Anderson et al., 2003). In the current study, the head was suspended in a sling with a fulcrum approximately 50 cm above the head, instead of being placed on sandbags as previously reported (Lewis et al., 1996;Van Den Heuvel et al., 2000;Anderson et al., 2003;Van Den Heuvel et al., 1998). This may have contributed to altered distribution of head acceleration along/about the anatomical axes, which may have affected the axonal pathology observed.
Nonetheless, axonal injury was detected via analysis of SNTF, despite a lack of APP+ve DAI. SNTF is formed following the cleavage of alpha-II spectrin by calcium-dependent calpain proteases (Johnson et al., 2016), due to mechanically mediated ion channel dysfunction and alterations to membrane permeability within the axon allowing massive calcium influx (Johnson et al., 2013). At 6 hrs post-injury in a porcine model of mild TBI, SNTF+ve axons were frequently distinct from injured axons detected via accumulation of APP, a distinction that was lost by 24 hrs post-injury (Johnson et al., 2016). Thus SNTF+ve axons at 4 hrs may progress to microtubule disruption and impaired axonal transport with APP accumulation at a later time-point. Indeed, the presence of only SNTF+ve axonal injury early, may be a marker of mild injury, as with more severe injury, acutely both APP and SNTF co-localize due to the more severe disruption of axon dynamics. SNTF may be a useful early biomarker of injury, particularly mild TBI, with serum SNTF increasing up to 92% within 24 hrs post-injury in clinical mild TBI (Siman et al., 2020;Siman et al., 2013). Plasma SNTF levels also correlate with white matter changes measured by neuroimaging, alongside persistent cognitive and sensory-motor impairment up to 3 months post mild-TBI (Siman et al., 2020;Siman et al., 2013). However, a consensus on whether SNTF levels can predict injury severity has yet to be reached. Simple correlations did not detect a relationship between SNTF and peak linear or angular acceleration and velocity; however, this study was not designed to intentionally induce large variability in head kinematics in order to investigate this relationship.
Although axonal injury was present, as detected via SNTF immunoreactivity, the only inflammatory marker that was altered was GFAP expression within the CSF, with no change in the number or percent area staining of microglia or astrocytes in the brain regions of interest examined nor in inflammatory cytokines within key white matter regions or the CSF. GFAP is an intermediate filament that forms part of the cytoskeleton of mature astrocytes and is a marker of astroglial injury (Yang and Wang, 2015). Increased GFAP within the CSF has been suggested to have utility as a biomarker of TBI, (Fraser et al., 2011;Huang et al., 2015;Stukas et al., 2023;Zwirner et al., 2021), with levels increasing with injury severity. Indeed, increases in CSF GFAP have been reported acutely following mTBI in both pre-clinical models (Huang et al., 2015;Mountney et al., 2017) and clinical studies (Neselius et al., 2012;Neselius et al., 2013). In the current study, increased CSF GFAP expression was not associated with a change in the number of the GFAP+ve cells within the examined brain regions. Early GFAP elevations within the CSF may instead be a marker of the primary injury causing disruption to the cytoskeleton of astrocytes, which may not be sufficient to induce loss of astrocytes in mTBI (Huang et al., 2015). At later time-points GFAP levels may increase in response to astrogliosis (Yang and Wang, 2015), with an increase in both astrocyte number and GFAP expression within individual astrocytes (Chen et al., 2014;Cheng et al., 2019;Hatakeyama et al., 2021). For example, at 24 hrs following injury in a mouse model of lateral fluid percussion, a modest increase in astrocyte number and GFAP immunoreactivity was seen in the cortex and hippocampus alongside an increase in CSF GFAP levels (Mountney et al., 2017). The 4 h time-point in our study may have been too acute to see an increase in microglia or astrocyte number post-injury. In a rodent model of diffuse TBI, accumulation of microglia/macrophages within white matter tracts was not maximal until 8-14 days post-injury with only modest changes at 24 h (Hellewell et al., 2010;Venkatesan et al., 2010). Clinically, no association was observed between inflammation, as measured by the number of CD68+ve cells and axonal injury was within white matter acutely (24-96 h), with increases in CD68+ve cells only observed at 5 days post-injury in areas of axonal injury (Oehmichen et al., 1999). It should also be noted that analysis of the number of IBA1+ve cells and % area may be too blunt a measure and that measures of activation status may be required. For example, increased microglial activation index was observed at 6 h post-injury in the diffuse porcine HYGE model, with microglial processes converging on APP+ve injured axons (Lafrenaye et al., 2015). Furthermore, morphological appearance is not sufficient to determine the function of microglia and future studies will be needed to examine the transcriptomic and proteomic responses within microglia and astrocytes in response to injury (Wang et al., 2023).
Surprisingly the increase in CSF GFAP was not associated with an increase in pro-inflammatory cytokine levels either within the CSF or within the key white matter regions examined. This contrasts with the increased pro-inflammatory cytokines previously reported following all severities of TBI both clinically and experimentally Feuerstein et al., 1998;Frugier et al., 2010;Di Battista et al., 2016;Baratz et al., 2015). Indeed, it should be noted that TNFα levels actually decreased in the TBI:normoxia and TBI:hypoxia animals, an effect not seen in shams. Clinically CSF measurements occur following more severe TBI were TNFα levels are increased within 24 hrs following injury (Clay Goodman et al., 1990;Morganti-Kossman et al., 1997). In mTBI where serum is typically sampled, a systematic review found only 2/8 studies reported an acute increase in TNFα following injury (Visser et al., 2022), suggesting that TNFα elevations acutely may be injury severity dependent. However, it must be acknowledged that levels of TNFα would be expected to be higher in CSF than in serum (Shiozaki et al., 2005). Nonetheless, why a decrease was seen in the CSF following injury here is unclear. The disparity in results may relate to the length of anaesthesia, as animals were anaesthetised for the duration of the experiment. Both ketamine (Bell, 2017) and isoflurane (Statler et al., 2006) are neuroprotective. Specifically, ketamine has anti-inflammatory properties with in vitro studies demonstrating that ketamine inhibits the production and release of pro-inflammatory cytokines including IL-1β, IL-6 and TNF-α in macrophages (Chang et al., 2010), microglial cells (Chang et al., 2009) and astrocytes (Yuhas et al., 2015). Similarly, prolonged isoflurane exposure has been shown to reduce the levels of serum cytokines TNF-α, IL-6 and IL-10 in response to lipopolysaccharide in mice (Fuentes et al., 2005;Fuentes et al., 2006) Thus, the extended period of anaesthesia may have influenced the secondary injury response to the TBI and attenuated the inflammatory response. The observed increase in GFAP within the CSF may therefore relate to the mechanical damage of astrocytes at the point of injury, rather than an ongoing secondary injury process. Further studies using shorter periods of anaesthesia and longer survival times are required.
Post-injury hypoxia had no effect on the development of axonal injury. This was unexpected given that hypoxia reduces Adenosine triphosphate (ATP) production and thus the ability to efflux calcium following TBI (Weber, 2012). Typically this would result in a build-up of intra-axonal calcium, which would be expected to perpetuate axonal damage and increase SNTF levels (Johnson et al., 2013;Pettus and Povlishock, 1996). Similarly, post-injury hypoxia did not exacerbate the inflammatory response, in contrast to previous work in a rodent diffuse model of TBI in which microglial and astrocyte number were significantly increased, alongside pro-inflammatory cytokine expression (IL-6, IL-1β and TNFα) within cortical homogenates (Hellewell et al., 2010;Yan et al., 2011). The hypoxia-driven increase in macrophage infiltration, and enhanced astrogliosis, co-localised with increased axonal damage post-traumatic hypoxia (Hellewell et al., 2010). Even a delayed model of hypoxia, where it was delivered 24 hrs following focal injury increased axonal injury and astrogliosis (Parikh et al., 2016). Secondary hypoxia following severe TBI in humans is also associated with prolonged cytokine production (Yan et al., 2014). In the current study, the degree and duration of hypoxia (30-35% reduction in PO 2 for 15 min) may have been insufficient and the 4 h time-point may have been insufficient to allow the effects of hypoxia to be observed. In a rodent model which found hypoxia exacerbated inflammation and axonal injury, arterial blood oxygenation saturation was reduced by ~ 50% for 30 min post-injury (Hellewell et al., 2010). Clinically, hypoxia postinjury is described as a SaO 2 < 92% or apnoea or cyanosis immediately after injury (Yan et al., 2014), with no mention of the length of the hypoxic period. However a 10 min period of hypoxia has also been utilized in rodent models, less than that used here (Plummer et al., 2018). Furthermore, neural mechanisms may differ between lissencephalic to gyrencephalic brains and post-traumatic hypoxia in a large animal model has not been reported prior to this study.
The only additive effect of post-traumatic hypoxia was on the acute Fig. 9. Examination of TNF-α, IL-1β and IL6 within the (A) internal capsule and (B) corpus callosum found no effect of injury or hypoxia. Data expressed as mean ± SEM, n = 3-4 per group. Fig. 10. CSF samples were analysed using ELISA to assess pro-inflammatory cytokines: GFAP (A), IL-6 (B), IL-1β (C) and TNFα (D), with samples taken prior to injury and at 4 h post TBI. Data expressed as mean ± SEM. n = 3-4 per group, *p < 0.05, **p < 0.001, ***p < 0.001, ****p < 0.0001. ABP response. Although hypoxia alone decreased BP, TBI:hypoxia animals showed the greatest derangement, failing to return to pre-injury ABP values by one hour post-injury. This was unexpected as acute hypoxia independently increases ABP in humans (Dixon et al., 2020). However, this result is supported by similar research in rodents, where TBI:hypoxia and sham:hypoxia animals had reduced ABP (Hellewell et al., 2010). Acute hypoxia causes increased cerebral blood flow via direct effects on the vascular cells of cerebral arteries and arterioles (Tomiyama et al., 1999). A hypoxia-induced reduction in ATP levels opens ATP channels on smooth muscle, causing hyperpolarisation and vasodilation (Taguchi et al., 1994). Additionally, hypoxia drastically increases nitric oxide and adenosine production locally, also promoting vasodilation, and decreasing arterial blood pressure (Umbrello et al., 2013). The typical ABP response following TBI has not been consistently reported. Transient hypertension followed by sustained hypotension is typical of experimental TBI and has been found in rodent, porcine and ovine models (Marmarou et al., 1994;Lewis et al., 1996;Butcher et al., 2007;McIntosh et al., 1989) and has been reported clinically following more moderate-severe forms of TBI (Simard and Bellefleur, 1989;Clifton et al., 1983). However, hypotensive episodes during resuscitation have also been observed following moderate-severe TBI (Manley et al., 2001). The potential mechanisms for an acute immediate decrease in ABP following TBI are poorly addressed in the literature; however, recent studies have suggested possible autonomic dysregulation to follow more mild forms of TBI as indicated by changes in heart rate variability (McDonald et al., 2020;Pertab et al., 2018;Esterov and Greenwald, 2017). Whether autonomic dysregulation was a factor in driving a reduction in ABP following TBI in our model requires further investigation with comprehensive cardiovascular monitoring studies.
This study had several limitations. There were low numbers of animals in each group. Furthermore, the peak resultant head accelerations, velocities, and head displacement produced was variable across animals, but this variability was similar across both injured groups, and similar to that of Anderson et al. (2003). The trajectory of the injury device, and therefore the vector of the applied load relative to the head coordinate system, was relatively consistent (Supplementary Video 1); the positioning of the head in the sling and the neck posture likely contributed to the kinematic variability. The angular and linear accelerations and velocities are reported in an anatomical coordinate system that differs from that of Anderson et al., but the resultant angular accelerations and velocities are not affected by this, and the resultant linear accelerations and velocities are only marginally affected by the change in coordinate system origin; data consistent with the Anderson et al., coordinate system are provided in Supplementary Fig. 6 and Supplementary Fig. 7.
The linear acceleration and velocity data are also provided at the brain centre of mass ( Supplementary Fig. 8, Supplementary Fig. 9), which is not coincident with the head centre of mass in sheep (Sharkey et al., 2022), to allow comparison with other models of brain injury. TBI research using nonhuman primates has described head accelerations at the CoM of the brain (estimated at the pineal gland, located at the notional "centre of the brain"), rather than at the CoM of the head (Abel et al., 1978). In-vivo human head accelerations and velocities are typically reported at/about the head centre of mass, consistent with the location of accelerometers in anthropometric test devices, the location of which is likely similar to the brain centre of mass.
In summary, in this animal model a hypoxic state of 15 min duration, induced immediately after head impact, hypoxia had no additional effect on the accumulation of inflammatory cells nor on axonal injury development. In this model at 4 h following injury, axonal injury was only detected by SNTF staining, suggested that early axonal injury is driven by calpain activation, and this may provide an early target for treatment intervention. No disruption in axonal transport, as detected by APP was noted. Further investigation is needed with longer observation durations to determine the fate of these axons to determine if SNTF+ve axons eventually show signs of disrupted axonal transport.

Transparency, Rigor, and reproducibility summary
The analysis plan was not formally pre-registered, but the senior author with primary responsibility for the analysis certifies that the analysis plan was pre-specified. Fourteen sheep were included within the experiment given the limitations in availability and cost of animals and randomly assigned to groups using a random number generator. All animals reached the assigned end-point of the study. Investigators who performed histological analyses were blinded to group by coding the slides, and all manual histological counts were conducted via two independent investigators with a < 10% SD between individual counts. All materials used to conduct the study were obtained from a widely available source, as stated within the Methods of the paper. The authors agree to provide the full content of the manuscript on request by contacting the corresponding author. Samples collected as part of the study are available to be used for future research without requirement for specific additional informed consent or regulatory controls. Please contact the corresponding author.

Funding statement
Funding was provided by the National Health and Medical Research Council (Australia-APP1145183) and the Neurosurgical Research Foundation. RDQ was supported, in part, by an Australian Research Council Discovery Grant (DP190101209; CFJ) during this study. CFJ was partially supported by a National Health and Medical Research Council (Australia) Early Career Fellowship during this study (APP1072387).

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.