Characterising spinal cerebrospinal fluid flow in the pig with phase-contrast magnetic resonance imaging

Background Detecting changes in pulsatile cerebrospinal fluid (CSF) flow may assist clinical management decisions, but spinal CSF flow is relatively understudied. Traumatic spinal cord injuries (SCI) often cause spinal cord swelling and subarachnoid space (SAS) obstruction, potentially causing pulsatile CSF flow changes. Pigs are emerging as a favoured large animal SCI model; therefore, the aim of this study was to characterise CSF flow along the healthy pig spine. Methods Phase-contrast magnetic resonance images (PC-MRI), retrospectively cardiac gated, were acquired for fourteen laterally recumbent, anaesthetised and ventilated, female domestic pigs (22–29 kg). Axial images were obtained at C2/C3, T8/T9, T11/T12 and L1/L2. Dorsal and ventral SAS regions of interest (ROI) were manually segmented. CSF flow and velocity were determined throughout a cardiac cycle. Linear mixed-effects models, with post-hoc comparisons, were used to identify differences in peak systolic/diastolic flow, and maximum velocity (cranial/caudal), across spinal levels and dorsal/ventral SAS. Velocity wave speed from C2/C3 to L1/L2 was calculated. Results PC-MRI data were obtained for 11/14 animals. Pulsatile CSF flow was observed at all spinal levels. Peak systolic flow was greater at C2/C3 (dorsal: − 0.32 ± 0.14 mL/s, ventral: − 0.15 ± 0.13 mL/s) than T8/T9 dorsally (− 0.04 ± 0.03 mL/s; p < 0.001), but not different ventrally (− 0.08 ± 0.08 mL/s; p = 0.275), and no difference between thoracolumbar levels (p > 0.05). Peak diastolic flow was greater at C2/C3 (0.29 ± 0.08 mL/s) compared to T8/T9 (0.03 ± 0.03 mL/s, p < 0.001) dorsally, but not different ventrally (p = 1.000). Cranial and caudal maximum velocity at C2/C3 were greater than thoracolumbar levels dorsally (p < 0.001), and T8/T9 and L1/L2 ventrally (p = 0.022). Diastolic velocity wave speed was 1.41 ± 0.39 m/s dorsally and 1.22 ± 0.21 m/s ventrally, and systolic velocity wave speed was 1.02 ± 0.25 m/s dorsally and 0.91 ± 0.22 m/s ventrally. Conclusions In anaesthetised and ventilated domestic pigs, spinal CSF has lower pulsatile flow and slower velocity wave propagation, compared to humans. This study provides baseline CSF flow at spinal levels relevant for future SCI research in this animal model. Supplementary Information The online version contains supplementary material available at 10.1186/s12987-022-00401-4.


Background
Cerebrospinal fluid (CSF) cushions the brain and spinal cord from mechanical injury, acts as a critical transport medium for nutrients and metabolites to maintain a homeostatic environment for neuronal cells, and provides a drainage system for interstitial fluid from the central nervous system (CNS) [1]. Although the CSF system remains relatively poorly understood, the diagnostic [2][3][4][5][6] and therapeutic [7,8] potential of CSF dynamics have become increasingly apparent. Free-flowing CSF is vital to brain and spinal cord health, and changes to CSF flow may adversely affect fluid balance and metabolism [9][10][11].
CSF flow in humans is predominantly oscillatory, with cranial and caudal pulsations along the spinal cord axis synchronous with the cardiac and respiratory cycles [12]. CSF pulsations can be measured with phase-contrast magnetic resonance imaging (PC-MRI). A bipolar gradient is applied in the pulse sequence to produce a phase shift, which is directly proportional to proton velocity in the flow-encoding direction [13], and is gated with cardiac activity to produce velocity data as a function of the cardiac cycle. PC-MRI studies of CSF flow in the cranium and spinal canal have been fundamental in improving understanding of CSF dynamics in normal and disease states [14][15][16][17][18]. CSF flow in the cerebral aqueduct and at the cranio-cervical junction are used to assist in the diagnosis of hydrocephalus [2][3][4]19] and Chiari malformation [5,6,20,21]. Although CSF dynamics in the spinal column may provide similar diagnostic capabilities, the spinal region remains relatively understudied. Several studies have characterised normal CSF dynamics along the spine in healthy humans [22][23][24][25] and non-human primates (NHP) [26]. Other studies have found that obstructions to the intrathecal space may cause changes to CSF dynamics in the context of cervical myelopathy [27][28][29] and in neurogenic claudication in lumbar spinal stenosis [30].
Traumatic spinal cord injury (SCI) may alter CSF flow due to the swollen spinal cord obstructing the spinal subarachnoid space (SAS) [31,32]. One study demonstrated CSF velocity changes at the level of injury in cervical SCI patients up to two years after injury [33], and another study of post-traumatic syringomyelia identified altered temporal features and lower peak velocities at the syrinx level in patients compared to healthy humans [34]. Pigs are emerging as a favoured large animal model for SCI research [35]. The more human-like size of the porcine spinal column [36], spinal cord, and intrathecal space [37,38] compared to rodents, provides the opportunity to study CSF dynamics using clinical imaging modalities and in-dwelling pressure sensors. The anterior-posterior diameter of the spinal canal at T10 is approximately 20 mm in humans, 10 mm in 40 kg domestic pigs [36], and 2.7-3.3 mm at T12-L1 in rats [39]. It is important to examine normal spinal CSF flow in the pig to provide a benchmark for future investigations of SCI. The aim of this study was to characterise normal CSF flow along the pig spine, examining two locations proximal, and two locations distal, to thoracic level T10, which is the standard spinal level for pre-clinical contusion models of SCI [40].

Animal ethics
This project was approved by the South Australian Health and Medical Research Institute Animal Ethics Committee (SAM 243 and SAM-22-031) and conducted in accordance with the Australian National Health and Medical Research Council Code of Care and Use of Animals for Scientific Purposes [41].

Animals
Magnetic resonance images (MRI) of fourteen healthy, female, Large White Landrace cross pigs (22-29 kg at imaging) were obtained. The animals were acclimatised in a purpose-built facility for 7-10 days prior to imaging. During this time they were provided enrichment toys, access to water, twice daily food rations, and administered once daily prophylactic antibiotics (Trimidine ™ 0.15gm/10 kg; Sulfadimidine 430 mg, Trimethoprim 86 mg). After imaging was completed, animals were recovered from anaesthetic, and maintained for further approved procedures on the same ethics protocol that are not reported herein, after which they were humanely killed.

Anaesthetic protocol
The animals were fasted overnight prior to anaesthesia and pre-medicated with medetomidine (0.02 mg/kg) and ketamine (7.5 mg/kg). Anaesthesia was induced using ketamine (11 mg/kg) and propofol (2 mg/kg), maintained with an intravenous protocol of ketamine (5.0-8.0 mg/ kg/hr), propofol (2.0-6.0 mg/kg/hr), and fentanyl (8.0-15.0 ug/kg/hr) and titrated as necessary towards the end of the imaging session. A propofol bolus (2.0 mg/kg) was administered when vital signs and/or physical indications suggested lightening of anaesthesia. Anaesthetics were administered via a venous catheter that was placed in the external jugular vein via surgical cut-down immediately prior to the MRI. The animals were intubated, ventilated at 17-21 breaths/min with 270-400 mL/min of oxygen, and heart rate, oxygen saturation and end-tidal carbon dioxide were monitored.

Magnetic resonance imaging
MRIs were performed on a 3T scanner (Siemens, Magnetom Skyra, Germany). Animals were placed in left lateral recumbency with two 18-channel body coils (Siemens, Germany) wrapped over their torso and neck. T2-weighted TSE (turbo spin echo) images of the cervical/thoracic and thoracic/lumbar spinal regions were acquired with the following parameters: sagittal slice thickness 4 mm, in-plane resolution 1.302 × 1.302 mm, repetition time (TR) 3500 ms, echo time (TE) 98 ms, 2 averages, 146° flip angle and 384 × 384 acquisition matrix. These images were used during scanning to prescribe the PC-MRI axial locations and orientations, and post-processed to measure spinal length (see below).

Phase-contrast magnetic resonance imaging acquisition
PC-MRI acquisition was performed with retrospective cardiac gating using a pulse oximeter attached to the tail. At spinal levels C2/C3, T8/T9, T11/T12 and L1/L2, a single axial slice orthogonal to the spinal cord in the sagittal and coronal planes, aligned with the centre of the adjacent intervertebral disc, was acquired ( Fig. 1A-G). Cranial flow was positively encoded (white pixels) and the number of acquired cardiac phases ranged from 14 to 30, selected according to the target R-R interval.
Acquisition duration per spinal level was between 3 and 9 min. Minimum TR/TE was selected within the pulse sequence. PC-MRI sequence parameters for the first ten animals included: field of view, 140 × 140 mm; TR, 46.39-59.84 ms; TE, 9.41 ms; in-plane resolution, 0.5469 × 0.5469 mm; slice thickness 5 mm; anterior-posterior phase-encoding direction; 15° flip angle. Encoding velocities (V ENC ) of 4 cm/s and 6 cm/s were selected during a pilot study and run consecutively at each spinal level. One animal was also scanned with an additional V ENC of 2 cm/s at each spinal level and the TR and TE changed to 55.61-69.02 ms and 12.41 ms, respectively. Following completion of the first ten animals, aliasing was detected in some animals at C2/C3 in the V ENC = 6 cm/s scans; therefore, for the last four animals (P011-P014), V ENC of 6 cm/s, 8 cm/s and 10 cm/s were used at that level. Because of this increase in V ENC , TR and TE changed to 59.84 ms and 8.04 ms, respectively, at C2/C3; all other parameters remained the same.

Phase-contrast magnetic resonance imaging post-processing
Post-processing of CSF flow images was undertaken by a single investigator (MAB) using Segment software (Version 3.2, Medviso, Lund, Sweden) [42]. At each spinal Fig. 1 Sagittal T2 TSE MRI of the A cervical and thoracic spine and B thoracic and lumbar spine from one representative animal with labelled spinal levels and planes of the phase-contrast MRI (PC-MRI) images. C-F Representative axial PC-MRI (phase) at peak systolic (caudal) and at peak diastolic (cranial) flow for each spinal level with scale bar (velocity; cm/s). G A schematic demonstrating which pixels are assigned within the ROI. Blue is the ventral SAS ROI; pink is the dorsal SAS ROI. TSE turbo spin echo, MRI magnetic resonance imaging, ROI region of interest, SAS subarachnoid space level, regions of interest (ROI) were manually defined at the border of the CSF flow signal in the SAS, on the image that displayed the greatest contrast between CSF flow signal and the surrounding tissues. ROIs were defined separately in the ventral and dorsal SAS (referred to as the SAS region), because of its potential relevance to spinal pathologies [11,33] and because in some animals flow signal was observed in one region and not the other. The dorsal region was split into two ROIs when distinct compartments were observed, to ensure only pixels with flow signal were selected (Additional file 1: Fig. S1). In the Segment software, pixel selection for flow estimation included only those pixels with their centroid inside the contour of the ROI (Fig. 1G). Motion of the spinal canal was not observed on the re-phased [43] (anatomical) image at any spinal level, and the ROIs were propagated to each temporal image. To assess intra-rater reliability, repeat ROIs were completed two months later by the same investigator, on ten randomly selected PC-MRIs, across all spinal levels and animals.
Eddy current offset correction with time-dependent linear compensation was applied by iteratively adjusting the threshold in Segment software until the sum of the net flow of the two ROIs was as close to zero as possible. The resulting residual net flow was determined as a percentage of the total stroke volume (mL/cycle) of both ROIs. This was performed on all scans that captured the full systolic and diastolic flux. Residual net flow in all animals was less than 9% (2.85 ± 2.47%; Additional file 1: Table S1). The mean offset magnitude applied at C2/C3 was 0.05 ± 0.05 mL/s, at T8/T9 was 0.03 ± 0.02 mL/s, at T11/T12 was 0.02 ± 0.02 mL/s, and L1/L2 was 0.03 ± 0.04 mL/s. Each ROI was inspected for aliasing artefact (i.e., maximum velocity exceeding the acquisition V ENC ), and aliasing was corrected with phase-unwrapping which was confirmed visually by examining the pixel-by-pixel velocity in each image. This was performed on 8 animals at C2/C3, 1 animal at T11/T12, and 2 animals at L1/L2. At most 5 consecutive cardiac phases were corrected (mean consecutive corrected phases: 2 ± 1), and no more than 14% of the total number of pixels in both ROIs were corrected at C2/C3, 2% at T11/T12, and 9% at L1/L2. The following parameters were derived: CSF velocity (cm/s) in each ROI for each cardiac phase acquired; CSF flow (mL/s) in each ROI for each cardiac phase acquired; stroke volume, defined as the sum of the caudal flow volume and the cranial flow volume in one cardiac cycle (mL/cycle); maximum cranial and caudal velocity (single pixel; cm/s) in each SAS region, over the cardiac cycle.
These data were further processed using a custom MATLAB program (Version R2020a, Mathworks Inc, Natick, MA). For dorsal SAS with two ROIs, flow data were summed, and mean velocity were averaged, to provide a single dorsal value for each outcome. Flow and mean velocity versus time data for each cycle were interpolated to 100 points using a cubic spline ("pchip" MATLAB function to preserve minima and maxima), and expressed as a percentage of the cardiac cycle to normalise these data by heart rate. The following parameters were derived: peak systolic (caudal direction) and peak diastolic (cranial direction) flow; and, mean velocity, defined as the mean velocity of all pixels within the ROI, which was used to calculate the time to peak mean velocity (systolic and diastolic; from t = 0 in the cardiac cycle).

Velocity wave propagation speed
CSF velocity wave speed (VWS) [44] was estimated for each animal as the distance between the C2/C3 and L1/ L2 spinal levels divided by the time taken for the velocity wave (peak systolic, peak diastolic) to travel that distance. The distance from C2/C3 to the L1/L2 intervertebral disc was measured across both T2 FSE scans (summed distance from C2/C3 to T11/T12, and T11/T12 to L1/L2) with a spline placed along the centre of the spinal cord on a sagittal view, using Materialise Mimics software (Version 22.0, Materialise, NV). For the diastolic velocity wave, the difference between time to peak mean diastolic velocity at C2/C3 and L1/L2 was calculated. For the systolic velocity wave calculation, time to peak mean systolic velocity at L1/L2 was adjusted: cardiac cycle duration minus time to peak mean systolic velocity at C2/C3 was added to the time to peak systolic velocity at L1/L2. Systolic velocity wave speed was calculated as the difference between time to peak mean systolic velocity at C2/ C3 and the adjusted L1/L2 value.

Statistical analyses
All statistical analyses were performed with SPSS (version 26, IBM Corporation, Armonk, NY). Intra-rater reliability of the ROI area, and the corresponding peak diastolic and systolic flow, was determined using intraclass correlation coefficients (ICC) of the absolute agreement between the repeated ROI segmentations. ICC values > 0.75 and < 0.9 were defined as good reliability, and values > 0.9 as excellent reliability [45]. Normality and homogeneity of variance were assessed for peak systolic flow, peak diastolic flow, and maximum cranial and caudal CSF velocity data using Shapiro-Wilk and Levene's tests. Linear mixed-effects models (LMMs) were developed to identify if there was an effect of the spinal level, SAS region (ventral and dorsal ROI), the spinal level*SAS region interaction, on peak systolic and diastolic flow, and maximum cranial and caudal CSF velocity. LMMs initially also included a fixed effect of temporal resolution (cardiac cycle duration divided by the number of cardiac phases acquired); however, it was not associated with any outcome measures, and was subsequently removed from the models. Post-hoc pairwise comparisons, with Bonferroni correction, were performed for LMMs with significant interactions (significance level of α = 0.05). Experimental data is presented as mean ± standard deviation (SD) in text, with the p-values corresponding to the LMMs, and complete LMM outcomes are available in the supplementary materials.

Results
Complete CSF flow data were obtained for 11 of 14 animals. CSF flow was not detected at any spinal level in 3 animals (P005, P011, and P013). In one animal (P003), CSF flow was not detected at T8/T9 and T11/T12. CSF flow was not detected in the dorsal SAS at T8/T9 in 4 of 11 animals (Additional file 1: Table S2). When CSF flow was not detected and a lower V ENC was not scanned, the spinal level and/or SAS region was excluded from all analyses and plots; zero flow could not be confirmed since it is possible that there was flow below a detectable threshold determined by the amount of phase noise present. Two animals (P004 and P012) had abnormal cardiac gating and optimised eddy current corrections could not be performed since the full systolic and diastolic flux was not captured (Additional file 1: Figure S2). Data for these animals were excluded from temporal analyses and plots, but included in peak flow values since the diastolic and systolic extremes were maintained and were within, or marginally outside the range of the remaining data (Additional file 1: Table S3). The mean heart rate for the duration of the scans was 101 ± 27 bpm (Additional file 1: Table S4).
For all levels except C2/C3, the dorsal ROIs were only slightly smaller in area than the ventral ROIs (Table 1). ROI area and associated peak diastolic and systolic flow, in the dorsal SAS and ventral SAS, demonstrated Table 1 ROI area (mean ± SD) for all animals with detected flow signal (n = 11), and intraclass correlation coefficients (ICC) for ROIs manually defined two months apart by the same investigator using a sample of ten PC-MRIs   (Table 1).

CSF flow
Bidirectional (cranial-caudal) CSF flow across the cardiac cycle was observed at all spinal levels (Fig. 2). At the thoracolumbar levels, the flow waveform had similar shape and amplitude for all animals. The standard deviation at C2/C3 across the cardiac cycle was larger than at the other spinal levels. The flow waveforms at C2/C3 start at the beginning of the diastolic pulse due to peripheral pulse gating [46]. Stroke volume at C2/ C3 was 0.09 ± 0.04 mL/cycle in the dorsal SAS, and 0.04 ± 0.03 mL/cycle in the ventral SAS. At L1/L2, stroke volume was 0.02 ± 0.01 mL/cycle in the dorsal SAS, and 0.04 ± 0.04 mL/cycle in the ventral SAS. Stroke volume and net flow data for all spinal levels are available in Additional file 1: Table S5.

Velocity wave propagation speed
The mean CSF velocity waveforms (Additional file 1: Fig. S3) demonstrated a phase shift from C2/C3 to L1/ L2, indicating propagation of the velocity wave in the caudal direction in both the dorsal and ventral regions (Fig. 5A, B). The mean distance between C2/C3 and L1/ L2 was 416 ± 24 mm. Peak systolic velocity at C2/C3 was close to coincident with peak diastolic CSF velocity at T8/T9, T11/T12 and L1/L2 (Additional file 1: Table S9). Peak systolic velocity at C2/C3 occurred at 70.0 ± 6.3% (392.9 ± 35.5 ms) of the cardiac cycle in the dorsal SAS and at 64.

Discussion
CSF dynamics along the spine have previously been characterised in humans and NHP (macaca fascicularis) [14,15,17,[22][23][24][25][26]. Large animal models are increasingly being used to investigate SCI [35,40]. The current study provides normative data for spinal CSF flow in experimentally naïve anaesthetised domestic pigs as a baseline and comparison for future SCI studies. The study provides evidence that spinal CSF flow in anaesthetised and ventilated pigs is lower, and has slower velocity wave propagation, than in healhy, conscious humans [22,44,47], but is similar to that in anaesthetised NHP [26].
Spinal CSF flow in anaesthetised pigs is considerably lower than in healthy, conscious humans across the cardiac cycle. CSF flow measured in the cervical (C2-C4) spine of awake, healthy humans at peak systole is 2.4-3.5 mL/s, and at peak diastole is 1.5-2.8 mL/s [15,19,22,48], which is approximately seven times greater than in the anaesthetised pigs. CSF flow in the lumbar spine (L1-L3) in healthy humans at peak systole is 1.07-1.26 mL/s, and at peak diastole is 0.29-0.58 mL/s [22,76], which is approximately eight and four times greater, respectively, than in the pigs. In addition, stroke volume in the cervical (C2) spine in healthy humans is approximately 0.5 mL/cycle [49,50], which is four times greater than in the pigs. In healthy conscious humans, peak mean velocity in the thoracic region (5.81 ± 1.42 cm/s; N = 14) [51] is greater than in the anaesthetised pig (Additional file 1: Table S10). The SAS is also larger in humans than the pig: for example, the T10 anterior-posterior diameter of the CSF (dural minus spinal cord diameter) in the domestic pig is 2.09 mm [1.70 mm-2.66 mm] (median and range; ultrasound measurements) [52], whereas in humans it is 7.4 ± 3.1 mm (mean ± SD; computed tomography contrast myelogram measurements) [53]. Together, these results suggest that higher CSF flow in conscious humans than anaesthetised pigs results, at least in part, from the combined effect of a larger SAS and greater CSF velocity in humans. CSF flow in the pig was similar to that reported for anaesthetised, but not mechanically ventilated, NHPs: NHP peak systolic flow in the cervical spine Fig. 4 Maximum CSF velocity in the cranial and caudal direction in the SAS at four spinal levels (n = 11). The red data is CSF flow in the dorsal SAS and the blue data is CSF flow in the ventral SAS. Significant differences (*) between spinal levels for the select SAS region, correspond to LMM post-hoc pairwise comparisons (p ≤ 0.05). Each data point is from one animal, large circle presents the mean ± one SD (solid line). CSF cerebrospinal fluid, SAS subarachnoid space, LMM linear mixed-effects models, SD standard deviation (0.5 ± 0.2 mL/s, C2/C3; estimated from graphical data), and in the lumbar spine (0.1 ± 0.05 mL/s, L3/L4; estimated from graphical data), was approximately the same as in the pigs. NHP cervical peak diastolic flow was nearly two times smaller than (0.2 ± 0.2 mL/s; estimated from graphical data), and lumbar flow was approximately the same as (0.1 ± 0.1 mL/s; estimated from graphical data), that in the current pig study [26]. In pigs, peak systolic CSF flow decreased from the cervical to the thoracic levels, but was unchanged at the thoracolumbar levels both dorsally and ventrally. In humans, peak systolic and diastolic flow decreases caudally to a minimum in the lumbar region [22]. Such a reduction across the spine may not have been observed in the current study because the thoracic (T8/T9) and lumbar (L1/L2) measurements encompassed a limited span (193 ± 20 mm). It is also possible that respiratory forces have a greater effect on CSF flow in the thoracic region compared to at the lumbar levels [23], and that no difference was observed between adjacent thoracic and lumbar sites due to mechanical ventilation of the pigs.
Although the relative contribution of respiratory and cardiac cycles to CSF dynamics is unclear, it is evident that physiological variations of both can alter CSF flow [54][55][56][57]. It is commonly thought that CSF travels cranially with inhalation, and caudally with exhalation, due to changes in intrathoracic pressure with spontaneous breathing [23,24,55,56]. Unlike in most non-acute clinical CSF flow studies reported, the animals in this study were anesthetised and ventilated, which causes positive intrathoracic pressure and produces increased pressure during inhalation and decreased pressure during exhalation (the opposite of spontaneous breathing). Positive intrathoracic pressure throughout the respiratory cycle may contribute to lower CSF flow. Mechanically ventilated rats had less movement of fluorescent tracer in the spinal SAS in the caudal direction compared to spontaneously breathing animals [54]. This suggests that respiratory conditions need to be carefully considered in CSF flow study comparisons. Mechanical ventilation can also influence arterial pulse pressure through complex cardiopulmonary interactions [58]. It is apparent that reduced arterial pulse pressure decreases CSF flow in the spinal perivascular spaces [59,60], but its effect on CSF pulsations in the SAS is unknown. There is limited and conflicting evidence that heart rate influences CSF flow: increasing heart rate resulted in increased bidirectional CSF flow velocities in a three dimensional computational fluid dynamics model of the SAS [57], while a study in rats showed that increased heart rate had little influence on CSF flow [54]. The heart rate range in this study (71-174 bpm) is larger than in the other animal studies (canines, 70-110 bpm [73] ; NHP, 92-132 bpm [26]). In this current study, two animals had notably higher heart rates (P004, 120-130 bpm; P012, 174 bpm), both with apparent abnormal cardiac gating. The animal (P006)  Table S3) had unremarkable physiological parameters (Additional file 1: Table S4). Anaesthesia may also contribute to CSF dynamics [61], and this is potentially due to its influence on blood pressure and partial pressure of carbon dioxide in the blood stream. Mean arterial pressure is thought to influence CSF flow in the cerebral perivascular spaces [62]; however, its effect on pulsating spinal CSF flow in the SAS is understudied. In one rat study there was no effect of mean arterial pressure on spinal SAS CSF flow [54]. In the current study, the animal that received a propofol bolus during the scan (P014; T11/T12 and L1/L2 scanned prior to bolus, and T8/T9 and C2/C3 after bolus) had abnormally higher peak systolic flow at T11/T12 and L1/L2 dorsally, T8/T9 and L1/ L2 ventrally and peak diastolic flow at T8/T9 dorsally. The propofol bolus was indicated because the animal was not deeply anaesthetised (as evidenced by physical movement), and was therefore likely to have higher blood pressure; however, blood pressure was not measured concurrently in this study. In addition, cerebral blood flow is influenced by the partial pressure of arterial carbon dioxide [63]. Hypercapnia increases cerebral blood flow and intracranial pressure [64], which may effect spinal CSF flow due to the potential interaction between intracerebral arteries and pulse propagation [65]. It is becoming increasingly clear that respiratory forces have a large influence on CSF flow; however, the potential contributions of other physiological parameters such as heart rate, blood pressure, and carbon dioxide in the bloodstream, need to be further investigated to establish their relative effect on CSF dynamics. These physiological parameters should be measured and recorded during PC-MRI acquisition of CSF flow in future studies.
Velocity wave propagation was detected in the caudal direction along the spinal axis in laterally recumbent pigs. Velocity wave propagation speed can be used to approximate pulse wave velocity (pressure wave propagation) since it has been shown that they are nearly identical under certain conditions [47]. Because of this, some studies use the term 'pulse wave velocity' rather than VWS, for the identical measure [26,44]. It has been hypothesised that CSF pressure waves originate in the intracerebral arteries and propagate in a caudal direction [65]; however, the origin of the pulse remains unresolved since there are other studies which suggest local sources [22,66,67]. In cardiovascular diagnostics, pulse wave velocity is a measure of vessel compliance. VWS has recently been reported for spinal CSF flow studies [44,47,68] where it likely reflects compliance of the spinal cord tissue, dura, and surrounding tissues [69]. A one-dimensional tube model of the spinal SAS has been used to show that increasing spinal compliance results in slower VWS and greater attenuation of the pulse [70]. As observed in these data (Fig. 5), spinal compliance can enable CSF to flow in opposite directions concurrently at spinal locations remote to each other (e.g. C2/C3 and L1/L2). In healthy conscious humans, VWS is three (4.6 ± 1.7 m/s) [47] to four (5.83 ± 3.36 m/s) [44] times faster than that measured in the pigs. In NHP, VWS is similar to the VWS estimated for the dorsal SAS in this study (1.13 m/s) [26]. In humans, compliance provided by the craniospinal compartment compensates for postural pressure changes [49,50]. The extent of this effect in quadrupeds is unclear. In the current study, VWS was calculated between C2/ C3 and L1/L2, rather than between each adjacent spinal levels. Because the lower spinal levels sampled were concentrated around T10, the increased distance, and therefore pulse transit times, between these locations should improve temporal accuracy but remove detection of region-specific VWS. Further study is necessary to elucidate the relevance of VWS along the spine, and locally, in healthy and diseased states.
Defining dorsal and ventral regions in the SAS is likely to be beneficial for future CSF flow investigations in the context of SCI, where SAS occlusion may not be uniformly distributed. The current study suggests that, in these pigs, CSF flow and maximum velocity were dependent on SAS region at C2/C3. While the majority of studies do not report CSF flow in separate SAS regions, a study using dynamic sagittal PC-MRI scans on cervical myelopathy patients found that grade 2 cervical stenosis was more frequently associated with interrupted flow patterns in the ventral or dorsal SAS [27]. A study of Chiari malformation in canines reported that syringomyelia was associated with lower peak velocity in the dorsal SAS (foramen magnum and C2/C3) but not in the ventral SAS [71]. Together, these observations suggest the potential importance of considering local CSF flow characteristics. In the current study, CSF flow was generally similar, but not identical, in the two regions, and at spinal level T8/T9, four animals had no flow signal in the dorsal SAS while CSF flow was observed in the ventral SAS. There is some evidence that CSF velocity is higher ventrally during both systolic and diastolic flow, in the human cervical SAS [72]; therefore, it is possible that in these animals flow in the dorsal SAS was not detected by PC-MRI because of lower velocity. In addition, the variability of maximum CSF velocity between animals in this study suggests that multi-V ENC scans should be run in the thoracolumbar spine. Maximum CSF velocity at C2/ C3 also exceeded the applied V ENC (6 cm/s) in 70% of the scans in the first ten animals. Aliasing can be difficult to detect prior to image post-processing, therefore "realtime" post-processing should be completed immediately after the scan to allow for immediate re-scanning as needed. A V ENC of at least 8 cm/s at C2/C3 is recommended to minimise the risk of aliasing, in pigs.
There are several limitations in this study. The sample size was limited; however, the number of animals used in this study is greater than that reported for similar animal studies characterising CSF flow in NHP (macaca fascicularis; N = 8) [26], and quantifying CSF velocity in canines (beagle, N = 6) [73]. In addition, the age and sex of the animals in this study were selected because of their relevance to the associated SCI experiments; it is not known if CSF flow characteristics change with animal age, size and sex. The spinal levels selected were of particular relevance to pig SCI contusion models, and may not be generalisable to levels of interest for the study of other spinal pathologies. Blood pressure was not measured concurrently with image acquisition, so blood pressure could not be identified as a physiological contributor to variability in CSF flow. Because the SAS was split into two regions with different flow, eddy current corrections were optimised for the total net flow of both regions, rather than for each region. When performing eddy current correction, we assumed that the net flow over a single cardiac cycle was zero. Although it is understood that the majority of CSF is formed cranially and net flow must occur since it is absorbed at the same rate, this assumption is likely acceptable because net flow is very low across a single cardiac cycle. Estimated CSF formation is approximately 0.067 mL/min in the pig (based on data from 25-30 kg female sheep [74]), and therefore, the volume of CSF produced (and absorbed) during one cardiac cycle (at 100 bpm) is approximately 0.67 µL. The cause of the apparent abnormal cardiac gating in two animals could not be identified; the pulse oximeter signal was not acquired for post-processing. Although the cardiac phase numbers were adjusted for heart rate to optimise competing demands of temporal resolution and signalto-noise ratio, the minimum number of phases selected in this study was relatively low and could be increased in future studies. The temporal resolution of the acquired cardiac phases varied from 16 to 41 ms, which may have affected the temporal outcome parameters [75]. In addition, the cause of undetected CSF flow in three animals could not be identified; however, technical failures and/or insufficient CSF flow have also been reported in human PC-MRI studies [22,24,44,68].

Conclusions
This study demonstrates that CSF flow in the pig can be detected and quantified with PC-MRI using a 3T MR scanner. Spinal CSF flow in the healthy, anaesthetised, ventilated domestic pig in lateral recumbency is lower, and has a lower velocity wave propagation speed, than in conscious humans. These data are a normative baseline at spinal levels relevant for future SCI research in this animal model, and provide a means to validate computational models of the pig SAS.
Additional file 1: Figure S1. Exemplar PC-MRI A magnitude B phase images at peak diastolic flow with two ROIs in the dorsal SAS dependent on flow signal within that region (P009). Table S1. Net flow (mL/ cycle) and percent residual flow (%) for each animal and spinal level, with means ± one standard deviation. Table S2. The number of ROI and area of ROI (mm 2 ) drawn in the dorsal or ventral SAS dependent on detectable CSF flow signal within that region, where 0 corresponds to no CSF flow detected in that region. Figure S2. Experimental flow data from P004 and P012 were excluded from temporal analyses and plots due to abnormal cardiac gating. Table S3. Peak systolic and diastolic flow values (mL/s) for each animal and spinal level, with means ± one standard deviation. Table S4. Physiological recordings and anaesthetic drug levels of each animal during the PC-MRI scan. Table S5. Stroke volumes (mL/cycle) for each animal and spinal level, with means ± one standard deviation. Table S6. Estimated marginal means with 95% CI for peak diastolic and systolic flow, and maximum cranial and caudal velocity from LMMs. Table S7. Summary of results from the LMM pairwise comparisons for peak systolic and diastolic flow and maximum cranial and caudal velocity. Figure S3. Mean CSF velocity in the dorsal and ventral region of the SAS at each spinal level. Table S8. Time to peak systolic and diastolic velocity (as a percentage of the cardiac cycle) and cardiac cycle duration (ms) for each animal and spinal level. Table S9. Peak mean systolic and diastolic velocity values (mL/s) for each animal and spinal level, with means ± one standard deviation.