‘True’ transient ischaemic attacks are characterized not only clinically, but also radiologically by a lack of corresponding changes on magnetic resonance imaging. During a transient ischaemic attack it is assumed that the affected tissue is penumbral but rescued by early spontaneous reperfusion. There is, however, evidence from rodent studies that even brief focal ischaemia not resulting in tissue infarction can cause extensive selective neuronal loss associated with long-lasting sensorimotor impairment but normal magnetic resonance imaging. Selective neuronal loss might therefore contribute to the increasingly recognized cognitive impairment occurring in patients with transient ischaemic attacks. It is therefore relevant to consider treatments to reduce brain damage occurring with transient ischaemic attacks. As penumbral neurons are threatened by markedly constrained oxygen delivery, improving the latter by increasing arterial O 2 content would seem logical. Despite only small increases in arterial O 2 content, normobaric oxygen therapy experimentally induces significant increases in penumbral O 2 pressure and by such may maintain the penumbra alive until reperfusion. Nevertheless, the effects of normobaric oxygen therapy on infarct volume in rodent models have been conflicting, although duration of occlusion appeared an important factor. Likewise, in the single randomized trial published to date, early-administered normobaric oxygen therapy had no significant effect on clinical outcome despite reduced diffusion-weighted imaging lesion growth during therapy. Here we tested the hypothesis that normobaric oxygen therapy prevents both selective neuronal loss and sensorimotor deficits in a rodent model mimicking true transient ischaemic attack. Normobaric oxygen therapy was applied from the onset and until completion of 15 min distal middle cerebral artery occlusion in spontaneously hypertensive rats, a strain representative of the transient ischaemic attack-prone population. Whereas normoxic controls showed normal magnetic resonance imaging but extensive cortical selective neuronal loss associated with microglial activation (present both at Day 14 in vivo and at Day 28 post-mortem) and marked and long-lasting sensorimotor deficits, normobaric oxygen therapy completely prevented sensorimotor deficit ( P < 0.02) and near-completely Day 28 selective neuronal loss ( P < 0.005). Microglial activation was substantially reduced at Day 14 and completely prevented at Day 28 ( P = 0.002). Our findings document that normobaric oxygen therapy administered during ischaemia nearly completely prevents the neuronal death, microglial inflammation and sensorimotor impairment that characterize this rodent true transient ischaemic attack model. Taken together with the available literature, normobaric oxygen therapy appears a promising therapy for short-lasting ischaemia, and is attractive clinically as it could be started at home in at-risk patients or in the ambulance in subjects suspected of transient ischaemic attack/early stroke. It may also be a straightforward adjunct to reperfusion therapies, and help prevent subtle brain damage potentially contributing to long-term cognitive and sensorimotor impairment in at-risk populations.
According to their new tissue-based definition, transient ischaemic attacks (TIAs) are characterized both clinically by focal neurological symptoms of ischaemic origin lasting <24 h, and radiologically by a lack of topographically congruent changes on diffusion-weighted imaging (DWI) or fluid-attenuated inversion recovery (FLAIR) MRI performed within 2 weeks of the clinical event ( Saver, 2008 ). Such ‘true’ TIAs account for 20–50% of all clinically-diagnosed TIAs ( Brazzelli et al. , 2014 ). The vast majority of TIAs are believed to be secondary to transient arterial occlusion from an upstream blood clot with rapid spontaneous fibrinolysis. During a TIA it is assumed that tissue perfusion is reduced below the penumbra threshold, causing the focal symptoms, but that early spontaneous reperfusion salvages the penumbra in whole or in large part, causing full clinical recovery within 24 h either with no MRI sequelae or with a very small asymptomatic DWI lesion that sometimes disappears secondarily ( Oppenheim et al. , 2006 ).
There is however recent evidence from rat studies that even brief middle cerebral artery (MCA) occlusion not resulting in tissue infarction can cause variably extensive selective neuronal loss (SNL) affecting both the striatum and the cortex or only the latter in case of proximal or distal occlusion, respectively (see Baron et al. , 2014 for review), associated with significant and long-lasting sensorimotor impairment ( Sicard et al. , 2006 ; Ejaz et al. , 2015 a ). SNL is characterized by patchy and/or layer-wise partial loss of neurons without loss of glial cells, other tissue constituents and extracellular matrix, and with normal standard MRI after the ischaemic event, and as such strikingly differs from infarction ( Baron et al. , 2014 ). Although there is no direct evidence that it occurs after a TIA in humans, SNL has been documented in patients with full/nearly-full clinical recovery following early reperfusion for severe stroke ( Saur et al. , 2006 ; Guadagno et al. , 2008 ; Carrera et al. , 2013 ), as well as in patients with permanent internal carotid artery occlusion or MCA occlusion and repeated TIAs ( Yamauchi et al. , 2007 , 2009 , 2011 ). It has been suggested that SNL, which is consistently accompanied by topographically congruent microglial activation lasting several weeks ( Baron et al. , 2014 ), may contribute to the increasingly recognized subtle cognitive impairment occurring in patients with TIAs, which can be temporary or long lasting ( Pendlebury et al. , 2011 ; Sivakumar et al. , 2014 ; van Rooij et al. , 2014 ). It has thus been suggested that accumulating SNL resulting from symptomatic (i.e. TIAs) or asymptomatic small emboli (e.g. lodging in non-eloquent brain areas) reduces the cognitive reserve in such patients, possibly ultimately contributing to cognitive decline, in association or not with mild Alzheimer pathology ( Baron et al. , 2014 ). If this is correct, TIAs should no longer be just considered as potential heralds of imminent stroke ( Rothwell et al. , 2011 ), but also as events causing ‘hidden’ tissue damage and hence facilitating permanent behavioural impairment in the longer-term. Given this paradigm shift, it is in turn important to consider treatments aiming to reduce brain damage from TIAs.
Given the inference of spontaneous recanalization, preventing tissue damage from TIAs should target penumbral protection until reperfusion occurs, so-called ‘penumbra freezing’ ( Fisher and Saver, 2015 ). Neurons in the penumbra being at risk from constrained oxygen delivery, increasing arterial O 2 content seems a logical approach. Hyperbaric oxygen as a potential therapy for ischaemic stroke has been tested in 11 small-scale randomized controlled trials but shown in one recent meta-analysis to have no significant benefit on mortality and no consistent effect on functional scales ( Bennett et al. , 2014 ). Accordingly, normobaric oxygen therapy (NBO) has been tested both in rodent models and in small-scale clinical trials. Critically, the extremely low levels of tissue oxygen pressure in the penumbra ( Crockard et al. , 1976 ; Harris et al. , 1987 ; Hoffman et al. , 1996 , 1999 ; Liu et al. , 2004 ; Hou et al. , 2007 ; Shin et al. , 2007 ; Baskerville et al. , 2011 ) are significantly improved by NBO despite the only mild increase in total arterial O 2 content ( Liu et al. , 2004 , 2006 ; Shin et al. , 2007 ; Baskerville et al. , 2011 ). In turn, this could maintain the penumbra alive until reperfusion occurs. Accordingly, beneficial effects of NBO on infarct volume have been repeatedly reported in rodent models ( Flynn and Auer, 2002 ; Singhal et al. , 2002 a ; Kim et al. , 2005 ; Liu et al. , 2006 ; Henninger et al. , 2007 ; Shin et al. , 2007 ; Esposito et al. , 2013 ; Jin et al. , 2013 ). However, in some studies the beneficial effect was only marginal ( Henninger et al. , 2007 , 2009 ), while lack of benefit has also been repeatedly reported ( Kim et al. , 2005 ; Beynon et al. , 2007 ; Hou et al. , 2007 ; Fujiwara et al. , 2009 ; Liang et al. , 2015 ). Consequently, there is currently no clear consensus on the benefits of NBO in rodent stroke models, although one caveat when critically assessing this literature is lack of, or too delayed, reperfusion in some reports. Indeed, it has recently become clear that significant protection of the penumbra cannot be efficient unless early reperfusion also occurs ( Fisher, 2011 ). This factor may in turn explain the limited benefit from early-administered NBO on 3-month clinical outcome despite reduced DWI lesion growth during therapy in the single randomized trial published to date ( Singhal et al. , 2005 ). Thus, testing the effects of NBO in very brief MCA occlusion mimicking true TIA is worth considering, but has not been reported so far. NBO is a particularly attractive therapy because it is clinically translatable and could be started at home or in the ambulance in appropriate patients.
Applying this straightforward mechanistic reasoning, we therefore carried out a translational study in spontaneously hypertensive rats, a strain representative of the TIA-prone population, specifically to test the hypothesis that NBO given right from the onset of brief MCA occlusion mimicking true TIA and until recanalization prevents SNL and sensorimotor impairment, and associated tissue inflammation. This was designed as a pragmatic, clinically-orientated study assessing these three major variables only, not the whole gamut of possible tissue damage subtypes and behavioural processes.
To this end, we assessed here the effects of NBO on SNL—a temporally stable process—by means of NeuN immunofluorescence at 28 days, while the effects on microglial activation—a time-dependent process ( Benavides et al. , 1990 ; Hu et al. , 2012 )—were assessed at two time points in the same animals, first at Day 14 in vivo using 11 C-PK11195, a reliable PET tracer for mapping microglial activation after temporary MCA occlusion in the rat ( Hughes et al. , 2012 ), and again post-mortem at Day 28 using IB4 immunofluorescence. Sensorimotor effects were assessed serially using both simple neurological observation and a behavioural task sensitive to subtle impairments. Finally, standard MRI sequences were obtained both immediately after reperfusion and at the time of killing.
Materials and methods
This study was approved by the University of Cambridge Ethical Review Panel. In accordance with the legislation of UK Animals Scientific Procedures Act 1986, the Ethical Review Board required that the study be designed so as to keep the number of animals used to a minimum, yet sufficient to obtain meaningful results. Accordingly, we estimated that the study should involve 12 subjects, six in the normoxia control group and six in the hyperoxia group. All surgical and imaging procedures were performed on anaesthetized ∼3–6-month-old male spontaneously hypertensive rats (∼300 g body weight). Subjects were randomized into the two groups, and all the animals used in this study will be reported below. The findings in the control group, which also served to validate our previously reported model of pure cortical SNL in spontaneously hypertensive rats ( Ejaz et al. , 2015 b ), have been reported ( Ejaz et al. , 2015 a ), showing normal in vivo MRI, sensorimotor impairment still incompletely recovered 4 weeks after the insult, and cortical SNL and microglial activation but no infarcts at post-mortem ( Ejaz et al. , 2015 a ).
Experiments were performed in freely breathing animals. Anaesthesia was induced with 4% isoflurane administered in a 0.3 l/min O 2 and 0.7 l/min N 2 O mix and maintained with 2% isoflurane during surgical procedures. Body temperature of the animals was monitored with a rectal probe and maintained at 37.0 ± 0.5°C using a heated pad throughout all surgical procedures. Blood oxygen saturation and heart beat were continuously monitored using a pulse-oximeter and remained within normal ranges throughout. In the hyperoxia group, the gas mixture administered via a nose cone was switched to 1 l/min pure oxygen on MCA occlusion time (see below).
Middle cerebral artery occlusion
Microclip distal temporary MCA occlusion was performed using the method described by Buchan et al. (1992) as implemented in our laboratory and detailed previously ( Supplementary material ) ( Takasawa et al. , 2007 , 2011 ; Hughes et al. , 2010 , 2012 ; Ejaz et al. , 2013 , 2015 a ). The MCA clip was removed after 15 min and the wound was closed. MCA reperfusion, visually evident on clip removal as immediate reflow distal to the clip ( Hughes et al. , 2010 ), was present in all 12 rats of this study.
In the hyperoxia treated group, the O 2 flow was increased to 1 l/min with no N 2 O at the same time as MCA occlusion and was maintained until clip release, while the protocol was left unchanged (0.3 l/min O 2 and 0.7 l/min N 2 O mix) for the animals of the normoxia group.
11 C-PK11195 PET
At Day 14 post-MCA occlusion, all 12 rats underwent 11 C-[R]-PK11195 PET applying the same procedures as previously reported ( Hughes et al. , 2012 ), except for using an improved scanner ( Supplementary material ).
Magnetic resonance imaging
MRI was carried out at Day 14 of reperfusion immediately after PET scanning, and included T 2 -weighted and DWI sequences, as previously detailed ( Ejaz et al. , 2015 a ) ( Supplementary material ). Given the previously reported 100% recanalization rate on magnetic resonance angiography in temporary distal microclip MCA occlusion as implemented in our lab ( Ejaz et al. , 2013 , 2015 a ), magnetic resonance angiography was not performed in this study.
Animals were single-housed on a 12-h light/dark cycle and had free access to water and standard rodent chow. Training/testing was performed in the light phase and animals were left in their housing cages during sessions. Animals received daily handling for at least 4 days before baseline testing to ensure accurate behavioural results. Neurological examination was carried out the day before surgery and at postoperative Days 1, 7, 14, 21, and 28 using Garcia’s Neuroscore. The modified Sticky Label Test was performed 1 day before surgery and postoperatively on Days 1, 3, 7, 11, 14, 18, 21, 25, and 28, blind to the subject’s belonging to the control or NBO group.
Garcia’s Neuroscore consists of motor, sensory, reflex, and observational tests to evaluate neurological deficits following MCA occlusion in rats ( Garcia et al. , 1995 ). It is scored on a scale from 3 to 18 (normal: 18; maximal deficit: 3), i.e. the lower the score the worse the deficit.
Subtle sensorimotor dysfunction following MCA occlusion was assessed using the modified Sticky Label test as previously described ( Sughrue et al. , 2006 ; Komotar et al. , 2007 ). This test is sensitive to subtle ischaemic damage even when the Neuroscore is normal ( Sicard et al. , 2006 ; Freret et al. , 2009 ) and, contrary to the standard version, uses a non-removable tape. As a result, non-stroked rats spend most of the 30-s sessions trying to remove it, with no habituation effect over time ( Sughrue et al. , 2006 ; Komotar et al. , 2007 ). A small patch of paper tape (2.5-cm long, 1.0-cm wide) is wrapped around the animal’s wrist contralateral to the ischaemic insult such that the tape sticks to itself and that the fingers protrude from the sleeve formed. The rat is placed in its home cage and the time spent attending to the stimulus, be it using the teeth or contralateral paw, is recorded. Animals are given five sessions per day, each observation period lasting for a maximum of 30 s. After each trial the tape is removed and animals are given a resting time of ≥3 min. Modified Sticky Label Test performance is calculated by dividing the time attending to the stimulus by 30 s, expressing the fraction of the observation period that the animal spends attending to the tape. The best two ratios on each day were averaged. To reduce noise, all daily time points for each week were collapsed into a single value. The results of the final day of presurgery training served as baseline for assessment of post-MCA occlusion performance. Following stroke, sensorimotor deficits make the rat spend less time attending to the tape than normal ( Freret et al. , 2009 ). Deficits on the SLT are thought to reflect a mix of subtle sensorimotor impairments including hemi-sensory neglect and extinction, forepaw motor deficit and impaired somatosensory integration ( Schallert et al. , 1982 ; Komotar et al. , 2007 ; Bouet et al. , 2009 ; Freret et al. , 2009 ).
On Day 28, the subject was perfusion-fixed, and brains removed, fixed, coronally sectioned and stained with NeuN and IB4 as described in detail elsewhere ( Ejaz et al. , 2013 ; Williamson et al. , 2013 ) ( Supplementary material ).
Evaluation of ischaemic damage
To quantitate the extent of ischaemic changes, we used the method described in detail elsewhere ( Ejaz et al. , 2015 a ). Briefly, on each digitized whole-brain section, two independent raters blinded to the subject’s experimental group manually traced any area with lack of NeuN-immunoreactivity or increased IB4 binding, using a computer mouse. For this analysis, eight coronal sections spanning the MCA territory were selected at bregmas 2.70 mm, 1.00 mm, −0.26 mm, −0.92 mm, −2.12 mm, −3.14 mm, −4.52 mm and −6.04 mm, and the sections for the different rats were presented in randomized order across groups (treated and controls) to the blind observers. ImageJ software was used to measure the cross-sectional surface area enclosed within the traced regions of interest on each section. Lesion volumes (mm 3 ) were then calculated using Cavalieri’s principle ( Cotter et al. , 1999 ), and summed across all regions of interest for each rat (to be referred to as ‘Vol-total’ below). As previously detailed ( Ejaz et al. , 2015 a ), the inter-rater agreement for lesion volume determination between the two independent raters was excellent, indicating strong reliability. Based on this, consensus regions of interest were agreed.
PET data postprocessing
Parametric images of 11 C-PK11195 non-displaceable binding potential (BP ND ) were produced using the basis function version of the simplified reference tissue model ( Gunn et al. , 1997 ). The ipsilateral cerebellum, manually defined on a symmetric spontaneously hypertensive rat MRI template ( Hughes et al. , 2012 ) and inverse warped to each individual T 2 -weighted MRI using SPM5 ( www.fil.ion.ucl.ac.uk/spm ), was used as the reference tissue ( Gerhard et al. , 2005 ; Price et al. , 2006 ). Each individual T 2 -weighted MRI was warped to the MRI template using SPM5 and this transformation was applied to the co-registered BP ND map to bring it to template space for regional analysis. Note that this method may generate negative BP ND values, which simply represent lower specific binding than in cerebellum.
An automated method for data analysis was then applied to the entire dataset without knowledge of the subjects’ group until after statistical analysis was complete. To this end, regions of interest delineated on the magnetic resonance template were used to obtain 11 C-PK11195 BP ND values from the BP ND maps ( Fig. 3 ). The eight coronal cuts of the MRI template matching as precisely as possible the eight Paxinos sections used for the immunofluorescence image analysis were first selected. Then, a set of 44 symmetrical regions of interest defined according to Paxinos and Watson’s (1996) cytoarchitectonic atlas covering the grey matter of the eight coronal sections (per hemisphere: 39 cortical regions of interest, four caudate/putamen regions of interest and one thalamic region of interest) were used, as detailed previously ( Ejaz et al. , 2013 ; Hughes et al. , 2010 , 2012 ). This region of interest template was then projected onto the corresponding slices of the co-registered BP ND maps to obtain the mean BP ND value for each region of interest. Then, for each pair of symmetrical regions of interest the (affected – unaffected) difference in BP ND was calculated. A mean ± standard deviation (SD) (affected – unaffected) BP ND difference across all 44 regions of interest, weighted by the region of interest volume, was then calculated for each rat and then across the six rats of each group. Finally, we also computed the mean BP ND difference for each region of interest across the six rats.
Statistical analyses were performed using SPSS (version 15, SPSS Inc., Chicago, USA) software. The modified Sticky Label Test data were analysed using one-way repeated measures ANOVA within each group assessing the effects of Time, followed if significant by Holm-Bonferroni post hoc tests corrected for multiple tests to assess each time-point relative to baseline. Then, a two-way repeated measures ANOVA was used to compare the NBO to the normoxic group, with again post hoc tests if a significant difference emerged. Regarding the PET data, the weighted mean BP ND (affected – unaffected) difference within each group was compared to neutral (i.e. zero, assuming no difference in PK11195 binding between the two hemispheres) by paired t -test, and then the two groups were compared using two-sample t -test. Also, the weighted mean BP ND difference across the 44 regions of interest for each rat was tested against neutral to determine individual statistical significance. Finally, each region of interest’s mean BP ND across the six rats was tested against neutral to assess the particular region of interest’s statistical significance within each group. Regarding the immunofluorescence data, the presence of lesions on NeuN or IB4 was tested both within- and between-groups using non-parametric tests, while lesion volumes were tested using t -tests. Pearson’s correlation coefficient was used to assess for relationships between modified Sticky Label Test performance and NeuN- or IB4-labelled lesion volumes. Results were considered statistically significant if two-tailed P < 0.05.
Arterial PO 2 measurements
In addition to the above experiments, and to determine the effects of NBO on PaO 2 levels, PaO 2 was measured in two additional spontaneously hypertensive rats in both the normoxic and hyperoxic conditions. PaO 2 was 121.5 mmHg (± 16; n = 9 measurements) and 478.5 mmHg (± 18; n = 6) in normoxia and hyperoxia, respectively, i.e. a ∼4-fold increase during NBO. Corresponding values for SaO 2 were 98.7 % (± 0.5%) and 100% (± 0%), respectively.
All 12 rats entered in the study completed the protocol without any complication or early death until the 28-day termination.
MRI: Day 14
No DWI or T 2 -weighted changes were observed in any rat ( Fig. 5 ).
Neurological score and behavioural testing
The Neuroscore was zero, i.e. normal, in both groups at all times, indicating no detectable neurological impairment.
Modified Sticky Label Test
The modified Sticky Label Test data and detailed statistical findings are shown in Fig. 1 . Following temporary MCA occlusion, control animals demonstrated highly significant ( P < 0.0001, repeated measures ANOVA) impairment in modified Sticky Label Test performance still present but recovering at the last assessment at 28 days. In contrast, NBO animals did not demonstrate any detectable modified Sticky Label Test deficit (nil significant, repeated measures ANOVA). Compared to controls, the NBO group was significantly less impaired for all post-MCA occlusion time-points ( P = 0.02). On post hoc tests, there was no significant difference between the two groups at baseline, but the performance was significantly better ( P range: 0.045–0.015) in the hyperoxia group for all time points until the end of the experiment.
11 C-PK11195: Day 14
In the control group, the individual weighted mean BP ND (affected – unaffected) difference ranged from +0.033 to +0.074, with a mean of +0.052 ± 0.015 ( n = 6), significantly higher than neutral ( t = 6.054, P < 0.005). In the hyperoxia group, the weighted mean BP ND (affected – unaffected) difference ranged from +0.018 to +0.075, with a statistically significant weighted mean difference, though less so than controls (+0.040 ± 0.023; t = 3.932, P < 0.02), with no significant difference between the two groups ( t = 1.078, P = 0.31). Figure 2 illustrates these data.
When tested individually for a significant difference between the two hemispheres, 5/6 rats were individually significant in the control group versus 3/6 in the hyperoxia group, again pointing to a small difference in favour of NBO. Likewise, when testing individual regions of interest for significant difference between the two hemispheres across the six rats, 19/44 were significant in the control group versus 4/44 in the hyperoxia group (located in the insula and primary somatosensory area), a significantly reduced proportion ( P = 0.002; χ 2 ). Overall therefore, there was significant though mild microglial activation in the affected hemisphere in both groups at Day 14, with a clear trend for reduced microglial activation in the hyperoxia group.
Figure 3 illustrates typical 11 C-PK11195 images in one control and one hyperoxia rat, together with the corresponding MRI and region of interest templates.
Immunofluorescence: Day 28
Abnormal areas were delineated by the two observers exclusively in the MCA occlusion hemisphere. In the affected hemisphere, areas showing loss of NeuN or high IB4 were consistently topographically congruent and were present in 5/6 normoxic animals. In contrast, in NBO rats a very small patch of absent NeuN binding was present in 1/6 rats, and increased IB4 areas in 0/6 rats. This difference in occurrence of lesions between the two groups was statistically different for IB4, and showed a trend for NeuN ( P = 0.015 and 0.08, respectively; Fisher’s test).
Regarding lesion volumes, the mean (± SD) Vol-Total values for NeuN and IB4 in controls were 1.41 ± 1.79 mm 3 and 1.33 ± 1.7 mm 3 , respectively, and 0.01 ± 0.02 mm 3 and 0.0 ± 0.0 mm 3 in NBO rats, respectively, a highly significant difference from the control group ( P = 0.005 and 0.002 for NeuN and IB4, respectively). Figure 4 illustrates these findings. Overall, therefore, these results indicate a near-total and total lack of SNL and microglial activation, respectively, at 28 days in the hyperoxia-treated group.
Figure 5 illustrates the NeuN and Ib4 immunofluorescence findings in one control and one hyperoxic rat, also showing their individual T 2 -weighted MRI (coronal sections) at 28 days, which as already stated, was normal in both subjects.
Relationship between sensorimotor scores and selective neuronal loss
To assess if the behavioural impairment in this model is related to cortical SNL, we examined the relationship between NeuN total lesion volume and modified Sticky Label Test performance across the control and NBO groups ( n = 12 rats). There was a significant correlation between NeuN Vol-Total data and peak modified Sticky Label Test deficit, which was relatively weak (τ = 0.47, P < 0.05) but in the expected biological direction, i.e. a greater behavioural deficit with greater neuronal loss ( Supplementary Fig. 1 ). Similar findings emerged for IB4 (data not shown).
One strength of this proof-of-principle mechanistic study is that each rat underwent an extensive protocol involving serial sensorimotor assessments, Day 14 PET and full post-mortem immunofluorescence, affording considerable power. Accordingly, despite the relatively small samples, the striking between-group differences in both tissue damage and sensorimotor deficit seem beyond chance finding.
This is the first study to document that, as per our hypothesis, NBO applied during very brief MCA occlusion mimicking ‘true’ TIA is able to almost completely prevent selective neuronal death. Consistent with this neuronal protection, NBO completely prevented the marked sensorimotor deficits present in normoxic control rats. Regarding microglial activation, another sequelae of ischaemic injury that may have deleterious effects on brain function and outcome independently of SNL ( Terasaki et al. , 2014 ), it was partly prevented on Day 14 based on PET data, and completely abolished at Day 28 based on post-mortem assessment in the same subjects. This was confirmed post hoc by lack of significant microglial activation on formal cell counting in the four regions of interest showing significantly increased PK11195 binding at Day 14 (data not shown). Thus, in contrast to normoxic rats, the mild microglial activation still present at Day 14 in NBO rats had completely resolved on post-mortem at Day 28, indicating NBO accelerated the expected post-injury resolution of microglial activation. Such a mild and transient microglial activation without SNL or sensorimotor counterpart in the NBO-treated rats may be explained for instance by isolated dendritic damage known to be present following brief MCA occlusion ( Tomimoto and Yanagihara, 1992 ; Zhang et al. , 2005 ). Overall, therefore, NBO was able to almost entirely prevent both the tissue damage and sensorimotor deficits characterizing control rats.
Before addressing our findings in detail, some methodological points deserve a brief note. First, NBO increased arterial PO 2 nearly 4-fold in our model, similar to data reported by other groups ( Kim et al. , 2005 ; Beynon et al. , 2007 ; Fujiwara et al. , 2009 ; Baskerville et al. , 2011 ; Sun et al. , 2011 ). Second, we used spontaneously hypertensive rats as this strain is particularly relevant to the population at risk of TIA, because of chronic hypertension-related cerebrovascular changes such as increased arteriolar stiffness, impaired autoregulation and less efficient pial anastomoses ( Coyle and Heistad, 1986 ; Amenta et al. , 2010 ; Leoni et al. , 2011 ). Because of these changes, spontaneously hypertensive rats have higher vulnerability to brain ischaemia ( Coyle, 1986 ; Duverger and Mackenzie, 1988 ; McCabe et al. , 2009 ; Letourneur et al. , 2012 ), and might therefore benefit more from NBO than their normotensive counterparts. It will therefore be important to assess whether NBO has the same beneficial effects in normotensive rats as in spontaneously hypertensive rats in TIA models. Third, that 11 C-PK11195 PET was able to detect in vivo the small, albeit highly significant, amount of microglial activation present in this TIA model indicates it has excellent sensitivity, expanding on our previous findings in a 45 min MCA occlusion model ( Hughes et al. , 2012 ). Fourth, although PK11195 has affinity for activated astrocytes in vitro , this is ∼5-fold lower than for microglial cells ( Banati, 2003 ), is hardly detectable on post-mortem brain tissue ( Banati et al. , 1997 ; Venneti et al. , 2009 ; Maeda et al. , 2011 ) including after focal ischaemia ( Myers et al. , 1991 ; Stephenson et al. , 1995 ; Rojas et al. , 2011 ), and seems even lower in vivo especially when using the R-enantiomer as done here ( Banati, 2003 ). Accordingly, most authorities are of the opinion that in vivo11 C-PK11195 binding almost exclusively maps to activated microglia/macrophages ( Stephenson et al. , 1995 ; Banati, 2003 ). Partly unpublished data from our previous study comparing 11 C-PK11195 to OX42 immunohistochemistry (a selective microglial activation marker), both obtained 14 days following 45 min MCA occlusion in spontaneously hypertensive rats ( Hughes et al. , 2012 ), further support this idea. Thus, 11 C-PK11195 binding highly significantly (r 2 = 0.656, P < 0.0001) correlated to OX42 staining density, but less strongly so (r 2 = 0.541) to GFAP (a marker of astrocytosis; both immunostains obtained from same subjects and regions of interest), and the latter correlation was not significant anymore (r 2 = 0.060, P = 0.115) after adjusting for OX42 binding, indicating it was largely due to the underlying strong OX42–GFAP topographical and statistical relationship (r 2 = 0.646; conversely, the 11 C-PK11195 versus OX42 correlation remained significant after adjusting for GFAP: r 2 = 0.294, P < 0.0001). Taken together, all of the above therefore strongly suggest that our findings of increased 11 C-PK11195 binding at Day 14 in both the normoxic and hyperoxic rats almost exclusively reflect microglial activation. Fifth, as detailed in the Introduction, this was a pragmatic translational study designed to test a simple and clinically-orientated question, namely whether NBO is able to prevent the SNL and sensorimotor impairment as well as the associated microglial activation that characterize our rodent TIA model. Accordingly, we did not investigate the whole spectrum of potential tissue and behavioural changes, and specifically cannot exclude that NBO may not for instance prevent mild astrocytosis, damage to dendritic trees and/or subtle cognitive impairment. Likewise, to quantitatively assess the effects of NBO on SNL and microglial activation across the ischaemic territory, we applied our previously validated template of eight 0.9 mm-apart coronal sections covering the MCA territory ( Hughes et al. , 2010 , 2012 ; Ejaz et al. , 2013 , 2015 b ), which based on independent visual assessment by two experienced readers blinded to subject group, disclosed a highly significant reduction in both SNL and microglial activation in NBO-treated versus normoxic rats. Again, we cannot exclude that some additional pathology may have been present in intermediate slices. However, post hoc assessment of intermediate coronal slices revealed no additional SNL or microglial activation in the treated rats, further supporting our findings.
To date, NBO has not gained widespread recognition as a promising therapy for acute stroke. This likely relates to the observation that in rodent models it does not invariably result in significant infarct volume reduction and alleviation of behavioural impairments, an impression reinforced by the largely negative single pilot randomized clinical trial of early NBO therapy, reporting transient clinical and imaging improvements but no benefit on 3-month outcome ( Singhal et al. , 2005 , 2007 ; Gonzalez et al. , 2010 ). Furthermore, the only other trial of NBO, the Stroke Oxygen Supplementation Study, was also negative ( Ali et al. , 2014 ; Roffe, 2014 ), although it has to be noted that its aim was not to prevent acute damage but to compensate for the hypoxic episodes that may occur in the first days following stroke, and accordingly NBO was started within 24 h of onset and strokes of any type, including haemorrhagic, were recruited. However, as will be seen below, in-depth scrutiny of the literature in fact reveals that the marginal overall benefit in rodent models is partly due to heterogeneity of effects with duration of the ischaemic insult. Thus, rather than adding to the confusion, our strongly positive findings with a brief MCA occlusion are largely consistent with the literature, and serve to emphasize the key point that the benefits from NBO are to be expected only in situations of rapidly occurring reperfusion.
Of the three studies that investigated the effects of NBO on infarct volume in permanent MCA occlusion, two found no significant effect ( Singhal et al. , 2002 a ; Kim et al. , 2005 ) and one a marginal though significant effect ( Henninger et al. , 2007 ). Likewise, no consistently significant effect of NBO was observed in rats subjected to ≥ 4 h temporary MCA occlusion ( Kim et al. , 2005 ; Liang et al. , 2015 ), which is considered equivalent in terms of infarct size to permanent occlusion. Similarly, using an embolic MCA occlusion model causing stable ischaemia, NBO started 30 min or 1 h after occlusion did not reduce infarct size ( Fujiwara et al. , 2009 ) or did so, but only marginally ( Henninger et al. , 2009 ). In contrast with this universal lack of significant benefit from NBO in permanent or long-lasting occlusion, NBO showed equipoise in rats subjected to 2 or 3 h MCA occlusion, with three studies reporting beneficial effects ( Singhal et al. , 2002 b ; Kim et al. , 2005 ; Henninger et al. , 2007 ) and three no effect ( Beynon et al. , 2007 ; Hou et al. , 2007 ; Liang et al. , 2015 ). Finally, all six published studies with occlusion times of 60 to 90 min reported significant infarct volume reductions, narrowly ranging from ∼35 to 50% ( Flynn and Auer, 2002 ; Kim et al. , 2005 ; Shin et al. , 2007 ; Esposito et al. , 2013 ; Jin et al. , 2013 ). Our study is the first to report the effects of NBO for occlusion times <60 min. That the effects were near-complete in our model would be in line with the above infarct volume reductions with MCA occlusion duration of 60–90 min. Interestingly, a similar dependence of effects from hyperbaric oxygen therapy on MCA occlusion duration has also been reported ( Lou et al. , 2004 ).
These benefits from NBO with transient MCA occlusion ≤3 h, and particularly ≤90 min, are in fact entirely consistent with an extensive literature reporting strongly positive effects of NBO on early markers of tissue hypoxia/ischaemia. As already noted, in rodent MCA occlusion models NBO has been shown to immediately and markedly increase the very low tissue oxygen levels prevailing in the penumbra in normoxic conditions ( Liu et al. , 2004 , 2006 ; Shin et al. , 2007 ; Baskerville et al. , 2011 ). Contrary to previous beliefs, NBO does not induce vasoconstriction in the ischaemic brain ( Singhal et al. , 2002 a ; Henninger et al. , 2007 ) and may even mildly improve penumbral perfusion ( Shin et al. , 2007 ; Baskerville et al. , 2011 ). Consistent with these effects, NBO improves ATP stores ( Sun et al. , 2011 ), ‘freezes’ the perfusion/diffusion mismatch and DWI lesion growth ( Singhal et al. , 2002 a ; Henninger et al. , 2007 ), and dampens ischaemic tissue acidosis ( Sun et al. , 2011 ), without increasing oxidative stress or tissue levels of oxygen radicals or metalloproteinase-9 ( Singhal et al. , 2002 b ; Kim et al. , 2005 ; Liu et al. , 2006 ). NBO also reduces the frequency of peri-infarct depolarizations ( Shin et al. , 2007 ), which strongly contribute to the demise of the penumbra. Finally, consistent with our findings, Esposito et al. (2013) reported that NBO reduced the expression of IBA1, a marker of microglial activation, and improved sensorimotor function 14 days after 100 min transient MCA occlusion.
Our findings suggest that NBO administered from the onset of a TIA, or by extension in subjects who recanalize early after stroke onset—be it spontaneously or following reperfusion therapy—could prevent or limit brain damage. This contention is supported by findings from the previously mentioned clinical trial that reported that, relative to normoxia, NBO prevented initial DWI lesion growth and occasionally resulted in lesion reversal; maintained the DWI/PWI mismatch; and prevented early tissue lactate rises, together with transient clinical improvement ( Singhal et al. , 2005 , 2007 ; Gonzalez et al. , 2010 ; Wu et al. , 2012 ). That NBO did not result in long-term functional benefits in this small-scale trial on patients not eligible for recanalization therapies may result from the fact that none of the nine NBO-treated patients spontaneously reperfused based on a 4-h follow-up MRI ( Singhal et al. , 2005 ). As already pointed out, therapies to ‘freeze’ the penumbra ( Fisher and Saver, 2015 ) can be truly beneficial only if reperfusion occurs early enough ( Fisher, 2011 ), and NBO is no exception to this general rule.Whether SNL and microglial activation are present in humans after a single true TIA is not known, but both processes have been reported in acutely penumbral but eventually surviving peri-infarct areas in stroke patients enjoying early recanalization and good-to-excellent clinical recovery ( Price et al. , 2006 ; Saur et al. , 2006 ; Guadagno et al. , 2008 ; Morris et al. , 2012; Carrera et al. , 2013 ). Furthermore, in such patients SNL was found to effectively interfere with neuronal function in the chronic stage ( Carrera et al. , 2013 ). Furthermore, cortical SNL without associated magnetic resonance changes prevails in patients with chronic symptomatic carotid or MCA occlusion/severe stenosis ( Yamauchi et al. , 2007 , 2009 , 2011 ), where TIAs are thought to be predominantly caused by embolic events exacerbating permanent haemodynamic insufficiency ( Moustafa et al. , 2010 , 2011 ). Overall, our findings are relevant to TIA and quickly recovering stroke, and NBO could be an important new therapeutic approach in this clinical scenario. In turn, by preventing subtle brain damage, NBO could maintain the brain’s ‘cognitive reserve’ and reduce the risk of eventual cognitive decline in this category of patients. Temporary ( Pendlebury et al. , 2011 ; Sivakumar et al. , 2014 ) or persistent ( van Rooij et al. , 2014 ) cognitive deficits have been reported in TIA patients, who appear particularly at risk of cognitive decline in the longer term.
Regarding potential clinical translation, one major point raised by our findings is: How early should NBO be started after TIA onset? In the present proof-of-concept study, we opted to start NBO as soon as the occlusion was completed, in order to maximize any potential benefit from this therapy. Although it is not unrealistic that, if properly trained, patients at high risk of TIA or their relatives carry a small oxygen bottle and use it at once in case a TIA occurs, in the real world at least a few minutes would elapse before oxygen therapy could be started. Accordingly, it would be critical to assess in rodent TIA models the effects of NBO started, for example, 5–10 min post-MCA occlusion. Few studies have so far looked at the effects of NBO as a function of time of administration. Using a 2 h MCA occlusion rat model, Singhal et al. (2002 a ) reported declining benefits from NBO when started at 30 min post-MCA occlusion relative to 15 min, and no benefit if started at 45 min. These findings not only emphasize the importance of the timing of NBO but also indicate that, consistent with our findings, the benefits from NBO in the clinical setting are likely to be significant only if administered within 45 min of clinical onset, and the earlier the larger the expected effects. Another point worth noting regarding our study’s clinical implications regards the intravenous administration of new-generation perfluorocarbons, which are reported to enhance oxygen content within the penumbra over and above NBO ( Deuchar et al. , 2013 ), as well as reduce final infarct size if started immediately ( Woitzik et al. , 2005 ) or 1 h ( Brown et al. , 2014 ) after permanent MCA occlusion ( Seiffge et al. , 2012 ), and might therefore constitute an alternative or additional therapy to NBO in future trials.
In the scenario of NBO administered at or soon after stroke onset, however, the ischaemic or haemorrhagic nature of the event is generally unknown, raising the question whether NBO might be detrimental in case of haemorrhagic stroke. This has been assessed in two rodent studies, both showing no haematoma enlargement and no increase, or even a decrease in peri-haematoma oedema and blood–brain barrier disruption ( Fujiwara et al. , 2011 ; Zhou et al. , 2015 ). Likewise, if NBO is started soon after onset of neurological deficit, it might be deleterious if continued into the reperfusion stage, by e.g. enhancing oxidative stress or haemorrhagic risk. However, one study has reported no deleterious effects of NBO in this scenario ( Singhal et al. , 2002 b ). Also, NBO started during the reperfusion phase has not been found to have any deleterious effect ( Eschenfelder et al. , 2008 ; Geng et al. , 2013 ), and may in fact be beneficial ( Flynn and Auer, 2002 ). Finally, might NBO interfere with t-PA (tissue plasminogen activator) and/or increase the risk of haemorrhagic transformation and haematoma? In three studies, NBO did not interfere with the beneficial effects of, and actually showed synergism with t-PA in reducing infarct volumes ( Fujiwara et al. , 2009 ; Henninger et al. , 2009 ; Liang et al. , 2015 ), while also showing protective effects against the deleterious effects of t-PA on the neurovascular unit including the blood–brain barrier ( Liang et al. , 2015 ). Likewise, Henninger et al. (2009) found no increase in haematoma volume or petechial incidence in rats given both NBO and t-PA in an embolic model, and two additional similar studies found that NBO in fact reduced post-thrombolytic intracerebral haemorrhage ( Liu et al. , 2009 ; Sun et al. , 2010 ). Thus, the safety profile of NBO seems reassuring in most realistic clinical situations, although further studies would be required to confirm this.
In conclusion, our findings document that NBO administered during brief MCA occlusion in a clinically-relevant rat model of ‘true’ TIA near-completely prevented neuronal death and microglial inflammation as well as sensorimotor impairment. Further studies using our model should assess delayed NBO and NBO continued after reperfusion, rat strains less vulnerable to ischaemia, and additional markers of tissue damage and behavioural impairment. Taken together with the available literature, our findings point to NBO appearing extremely promising for brief MCA occlusions. The particular attractiveness of NBO is its potential clinically translatability as a straightforward adjunct to reperfusion therapies, e.g. it could easily be administered in the ambulance in acute-onset focal neurological deficits suspected of stroke. Furthermore, patients at-high risk of TIAs/stroke could be equipped and trained (or their relatives) to immediately administer oxygen through a face mask in case of acute-onset focal symptoms. Randomized trials comparing the clinical and tissue outcome of subjects starting NBO versus room air could ultimately test this hypothesis.
middle cerebral artery
normobaric oxygen therapy
selective neuronal loss
transient ischaemic attack
The authors are grateful to P. Simon Jones for help with statistical analyses at the revision stage.
EU Grant (EUSTROKE Health-F2-2008-2022131). J.V.E. was funded by a grant from Theracur Stiftung; D.J.W. was funded by an MRC collaborative grant (G0600986).
Supplementary material is available at Brain online.
- magnetic resonance imaging
- transient ischemic attack
- cerebrovascular accident
- middle cerebral artery occlusion
- ischemic stroke
- oxygen therapy
- reperfusion therapy
- brain injuries
- rats, inbred shr
- cognitive impairment
- symptom onset