Cerebral oxygenation during locomotion is modulated by respiration

In the brain, increased neural activity is correlated with increases of cerebral blood flow and tissue oxygenation. However, how cerebral oxygen dynamics are controlled in the behaving animal remains unclear. We investigated to what extent cerebral oxygenation varies during locomotion. We measured oxygen levels in the cortex of awake, head-fixed mice during locomotion using polarography, spectroscopy, and two-photon phosphorescence lifetime measurements of oxygen sensors. We find that locomotion significantly and globally increases cerebral oxygenation, specifically in areas involved in locomotion, as well as in the frontal cortex and the olfactory bulb. The oxygenation increase persists when neural activity and functional hyperemia are blocked, occurred both in the tissue and in arteries feeding the brain, and is tightly correlated with respiration rate and the phase of respiration cycle. Thus, breathing rate is a key modulator of cerebral oxygenation and should be monitored during hemodynamic imaging, such as in BOLD fMRI.


25
In the brain, increased neural activity is correlated with an increase of cerebral blood flow and 26 increased tissue oxygenation. However, how cerebral oxygen dynamics are controlled in the 27 behaving animals remains unclear. Here, we investigated to what extent the cerebral oxygenation 28 varies during natural behaviors that change the whole-body homeostasis, specifically exercise.

29
We measured oxygen levels in the cortex of awake, head-fixed mice during locomotion using 30 polarography, spectroscopy, and two-photon phosphorescence lifetime measurements of oxygen 31 sensors. We found that locomotion significantly and globally increases cerebral oxygenation, 32 specifically in areas involved in locomotion, as well as in the frontal cortex and the olfactory bulb.

33
The oxygenation increase persisted when neural activity and functional hyperemia were blocked, 34 occurred both in the tissue and in arteries feeding the brain, and was tightly correlated with 35 respiration rate and the phase of respiration cycle. Thus, respiration provides a dynamic pathway 36 for modulating cerebral oxygenation. 37 hemodynamic response function (HRF) 25,29 , which is the linear kernel relating locomotion events 88 to observed changes in CBV and CBF (Supplementary Fig. 1). Both measures showed a 89 decrease in CBV and CBF in the FC during locomotion (Fig. 1d, Supplementary Fig. 1). This 90 shows that locomotion and the accompanying cardiovascular changes do not drive global 91 increases in CBF/CBV, rather CBF/CBV increases are under local control. This lack of non-92 specific flow increase in the cortex during locomotion is likely because of autoregulation of the 93 feeding arteries at the level of the circle of Willis and increased blood flow to the muscles 30 .

94
To assess neural activity during locomotion, we measured local-field potential (LFP) and 95 multi-unit activity (MUA) in a separate group of 7 mice (6 sites in FL/HL and 4 sites in FC) using 96 multi-channel linear electrodes (Fig. 1e). We used electrophysiological measures of neural 97 activity, as they are more sensitive than calcium indicators (which fail to detect about half of the 98 spikes even under ideal conditions 31 ), and do not disrupt normal neural activity as genetically 99 encoded calcium indicators can do 32-34 . Since gamma-band (40-100 Hz) power in the LFP has 100 been observed to be the strongest neural correlate of hemodynamic signals in rodents 25,35 , dependence of resting PtO2 in awake mice, with smaller oxygenation in surface layers and greater 138 oxygenation in deeper layers in both FL/HL and FC (Supplementary Fig. 4a, b). Resting PtO2 139 was similar at each cortical depth in both FL/HL and FC (Supplementary Fig. 4a, b). These 140 results, together with the observation that resting PtO2 is similar in somatosensory cortex and the 141 olfactory bulb glomerular layer 4 , indicate that the spatial distribution of oxygen in the brain under 142 normal (non-anesthetized) physiological condition is homogenous. Locomotion produces large, 143 sustained dilation of arteries 48 and increases in CBF and CBV 13,26 in the somatosensory cortex.

144
These locomotion-induced dilations were not due to systemic effects, as they have been shown 145 to be unaffected by drugs that do not cross the blood brain barrier that increase or decrease the 146 heart rate 49 and are blocked by the suppression of local neural activity 50 . The locomotion-induced 147 dilations are comparable in magnitude to those elicited by episodic whisker stimulation 25 which is 148 known to elevate oxygenation, so one would expect increases in tissue oxygenation in FL/HL 149 during locomotion. As anticipated, we observed increases in PtO2 during locomotion in FL/HL in 150 all layers (Fig. 2b-d). Because the supply of blood to FC does not increase, but the neural activity 151 does, one would expect a decline in tissue oxygenation during locomotion. Surprisingly, we also 152 observed a very similar PtO2 increase in FC ( Fig. 2b-d) to that observed in the FL/HL, despite 153 small decreases in CBV or CBF, and an increase in neural activity. The elevation of PtO2 in FC 154 during locomotion suggests that other factors can increase oxygenation in the brain.

155
Polarographic probes provide measures of oxygen tension over a small region of brain 156 tissue, and the response may be affected by the vasculature type and density 51-53 surrounding the 8 polarography reports average oxygen concentration in the tissue near the electrode. The oxygen 164 levels in the tissue will differ from that in the blood somewhat due to the constraints of oxygen 165 diffusion from the blood into the tissue and ongoing metabolic processes in the neurons and glial 166 cells. Using the cerebral oxygenation index (HbO-HbR) 56 , the spectroscopic measures of 167 hemoglobin oxygenation were similar to measurements from the tissue using polarographic 168 probes: both methods yielded an increase in oxygenation during locomotion in both FC and FL/HL 169 ( Fig. 2f, g). These oxygenation changes persisted even when the heart rate increase associated 170 with locomotion was pharmacologically blocked or occluded (Supplementary Fig. 5b, d, e), 171 indicating they were not driven by the increased cardiac output during locomotion.

172
Moreover, the locomotion-induced elevation in oxygenation were present in the 173 parenchyma, arterial and venous blood (Fig. 2f). As oxygen levels in the brain strongly depends 174 on the arterial oxygen content 9 , we made direct measurements of oxygen partial pressure in the 175 center of pial arteries (PaO2) using two-photon phosphorescence lifetime microscopy (2PLM, Fig.   176 2h) 4,8,9 , with a new phosphorescent probe (Oxyphor 2P) which has a very high brightness, 177 improving measurement speed and imaging depth 57 . We asked if the oxygen levels increased in 178 the center of the large pial arteries that supply blood to the brain. As the blood in these arteries Our observation that locomotion induced localized blood flow/volume increases, but cortical-wide 189 increases in brain oxygenation, led us to hypothesize that the activity-dependent vasodilation may 190 not be necessary for an increase in oxygenation. To test this, we pharmacologically blocked the 191 glutamatergic and spiking activity by infusing/superfusing a cocktail of 6-cyano-7-nitroquinoxaline-192 2,3-dione (CNQX, 0.6 mM), (2R)-amino-5-phosphonopentanoic acid (AP5, 2.5 mM) and 193 muscimol (10 mM) to suppress local neural activity. We first infused a cocktail of 194 CNQX/AP5/muscimol via a cannula into FL/HL 25 , while concurrently monitoring neural activity,

195
CBV and blood oxygenation (n = 4 mice, Fig. 3a). The cocktail infusion suppressed resting 196 gamma-band (40-100 Hz) LFP power by 80 ± 12% and spiking activity by 82 ± 3% relative to 197 vehicle infusions (Fig. 3b). Similarly, the standard deviation (SD) in gamma-band LFP power 198 fluctuations during resting periods, an indicator of spontaneous neural activity levels, was 199 decreased by 75 ± 18% in the gamma-band power and by 85 ± 6% in the MUA amplitude. To 200 quantify the blood volume responses, we selected a semicircular region of interest (ROI) centered 201 on the cannula and with a radius specified by the distance between the electrode and cannula 202 ( Fig. 3a), to ensure the ROI only included suppressed cortex 25 . Accompanying this neural activity 203 blockade, baseline reflectance from the ROI increased (indicating decreased CBV, data not 204 shown), and the locomotion-evoked decrease in reflectance (vasodilation) was almost completely 205 suppressed ( Fig. 3b-d), consistent with our previous study 25 showing that intracerebral infusion blocked, the locomotion-evoked increase in differences of oxy-and deoxygenated hemoglobin 215 concentration (HbO-HbR) persisted, though the increase was smaller ( Fig. 3b-

219
We further studied the effects of the suppressed vasodilation on oxygen responses in the 220 tissue in a separate set of mice using polarographic electrodes (n = 9 mice, 5 in FC and 4 in 221 FL/HL). We topically applied a cocktail of CNQX/AP5/muscimol to the cortex, while measuring 222 spontaneous and locomotion-evoked neural activity and PtO2 in the superficial cortical layers 223 (100-200 µm below the pia). The efficacy of the cocktail in suppressing neural activity was 224 monitored with two electrodes spanning the oxygen measurement site 25,35 (Fig. 3e)

242
Respiration drives changes in cerebral and blood oxygenation 243 One possible driver of the increases in cerebral oxygenation is the increase in respiration during 244 locomotion. Changes in respiration affect blood oxygen levels in the carotid artery 61,62 in 245 anesthetized animals, and in humans, inhalation of 100% oxygen can elevate brain oxygen 246 levels 63 . However, it is not known if normal fluctuations in respiration rate can impact cerebral 247 oxygenation during normal behaviors. We tested whether respiration was correlated with 248 oxygenation during locomotion by simultaneously measuring cortical tissue oxygenation and 249 respiration (Fig. 4a). Locomotion was accompanied by a robust increase in respiratory rate ( Fig.   250 4a, Supplementary Fig. 6), and fluctuations in respiratory rate on the time scale of seconds were 251 linked to fluctuations in PtO2 (Fig. 4a). We quantified how well the fluctuations of respiratory rate 252 and gamma-band (40-100 Hz) LFP power (which has been shown in previous studies to be the previous reports showing decreases in the power of these bands during voluntary 265 locomotion 40 (also see Supplementary Fig. 8). Because cortical excitability and respiratory rate 266 are correlated during locomotion (likely due to the reciprocal connections between respiratory and 267 modulatory regions 65 ), we sought to disentangle their respective contributions to cerebral 268 oxygenation using partial coherence analysis 66 . We found that the coherence between respiratory 269 rate and PtO2 was not due to the co-varying neural component (Supplementary Fig. 9b), nor 270 was the coherence between gamma-band power and PtO2 affected by removing the respiratory 271 rate contribution (Supplementary Fig. 9c). Thus, the partial coherence analysis indicates that 272 respiration and neural activity (and likely vasodilation) affect tissue oxygenation independent of 273 each other.

274
The correlated fluctuations in respiratory rate and PtO2 suggests that the oxygen tension 275 of arterial blood should also track the respiratory rate. To test this, we simultaneously monitored 276 respiration and PaO2 in the pial arteries using 2PLM. In mice with irregular respiration, where 277 respiratory rate transients of a few seconds occurred without locomotion, PaO2 followed 278 respiration rate fluctuations (Fig. 4f), showing that changes in respiration rate can alter the 279 oxygenation of the arterial blood entering the cortex.

280
We then asked if PaO2 tracked the phase of respiration, that is, whether the concentration 281 of oxygen in the blood entering the brain fluctuated in phase with the inspiration-expiration cycle.

282
This requires measuring PaO2 at rates high enough (> 5 Hz) to capture fluctuation in PaO2 due to 283 respiration (nominally 2.5 Hz). As measurement of PaO2 with the 2PLM method is based on the 284 lifetime of the phosphoresce decay of the dye, accurate quantification of the oxygen concentration 285 requires averaging of decays 57 , which amounted to ~3000 decays at our laser power 286 (corresponding to ~0.75 s of data), too slow to capture inspiration-expiration linked changes in and binned according to their place in the phase of the respiratory cycle (Fig. 4g), analogous to 290 how erythrocyte-related transients can be detected in the capillaries 4,8 , or analyzing the signal in 291 the frequency domain. In a few animals with long bouts of highly regular respiration rate (average 292 frequency 2.5 Hz, SD ≤ 0.6 Hz, average frequency/SD > 4) fluctuations of PaO2 tracked the 293 respiratory cycle [4 out of 7 arteries (3 in the cortex, and 1 in the olfactory bulb) in 4 mice] (Fig.   294 4h-j). These arteries showed oscillations in PaO2 at the frequency of respiration that were 295 significantly larger than would be expected by chance (reshuffling test, see Methods). This shows 296 that the arterial blood flowing to the brain is not saturated at rest. It also shows that the oxygen 297 tension in the blood tracks sub-second respiration dynamics, so increase in respiration can drive 298 rapid increases in systemic blood oxygenation that will impact the brain oxygenation.

299
Computational modeling indicates respiration contributes to tissue oxygenation 300 Using computer simulations, we then asked what the relative contributions of increased arterial 301 oxygenation and vasodilation were to changes in PtO2. Recent work has shown that substantial 302 oxygenation exchange occurs not only at capillaries, but also around the penetrating arteries in 303 the cortex 7,67 . To better understand how increase in blood oxygenation impact tissue oxygenation 304 around arterioles, where the simple geometry of the vasculature allows us to better capture the 305 dynamics of oxygenation changes due to vasodilation and systemic oxygenation changes, we 306 created a Krogh cylinder model of a penetrating artery in the cortex 68 (Fig. 5a). For this model, arterial oxygenation to tissue oxygenation changes in both FC and FL/HL. In FL/HL, the large 315 increase in CMRO2 during locomotion were counteracted by increases in arterial oxygenation due 316 to vasodilation and increase in arterial oxygenation. In FC, the small increase in CMRO2 and 317 vasoconstriction was totally offset by the increase in arterial oxygenation (Fig. 5d). These 318 simulations show that respiration plays an important role in modulating tissue oxygenation. The 319 increase in arterial oxygenation will also increase the oxygen tension in the tissue around the 320 capillary bed 69 , though the actual changes will depend on the details of the capillary geometry 321 and the movement of individual red blood cells, which is hard to capture without detailed 322 anatomical models, and will depends on the details of flow dynamics. These simulations show 323 that increased arterial oxygenation that accompanies increases in respiration can lead to 324 increases in tissue oxygenation, even in brain regions showing vasoconstriction. Taken together, 325 the experimental data and the simulation support the notion that increases in respiratory rate play 326 an important role in regulating cerebral oxygenation.

328
We observed increases in cerebral tissue and blood oxygenation when respiration increased both 329 at rest and during bouts of voluntary exercise. We also saw increases in tissue and blood 330 oxygenation during locomotion when local neural activity was suppressed and vasodilation was 331 blocked, conditions where we would expect a decrease in oxygenation. Note that while the 332 changes in tissue and arterial oxygenation had similar dynamics (sustained increases in oxygen 333 during locomotion), the oxygen increases measured spectroscopically were largest close to the 334 onset of locomotion. This is likely because the spectroscopic imaging samples from arteries, 335 capillaries, and veins. As the veins will be deoxygenated by the increased metabolic rate during 336 periods of sustained neural activity, this will tend to reduce the measured oxygen change in the 337 spectroscopic studies as compared to the polarography measurements, which primarily report 338 tissue oxygen concentrations. Oxygen levels in the large arteries rose following increases in respiration both at rest and during exercise, and tracked the inspiration-expiration phase, showing 340 that the oxygenation levels of the blood coming into the brain can be modulated by respiration 341 both during rest and locomotion.

342
Respiration is not the only physiological change that accompanies exercise, and it bears 343 considering other mechanisms that could account for the cerebral and arterial oxygenation 344 changes seen here. Exercise causes large changes in cardiac output and blood pressure, and 345 can be accompanied by changes in blood CO2 and lactate levels, but we think they are unlikely 346 to be the cause of the nonspecific increase in cerebral oxygenation that we saw here. First, for

379
The role of increased respiration in increasing brain oxygenation during behavior observed 380 here is likely facilitated by the reciprocal connections between respiratory centers and the locus 381 coeruleus 65,77 and other brain regions involved in arousal 78,79 . Consistent with a tight interplay 382 between respiration and metabolic demand in the brain, activation of the locus coeruleus, which 383 will cause increases in alertness, and also causes concomitant increases in neural activity and 384 blood flow in the cortex 80 . This tight interplay at the behavioral and anatomical levels between

628
Cerebral blood flow measurements using laser Doppler flowmetry were performed in 5 male mice.

629
Tissue oxygenation measurements using spectroscopy (using alternating 470 nm and 530 nm 630 illumination) were conducted in 11 male mice (4 mice were implanted with cannula and electrode).

726
Precision Instruments) and then sampled at 30k Hz (PCI-6259, National Instruments). The oxygen 727 signal in these experiments was recorded at a depth of ~100-200 µm.

728
At the end of the experiment, the mouse was deeply anesthetized, and a fiduciary mark 729 was made by advancing an electrode (0.005" stainless steel wire, catalog #794800, A-M systems)

730
into the brain with a micro-manipulator to mark the oxygen measurement site.

734
Engineering) placed near the mouse's nose (~ 1 mm), with care taken to not contact the whiskers.

735
Data were amplified 2000x, filtered below 30 Hz (Model 440, Brownlee Precision), and sampled 736 at 1000 Hz (PCI-6259, National Instruments). Downward and upward deflections in respiration 737 recordings correspond to inspiratory and expiratory phases of the respiratory cycle, respectively 738 ( Fig. 4a). We identified the time of each expiratory peak in the entire record as the zero-crossing 739 point of the first derivative of the thermocouple signal.

740
Laminar electrophysiology. Laminar electrophysiology recordings were performed in a separate 741 set of mice (n = 7, Fig. 1e). On the day of measurement, the mouse was anesthetized using 742 isoflurane (in oxygen, 5% for induction and 2% for maintenance). Two small (1x1 mm 2 ) 743 craniotomies were performed over the frontal cortex (1.0 to 2.5 mm rostral and 1.0 to 2.5 mm 744 lateral from bregma) and FL/HL representation in the somatosensory cortex (0.5 to 1.0 mm caudal 745 and 1.0 to 2.5 mm lateral from bregma) over the contralateral hemisphere (Fig. 1e), and the dura 746 was carefully removed. The craniotomies were then moistened with warm saline and porcine 747 gelatin (Vetspon). After this short surgical procedure (~20 minutes), the mouse was then 748 transferred to the treadmill where it was head-fixed. Measurements started at least one hour after 749 the cessation of anesthesia 24,83 .

750
Neural activity signals were recorded using two linear microelectrode arrays (A1x16-3mm-  Instruments). To verify that the dynamics observed after drug infusion were not due to changes 788 of peripheral cardiovascular system 84,85 , we also injected water, atenolol (2 mg/kg body weight) 789 and glycopyrrolate (0.5 mg/kg body weight) 49 intraparietal in the same mouse, and the 790 hemodynamic response was measured described as above (Supplementary Fig. 5).

791
Brain oxygen measurement using two-photon phosphorescent lifetime microscopy. A complete 792 description of 2PLM can be found in previous reports 4,9,86 . In brief, the oxygen sensor Oxyphor

856
(the minimal temporal resolution at 20kHz sampling rate). A histogram of peak-to-trough times 857 was fitted as a sum of two Gaussian distributions (Supplementary Fig. 2a, f), and a receiver 858 operator characteristic curve was used to segregate spikes in a given bin as either FS or RS 859 waveforms using a 95% probability of belonging to a group as the inclusion threshold. Spikes not 860 reaching the inclusion threshold for either group were not included in the analysis. Fast spiking

861
(FS) waveforms (Supplementary Fig. 2b, g) were characterized by short durations between 862 action potential peak and peak of hyperpolarization, peak-to-trough-duration as described

879
For spontaneous correlations, only periods of rest lasting more than 30 seconds were used, with 880 a four-seconds buffer at the end any locomotion event. We also calculated the correlations using 881 all the data including periods with locomotion. To check the spatiotemporal distribution of the 882 correlation, we calculated cross-correlogram between PtO2 and LFP power in each frequency 883 band (Supplementary Fig. 7). Briefly, LFP signals were separated into frequency bands (~1 Hz 884 resolution with a range of 0.1-150 Hz) by calculating the spectrogram (mtspecgramc, Chronux 885 toolbox) 90 , and then we calculate the temporal cross-correlation between power in each frequency 886 band and the oxygen concentration (xcorr, MATLAB). Positive delays denote the neural signal 887 lagging the oxygen signal. The oxygen tension and neural activity were both low-pass filtered 888 below 1 Hz before calculating the cross-correlation. The temporal cross-correlation between 889 respiratory rate and oxygen signals was also calculated over a similar interval (xcorr, MATLAB).

890
Statistical significance of the correlation was computed using bootstrap resampling 94 from 1000 891 reshuffled trials.

892
Arterial oxygen tension changes during the respiration cycle. To evaluate the arterial oxygen 893 tension change within the respiration cycle, we selected oxygen measurements during periods 894 with regular respiratory rate (average frequency 2.5 Hz, SD ≤ 0.6 Hz). The phosphorescent decays were aligned according to their place in the phase of the respiratory cycle (Fig. 4 g). To 896 further determine whether the fluctuations of oxygen tension was induced by respiration, we 897 calculated the power spectrum of arterial oxygen tension, and determined the peak frequency in 898 the power spectrum. Statistical significance of this peak was calculated by reshuffling the arterial 899 oxygen measurements 94 , and the 95% confidence interval was calculated using 10000 reshuffled 900 trials.

901
Ordinary coherence and partial coherence. We used coherence analysis 95 to reveal correlated where U MM<W and U NN<W is the auto-spectra associated with the residual part of x and y after 914 removing the part coherent with z, respectively. U MN<W is the cross-spectrum between the residual 915 part of x and y after removing the part coherent with z. If all the networks are connected, partial 916 coherence will be between zero and the level of the ordinary coherence. If the connection behaves in an asymmetric manner, i.e., signal z affects x and y differentially, the coherence between two 918 signals may increase after partialization (Supplementary Fig. 9a).

919
Statistical analysis. Statistical analysis was performed using Matlab (R2015b, Mathworks). All 920 summary data were reported as the mean ± standard deviation (SD) unless stated otherwise.

921
Normality of the samples were tested before statistical testing using        Coherence (0 -0.5 Hz) n = 9 n = 9