Phasic oxygen dynamics underlies fast choline-sensitive biosensor signals in the brain of behaving rodents

Fast time-scale modulation of synaptic and cellular physiology by acetylcholine is critical for many cognitive functions, but direct local measurement of neuromodulator dynamics in freely-moving behaving animals is technically challenging. Recent in vivo brain measurements using choline oxidase (ChOx)-based electrochemical biosensors have reported surprising fast cholinergic transients associated with reward-related behavioral events. However, in vivo recordings with conventional ChOx biosensors could be biased by phasic local field potential and O2-evoked enzymatic responses. Here, we have developed a Tetrode-based Amperometric ChOx (TACO) sensor enabling minimally invasive artifact-free simultaneous measurement of cholinergic activity and O2. Strikingly, the TACO sensor revealed highly-correlated O2 and ChOx transients following spontaneous locomotion and sharp-wave/ripples fluctuations in the hippocampus of behaving rodents. Quantitative analysis of spontaneous activity, in vivo and in vitro exogenous O2 perturbations revealed a directional effect of O2 on ChOx phasic signals. Mathematical modeling of biosensors identified O2-evoked non-steadystate ChOx kinetics as a mechanism underlying artifactual biosensor phasic transients. This phasic O2-dependence of ChOx-based biosensor measurements confounds phasic cholinergic dynamics readout in vivo, challenging previously proposed ACh role in reward-related learning. The discovered mechanism and quantitative modeling is generalizable to any oxidase-based biosensor, entailing rigorous controls and new biosensor designs.

The data are given as the mean ± CI (95%) The number of sensors tested is given in parentheses.
Acetylcholine can suppress the occurrence of SWRs (Norimoto et al., 2012;Vandecasteele et al., 23 2014) and positively correlates with arousal Marrosu et al., 1995;Reimer 24 et al., 2016;Teles-Grilo Ruivo and Mellor, 2013;Teles-Grilo Ruivo et al., 2017), which typically 25 coincides with hippocampal theta oscillations and manifests as locomotion in freely-moving animals 26 (Buzsáki, 2002;Gu et al., 2017). Thus we tested whether these events could correlate with fast 27 changes in putative cholinergic activity reflected by COA. hippocampal SWRs were detected from a 28 silicon probe site in the CA1 pyramidal layer during NREM sleep. Average raw biosensor signals 29 triggered on the peak of SWRs showed a prominent peak in all TACO sensor's sites ( Figure 2D, top). 30 The similarity of peak amplitudes suggests an LFP-related origin of these currents, which was 31 virtually absent in the cleaned signals (Figure,2D,middle), and reflects the slow profile of the sharp 32 wave. The high magnitude of this LFP signal contaminating COA measurements emphasizes the 33 importance of the common-mode rejection approach in revealing putative cholinergic dynamics that differential signals showed a peak lagging the SWR by ~3 s. Importantly, this peak was not present in 1 pseudo-sentinel sites, supporting an authentic change in COA ( Figure 2D, top and middle). 2 In our recordings, high-frequency bursts captured with electrophysiology and amperometry typically 3 coincided in time and exhibited similar spectral features ( Figure S2A). Accordingly, the continuous 4 power-power correlation between amperometry-derived current and silicon probe-derived LFP was 5 high (>0.7, Figure S2B) across a wide frequency range. The very similar average power spectra 6 triggered to SWRs ( Figure 2D, bottom and Figure S2C) and the high co-occurrence of SWRs detected 7 with both modalities ( Figure S2D) further prove the ability of our amperometric system to reliably 8 record high-frequency LFP-related currents. Although theoretically expected, this capability has not 9 been experimentally documented before. 10 Bouts in locomotion, detected as peaks in the rat running speed, were associated with transient 11 increases in theta power and were also followed by a peak in putative Ch signal, observed in the clean 12 biosensor signal but not in the pseudo-sentinel subtraction control ( Figure 2E). 13 These results highlight the usefulness of our multichannel pseudo-sentinel approach to discriminate 14 between authentic changes in COA and interferent signals in the brain. Notably, phasic changes in 15 COA were associated with transient changes in arousal or exploratory behavior and with SWRs, 16 events that are critical for memory encoding and consolidation, respectively (Buzsáki, 2002(Buzsáki, , 2015. 2 Both probes were attached to microdrives. Top right panel shows a segment of a raw signal (low-pass filtered at 3 1 Hz) and the resulting clean putative Ch and LFP components upon common-mode rejection, recorded during spontaneously running on a treadmill ( Figure 3A bottom). Strikingly, we found that phasic COA 23 dynamics typically matched the simultaneously recorded O 2 fluctuations, which were generally 24 related to changes in behavioral state ( Figure 3B). In line with the freely-moving data ( Figure 2E), 25 head-fixed running bouts were temporally correlated with an increase in the power of theta 26 oscillations as well as with delayed phasic increases in both COA and O 2 , peaking a few seconds later 27 ( Figure 3B and C). This lag was significantly greater than zero in all recording sessions (two-sample 28 t-test, p<0.01), averaging 3.85 ± 2.04 s (n = 5) for COA and 5.04 ± 3.12 s (n = 5) for O 2 . Notably, 29 running-related isolated COA or O 2 peaks were rare, with the vast majority of the events showing 30 either no identifiable change or co-occurrence of the two transients ( Figure 3D). Amplitudes of COA 31 peaks were significantly correlated with those of O 2 (p<0.001) when pooling together the events from 32 all recordings, which allowed sampling across the full O 2 amplitude range ( Figure 3E). Phasic COA 33 correlated more consistently with O 2 than with theta power or speed ( Figure 3F). Moreover, running-34 bout-related COA and O 2 peak lags were significantly correlated (p<0.001, Figure 3G) and, 35 interestingly, within most sessions (3 out of 5), COA peaked significantly earlier than O 2 (p<0.05).
In summary, these results indicate a strong correlation between phasic COA and O 2 in the 1 hippocampus of head-fixed mice following locomotion bouts. 2 Phasic COA and oxygen signals follow clusters of sharp-wave/ripples during 1 immobility 2 Hippocampal SWRs are critical for memory consolidation and their occurrence has been proposed to 3 be anti-correlated with cholinergic activity in the hippocampus Norimoto 4 et al., 2012;Vandecasteele et al., 2014). However, our freely-moving data showing COA transients 5 following SWRs contradicts this prediction, posing questions on the factors driving the biosensor 6 response during these events. Thus, we investigated whether the SWR-related response of 7 immobilized ChOx was correlated with extracellular O 2 in head-fixed mice during periods of 8 quiescence. 9 Remarkably, on average, SWR events were followed by fast transients in both COA and O 2 ( Figure  10 4A). Both COA and O 2 peak amplitudes correlated best with ripple power integrated over a period of 11 ~2 seconds lagging them by 3-4 seconds ( Figure 4B and C). Similarly, both SWR count and summed 12 ripple power integrated in a 2-second window positively correlated with the delayed amplitude of 13 both COA and O 2 transients ( Figure 4D-E and Figure S3). 14 These findings might indicate the contribution of a time-constant related to the sensor response and/or 15 to a relatively slow physiological process by which SWRs recruit cholinergic activity or a local 16 hemodynamic response leading to O 2 increase. Indeed, functional MRI has reported SWR-triggered 17 increases in BOLD signal in the primate hippocampus, reflecting a local tissue hemodynamic 18 response at a time-scale matching O 2 transients observed here (Ramirez-Villegas et al., 2015). 19 Importantly, the amplitudes of SWR-associated COA and O 2 phasic transients were consistently 20 correlated within all recordings (n=3, p<0.001, Figure 4F). Similarly, lags of these transients to SWR 21 were significantly correlated ( Figure 4G, H, I). 22 Together, the data indicate correlated phasic profiles of COA and extracellular O 2 in response to 23 SWRs ( Figure 4I), especially when they happen in clusters. As in the case of locomotion bouts, at this 24 stage one could not rule out the contribution of neither phasic Ch or O 2 as the trigger for the peaks in 25 COA. The putative cholinergic origin of this response would, nevertheless, be surprising given the 26 suppressive effect of ACh on SWR occurrence (Norimoto et al., 2012;Vandecasteele et al., 2014). 27 Thus, in light of the O 2 transients that accompany the rise in COA, these observations cast doubt on 28 the validity of the putative SWR-triggered cholinergic response.

14
Group statistics on the lags of ChOx and O 2 peaks relative to SWRs. Each dot is the average from one 15 recording. Bars represent means ± CI. (H) Lags of ChOx peaks as a function of O 2 peak lags relative to SWRs.

16
The parameters were significantly correlated (r spearman =0.56, p<0.0001, n=1067). (I) Summary of correlations Interactions between COA and oxygen are not sensitive to ongoing 1 hippocampal dynamics and depend on the time-scale 2 The correlation between COA and O 2 following locomotion and SWRs may reflect interaction 3 between the two signals or result from the coincident recruitment of cholinergic and hemodynamic 4 responses. While a consistent relationship between the two variables is expected in the first case, 5 irrespective of ongoing network dynamics, the same may not happen in the latter. To get insights into 6 this question, we analyzed an additional category of events, consisting of fast O 2 transients detected 7 outside the time-windows surrounding SWRs and peaks in locomotion. Under this condition, the 8 average rate of O 2 peaks occurrence was only 18% of the rate computed from total O 2 peaks ( Figure  9 S4A), emphasizing the strong modulatory effect of hemodynamics and respiration on O 2 , potentially 10 evoked by SWRs and locomotion, respectively (Leithner and Royl, 2014;Ramirez-Villegas et al., 11 2015;Zhang et al., 2019). Remarkably, virtually all (>95%) of these events had an associated COA 12 transient. The dynamics of COA and O 2 were typically similar ( Figure 5A), with COA peaks lagging, 13 on average, from -0.09 s to 0.48 s (n=3 recordings) relative to O 2 . Signal amplitudes were 14 significantly correlated both when pooling together all events (p<0.001, Figure 5B) or within all 15 individual sessions (p<0.01). The amplitude correlation coefficients were in the range of those 16 obtained for O 2 peaks associated with locomotion and SWRs ( Figure 5C), but the average amplitudes 17 of O 2 and corresponding COA signals were in the sub-micromolar range, comparable to those 18 associated with SWRs ( Figure 5D). Overall, although the amplitude range of locomotion-related 19 peaks was wider, it is notable that the whole data fit to a COA/O 2 relationship that seems to follow the 20 same model across recordings and event types ( Figure 5D). These observations are therefore 21 compatible with an interaction between COA and O 2 , although contribution of third-party factors 22 could not be definitely excluded at this stage. 23 Besides the amplitude of O 2 transients, the temporal profile of O 2 rise might influence the shape and 24 amplitude of associated ChOx responses and thus provide further hints on the causality and 25 directionality of COA-O 2 interactions. Thus, we detected O 2 peaks at different frequency bands 26 (ignoring their correlation with SWRs or speed), resulting in a spectrum of O 2 rise times from less 27 than 2 to 14 s (please see Methods section for details). Across all recording sessions, we consistently 28 observed that the COA peak anticipated O 2 (negative lag) as O 2 rise time increased ( Figure 5E,F and 29 Figure S4B). Maximal COA lags averaged 0.38 ± 0.42 s and were associated with fast O 2 rises, 30 lasting 1.4-2.6 s ( Figure 5F and G). Furthermore, the slope of COA vs. O 2 amplitudes was time-scale 31 dependent, peaking for O 2 transients that took 2.3 to 6.2 s to rise ( Figure 5E, G and Figure S4C amplitude as a function of O 2 rise, the amplitude of corresponding COA peaks was, on average, nearly 1 time-scale independent for O 2 rise times in the range of 3-14 s ( Figure S4D). 2 In general, these results provide important insights into the interaction between COA and extracellular 3 O 2 in the hippocampus. The advancement of COA relative to O 2 is apparently compatible with a Ch-4 O 2 directionality, possibly caused by ACh-evoked changes in local blood flow (Takata et al., 2013). 5 However, the diffusional delay underlying such a mechanism is expected to be constant as a function 6 of time-scale. Moreover, the time-scale dependence of the relative amplitude of enzyme response is 7 hard to interpret in light of a Ch-O 2 directionality. Alternatively, non-steady state ChOx responses to 8 O 2 transients are fully compatible with our observations. Such modulation of sensor response is, by 9 definition, phasic and is expected to be amplified within an optimal, probably short, time-window. Remarkably, immobilized ChOx exhibited robust phasic responses to exogenous O 2 transients, 7 regardless of the time-scale of O 2 change ( Figure 6A and Figure S5A). Amplitudes of COA and O 2 8 were significantly correlated (p<0.0001, Figure 6B). However, similarly to spontaneous dynamics, the 9 CAO/O 2 amplitude ratio of single events (equivalent to CAO/O 2 slope in spontaneous data) 10 significantly decreased with O 2 rise time in all experiments (negative correlation, p<0.05, Figure 6C). 11 The decrease in COA amplitude was accompanied by the advancement of its peak relative to O 2 12 (p<0.001 in all recordings, Figure 6D). Thus these results qualitatively recapitulate our observations 13 for the spontaneous COA-O 2 interactions, reinforcing the putative O 2 -COA directionality. 14 Next, we manipulated hippocampal O 2 levels non-invasively, through inhalation, to further confirm 15 O 2 causality. We exposed mice to a pure O 2 stream during variable periods (4-30 s) in order to 16 generate different O 2 transients. Like in the case of local O 2 delivery, we observed reproducible 17 changes in COA in response to exogenous O 2 ( Figure 6E and FigureS5B). The data showed a 18 significant correlation between peak amplitudes (p<0.0001, Figure 6F). Importantly, both the COA/O 2 19 amplitude ratio and COA peak lag significantly decreased as a function of O 2 rise time (p<0.05), 20 corroborating the conclusions from local O 2 manipulation. 21 Overall, O 2 inhalation tended to generate slower and smaller O 2 transients (and perhaps more 22 physiological) than those by local application. Despite the differences in magnitudes and time-scales, 23 the tested correlations are consistent across the two paradigms ( Figure 6I  Given the in vivo causality between O 2 and COA, we sought a detailed investigation of the biosensor 2 O 2 -dependence in vitro, in order to get mechanistic insights into the relationship between these 3 signals. The in vitro tests were based on step additions of known O 2 concentrations in the presence of 4 a background Ch concentration (5 µM) representative of average brain extracellular Ch tonic levels 5 (Brehm et al., 1987;Garguilo and Michael, 1996;Parikh et al., 2004). Unlike previous studies, this 6 allowed us to get a clear distinction between phasic vs. steady-state (tonic) sensor responses. 7 Notably, upon removal of O 2 from solution and in the presence of background Ch, most biosensors 8 responded to consecutive O 2 steps with a fast transient (phasic component) before reaching a steady-9 state ( Figure 7A). Both phasic and tonic components decreased with O 2 baseline, but not equally. The 10 phasic response was usually still prominent upon the exhaustion of the tonic component ( Figure 7A). 11 We quantified these differences by fitting the Michaelis-Menten equation to the tonic changes and the 12 Hill equation to the cumulative phasic peaks, as the latter did not appear to follow a pure Michaelis-13 Menten profile ( Figure 7B). The resulting phasic K 0.5 O 2 values for phasic responses were, on average, 14 one order of magnitude larger than tonic KmO 2 , reinforcing that the phasic component vanishes at 15 much larger O 2 baselines than tonic responses (p<0.0005, Figure 7C). Importantly, the phasic ChOx 16 response was not matched by a fast O 2 transient following each addition, as O 2 raised considerably 17 slower than the peak in COA ( Figure 7D). 18 Next, we assessed how the coating composition and its physical properties could modulate the sensor 19 O 2 -dependence. First, we calculated the biosensor efficiency (ratio of Choline vs. H 2 O 2 sensitivities), 20 as a proxy to the enzyme loading in the coating and plotted it against tonic KmO 2 app ( Figure 7E). The 21 result revealed a decreasing trend ( Figure 7E), suggesting that biosensors with a high enzyme loading 22 have low sensitivity to tonic O 2 changes (low-KmO 2 app ). Strikingly, the biosensors with the lowest 23 tonic KmO 2 app exhibited the highest phasic peaks ( Figure 7F), with maximal phasic responses to O 2 24 decreasing exponentially as a function of KmO 2 app ( Figure 7G). 25 In order to further detail on the effect of O 2 baseline on non-stationary COA, we split the sensor 26 calibrations into two groups according to their tonic O 2 -dependence. As already suggested in Figure  27 7D, we confirmed that the larger phasic responses in the low-KmO 2 app vs. high-KmO 2 app groups could 28 not be attributed to differences in the underlying O 2 profiles. Oxygen transients after each addition 29 were negligible and did not significantly differ across KmO 2 app groups (p=0.998, Figure 7H). 30 Noteworthy, in the low-KmO 2 app group, the highest COA peak was achieved in response to the second 31 O 2 addition (10 µM of cumulative [O 2 ]) rather than to the first in 6 out of 7 calibrations (marginally 32 significant difference between 5 and 10 µM O 2 responses, p=0.076, paired t-test). Furthermore, the 33 same COA peaks had the longest decay across the two KmO 2 app groups (p<0.0005, Figure 7I). These 34 non-monotonic profiles with respect to [O 2 ] reflected the deviation of the cumulative phasic COA vs. calibration fittings (e. g. Figure 7B) were above 2 for sensors with low KmO 2 app and significantly 1 decreased towards 1 as KmO 2 app increased (p<0.05). These observations suggest a cooperative 2 mechanism that enhances phasic responses as the O 2 baseline increases, in biosensors with low tonic 3 In summary, we show that, under a Ch background, ChOx biosensors respond to O 2 with a transient 5 increase in enzyme activity before reaching a steady-state. Importantly, phasic and tonic components 6 were apparently mutually exclusive and their relative magnitude was sensitive to the properties of the 7 enzyme coating, namely enzyme loading. Sensors with high enzyme loading have a low tonic O 2 -8 dependence and show large phasic responses that seem to be potentiated by the O 2 baseline. 9 A possible explanation for the sensors' phasic and tonic components and their anti-correlation is the 10 local consumption of Ch within the coating milieu, boosted by phasic O 2 . In that case, a large Ch 11 depletion, in sensors with high enzyme loading, would blunt the tonic ChOx response to O 2 while 12 sparing the initial non-steady-state peak in enzyme activity. 13  Modeling in vitro biosensor responses reveals the mechanisms underlying tonic 1 and phasic oxygen-dependence 2 In order to provide a theoretical ground for our in vitro observations, we have simulated the behavior 3 of biosensors in calibration conditions. We numerically solved a system of partial differential 4 equations describing the diffusion of Ch and O 2 in the coating and their interaction with the enzyme, 5 leading to H 2 O 2 generation. 6 To mimic our experimental calibrations, we simulated sensor responses to 5 µM step increases in O 2 7 (starting from zero) under a constant level of 5 µM Ch in the bulk solution. Remarkably, in line with 8 our experimental findings, the model predicted phasic and tonic components of sensor response to O 2 9 whose magnitude depended on O 2 baseline ( Figure 8A). Sensors with high enzyme loading showed 10 higher phasic peaks and tonic responses that saturate at lower O 2 baselines than sensors with a low 11 enzyme amount ( Figure 8A). To get a more resolved characterization of biosensor's O 2 dependence, 12 we next generated response curves at 1 µM O 2 steps, until saturation of enzymatic H 2 O 2 generation 13 was nearly reached (see Methods). By simulating a range of coating thicknesses and enzyme 14 concentrations, we found that both parameters decreased KmO 2 app of tonic responses ( Figure 8B) and 15 increased the magnitude of phasic peaks ( Figure 8C). Interestingly, our model predicts that particular 16 combinations of coating thickness and enzyme concentration can be used to optimize sensor 17 sensitivity to Ch (at saturating O 2 levels) ( Figure S6A). Yet, such a strategy is not expected to 18 concomitantly reduce phasic and tonic O 2 dependence, which seem to be mutually exclusive. In To get further clues into the factors shaping sensors' O 2 -evoked responses, we analyzed the 26 concentration dynamics of Ch and O 2 in the coatings during simulated calibrations. We observed that, 27 under high enzyme loading, Ch is rapidly depleted in the coating as O 2 levels in solution increase 28 ( Figure 8E). This effect is less pronounced in sensors with low enzyme loading ( Figure 8F). 29 Interestingly, significant O 2 consumption, observed mainly in coatings highly loaded with enzyme, 30 was stronger for low O 2 levels before reaching saturation of the sensor tonic response ( Figure 8G). 31 These observations suggest that depletion of Ch in the enzyme coating is the limiting factor that 32 shapes sensors' tonic responses to O 2 . 33 To further investigate phasic responses under high enzyme loading, in addition to substrate profiles, concentrations of E red BA and E red GB change oppositely as O 2 is increased, which maintains the sum 1 of both intermediates relatively stable ( Figure S6B). At 1 µM buffer O 2 steps, the profiles of reduced 2 intermediates, Ch and O 2 at the electrode surface, suggest that phasic biosensor peaks result from a 3 combination of multiple factors ( Figure 8H). As expected, the sensor's tonic response steeply decays 4 upon Ch depletion in the coating. This decrease is accompanied by a sharp rise in the instantaneous 5 ΔO 2 at the electrode surface evoked by O 2 increments in the bulk solution. As the rate of O 2 6 consumption depends on the concentration of reduced enzyme-bound complexes, the increasing 7 profile of ΔO 2 results, in a first stage, from the depletion of the E red BA complex, followed by the 8 decrease in E red GB ( Figure S6B). In turn, the amount of E red BA and E red GB depends on both Ch and 9 O 2 , which leads to a summed profile that is shifted to the right relative to the sensor's tonic response.

3
The ΔO 2 is the initial rise in O 2 following each O 2 increment in solution (at a lag of 0.3 s).
for the highly sensitive and unbiased simultaneous measurement of putative cholinergic activity and 3 O 2 dynamics in the brain. Our approach, based on the differential plating of recording sites to create 4 pseudo-sentinel channels outperforms previous common-mode rejection strategies, which were 5 limited by diffusional cross-talk (Burmeister et al., 2003;Parikh et al., 2007;Santos et al., 2015). This 6 strategy allowed us to substantially reduce the size and increase the spatial confinement of recording 7 sites by using a 17 µm wire tetrode as the biosensor electrode support. Our recordings in freely-8 moving and head-fixed rodents reveal the usefulness of this compact multi-site design to clean 9 artifacts from sensor signals and assess the correlation between the activity of the immobilized 10 enzyme and brain extracellular O 2 on a fast time-scale. Importantly, this method can be generalized to 11 improve the selectivity and address the in vivo O 2 -dependence of any oxidase-based biosensor. 12 By simultaneously measuring the activity of immobilized ChOx and extracellular O 2 in the 13 hippocampus of behaving mice, we found that fast biosensor signals correlate in amplitude and time 14 with O 2 transients evoked by behavioral and network dynamics events exemplified by locomotion 15 bouts and SWRs. Notably, the relationship between COA and O 2 profiles was apparently not sensitive 16 to the underlying neurophysiological or behavioral context and was preserved during periods without 17 appreciable SWR incidence or locomotion. By using two different methods to manipulate 18 extracellular O 2 , we show that O 2 fluctuations in the physiological range can drive phasic COA. 19 Remarkably, the time-scale dependence of biosensor response amplitude and lag relative to 20 exogenous O 2 qualitatively matched that of spontaneous profiles, suggesting that the same 21 directionality happens in spontaneous conditions. 22 Locomotion-related O 2 elevations in head-fixed mice have been recently shown to be modulated 23 mainly by respiration rate (Zhang et al., 2019), whereas SWR-evoked O 2 peaks have been indirectly 24 inferred by fMRI and likely result from neurovascular coupling (Ramirez-Villegas et al., 2015). Thus 25 our study provides a link between the neurophysiological or systemic mechanisms that modulate brain 26 O 2 levels and the response of ChOx-based biosensors in vivo. As it is an intrinsic component of any 27 behavioral task, our results highlight the importance of controlling for O 2 -evoked biosensor signals 28 related to locomotion or movement. It is thus likely that, in reward-related tasks, locomotion related to 29 reward retrieval elicits few seconds delayed phasic changes in O 2 that drive transient increases in 30 COA. Likewise, a high incidence of SWRs in reward locations, reflecting a consummatory state 31 (Buzsáki, 2015), might trigger O 2 transients and, in turn, phasic ChOx biosensor responses. These two 32 examples provide alternative explanations for previously reported cholinergic transients inferred from 33 COA signals in the prefrontal cortex and hippocampus of rodents engaged in cognitive tasks (Howe et 34 al., 2017;Parikh et al., 2007;Teles-Grilo Ruivo et al., 2017). Importantly, since the rate of the expected to decrease following experimental controls that have been aimed at inhibiting or removing 1 cholinergic inputs (Parikh et al., 2007). 2 Our in vitro characterization of the biosensor O 2 dependence provided critical insights to interpret the 3 in vivo relationship between COA and O 2 . We found robust O 2 -evoked phasic responses whose 4 amplitude was anti-correlated with sensors' tonic O 2 dependence. Interestingly the phasic peaks 5 decreased with O 2 baseline but were detected even under relatively high O 2 levels, suggesting a high 6 likelihood of such responses to occur in vivo. Thus, our data emphasize the impact of O 2 -evoked non-7 steady-state biosensor dynamics on fast time-scale in vivo measurements. This has not been described 8 in previous studies partly because the experimental procedures used to study O 2 -dependence were 9 unable to unmix tonic and phasic components of sensor response (Baker et al., 2017;Burmeister et 10 al., 2003;Dixon et al., 2002). Instead of generating a continuous O 2 increase, we induced step estimates, these conclusions can be extrapolated to sensor geometries and enzyme coating 31 compositions that we have not covered. Furthermore, mathematical model of the biosensor described 32 here provides a rigorous approach for exploration and optimization of the design of any future 33 enzymatic biosensors and investigation of their behavior under non-stationary in vivo conditions. 34 The phasic O 2 -evoked COA signals described in vitro and predicted by mathematical modeling 35 provide crucial information to interpret the time-scale dependence of in vivo sensor dynamics 36 triggered either to spontaneous or exogenously evoked O 2 peaks. In light of those results, the temporal advancement and amplitude drop of biosensor transients relative to O 2 , as O 2 rises for longer periods, 1 is compatible with a major contribution of the phasic component of biosensor's O 2 -dependence. This 2 observation suggests that, in vivo, our biosensors operated in a regime close to saturation of the tonic 3 response and highlight the effect of non-steady-state ChOx biosensor responses to O 2 , which can be 4 erroneously attributed to Ch. 5 Our observations disfavor the quantitative optimization of coating properties as a strategy to reduce 6 sensors' O 2 dependence. Instead, we anticipate that strategies that increase O 2 accumulation in the 7 enzyme coating (Njagi et al., 2008) might have relative success, although the high O 2 levels required 8 to cancel phasic responses are hard to reach passively. Alternatively, some oxidase-based biosensors 9 for in vitro applications have incorporated an electrochemical actuator that enables local manipulation 10 of O 2 concentration based on water electrolysis (Park et al., 2006). However, applying such design in 11 vivo would require miniaturization and separation between the O 2 generation compartment and the 12 brain to avoid electrolytic tissue damage. Furthermore, in addition to the main O 2 confound, one 13 cannot completely rule out a possible modulation of COA in vivo by factors that affect enzyme 14 conformation, including temperature and pH (Hekmat et al., 2008). Although the enzyme is poorly 15 sensitive to the modest variations of these factors in vivo (Csernai et al., 2019;Venton et al., 2003), it 16 would be relevant to characterize their potential contribution to COA dynamics in future studies. A 17 further validation of ChOx-based measurements could be achieved by confronting the dynamics of 18 COA with that of cholinergic signals measured with other sensing approaches, under the same 19 experimental conditions. The latter technics include optogenetically-taged single unit recordings or 20 fluorescence reporters, which have previously revealed fast cholinergic dynamics related to arousal, 21 sensory sampling, negative reinforcements and unexpected events (Eggermann et al., 2014;Hangya et 22 al., 2015;Lovett-Barron et al., 2014;Reimer et al., 2016). 23 In summary, our results suggest that ChOx biosensor signals in vivo are composed of a mixture of O 2 -24 related artifacts and true cholinergic dynamics. The weight of each factor depends on the time-scale, 25 with slow state-related changes reflecting cholinergic dynamics with low O 2 -related contamination 26 and fast transients, except for a minor fraction (less than ca. 5% in our data), being caused by phasic 27 O 2 fluctuations. We show that O 2 transients can be triggered by cognitively-relevant events, such as 28 locomotion and periods of high SWR incidence, and confound the ChOx-based measurement of 29 cholinergic activity. Thus, our study reveals a previously ignored phasic O 2 -dependence of ChOx-30 based biosensors which is critical to control for, particularly in the case of fine time-scale 31 measurements of ACh. 32 Importantly, our conclusions can probably be generalized to other oxidase-based biosensors that have 33 been used to measure neurotransmitters or metabolically-relevant molecules in the brain (Chatard et 34 al., 2018;Dixon et al., 2002;Hascup et al., 2013;McMahon et al., 2007). The exact extent of phasic 35 and tonic O 2 dependence would depend on the particular enzyme kinetics and on the basal 36 extracellular concentrations of analyte relative to the magnitude of changes in the brain.
Solutions were prepared in ultra-pure deionized water (≥18MΩ.cm) from a Milli-Q water purification 4 system. 5 Tetrode fabrication and platings 6 The microelectrode support material was a 17 µm diameter Platinum/Iridium (90/10) wire insulated 7 by a polyimide coating (California Fine Wire Company). Tetrodes were fabricated using standard 8 methods (Gray et al., 1995). Briefly, four wires were twisted together and heated to melt the 9 insulation, creating a stiff bundle of twisted wires. The wires' insulation at the untwisted ending of the 10 tetrode was then removed and the tetrode was inserted in a silica tube (150 µm inner diameter), which 11 was glued to a holder that allowed easy manipulation of the tetrode. Next, the untwisted endings of 12 the tetrode wires were soldered to the pins of an adapter fixed to the tetrode holder, allowing 13 connection to the potentiostat's head-stage. Finally, the twisted ending of the tetrode was cut using 14 micro-serrated stainless-steel scissors. 15 Tetrode surface treatments and platings were performed with a portable potentiostat (EmStat 3, 16 PalmSens BV), using a freshly-prepared Ag/AgCl wire (125 µm diameter, WPI inc.) as pseudo-17 reference electrode. Prior to platings, electrode surfaces were cleaned by swirling the tetrode tip in 18 isopropanol followed by an electrochemical treatment in PBS. For that purpose, we applied 70 cycles 19 of a square wave with a first step at +1.2 V for 20 seconds followed by a 4 seconds step at -0.7 V. All 20 tetrode sites were then gold-plated in a 3.76 µM aqueous solution of tetrachloroauric acid by applying 21 impedances were checked after gold-plating using a nanoZ impedance tester (Multichannel Systems, Choline oxidase immobilization was performed as previously described (Santos et al., 2015). Briefly, 28 a 0.5% (w/v) chitosan stock solution was solubilized in saline (0.9% NaCl) under stirring at pH 4-5, 29 adjusted by addition of HCl. After solubilisation, the pH was set to 5-5.6 by stepwise addition of saline. The tetrode tip was coated by multiple dips (10-15) in a small drop of ChOx immobilization 3 mixture, created using a microliter syringe (Hamilton Co.). The microelectrode and syringe were 4 micromanipulated under a stereomicroscope. The coating procedure was stopped when the 5 chitosan/protein matrix was clearly visible under the microscope. 6 Following enzyme immobilization, tetrode site's response to Ch was tested and meta-7 phenylenediamine (m-PD) was electropolymerized on the pair of sites with the highest sensitivity 8 (please see Biosensor calibrations sub-section). Electropolymerization was performed in a nitrogen 9 bubbled oxygen-free PBS solution of 5 mM m-PD by DC amperometry at +0.6 V during 1500 s. The 10 biosensors were stored in air and calibrated on the day after m-PD electropolymerization. 11

Biosensor calibrations
12 All in vitro tests were done in a stirred calibration buffer kept at 37 ºC using a circulating water pump 13 (Gaymar heating/cooling pump, Braintree Scientific, Inc., USA) connected to the calibration beaker. 14 Routine calibrations after enzyme immobilization and m-PD electropolymerization steps were 15 performed by amperometry at a DC potential of +0.6 V vs. Ag/AgCl pseudo-reference electrode. 16 After stabilization of background current in PBS, sensors were calibrated by three consecutive 17 additions of 10 µM Ch followed by 4.9 µM H 2 O 2 . In the case of complete (m-PD polymerized) 18 biosensors, the response to 1 µM DA and 100 µM AA was also tested. Voltammograms of H 2 O 2 were 19 done by consecutive additions of 4.9 µM H 2 O 2 at different applied DC voltages. 20 In vitro O 2 tests were carried out in a sealed beaker. After addition of 5 µM Ch, the calibration buffer µM) were added to the medium from an O 2 -saturated PBS solution previously bubbled with pure O 2 23 during 20 min. Biosensor response to O 2 additions was measured at +0.6 V vs. Ag/AgCl. In a set of 24 calibrations, O 2 was concurrently measured by polarizing a gold-plated channel at -0.2 V, while 25 keeping the remaining at +0.6 V. To obtain O 2 voltammograms, calibrations consisting in 3 additions 26 of 5 µM O 2 were performed at different applied DC potentials. In these experiments, O 2 was purged 27 from the solution after each voltage step. 28

Experimental model and subject details 29
Freely-moving recordings were performed on a 6 months old Long-Evans rat and head-fixed 30 recordings were done in a total of six 3-7 month old C57BL/6J mice. All experimental procedures 31 were established, and have been approved in accordance with the stipulations of the German animal 32 welfare law (Tierschutzgesetz )(ROB-55.2-2532.Vet_02-16-170). silicon probe array (A1x32-7mm-100-1250-H32, NeuroNexus Technologies, Inc) in the rat brain. 3 The general procedures for chronic implantations of electrode arrays have been described in detail 4 (Vandecasteele et al., 2012). Prior to surgery, the tetrode biosensor and the silicon probe array were 5 attached to home-made microdrives. Silicon probe's sites were then gold-plated until impedances at 1 6 kHz decreased below 200 kΩ (Ferguson et al., 2009). Anesthesia was induced with a mixture of 7 Fentanyl 0.005 mg/kg, Midazolam 2 mg/kg and Medetomidine 0.15 mg/kg (MMF), administered 8 intramuscularly. The rat was continuously monitored for the depth of anesthesia (MouseStat, Kent 9 Scientific Corporation, Inc.). After the MMF effect washed out, anesthesia was maintained with 0.5-10 2% isoflurane via a mask, and metamizol was then subcutaneously administered (110 mg/kg) for 11 analgesia. The tetrode biosensor was implanted in the cortex above the right dorsal hippocampus (AP 12 -3.7 mm, ML -2.5 mm, DV -1.2 mm, relative to bregma) and the silicon probe array was implanted at 13 0.8 mm posterior from it, spanning most cortical and hippocampal layers (AP -4.5, ML -2.4, DV -14 3.4). The microdrives were secured to the skull with a prosthetic resin (Paladur, Kulzer GmbH). An 15 Ag/AgCl (125 µm thick) silver wire coated with Nafion (Hashemi et al., 2011) was inserted in the 16 cerebellum and served as the pseudo-reference electrode for electrochemical recordings. The ground 17 for electrophysiology was a stainless-steel screw implanted at the surface of the cerebellum. To 18 reduce line noise, this Ag/AgCl wire was shorted with the electrophysiology ground at the input of the 19 electrochemical head-stage. 20 Mice used in head-fixed recordings were implanted with a head-post. Anesthesia followed the same 21 procedures as in rats. A mixture of 0.05 mg/kg Fentanyl, 5 mg/kg Midazolam and 0.5 mg/kg 22 Medetomidine was administered intraperitoneally to induce anesthesia, which was later maintained 23 with isoflurane and Metamizol (200 mg/kg). A craniotomy was made above the dorsal hippocampus 24 and a Nafion-coated Ag/AgCl wire was implanted in the cerebellum. Depending on the head-post 25 configuration, it was cemented either to the back of the skull above the cerebellum or above the 26 hemisphere contralateral to the craniotomy, using UV-curing dental cement (Tetric EvoFlow, Ivoclar 27 Vivadent AG). Finally, the craniotomy and surrounding skull were covered with a silicone elastomer 28

(KWIK-CAST, World Precision Instruments Inc.). 29
Electrochemical and electrophysiological equipment and recordings 30 Amperometric measurements were performed using either a 4-channel (MHS-BR4-VA) or a 8-31 channel (MBR08-VA) potentiostat connected to 4-or 8-channel miniature head-stages, respectively 32 (npi electronic GmbH, Germany). In addition to providing a higher channel count, the MBR08-VA 33 allowed independent control of the potential applied to each channel. This feature enabled from the head-stage was amplified and digitized at 30 kHz and stored for offline processing using the 1 Open Ephys acquisition board and GUI (Siegle et al., 2017). 2 Freely-moving electrochemical recordings were done using the MHS-BR4-VA potentiostat and the 3 corresponding 4-channel miniature head-stage. Electrophysiological signals were pre-amplified using 4 a 32-channel head-stage with 20x gain (HST/32V-G20, Plexon Inc.) which was connected to a 5 multichannel acquisition system (Neuralynx Inc). Data was acquired at 32 kHz and stored for offline 6 processing. Both head-stages were connected to the respective recording systems via light and flexible 7 cables suspended on a pulley so as not to add weight to the animal's head. The tetrode biosensor was 8 gradually lowered through the cortex until it reached the hippocampal CA1 pyramidal layer. Correct 9 targeting was assessed based on brain atlas coordinates and by the identification of hippocampal 10 ripples. Recordings were performed in a square open-field arena (1.5 m x 1.5 m), where the animal 11 could sleep or explore the environment at will. Chocolate sprinkles were occasionally spread on the 12 maze to enforce exploratory behavior. The position of the rat head was derived from small reflective 13 markers attached to the chronic implant. A motion capture system consisting of multiple infrared 14 cameras (Optitrack, NaturalPoint Inc.) was used to 3D-track the markers with high spatio-temporal 15 resolution (data acquired at 120 Hz). 16 Head-fixed recordings in mice were performed using the MBR08-VA potentiostat and respective 17 head-stage. After fixing the mouse, the layer of silicone elastomer protecting the craniotomy was 18 removed. The dura matter above the target brain region was removed and the tetrode biosensor was 19 slowly inserted through the cortex until the hippocampal CA1 pyramidal layer was reached. Accurate 20 targeting was assessed according to brain atlas coordinates and/or by the online identification of 21 hippocampal ripples in the recording. In 5 out of 10 recording sessions mice were head-fixed on a 22 cylindrical treadmill. Movement was quantified based on the video optical flow arising from treadmill 23 rotation using Bonsai (Lopes et al., 2015). In the remaining sessions, mice were head-fixed on a 24 rotating disc which encoded its turns. The analog signal from the disc encoder was fed into the Open 25 Ephys acquisition board and used to quantify mice locomotion. 26

Data analysis
27 Raw recordings were preprocessed by low-pass filtering and resampling at 1 kHz. All data analysis 28 was done in Matlab using custom-made functions (MathWorks). 29

In vitro sensor responses 30
In vitro analysis of biosensor responses was performed on 10 Hz downsampled data, low-pass filtered 31 at 1 Hz. Sensitivities to Ch and O 2 were determined by linear-regression of the responses to the first 3 32 analyte additions, whereas the sensitivities to H 2 O 2 and interferents were estimated from a single 33 addition. Minimal selectivity ratios were estimated by dividing the lower limit of the 95% confidence interval (CI) of Ch sensitivity by the upper limit of the CI of interferent responses. Following pseudo-1 sentinel subtraction, the biosensors' limit of detection (LOD) for Ch was calculated as the Ch 2 concentration corresponding to 3 times the baseline standard deviation (SD). The T 50 and T 90 response 3 times were defined as the time between the onset of current increase in response analyte and 50% or 4 90% of the maximum current, respectively. 5 Artifact cancellation by common-mode rejection 6 In vivo electrochemical signals from Ch-or O 2 -sensitive sites were cleaned by subtraction of the 7 respective 1 kHz data by the pseudo-sentinel (Au-Pt/m-PD) channel upon a frequency-domain 8 correction. The latter procedure has been described in detail and optimizes common-mode rejection 9 by correcting phase and amplitude mismatches between channels arising from slight variations in 10 impedance (Santos et al., 2015). This correction was based on the estimation of a transfer coefficient 11 Local-field potential-related power spectrograms were computed using custom-made Matlab 30 functions based on multi-taper analysis methods (Mitra and Pesaran, 1999). Separation of brain states 31 in freely-moving recordings was based on LFP spectral features and behavior. Active wake states 32 were defined as periods when the animal moved vigorously and continuously (>30 s) and showed a 33 prominent LFP spectral peak in theta range (6-10 Hz). Quiet wakefulness or immobility was defined bouts). Long periods (>1 min) without movement and without prominent theta, rather showing high 1 delta power (1-4 Hz) were ascribed to NREM sleep. Rapid eye movement sleep was detected as 2 periods showing a sustained theta band (>30 s) and negligible movement. 3 Sharp-wave/ripples and related biosensor signals 4 To detect SWRs, the wide-band electrochemical or electrophysiological signal was band-pass filtered 5 (120-200 Hz), squared and smoothed with a 4.2 ms standard deviation and 42 ms wide Gaussian 6 kernel. The square root of this trace was then used as the power envelope to detect oscillatory bursts. 7 The events exceeding the 98 th percentile of the power envelope distribution, having at least 5 cycles 8 and lasting less than 200 ms were detected as ripples. For the analysis of correlations between 9 integrated ripple power and SWR-triggered sensor signals, a ripple power envelope was obtained 10 upon Hilbert-transforming the band-pass filtered electrochemical signal (ripple band,. 11 Different ripple integration times were obtained by smoothing the power envelope with moving 12 average windows of different lengths. Correlations were then computed between smoothed ripple 13 power envelopes at SWR times and the corresponding changes in COA or O 2 (relative to their 14 baseline value 1s prior to SWRs) at different lags from SWRs. 15 The amplitudes of SWR-related COA and O 2 were obtained from the difference between the values at 16 SWR lags corresponding to peaks and onsets, extracted from average SWR-triggered traces. 17

18
To detect locomotion bouts in freely-moving recordings, speed was computed from the derivative of 19 low-pass filtered position (0.5 Hz). Speed was then band-pass filtered (0.02-0.2 Hz) and locomotion 20 bouts were detected as peaks in speed that exceeded a manually-set threshold. In head-fixed 21 recordings on the disc, mouse locomotion was derived from its rotation in 1 s bins. When head-fixed 22 on the treadmill, mouse locomotion was quantified based on the optic flow from a recorded video, 23 choosing a region of interest that covered only a moving part of the treadmill. The signal was 24 resampled to 1 Hz in order to match the sampling rate of locomotion on the rotating disc. Locomotion 25 bouts were detected as peaks on the band-pass filtered speed (0.02-0.2 Hz) that exceeded a manually-26 defined threshold. 27 The amplitude of locomotion-bout-associated COA and O 2 signal transients was calculated based on 28 the difference between the values at manually-determined times of transient onsets and peaks, 29 associated with each event. Likewise, the times of locomotion bout onsets (used to calculate speed 30 change) and the associated peaks in theta power were manually defined based on visual inspection of 31 speed time courses and LFP spectrograms, respectively.

Oxygen-related ChOx transient signals analysis 1
The amplitude of broad-band spontaneous and exogenously-induced O 2 transients and COA transients 2 associated with them was calculated, for each event, as the difference between the peak value and that 3 at semi-automatically defined time of the transient's onset.
filtered in a frequency range [f/2 f] with corner frequency f, defining each filter, varying from 0.05 to 7 0.5 Hz. Peaks of the transients were then detected in both band-passed COA and O 2 signals, selecting 8 the events that exceeded half of the maximal peak amplitude. In each frequency band, COA transients 9 used for the analysis were restricted to be within 0.75/f of any O 2 peak. The amplitudes of O 2 and 10 COA transients were defined as peak magnitudes derived from the band-pass filtered signals. For each 11 filter band, the linear relationship between COA and O 2 peak amplitudes was estimated as a slope of 12 the linear model fit to all detected event pairs for this filter. To interpret the nonlinear shape of the O 2 13 transients captured by each filter, we extracted rise time (from trough to peak) from the median peak 14 aligned O 2 transients and used these values in Figures 4E-H. 15 The transient onsets and peaks of exogenously-induced changes in O 2 and CAO were manually 16

detected. 17
Modeling biosensor responses in vitro 18 We simulated biosensor responses in vitro by numerically solving a system of partial differential 19 equations that describe the diffusion of the substrates Ch and O 2 in the enzyme coating and their 20 interaction with the enzyme, leading to product formation. 21 The buffer solution where the biosensor was placed for calibration is a free-flow environment, in 22 which the concentrations of Ch and O 2 are constant over time. Therefore, considering R is the coating 23 thickness, at the boundary between the biosensor coating and the calibration buffer Ch is kept 24 constant at 5 µM during the calibration.
Oxygen is changed in steps (x), starting from 0, during the calibration, so between O 2 step increases 29 we have: (3) 32 Enzyme substrates (Ch and O 2 ) diffuse from the bulk solution into the enzyme layer, eventually 1 reaching the electrode site, whereas H 2 O 2 is locally generated and diffuses within the coating. As the 2 size of our recording sites is very small, this process is better described by a spherical diffusion 3 equation: 4 (4) where S represents substrates or H 2 O 2 concentration, D S is the respective diffusion coefficient and r is 5 the distance to the electrode surface. 6 In order to simulate realistic two-substrate biosensor responses, we modeled the formation of enzyme 7 intermediate complexes resulting from Ch binding to the enzyme and O 2 oxidation reactions, which 8 have been described in detail (Fan and Gadda, 2005) (Scheme 1). Briefly, enzyme-bound Ch (ECh), 9 which is in equilibrium with the free reactant species (E and Ch), undergoes a chemical step leading 10 to the reduction of the FAD enzyme prosthetic group. In this nearly irreversible step, Ch is converted 11 to betaine aldehyde, which remains mostly enzyme-bound (E red BA). The first step in which H 2 O 2 is 12 produced results from the oxidation of FAD red by O 2 (E ox BA) followed by a second chemical step in 13 which FAD ox is reduced by betaine aldehyde. The resulting enzyme-bound glycine betaine (E red GB) is 14 then oxidized by O 2 , producing H 2 O 2 . The reaction cycle is completed with release of glycine betaine 15 bound to FAD-oxidized enzyme (E ox GB). 16 The instantaneous change in the concentration of enzyme substrates, free enzyme, enzyme-bound 17 intermediate complexes and reaction products can then be described by the following system of Note that we ignored the kinetics of glycine betaine formation, as it is not relevant regarding the 1 sensor signal transduction and it would not affect the concentrations of any reaction species. 2 In the beginning of the simulated calibration, there is no Ch in the coating, so all enzyme molecules 3 are free, at a concentration that equals the total enzyme concentration ([E] 0 ).
where F is the Faraday's constant and πr e 2 is the electrode surface area. 24 Considering the initial conditions described above the system of partial differential equations was 25 numerically solved by discretization in space and time (time and space steps dt = 0.1 ms and dr = 1 26 µm, respectively) using the finite difference approximation method (Baronas et al., 2009). 27 The values ascribed to the variables used in the model are summarized in Table 2. The rate constants 28 corresponding to the reaction mechanism in Scheme 1 were extracted from the corresponding 29 literature (Fan and Gadda, 2005). An enzyme concentration in the coating of 263 uM was estimated 30 from its concentration in the mixture used for coatings, ignoring drying effects upon coating. 31 Variations in enzyme concentration were simulated around that value. We estimated the diffusion 32 coefficients of enzyme substrates and H 2 O 2 in the coating taking into account the free diffusion 33 coefficients in solution multiplied by a hindrance factor α. The latter was set at 0.8, considering the 34 expected effect of macromolecular crowding on diffusion, for protein concentrations in the range of Scheme 1. Minimal kinetic mechanism of Choline Oxidase. (E) Free enzyme, (ECh) enzyme-bound Ch, 1 (E red ) enzyme with reduced FAD co-factor, (E ox ) enzyme with oxidized FAD co-factor, (BA) betaine aldehyde, 2 (GB) glycine betaine.

Supplementary Materials
for both factors in ChOx and O 2 data). The same applied for recording 2 (p<0.0001 and F>12 for 7 both factors in ChOx and O 2 data). As for recording session 3, the SWR count effect was significant 8 for both sensor signals (p<0.05 and F>3.4) and the summed ripple power significantly affected O 2 9 (p<0.05 and F>4), but was not consistently related to ChOx transients amplitude (p=0.07, F=2.71). 10 occurrence rate of O 2 peaks during either whole recordings or periods outside SWRs and locomotion 2 bouts (events occurring within -5 s to +1 s from SWRs or within -14 s to + 4 s from peaks in speed 3 were excluded). The difference between O 2 peak rates was significant (p<0.01, paired t-test). (B-C) 4 plots display the same statistics as in an example on Figure 5E, but for all recordings (B) Lag of ChOx 5 activity peaks relative to O 2 for changes in O 2 detected with varying rise time (from whole recordings,