Neonatal brain injury causes cerebellar learning deficits and Purkinje cell dysfunction

Premature infants are more likely to develop locomotor disorders than term infants. In a chronic sub-lethal hypoxia (Hx) mouse model of neonatal brain injury, we recently demonstrated the presence of cellular and physiological changes in the cerebellar white matter. We also observed Hx-induced delay in Purkinje cell (PC) arborization. However, the behavioral consequences of these cellular alterations remain unexplored. Using the Erasmus Ladder to study cerebellar behavior, we report the presence of locomotor malperformance and long-term cerebellar learning deficits in Hx mice. Optogenetics experiments in Hx mice reveal a profound reduction in spontaneous and photoevoked PC firing frequency. Finally, treatment with a gamma-aminobutyric acid (GABA) reuptake inhibitor partially rescues locomotor performance and improves PC firing. Our results demonstrate a long-term miscoordination phenotype characterized by locomotor malperformance and cerebellar learning deficits in a mouse model of neonatal brain injury. Our findings also implicate the developing GABA network as a potential therapeutic target for prematurity-related locomotor deficits.

Adaptive cerebellar learning is tested by comparing ladder rung steptimes between Hx and Nx groups during training and challenge phases of the learning paradigm. While training sessions measure basal steptimes, challenge sessions measure steptimes when an obstacle (US) is presented, paired with a 250-millisecond preceding warning tone (CS). A higher postperturbation steptime during session 5 indicates the start of associative learning, where mice are beginning to adapt their movements to avoid the obstacle. In subsequent sessions, normal mice learn to adapt their movements by associating the US and CS, increasing movement speed to avoid the obstacle, thus resulting in a decrease in post-perturbation steptimes.

Optogenetic targeting of PCs in the cerebellar cortex
Viral injections were performed stereotactically into Pcp2-cre (Mpin) animals at P9, at three injection sites in the simple lobule, bilaterally: anterior/posterior -5.6 mm / -5.4 mm / -5.5 mm; medial/lateral -2.1 mm/ -2.0 mm/ 2.3 mm from the bregma; dorsal/ventral -0.35 mm from the dura. During the entire injection procedure, mice were placed under 2% isoflurane anesthesia in oxygen delivered from a precision vaporizer (Harvard Apparatus). A total of 1.0 µl viral solution was delivered using a 26-gauge micro syringe (Hamilton, USA). After injection, craniotomy holes were covered with styptic power (Kwik stop, ARP laboratory). The scalp was repositioned using tissue adhesive (3M Vetbond, 3M) and ketorolac was administered subcutaneously at a dose of 5 mg per kg body weight. Pups were then returned to their foster mother (female CD1 mouse) until subsequent steps in our experimental protocol. Although protocols by other groups recommend 14 days viral incubation post-injection 10 , our experimental paradigm required only 4 days of incubation for robust viral labeling (Supplementary fig. 1b). For adenoviral injections involving hypoxic groups, we used a slightly modified neonatal hypoxia treatment (10.5% O 2 from P3 through P7).
At 4 and 14 days post-injection, animals in each group were anesthetized with Isoflurane and trancardially perfused with 4% paraformaldehyde solution. Brains were dissected out and postfixed overnight in PBS. This was followed by incubation with 30% sucrose solution overnight. 40 µm thick coronal cerebellar slices were prepared using a vibratome (Thermo Scientific Microm HM 430). Cerebellar slices were first saturated with 20% goat serum for an hour and then incubated overnight with Rabbit anti-Calbindin antibody (1:2000 concentration, catalog no. CB-38a, Swant, Switzerland) on a shaker. Slices were then washed thrice with PBS, followed by secondary incubation with Alexa Fluor conjugated AffiniPure Goat Anti Rabbit IgG (Jackson Immuno Research code 111-605-144) for two hours at room temperature. Next step was phosphate buffer washing and the slices were mounted on slide with DAPI containing mounting solution (DAPI Fluoromount-GSouthern Biotech, USA). Low magnification (20X) images were obtained using a BX61 fluorescent microscope (Olympus). Virus-infected cells appeared red under green (λ=587nm) excitation light due to mCherry fluorescence. In the representative figures, anti-Calbindin labeled cells were pseudocolored as yellow (Supplementary fig. 1b). Both 4 days and 14 days post-virus-injection groups contained 3 animals each. From each mouse, eight sections (each section 40 µm thick) were randomly picked, and three 1 mm 2 areas were counted under the microscope for virus expression from each slice. Virus expression was defined as the percentage of mCherry + PCs which also co-label as anti-Calbindin + (Supplementary fig. 1c). Our quantification confirmed that only 4 days of post-injection incubation period was sufficient for ~60% virus expression in PCs of Pcp2-cre mice. Comparison of mean percentages of mCherry + /anti-Calbindin + double-labeled cells at 4 days and 14 days post-injection was statistically not significant (Supplementary fig. 1c, Two sample t test, t Statistic = -0.83851, DF = 17, Prob>|t| = 0.41338). Finally, to identify the extent of nonspecific viral labeling of cells in the cerebellar cortex, we counted mCherry+/calbindin-cells in the molecular layer. Confirming an earlier report 11 , we did notice non-specific viral expression in non-PC cells in the molecular layer in both Nx and Hx groups (Supplementary fig. 3 a,b, arrowhead). However, using our particular injection protocol which targeted the simple lobule, this non-specific labeling was exceedingly rare (2 non-PCs:133 PCs from n = 3 animals in Nx and 4 non-PCs:161 PCs from n = 3 animals; < 0.1 cells per (100 µm) 3 in Nx and < 0.2 cells per (100 µm) 3 in Hx;) and not significantly different between Hx and Nx groups (Supplementary fig.  3c, Unpaired t-test, two-tailed, t = 0.7686, df = 4, P = 0.4850; Supplementary fig. 3d, Fisher's exact test, two-tailed, P = 0.6938).
In a recent paper referenced above, the authors crossed the Pcp2-Cre Mpin line with the ChR2-EYFP line, and showed evidence for direct and robust photocurrents in 21% of MLIs in Pcp2-Cre-Mpin-ChR2-EYFP mice, indicating non-specificity in photoresponse to blue light stimulation 11 . As seen in Supplementary fig. 4, using in vitro optogenetics, we did not notice any light-evoked photoresponses in MLIs.
Since we obtained robust mCherry reporter expression in PCs of Pcp2-cre mice, we proceeded with optogenetic experiments combined with in vivo extracellular multielectrode array recordings from virus-injected mice at ages P13, P21, P30, and P45 in both Nx and Hx groups. For the pharmacological intervention group, a separate set of Nx and Hx mice at P30 and P45 were used. The stereotactic placement of electrodes is shown marked with red dots in Supplementary fig. 1d (right, image) in a coronal section of mouse cerebellum (Mouse Brain Library: C57BL/6J Coronal -Section 26) 12 .
In a set of calibration experiments with in vivo stimulation and recording from cerebellar PCs of Pcp2-cre mice, we tested a range of stimulation protocols that included photo-evoked spike activity at 1 Hz, 5 Hz, 10 Hz, 15 Hz, 20 Hz, 25 Hz, 30 Hz, 40 Hz up to 100 Hz (Supplementary fig. 1e). In order to determine the optimal stimulation paradigm 13 , we used stimulation duration from 1 ms to 100 ms (Supplementary fig. 1f). From this calibration, we concluded that cerebellar PCs with ChR2 displayed high fidelity (> 80%) with 25 Hz 473 nm blue LED pulses of 30 ms duration (Supplementary fig. 1g). Thus, we applied 25 pulses of blue light with a pulse width of 30 ms, and an inter-pulse interval of 10 ms [25 pulses × (30 + 10) = 1000 ms total optostimulation period] as part of our optimized paradigm for all in vivo optogenetics experiments.
Stimulation and simultaneous in vivo extracellular recording were acquired as represented in Supplementary fig. 1d, using 40 KHz OmniPlex D (version 1.11, Plexon, USA) neural data acquisition system, and preamplified using MiniDigi preamplifier (16 channels, Plexon USA). Recorded spike trains were then sorted offline using a combination of waveform template matching and primary component analysis (PCA) cluster method in Offline Sorter (version 3.0, Plexon, USA), and further analyzed in Neuroexplorer version 5.0 (Plexon, USA). PC simple spikes and complex spikes were detected independently. We performed firing frequency analysis over a two second time window (one second before and one second after optical stimulation) for all groups. For firing frequency analysis, only those PCs which had a mean post-optostimulation firing rate (optostimulation onset [t = 0] to 1000 ms) higher than the threshold defined as mean pre-optostimulation frequency (-1000 ms to optostimulation onset) + 1 standard deviation (S.D.) were selected. Regularity of PC spike firing over a time-period of 12 5 seconds was measured as coefficient of variance (CV). The CV was calculated as a ratio of the S.D. of interspike interval (ISI) to the mean of ISI of a given cell. Short term rhythmicity was measured as CV2 as per the following previously published 14 equation: For each PC, ISI values were normalized to the maximum ISI to avoid the effect of inter-PC differences in firing rates. For spike profile analysis of individual simple spikes, we plotted normalized distributions of spike duration, and slopes of depolarization and repolarization spike phase recorded from all units in each experimental group. Gaussian fits were estimated for each distribution. For complex spike detection, we used a voltage threshold applied to the primary spike in high-band-pass-filtered data 15 . Following automated sorting, each complex spike was further verified by visual inspection for spike shape, and the presence of spikelets. Firing pattern analysis was performed similar to analysis for simple spike data.
After obtaining whole-cell recording, we delivered a brief (10ms) or longer (several seconds) pulse of high-intensity LED light (470nm, Mightex, Pleasanton, CA) to stimulate PCs. The LED light was delivered via the optical path of the BX51W1 upright microscope.

PCs are less excitable to current injection in Hx mice
In our in vitro data, although we didn't find significant differences in current responses between virus-injected and non-injected Nx and Hx groups (Supplementary fig. 2b-d), we observed a difference in the injected current to elicited firing frequency relationship between Nx and Hx (Supplementary fig. 2e-f). We compared the best exponential fit [Y = Y0e (kX) ] of currentfrequency data from Nx and Nx + ChR2 PCs and observed that data from both groups were fitted to a single exponential curve (black curve, Supplementary fig. 2e) since the Y0 and k constants in the individual fits were not significantly different between these groups. Similarly, Hx and Hx + ChR2 groups also were fitted to a single exponential curve (red curve, Supplementary fig. 2f). Thus, the current-frequency within Hx and Hx + ChR2, or Nx and Nx + ChR2 groups does not change on the basis of ChR2 expression. However, when we compared between Nx and Hx groups, we observed that the current-frequency fit constants are significantly different from each other (Extra sum-of-squares F test, F (2,512) = 10.75, P < 0.0001).

PC spike profiles are altered in Hx mice and partially rescued following Tiagabine treatment
PC spike profiles are important indicators of cerebellar-cortex function and consequently locomotor behavior. As a corollary, abnormalities in PC spike profiles are potentially related to cerebellar dysfunction and locomotor deficits 17,18 . Therefore, we performed an analysis of PC simple spike profiles from P13 and P21 mice (Supplementary fig. 6,a-d, see below). In P13 Nx mice, mean spike duration is distributed across two peaks at 0.93 ms and 1.25 ms (Supplementary fig. 6b, black). At P21, only one peak is visible in Nx mice at 0.29 ms (Supplementary fig. 6b, blue). However, in the P21 Hx group, two peaks are seenat 0.49 ms and 0.93 ms (Supplementary fig. 6b, wine). The slower mean spike duration for the first peak, and the presence of a second peak at 0.93 ms suggests a spike profile developmentally similar to P13 Nx group. Similarly, depolarization phase slopes for Hx P21 (Supplementary fig. 6c, wine) are closer to Nx P13 (black), than to Nx P21. However, repolarization slopes of spikes are not substantially different between Hx P21 and Nx P21, when each is compared to Nx P13 (Supplementary fig. 6d). Taken together, our analysis of in vivo PC basal firing patterns, and spike profile analysis at P13 and P21 indicates that Hx results in a profound dysmaturation of electrophysiological properties of PCs during postnatal development.
Spike profile analysis indicated that PC simple spikes detected in Tiagabine-treated animals have a profile that is somewhat improved compared to Hx animals (Supplementary fig. 6e, Tiagabine-treated Hx animals, green linemean profile, light greenrepresentative profiles; compare to Hx animals, red linemean profile, light redrepresentative profiles, and Nx animals, black linemean profile, light grayrepresentative profiles). Mean spike duration in Tiagabine-treated Hx animals indicates the presence of two peaks at 0.35 ms and 0.56 mslikely indicative of a partial shift towards a normal profile (Supplementary fig. 6f, compare to Nx peak at 0.26 ms and Hx peak at 0.64 ms). Similarly, repolarization phase slopes are closer to Nx animals in Tiagabine-treated Hx animals (Supplementary fig. 6h), however, depolarization slopes are shifted to the right of Hx animals (Supplementary fig. 6g).

SUPPLEMENTARY DISCUSSION
A major caveat regarding our misstep data is the contribution of brain regions other than the cerebellum, such as the brainstem, limbic system, spinal cord, basal ganglia, and cerebral cortex 19 in modulating locomotor behavior. Since our animal model is a global hypoxia model aiming to recapitulate most of the clinical hallmarks of neonatal birth injury, multiple brain regions associated with locomotor behavior could be potentially altered. Interestingly, while many studies identify the cerebellum as being particularly vulnerable during the perinatal brain injury time window, other locomotor-related brain regions such as the brainstem 20 and pre-motor regions 21 have also been suggested to undergo modest structural alterations associated with prematurity. Likely non-cerebellar candidate brain regions affected by neonatal brain injury could most likely comprise the major basal ganglia nuclei (caudate, putamen, nucleus accumbens, and pallidum) and the mesencephalic locomotor region (MLR) of the brainstem, which have long been implicated in locomotor control. In fact, a recent longitudinal study indicates that smaller volumes of basal ganglia nuclei in very preterm infants are associated with long-term neurodevelopmental motor deficits at 7 years of age 22 . This study also showed association of smaller neonatal thalamic volumes with motor deficits, suggesting that the entire cerebello-thalamo-cortical pathway 23 may be disrupted due to neonatal injury. Thus, although we cannot exclude the possibility that other brain regions might also be involved in the locomotor abnormalities observed in Hx mice, the cerebellum is an important component of the locomotor control circuit and plays a fundamental role in regulating this behavior.
While locomotion itself involved the recruitment of multiple regions of the CNS, adaptive responses during locomotion are cerebellar-dependent. In our conditioned learning paradigm (Figure 1b, c), we have set the delay between CS and US to 250ms. This timing is important to measure "true" adaptive cerebellar learning. Classical eyeblink conditioning experiments have measured the percentage of conditioned learning to peak at a delay of ~250ms 24 , which is well correlated to single PC firing rates in trained adult mice 25 .
In our study, we also tracked behavioral changes across two different ages. We noticed that juvenile mice are faster at adapting to an obstacle on the ladder than older mice (compare mean session 5 steptime in Figure 3a and Figure 4a). In contrast to our results, a human study using the cerebellar-dependent split belt treadmill paradigm suggested that timing of adaptation is slower in young children when compared to adolescents and adults 26 . These opposite results could be due to a host of different reasonsincluding inter-species differences, or differences in task design. Nevertheless, conclusions from our study regarding Nx mice more broadly agree with the literature on decreasing functional neural plasticity in specific brain regions as animals mature 27,28 .
In addition to changes in adaptive learning measured using steptime, we also noticed changes in stepping patterns in Hx mice across age ( Figure 5). One possible way to explain differential changes in stepping patterns in Hx mice may have to do with risk-taking behavior, which has been correlated to cerebellar activation in fMRI studies 29,30 . Risk-taking behavior spikes at adolescence and then trails off into adulthood 31 . P45 Nx mice (equivalent to adolescence/young adulthood) therefore use a smaller percentage of short steps since this is correlated with higher risk 32 . In comparison, P45 Hx mice display higher percentage of short steps. This result is in line with human studies measuring outcomes for young adults born prematurely, where risk-taking behavior was markedly less pronounced 33 . Thus, reductions in risk-taking, and putatively correlated stepping patterns may be part of a larger "preterm phenotype" rather than solely due to cerebellar dysfunction 34 .
In terms of pathophysiology of the cerebella of Hx animals, our observation of reduction in simple spike activity may also be related to increase in spike width in Hx animals (Supplementary fig. 6), as has been observed in PCs of mice lacking fast-acting voltage gated potassium channels 35 . Finally, in addition to changes in spike discharge and ionic equilibrium, morphological alterations may also drive behavioral changes. Since Hx disrupts the early postnatal timeline, developmental phenomena such as PC dendritic arborization may be a contributing factor. In our previous report, we observed a marked delay in dendritic arborization of PCs of Hx animals 1 . Previous work by another group has implicated PC dendrite maturation as a contributing factor to spontaneous firing 36 . An Hx-induced delay in dendritic arborization or maturation is thus consistent with our in vivo data showing a drastic reduction in spike frequency. This reduction in spike is not directly dependent on abnormal MLI function, since our analysis indicates that MLIs do not significantly and directly contribute to opto-evoked spike activity in PCs (Supplementary fig. 4 b).
In terms of complex spike firing, we noticed that optostimulation results in increased firing. While further experiments are needed to determine the exact mechanism behind this increase, our latency analysis of optostimulation evoked complex spike activity (Supplementary fig. 9) is in line with previous reports which employed ChR2-mediated stimulation in PCs, supporting the model of disinhibition of the inferior olive 37 . Statistical tests represented in i-j are Two-way ANOVA followed by Sidak's multiple comparison test. Asterisks represent * P < 0.05, *** P < 0.001, **** P < 0.0001, 'n.s.' is not significant. Error bars in e-f represent SEM.