Neurobehavioral impairments predict specific cerebral damage in rat model of subarachnoid hemorrhage

Subarachnoid hemorrhage (SAH) is a severe form of stroke that can cause unpredictable and diffuse cerebral damage, which is difficult to detect until it becomes irreversible. Therefore, there is a need for a reliable method to identify dysfunctional regions and initiate treatment before permanent damage occurs. Neurobehavioral assessments have been suggested as a possible tool to detect and approximately localize dysfunctional cerebral regions. In this study, we hypothesized that a neurobehavioral assessment battery could be a sensitive and specific early warning for damage in discrete cerebral regions following SAH. To test this hypothesis, a behavioral battery was employed at multiple time points after SAH induced via an endovascular perforation, and brain damage was confirmed via postmortem histopathological analysis. Our results demonstrate that impairment of sensorimotor function accurately predict damage in the cerebral cortex (AUC: 0.905; sensitivity: 81.8%; specificity: 90.9%) and striatum (AUC: 0.913; sensitivity: 90.1%; specificity: 100%), while impaired novel object recognition is a more accurate indicator of damage to the hippocampus (AUC: 0.902; sensitivity: 74.1%; specificity: 83.3%) than impaired reference memory (AUC: 0.746; sensitivity: 72.2%; specificity: 58.0%). Tests for anxiety-like and depression-like behaviors predict damage to the amygdala (AUC: 0.900; sensitivity: 77.0%; specificity: 81.7%) and thalamus (AUC: 0.963; sensitivity: 86.3%; specificity: 87.8%), respectively. This study suggests that recurring behavioral testing can accurately predict damage in specific brain regions, which could be developed into a clinical battery for early detection of SAH damage in humans, potentially improving early treatment and outcomes.


Introduction
Aneurysmal subarachnoid hemorrhage (SAH) is a devastating disease that frequently results from the sudden rupture of an intracranial aneurysm, with high rates of death or severe disability [1,2]. Survivors of SAH often experience signi cant neuropsychological impairment, including sensorimotor de cits, cognitive slowing, memory impairment, and language dysfunction [1,[3][4][5][6]. Even those with favorable functional outcomes may develop disabling constitutional impairments such as fatigue, depression, and anxiety [7,8]. These de cits are attributed to both early and ongoing SAH-induced damage [9,10]. Despite tremendous research efforts spanning decades, nimodipine remains the sole FDA-approved treatment for SAH patients, with marginal bene ts at best [11,12]. No other agents have been shown to reliably improve outcomes for SAH patients [13]. As the search for effective therapies for SAH continues, early detection of salvageable tissue may help guide targeted interventions aimed at preventing long-term de cits.
Historically, angiographic vasospasm has been regarded as the primary cause of SAH-induced damage images to binary black and white pixels. On H&E staining healthy brain tissue displays a left-skewed pixel intensity distribution re ecting predominantly basophilic staining while eosinophilic staining lesioned tissue has a right-skewed distribution. Anatomic regions were manually segmented using the Waxholm Space atlas [24] to allow for region-to-region comparison. A prespeci ed minimum pixel intensity threshold (145-254, black) was used, derived from the average intensity of visually identi ed lesion areas on prior sets of training slides. Values below threshold (1-144, white) were considered the color intensity of healthy brain tissue based on mean values in corresponding anatomic regions of sham animals. Total ROI was manually selected based on area above threshold. The Wand tool was used to select densely packed areas of potential damage, de ned as severe potential damage. ROI containing pixel density below threshold for severe potential damage and above 144 was de ned as moderate potential damage. ROI of severe and moderate potential damage were quanti ed as area percentage of the total anatomic area.
After ROI identi cation, H&E-stained slides were imaged using EVOS M7000 (Thermo Fisher Scienti c, Waltham, MA) at low power (20x = 0.2mm 2 ) and high power (40x = 0.05mm 2 ). All animals from each group underwent quanti cation of regional cellular health. Five high-power elds were selected at random from within each ROI in SAH group animals and corresponding regions in sham animals. Cellular morphology was compared between groups, and cells categorized as healthy/unhealthy using regional rodent pathology references [25]. To account for loss of severely damaged cells by 30D, the area density of healthy cells was also quanti ed for each region. Due to the lack of established de nitions for SAHinduced lesions, existing literature on ischemic tolerance was used to de ne lesion severity. A level of 20% unhealthy cells was set as the baseline for moderate damage based on established CBF thresholds for signi cant prolonged ischemia [26]. As brain tissue can tolerate marked reductions in blood ow for brief periods of time [26,27], we de ned severe damage as a region with greater than 70% unhealthy cells [28]. ROI from SAH and sham animals were assessed and classi ed as predominantly healthy (less than 20% unhealthy cells), moderately damaged (20-70% unhealthy cells), and severely damaged (> 70% unhealthy cells).
Statistical analysis: Statistical tests were performed using the Statistics-and-Machine-Learning Toolbox in MATLAB (version 2022a, MathWorks, Inc. Natick, MA), OriginPro (version 2022b 64-bit, OriginLab, Northampton, MA) and GraphPad Prism (version 9, GraphPad Software, Boston, MA). As all comparisons were between SAH and sham groups, effect sizes and comparisons were performed using a mixedeffects model with an ANOVA. For all tests, group and day were treated as xed effects while the subject was treated as a random effect. For object exploration time, the effect of treatment as well as the interaction between treatment and object investigated were tested. Post-hoc analysis was performed with the Tukey test. Results are presented as mean ± standard deviation (SD). Statistical signi cance was set to P < 0.05. To measure the sensitivity and speci city of early neurobehavioral de cits to predict long-term regional damage, behavioral test results at earliest assessment were compared to the damage observed at 30D after SAH. The degree of impairment observed in each behavioral assessment parameter was normalized to sham animal performance and compared to the average percentage of healthy cells in associated cerebral regions of the same animal and correlation coe cients calculated. For each region, the best correlated neurobehavioral impairment was selected for receiver operator characteristic (ROC) analysis. Behavioral test accuracy was reported as area under the ROC curve (AUC), and for each assessment a cut-off value with maximum sensitivity and speci city for detecting region-speci c cellular damage was identi ed.
For the rotarod test, there was a signi cant decrease in riding time in SAH animals compared to sham Sensorimotor assessments corresponded well with damage observed in regions implicated in sensorimotor function. The A-NSS was highly correlated with cortical damage (R 2 = 0.907) but less with striatal damage (R 2 = 0.736). The Garcia scores correlated equally with damage to the cortices (R 2 = 0.849) and striatum (R 2 = 0.826). Rotarod maximum speed correlated better with damage to the cortices (R 2 = 0.875) than striatum (R 2 = 0.808), while rotarod riding time correlated better with striatal health (R 2 = 0.819) than cortical health (R 2 = 0.679). The A-NSS was selected for ROC analysis (Fig. 9A) demonstrating that a cut-off score of 2 for the A-NSS provided optimal sensitivity and speci city for predating severe cortical damage (AUC: 0.905; 95% CI: 0.742-1.00; sensitivity: 81.8%; speci city: 90.9%). The Garcia score was best correlated with striatal health by ROC analysis (Fig. 9B) demonstrating a cutoff score of 13 for optimal sensitivity and speci city (AUC: 0.913; 95% CI: 0.750-1.00; sensitivity: 90.9%; speci city: 100%).

Memory impairment predicts hippocampal damage
The novel object recognition test (Fig. 4A) showed that the SAH group had a signi cant decrease in time interacting with either object (Fig. 4B; 24H: 3.49 ± 2.39 s, 72H: 6.62 ± 5.18 s, 7D: 4.37 ± 5.80 s, 14D: 6.27 ± 5.02 s, 21D: 5.59 ± 6.35 s, 28D: 4.30 ± 2.90 s; F = 6.2015, P = 0.014, mixed-effects ANOVA for treatment and object). Importantly, SAH animals showed a signi cant reduction in the interaction with the novel object relative to the familiar object (F = 19.78, P = 1.84x10 − 5 , mixed-effects ANOVA for treatment and object), with a signi cant decrease in discrimination ratio ( Fig Memory assessments corresponded well with damage to the different areas of the hippocampus. For recognition memory, the NOR discrimination ratio correlated with CA1 health (R 2 = 0.737) but less so with DH (R 2 = 0.411). Comparing the discrimination ratio obtained via NOR to CA1 health (Fig. 9C), the ROC analysis demonstrated that a discrimination ratio cut-off value of 0.5 had optimal sensitivity and speci city (AUC: 0.902; 95% CI: 0.699-1.00; sensitivity: 74.1%; speci city: 83.3%). For working memory, the percentage of novel arm entries was found to correlate better with CA3 health (R 2 = 0.708) than with DG health (R 2 = 0.392). Spontaneous alternations were observed to correlate well with CA3 health (R 2 = 0.807) but not with DG health (R 2 = 0.194). As the best correlated measure of working memory, spontaneous alternations and CA3 health were compared via ROC analysis. A cut-off value of 60% for spontaneous alternations was found to maximize sensitivity and speci city ( Fig. 9D; AUC: 0.746; 95% CI: 0.576-0.928; sensitivity: 72.2%; speci city: 58.0%), although test performance was worse than recognition memory.

Affective impairment predicts subcortical damage
For the EM test (Fig. 6A) and increase in immobile time between 21D and 28D time points (t = 3.98, P = 0.042, Tukey test). Our results suggest that depression begins as early as 7 days following SAH and worsens over the chronic timeframe.

Discussion
Detecting patients at risk for delayed neurologic deterioration is a major challenge in the clinical care of SAH [30]. This period of ongoing injury is regarded as the most signi cant reversible factor in a patient's recovery [31]. While vessel imaging techniques can identify large vessel vasospasm, other critical drivers of SAH-induced damage and dysfunction are di cult to detect [18, 32,33]. Early identi cation and prediction of ongoing cerebral damage may be crucial for the development of effective therapies for SAH, but no reliable method currently exists to do so [18]. Neurobehavioral de cits have been linked to in ammation, white matter disruption, and other factors that contribute to SAH-induced damage, suggesting that they could serve as a harbinger of permanent cerebral injury [20,21]. Furthermore, comprehensive behavioral assessments have the potential to detect subtle signs of cerebral damage that are independent of more obvious neurological de cits and can be detected earlier than some imaging methods may reliably identify regional damage [34]. These behavioral assessments provide an opportunity for early and sensitive detection of dysfunctional cerebral regions consistent with SAHinduced damage. Our experimental results have shown that early neurobehavioral de cits accurately predict cellular injury within brain regions. Speci cally, sensorimotor impairment, recognition memory impairment, anxiety-like and depression-like behaviors predicted cellular injury in the cerebral cortex and striatum, hippocampus, amygdala, and thalamus, respectively, with good accuracy (> 90%). Spatial memory impairment predicted hippocampal damage with acceptable accuracy (~ 75%). No de cit was found to predict damage to the hypothalamus.
Sensorimotor de cits were found to be sensitive and speci c indicators of SAH-induced regional damage, involving complex processes in the cortex, basal ganglia, thalamus, white matter tracts, and brainstem loci [35]. In this study, severe damage was observed in the cortex, basal ganglia and thalamus, with moderate potential damage ranging from 15.7% in the cortices to 0.3% in the thalamus. Similarly, cellular health was observed to be severely damaged in all animals but ranged from an average of 15.7% in the cortex to 24.9% in the striatum. Most parameters of sensorimotor assessments correlated well with cellular damage in cortical and subcortical regions known to be involved in sensorimotor function. Speci cally, the 24H A-NSS and Garcia scores were the best indicators of damage to the cortices and striatum, respectively, with accuracies greater than 90%. These ndings suggest that sensorimotor impairment is highly sensitive and speci c for damage in the cortex and striatum. While associations between sensorimotor impairments and cellular health in the acute phase of SAH have been reported by others [36], our study provides novel ndings of similar correlations in the chronic phase. While the study did not investigate the mechanisms underlying cellular damage, it is possible that the observed sensorimotor de cits are caused by a combination of early ischemia and ongoing in ammation, microvascular spasm and/or microthrombosis [37][38][39]. These factors may contribute to sensorimotor impairment beyond ischemic infarctions and highlight the complex nature of SAH-induced damage.
In this study, memory de cits were correlated with damage in associated hippocampal regions, with the NOR test appearing more accurate than the Y-maze test. There is wide agreement that CA1, CA3 and the dentate gyrus play essential roles in encoding memory [40][41][42][43][44]. When assessed histologically, most regions of the hippocampus displayed a greater degree of moderate potential damage than severe damage, with only the DH having a percentage of healthy cells below 30%. Signi cant early memory impairment was observed, which worsened over time and correlated with damage in the CA1 and CA3 regions. The NOR test accurately predicted cellular damage in CA1, while the Y-maze was somewhat less accurate in predicting damage to CA3. Previous studies have demonstrated that selective knockout of hippocampal neurons causes speci c patterns of memory impairment [41,45]. However, others have demonstrated a variable relationship between neurobehavioral de cits and cellular loss in SAH models [46,47], with some proposing that memory de cits are driven by alterations in long-term potentiation rather than direct cellular damage [20,48]. Our results suggest that the NOR test is most sensitive and speci c for detecting hippocampal damage associated with SAH, which alongside likely alterations in neuroplasticity led to long-term memory impairment.
We found that measures of anxiety-like behavior were highly sensitive and speci c predictors of damage to the amygdala. Although anxiety is a complex psychological condition, the amygdala is known to play a role in pathologic anxiety in various disease states [49,50]. Here we predominantly observed moderate potential damage as well as moderate cellular damage in the amygdala. Early anxiety-like behavior was present in the EM test although it showed a trend towards improvement over time. Conversely, the OF test did not demonstrate signi cant anxiety-like behavior at any time point. The percentage of open arm entry for the EM demonstrated a moderate (> 70%) correlation with and accuracy for predicting damage to the amygdala. While anxiety-like behavior has been frequently reported in experimental models of SAH, no preclinical studies have directly linked SAH-related anxiety to cellular damage in the amygdala [51][52][53].
However, anxiety is a common feature in patients who survive SAH [54], and clinical studies have demonstrated dysfunction in the amygdala in SAH patients [55]. This reinforces that behavioral assessments of anxiety-like behavior may be sensitive and speci c for detecting dysfunction and damage affecting the amygdala and may provide a useful tool for predicting the development of damage in the amygdala.
Depression is a common and disabling condition that often occurs after SAH [56], and our study found that measures of depression-like behavior are both sensitive and speci c for predicting damage in associated cerebral regions. Although the anatomic correlates of post-SAH depression are not well understood, studies in patients with SAH have demonstrated dysfunction in the cingulate cortex, which has also been observed in preclinical and clinical studies of depression [55,57,58]. Additionally, the thalamus and hypothalamus have been implicated in depression in humans [59,60], although their contribution to depressive behavior in SAH is less clear. Here the hypothalamus demonstrated a large degree of moderate and severe potential damage and the thalamus contained almost entirely severe potential damage. Cellular health was observed to be severe in both regions. SAH animals displayed severe and worsening depression-like behavior in the PFS, consistent with the damage observed in these structures. The degree of immobile behavior correlated with thalamic health, accurately predicting longterm damage to the thalamus. Other authors have found that animals that lacked signi cant depressionlike behavior following SAH also had no measurable damage in the thalamus [61], supporting the utility of behavioral tests of depression-like behavior in identifying damaged cerebral regions.
We show that a comprehensive battery of behavioral assessments is a sensitive means to identify and predict damage and dysfunction in speci c regions after SAH. The sensitivity and speci city of any individual behavioral assessment might vary, but applying multiple combined assessments interrogating the same region can increase sensitivity or speci city [62]. For example, combining the A-NSS and Garcia assessments in parallel allows for a combined sensitivity of 98.8% for detecting damage in structures critical for sensorimotor function, at the expense of a reduced speci city of 89.6%. This technique may be particularly useful with tests of anxiety-like and depression-like behaviors, which can be more affected by variability in animal behavior than other behavioral assessments [52,63]. While the accuracy of the EM assessment of anxiety was good, the sensitivity for detecting damage in the amygdala could be augmented via combination with another assessment such as the light/dark box test [64], although this test has not been applied in SAH models. In clinical practice, delayed, new-onset sensorimotor de cit usually portends impending irreversible injury, and its rapid recognition allows for treatment that greatly improves neurological outcomes after SAH [65,66]. While the comprehensive behavioral battery used in this study would likely be effective in detecting and localizing SAH-induced damage, it may not be feasible for regular clinical use. Therefore, it is crucial to identify the most sensitive and speci c behavioral tests for damage to speci c brain regions. By linking each at-risk region to a speci c behavioral test parameter that accurately predicts long term damage, clinical exam correlates of these test parameters could be identi ed and re ned to reliably detect reversible damage before it becomes permanent. These techniques may enable the rapid and accurate identi cation of patients at risk for irreversible injury, thereby improving outcomes after SAH.
Several limitations exist within this study. H&E staining is an established means of quantifying general cellular damage following neurologic insults [67-69], however the use of cell-speci c staining techniques may provide more detailed information regarding the mechanisms underlying ongoing cellular damage. Histological analysis was chosen at the 30-day point to better capture severe damage resulting from signi cant early injury, as damage induced by SAH is known to occur early and progress over a prolonged period of time [33]. The analysis of groups beyond the 30-day point, such as 60 or 90 days, may provide additional insights into the evolution of regional injury and the potential for spontaneous recovery.
Furthermore, while many neurobehavioral de cits could be detected via multiple assessments, some de cits are challenging or impossible to assess in combination. For example, the sucrose preference test of depression-like behavior requires food and water deprivation [70], which will invalidate other assessments done in the same period.

CONCLUSION
Patients who survive SAH often experience disabling impairments, and currently available clinical techniques have limited ability to detect damage before it becomes irreversible. In this study, we have shown that a comprehensive battery of behavioral assessments is a highly accurate early warning for damage affecting speci c cerebral regions. Speci cally, behavioral tests of sensorimotor function, memory, anxiety-like and depression-like behaviors were found to predict damage in the cortex and basal ganglia, hippocampus, amygdala, and thalamus, respectively. Our ndings suggest that these behavioral assessments can detect regional damage with acceptable accuracy, sensitivity, and speci city, with the notable exception of the hypothalamus. It is possible that combinations of behavioral assessments could further augment the diagnostic sensitivity or speci city of the assessments used here. Additional research into cell-speci c damage and the evolution of SAH-induced injury may better characterize the threshold of behavioral assessments required to detect damage. Armed with this knowledge, clinicians may be able to intervene with targeted therapies to prevent permanent damage and improve long-term outcomes.

Con ict of Interest
The authors declare that the research was conducted in the absence of any commercial or nancial relationships that could be construed as a potential con ict of interest.

Data Availability
The authors declare that all supporting data are available within this article. Figure 1 Conceptual diagram of experimental protocols. A: Timeline of behavioral testing. After a pre-training period before SAH induction or sham surgery, damage in the brain regions affected by SAH is assessed with a behavioral battery in which the indicated tests measure sensorimotor, memory and affective domains. (B) Subarachnoid hemorrhage is induced via endovascular perforation at the internal cerebral artery (ICA) bifurcation using a sharpened suture, after which hematoma forms in the basal subarachnoid space. Other abbreviations: ACA, anterior cerebral artery; BA, basal artery; MCA, medial cerebral artery (C) The top panel shows the early brain injury (EBI), which causes severe damage (red), and is associated with hematoma formation in the basal subarachnoid space damaging the piriform cortex, as well as increased intracranial pressure (ICP) damaging the motor, somatosensory, and piriform cortices, thalamus, and striatum. The lower panel shows the delayed cerebral ischemia (DCI), with more moderate damage (blue) developing in the hippocampus, white matter tracts, amygdala, and periventricular nuclei.

Figure 2
Distribution of damage severity following SAH. A: At 30 days following SAH, damage was observed to occur in multiple areas including the cortices, hippocampus, thalamus, hypothalamus, striatum and amygdala, with both severe damage (red) de ned as more than 70% unhealthy cells and moderate damage (blue) de ned as 20-70% healthy cells present in most regions assessed. Black bar = 5mm. B: The relative distribution of severe and moderate potential damage in different brain regions was quanti ed. Most regions displayed 10-30% area of potentially unhealthy regions, with a mixture of severe and moderate potential damage generally favoring moderate damage in all regions except the thalamus and striatum. C: The density of healthy cells corresponded to regions of potential damage observed previously. The cortex, hypothalamus, and amygdala exhibited both moderate and severely decreased healthy cell density. Hippocampal regions CA1, CA3 and DG contained both moderate and severe reductions in healthy cell density, while the DH region was entirely severely damaged. The striatum and thalamus also showed severe reductions in healthy cell density. and sensorimotor ability. D: Beam-walk score, a measure of coordination and sensorimotor function, was elevated in rats following SAH, indicating impaired sensorimotor function. SAH rats displayed signi cantly worsened sensorimotor function at all time points relative to sham baseline. E: Severe levels of cellular death (red outline) were observed in the motor and somatosensory cortices, along with the striatum. Additionally, regions of moderate damage were also observed in the motor, somatosensory and piriform cortices (blue outline) as well as the striatum and nucleus accumbens regions of the basal ganglia. Black bar = 100µm; white bar = 25µm; red arrowheads = unhealthy cells; blue arrows = examples of healthy cells; solid outline = 20x = 0.2mm 2 ; dashed outline = 40x = 0.01mm 2 ; **** P < 0.0001, mixedeffects ANOVA. SAH induces working reference memory dysfunction on Novel Object Recognition (NOR) assessment associated with predominantly moderate cellular damage in hippocampal regions CA1 and the dentate hilus (DH). A: Representative traces demonstrating reduced movement and interaction in SAH animals (square = novel object, circle = familiar object). B: The amount of time SAH rats spent exploring either object was diminished, suggesting impaired sensorimotor and exploratory function. SAH rats also spent less time investigating the novel object relative to time spent investigating the familiar object. C: Discrimination ratio was decreased in animals following SAH, consistent with impaired reference memory. D: Total distance moved during the NOR task was reduced in SAH rats. E: Examination of the DH and CA1 regions of the hippocampus revealed presence of severe damage (red outline) and moderate damage (blue outline) with a greater relative proportion of moderate damage. Black bar = 100µm; white bar = 25µm; red arrowheads = unhealthy cells; blue arrows = examples of healthy cells; solid outline = 20x = 0.2mm 2 ; dashed outline = 40x = 0.01mm 2 ; * P < 0.05, **** P < 0.0001, mixed-effects ANOVA. SAH induces working and reference spatial memory dysfunction and decreases exploratory and locomotor behavior on Y-Maze assessment in association with damage in the hippocampal CA3 and dentate gyrus (DG) regions. A: Representative traces illustrating impaired working spatial memory B: Representative traces demonstrating impaired reference spatial memory. C: Decreased entries of the previously closed arm were observed following SAH, consistent with impaired reference spatial memory. D: Percentage of complete spontaneous alternations was decreased in SAH animals, indicating impaired working spatial memory. E: Rats in the SAH group had fewer total entries into arms, consistent with impaired locomotor and exploratory function. F: Total distance travelled during the Y-Maze assessment decreased, indicating exploratory and locomotor function impairment following SAH. F: A small degree of severe damage (red outline) and larger proportion of moderate damage (blue outline) were observed in the CA3 and DG regions of the hippocampus following SAH, which are closely related to formation and retrieval of spatial memory. Black bar = 100µm; white bar = 25µm; red arrowheads = unhealthy cells; blue arrows = examples of healthy cells; solid outline = 20x = 0.2mm 2 ; dashed outline = 40x = 0.01mm 2 ; * P < 0.05, ** P < 0.01, *** P < 0.001, mixed-effects ANOVA. anxiety. D: Decreased locomotor function was observed in SAH rats, as shown by decreased total distance travelled. E: Velocity of movement was also decreased following SAH, indicating impaired locomotor function. F: Moderate damage (blue outline) was observed in the hypothalamus and amygdala following SAH induction, with severe damage (red outline) predominantly observed in the thalamus. Black bar = 100µm; white bar = 25µm; red arrowheads = unhealthy cells; blue arrows = examples of healthy cells; solid outline = 20x = 0.2mm 2 ; dashed outline = 40x = 0.01mm 2 ; **** P < 0.0001, mixedeffects ANOVA.  subcortical and white matter injury. A: SAH group animals showed a signi cant decrease in highly mobile behavior, indicating a more depressed state. B: Following SAH, animals displayed a signi cant increase in immobile behavior consistent with depression. C: Severe damage (red outline) was observed following SAH in the thalamus, along with moderate damage (blue outline) in the hypothalamus and ventromedial prefrontal cortex (VMPC). Loss of con uence was observed in the corpus callosum (yellow outline). Black bar = 100µm; white bar = 25µm; red arrowheads = unhealthy cells; blue arrows = examples of healthy cells; solid outline = 20x = 0.2mm 2 ; dashed outline = 40x = 0.01mm 2 ; ** P < 0.01, **** P < 0.0001, mixedeffects ANOVA. Figure 9