Complement Inhibition Reduces Post-Hemorrhagic Hydrocephalus in Mouse Neonatal Germinal Matrix Hemorrhage

Germinal matrix hemorrhage (GMH) is a devastating disease of infancy that results in intraventricular hemorrhage, post-hemorrhagic hydrocephalus (PHH), periventricular leukomalacia and neurocognitive decits. There are no curative treatments and limited surgical options. We developed a novel mouse model of GMH and investigated the role of complement in PHH development. Methods We utilized a neonatal mouse model of GMH involving injection of collagenase into the subventricular zone of post-natal day four (P4) pups. Animals were randomized into four experimental arms: Naïve, sham injured, injured and vehicle (PBS) treated, and injured and CR2Crry-treated (a pan-complement inhibitor). Histopathologic and immunouorescence analyses were performed at P14 with a focus on parameters of neuroinammation and neuroprotection. Survival was monitored through day 45, prior to which cognitive and motor function was analyzed. The complement inhibitor CR2Crry, which binds C3 complement activation products, localized specically in the brain following systemic administration after GMH. Compared to vehicle treatment, CR2Crry treatment reduced PHH and lesion size, which was accompanied by decreased perilesional complement deposition, decreased astrocytosis and microgliosis, and the preservation of dendritic and neuronal density. Progression to PHH and neuronal loss was linked to microglial phagocytosis of complement opsonized neurons, which was reversed with CR2Crry treatment. Complement inhibition also improved survival and weight gain, and improved motor performance and cognitive outcomes measured in adolescent GMH mice. Complement novel for


Introduction
Germinal matrix hemorrhage (GMH) is the most common neurologic pathology in neonates, estimated at 3.5 per 1000 live births. [1] It is caused by disruption of fragile vasculature in the highly vascular subventricular zone (SVZ). Risk factors for GMH include pre-term birth (<32 weeks) or low weight (<1,500g), with rates of 20-40% in these infant groups. [2] Germinal matrix hemorrhage often leads to intraventricular hemorrhage (IVH), resulting in post-hemorrhagic hydrocephalus (PHH) and periventricular leukomalacia. [3] These two progressive pathological processes negatively impact neurodevelopmental processes and are highly associated with development of cerebral palsy, with a rate of 30-42% of signi cant disability following severe GMH-IVH. [4,5] Furthermore, patients with severe GMH-IVH have an approximately 90% rate of associated morbidity and mortality within two years. [5] There are no medical treatments for GMH or its sequalae.[6] A currently used procedure of surgical cerebrospinal uid (CSF) diversion can mitigate the effects of PHH, but it does not cure the neurological disability caused by progressive damage from the hemorrhage, and there is life-long surgical morbidity in up to 90% of patients, including surgical infections and shunt malfunctions. [7,8] As in traumatic brain injury (TBI) and stroke, brain injury following GMH involves an unpredictable primary insult, followed by secondary injury driven, at least in part, by neuroin ammation. [9] The primary injury of GMH cannot be prevented by intervention due to the immediate and unexpected mechanical trauma of a hemorrhagic mass, which is accompanied by a surrounding ischemic insult. On the other hand, secondary injury involves a progressive injury that extends into neighboring tissue, with breakdown of the blood brain barrier and resultant cytotoxic edema. [9] There is evidence that in ammation and increased cytokine and chemokine release in the setting of GMH contribute to the propagation of the secondary injury and are associated with subsequent periventricular leukomalacia and PHH. [10,11] A central component of in ammatory cascades is the complement system, [12][13][14] although the role of complement in GMH has never been investigated. The complement system is a component of both innate and adaptive immunity, and there are three main activation pathways; the classical, lectin, and alternative pathways. [15] All pathways converge at the cleavage and activation of C3, leading to the generation of C3 opsonins (C3b, iC3b, C3d), anaphylatoxins (C3a and C5a), and nally the cytolytic membrane attack complex (MAC). Complement activation in TBI and stroke has been well studied, [12,13] and complement is known to be involved in an ongoing in ammatory response that is implicated in secondary injury. In the settings of TBI and stroke, C3 opsonins can facilitate the phagocytosis of synapses and neurons, the anaphylatoxins can recruit and activate immune cells and may be have both injurious and reparative roles, and the MAC can result in cell lysis, including lysis of red blood cells that contribute to the dispersion of oxidative molecules such as heme into brain tissue and CSF. [12,13,16,17] In the current work, we describe the development of a novel neonatal mouse model of GMH with high post-operative survivability and with high rates of PHH development. We used this model for a foundational investigation into the role of complement in the development of PHH following GMH. Our data demonstrate a central role for complement in post-GMH pathology, and we show that a clinically relevant approach of complement inhibition post-GMH has a major impact on brain injury and in ammation, with subsequent bene t in terms of animal growth and longer-term motor and cognitive function. This work also establishes a direct relationship between complement activation and the development of secondary, post-hemorrhagic hydrocephalus. In this study we used the complement inhibitor CR2Crry, a previously characterized inhibitor of C3 activation that speci cally targets to sites of complement activation and C3d deposition. [18] As previously shown, localizing complement inhibition to sites of complement activation signi cantly increases bioavailability and e cacy, and obviates the need to systemically inhibit complement, thus leaving the systemic physiological functions of complement, including host-defense, intact. [18,19] Materials And Methods

Study design
The study design and work ow are shown in Figure 1. Animal groups in this study were: Wild-type Naïve (no injury, no treatment), Sham (PBS injection in the SVZ in place of collagenase, no treatment), Vehicle (Collagenase injection in the SVZ, with intraperitoneal PBS treatment), and CR2Crry treated (Collagenase injection into the SVZ, with intraperitoneal CR2Crry treatment). Prior to the surgical procedures for GMH, animal breeders were randomly assigned to groups. Randomization was performed by an external lab personnel and was dependent on litter sizes at P1 of life in order to satisfy the numbers across groups.
To minimize confounders, a single lab person performed the surgeries, treatments, testing, and scoring and was blinded to group allocations for the duration of the study. Except where indicated, surgical injection into the SVZ was conducted on post-natal day 4 (P4). Animals were excluded if they died within 24 hours of surgery (<40% of animals). Study Endpoints were P14 (10 days post-injury) for subacute outcome analysis, including histology, and P45 (41 days post-injury) for animal survival study and cognitive tasks. For the P45 cohort, animal gender was identi ed after P8. In vehicle group, there were 10 females, 9 males, and 8 unknown due to death before P8. In the CR2Crry cohort, there were 5 females, 6 males, and 8 unknown due to death before P8.

Animal husbandry and care
All animal rearing, care, procedures and euthanasia were approved by the Institutional Animal Care and Use Committee at our institution. Wild-type C57BL/J mice (Jackson Laboratory, ME, USA) were obtained at age P30 and acclimatized for 1 week. Animals were then mated in pairs. Cages were cleaned weekly and corn cob bedding provided. All mice housed in the facility were exposed to 12 hours light/dark cycles. Mice received access to food and water ad libitum, while pregnant females received a high fat diet as recommended by the institutional veterinarian. All tests and experiments were conducted during the light cycle. Pregnancy and litter checks were performed daily. On day of initial injury induction (P4), male parent mice were removed and separated from the litter. Following surgery, pups were placed on a heating pad for 30 minutes, then reunited with the mother. The total handling time of pups away from the mother was approximately 45 minutes. They were then monitored for an additional 60 minutes to ensure care of the pups by the mother. All animals were then returned to the mouse housing facility.

Recombinant proteins and treatment paradigm
CR2Crry was prepared as previously described.
[18] Both CR2Crry and PBS used for intraperitoneal (IP) treatment of animals was endotoxin-free. Complement inhibitory activity of the recombinant protein was veri ed using a zymosan assay, as previously described. [18,20] Animals in the CR2Crry treatment group were treated IP at 10 mg/kg, a dose previously determined to be optimal in other models. [20] Two treatment time points were used in this study: 1 and 24 hours post injury. Following the rst treatment in the 1-hour group, IP injections of CR2Crry were then administered at P7, P10, and P13 for a total of 4 doses. In the 24-hour treatment group, the rst dose of CR2Crry was given 24 hours following injury, then at P7, P10, and P13. The vehicle group was treated IP with PBS 1-hour post collagenase injection, then at days P7, P10, and P13. To examine tissue targeting and the tissue half-life of CR2Crry, the protein was labeled with a uorescent marker (CF dye 92221, Biotium, Fremont, CA) per the manufacturer's protocol, and administered i.p. to neonates 1 hour after induction of GMH or to control animals. Live animal uorescence tomography (Maestro II, PerkinElmer, Waltham, MA, USA) was performed at 24 h, 48 h, 72 h, 96 h and 7 days after the single dose injection. Relative CR2Crry brain deposition was quanti ed by measuring signal intensity within the brain using NIH ImageJ (FIJI) integrated density.
Germinal matrix hemorrhage injury model and lesion grading system Clostridium-derived collagenase (Type VII-S collagenase, C2399, Sigma-Aldrich) was injected into the SVZ of mouse pups at P4 to induce direct spontaneous non-traumatic vessel rupture with intracerebral hemorrhage in the region of the germinal matrix and SVZ. P4 mouse pups were placed on a cooled platform to induce cryo-anesthesia as previously described. [21] A Hamilton 32-gauge needle (Model 80008, Hamilton Co., NV, USA) was used to puncture the pup's right sided scalp and skull with the following conditions: 1mm posterior to the eye, 1 mm superior to the orbit, and 1 mm deep to reach the periventricular zone (Fig. 1). The injection contained 0.5 units (0.5uL) collagenase for GMH groups, and PBS for the Sham group. The location of injection was chosen at the level of the periventricular region to induce parenchymal microvascular disruption ( Fig. 1, 2a). To optimize injections and reduce variability, the authors trained with Evans blue (EB) dye injection into the SVZ at the aforementioned parameters. Animal brains were inspected 24 hours after injection to con rm correct location of injections. At least 80% accuracy was required to perform the experiments. For this experimental design, a single researcher is recommended for all injections to maintain consistency.
Following collagenase injection, animals were placed on a heating pad for 30 minutes, and then returned to their cages. Once the entire litter had completed the procedure and had been returned to their cage, the mother was returned to the cage and monitored for interaction with the pups. The cage was kept on the heating pad for an additional 1 hour, then returned to the mouse housing facility. Sham PBS injections were performed to ensure that hemorrhage was a result of collagenase injection and not from mechanical insertion of the needle. Animals were sacri ced at P14 (subacute outcomes) and P45 (chronic outcomes). In the process of developing the mode, collagenase injections were also performed in P2 and P3 animals (procedural survival is shown in Fig. 2c).
The mechanism of injury in this model is the equivalent of clinical GMH grade 4 injury (intraparenchymal hemorrhage). Thus, an animal-speci c injury grading system was developed to establish a distinction between parenchymal injury, ventricular involvement, and PHH (refer to Fig. 2d). Scale 0 = No lesion or ventricular enlargement. Scale 1 = Lesion volume <30% of hemispheric cortical tissue ipsilateral to injury site without ventricular involvement. Scale 2 = Lesion volume >30% of hemispheric cortical tissue ipsilateral to injury site without ventricular involvement. Scale 3 = Lesion extending into the ipsilateral ventricle with no ventricular enlargement. Scale 4 = Lesion extending into the ipsilateral ventricle coupled with unilateral ventriculomegaly. Scale 5 = Lesion extending into both ventricles resulting in global hydrocephalus.

Cognitive performance assessment
Barnes maze was used to assess spatial learning and memory after GMH as previously described. [22] During the spatial learning phase, animals were trained beginning at P30 for 5 consecutive days with 2 trials per day spaced 60 minutes apart. Mice were given a two-day break, then re-tested using one trial for retention memory. All tasks were recorded using the Noldus EthoVision XT 13.0 system. Outcome measures included total distance traveled, latency to rst poke (mouse peeking into the escape hole without entry), latency to escape hole entry, velocity, errors recorded (mouse peeking into holes other than escape hole), and time spent at different quadrants of the maze. To assess fear-conditioning and learning memory, the passive avoidance task was used as previously described. [23] Mice were acclimatized at P40 and tested at P44.

Motor performance assessment
Gait analysis was performed using the automated CatWalk XT 10.6 system (Noldus Co., VA, USA). Mice were placed at the edge of an illuminated walkway monitored by an underlying camera. Using light scattered from contact between the animals' paws and glass, several computed parameters for each limb were tracked, including print area, limb contact intensity (average and maximum), gait consistency, and other gait-related parameters. The average from three runs was used for trail calculations for each animal. The run duration allotted was between 0.5 seconds and 8 seconds, and any run outside the given parameters was excluded. Any run that had more than 40% variation from start to nish was excluded. At P30, all experimental groups performed this task: naïve, vehicle, and CR2Crry-treated animals. Given the numerous output parameters of the device, a standardized Combined CatWalk Index (CCI) was applied as previously described. [24] In brief, parameters within each limb were assigned weighted signi cance, and the nal CCI represents the overall performance of the animal during each run.
Tissue processing and histologic analyses Animals in the subacute study were sacri ced at P14. Following euthanasia, cardiac perfusion was performed with cold PBS followed by 4% Paraformaldehyde mixed in PBS. Brains were then carefully extracted and placed in 4% Paraformaldehyde solution overnight at 4 C. The brains were then moved to a new vial with 30% sucrose mixed with 4% Paraformaldehyde in PBS. For tissue cutting, the brains were embedded in Tissue-Plus Optimal Cutting Temperature (OCT) compound (23-730-571, Fisher Healthcare) and frozen. At time of cutting, brains were cut in 40 µm coronal sections using a freeze-mount cryostat.
The complete brain was collected in 12-well plates and kept in PBS-lled wells until histologic analysis. For Nissl staining, serial brain sections 200 µm apart were mounted on a slide and stained using cresyl violet as previously described. [25] For ventricular and lesion volume measurements, 8 serial Nissl-stained brain sections 200 µm apart and 40 µm thick were used to reconstruct the total lesion volume. 4x magni cation images of each slice were acquired using a Keyence BZ-X710 microscope (Keyence Co, Itasca, IL, USA). Two, independent blinded observers calculated the lesion and ventricular areas using NIH ImageJ (FIJI). The average of both observers was reported. 2D analyzed images were reconstructed into 3D volumetric output les to measure brain lesion and ventricular volumes using "Free-D" software. [26,27] Immuno uorescence staining and imaging Mid-hippocampal and mid-ventricular regions were identi ed by stereometric measurement using a mouse brain atlas, followed by standard immuno uorescent (IF) staining as previously described.
[28] All imaging and analysis were performed by a blinded lab personnel. High-resolution imaging was performed using a Zeiss LSM 880 confocal microscope (Zeiss, Carl Zeiss Microscopy, LLC, White plains, NY, USA) at 40x zoom with water-media overlay and using the Z-stacking feature of the microscope. Images were deconvoluted using the ZEN 2.5 software (Zeiss) and reconstructed in 3D plane. Distance from lesion edge was calculated for GFAP and NeuN analysis using ZEN 2.5 software, spectrum analysis. MAP2 arborization was calculated using spectrum analysis on ZEN 2.5 software and quanti ed using MATLab software (MathWorks, Inc. Natick, MA, USA). GFAP and Iba-1 perilesional signal intensity were calculated as the mean grey value (average signal intensity per pixel) using NIH ImageJ. All GFAP and Iba-1 staining was performed with negative control images (secondary antibodies only) in order to correct for underlying auto-uorescence. Fluorescence-based analysis was performed rather than cell counting due to high cell density and clumping in the proximity of the injury site.
Colocalization analysis was performed using Imaris (Oxford Instruments, Concord, MA, USA) for 3D image reconstruction and quanti cation. Neurons were quanti ed per eld of view on Imaris.
Colocalization of C3/NeuN/Iba-1 was performed by spot-to-surface interaction and reported as a percent of total neurons within the eld. For internalization of C3 or NeuN, manual quanti cation of partial or fully internalized particles by Iba-1 cells was quanti ed. Total internalized NeuN was reported as a percent of total neurons in the eld. Primary antibodies used for staining were: anti-C3 (

Statistical analysis
Experimental sample size was determined using Power analysis and sample size estimation, performed through G*Power 3.1.9.2 tool (Franz Faul, Kiel University, Germany). Barnes maze performance was chosen as a reference test to calculate the effect size (Estimated mean and SD). Higher or comparable effect size was also expected for the remaining tests. A calculated effect size (d) of 2.0 was anticipated when comparing GMH mice to naïve and 1.6 when comparing vehicle to CR2Crry in the treatment group based on our preliminary studies with GMH. Therefore, we used an effect size of 1.6 for our power analysis for these aims. Two-tailed analysis with signi cance level α = 0.05 was considered and then a corrected αc = α/(number of primary comparisons)=0.05/(2 primary comparisons)= 0.025. Ratios of group numbers was considered to be 1 (N1/N2) with equal number of mice per group. The result of analysis reveals a sample size of 8 evaluable mice per group with an actual computed power of 84%. To ensure su cient number of evaluable animals is available, we corrected for potential 40% mortality/exclusion of animals in all studies. Thus, a nal number of 12 animals would be required per experimental group to satisfy the necessary minimum. Finally, in order to maintain animal litter continuity, litters were randomized into experimental groups rather than individual pups.
Was performed using GraphPad Prism 8.0 (GraphPad Software, San Diego, CA, USA

Results
Germinal matrix hemorrhage model and induction of post-hemorrhagic hydrocephalus in neonatal mice We developed a novel neonatal mouse GMH model with relevance to human clinical disease in several aspects, including pathophysiology of insult, post-hemorrhagic survival rate, temporal pro le of pathology, rate of post-hemorrhagic hydrocephalus, and motor/cognitive delay and de cits. Surgical details of the model are in materials and methods, and a work ow of the surgery and our subsequent complement inhibitor treatment paradigm is presented in gure 1. Brie y, at post-natal day 4 (P4), mouse pups were removed from their mothers and anesthetized on ice after de ning syringe insertion location.
Pups were then injected with 0.5 µl collagenase directly into the subventricular zone (SVZ) as annotated in gure 2a. After 1 hour on a heat pad, pups were returned to their mother. This procedure resulted in a lesion and deposition of blood products in a periventricular pattern, as shown in Nissl stains of brain sections collected 24 hours after injury (i.e. P5) (Fig. 2a). Of note, blood products were not only deposited in the ipsilateral hemisphere of needle insertion but were also evident along the ependymal layer of the contralateral ventricle (shown with a black arrow in Fig. 2a), indicating intraventricular hemorrhage, a feature of human GMH. The lesion and blood product deposition were a result of the collagenase injection and not mechanical insertion of the needle, as brain samples collected from PBS-injected animals (designated as sham) presented no lesion, blood deposition, or enlargement of the ventricles (hydrocephalus), in contrast to brains from collagenase-injected animals (Fig. 2b). In addition, unlike collagenase-treated animals, neither sham animals nor naïve non-injured pups showed any denudation of the ependymal lining in the lower border of the ventricles, shown in high resolution Nissl-stained images (Fig. 2b). The decision to induce injury at P4 was based on initial studies in which we found that collagenase treatment on P2 or P3 resulted in unacceptably high mortality rates (Fig. 2c). A grading system from 0 to 5 was developed to characterize the severity phenotypes of brain injury and hydrocephalus, with scale 5 corresponding to global hydrocephalus. Details of the scoring system are described in the methods section, and representative images corresponding to injury scale are shown ( Fig. 2d). Using Ter-119 red blood cell (RBC) stain, we show hematoma along the edges of the lesion three days after injury (Fig. 2e). Of note, the central portions of the hematoma get washed away as an artifact of the staining process, but the RBC stain shows layering of RBC's along the border of the lesion. Complement deposition was evident in the perilesional hemisphere with diffuse deposition along the lesion border as depicted by immuno uorescence staining for C3 (Fig. 2f), thus providing justi cation to explore the role of complement in the context of GMH and its post-hemorrhagic sequalae.
Targeting and tissue distribution of CR2Crry in neonates after induction of GMH The complement inhibitor CR2Crry has been shown to bind deposited C3 activation fragments ( [20]), which occurred at sites of injury in brains of collagenase injected mice (see above, Fig. 2e). To examine targeting speci city and whole-body distribution after systemic administration of CR2Crry, we administered uorescently labeled CR2Crry via i.p. injection after induction of GMH. Live animal uorescence tomography showed an initial systemic distribution of CR2Crry, with subsequent localization of the drug to the brains of GMH mice, but not to brains of control animals with no GMH (Fig. 3). Furthermore, quanti cation of uorescence intensity revealed a CR2Crry tissue half-life of about 3 days in the brains of GMH mice, which we used as the interval between CR2Crry treatments in the therapeutic paradigm below.

Complement inhibition reduces lesion size and hydrocephalus in injured pups
To investigate the role of complement in GMH-induced pathology in a clinically relevant setting, we treated pups with CR2Crry or vehicle starting at either 1 or 24 hours after collagenase injection, and every 3 days (the tissue half-life) thereafter until sacri ce at P14 (refer to Fig. 1). Nissl stains of midhippocampal and ventricular regions from the vehicle group collected at P14 demonstrated varying degrees of parenchymal lesion along with high rates of associated intraventricular hemorrhage and PHH; no brains from the vehicle-treated group scored scale 0. On the other hand, 28% of 1-hour CR2Crry-treated animals were scale 0, and 17% of 24-hour CR2Crry-treated animals were scale 0. Global hydrocephalus occurred in 61% of brains from vehicle treated animals (scale 5), compared to only 7% and 22% in the CR2Crry 1-hour and 24-hour treatment groups, respectively ( Fig. 4a and b). At P45 (41 days after injury), PHH was 75% in vehicle and 33% in the CR2Crry group (P<0.05) (Fig. 4f).
The lesion and ventricular volumes of the experimental groups were quanti ed using serial Nissl stained sections through each brain. Both ventricle and lesion volumes were decreased with CR2Crry treatment (both 1-hour and 24-hour treatments) compared to vehicle-treated animals ( Fig. 4c and d). There was no signi cant difference between the 1 and 24-hour CR2Crry treatment groups, and in subsequent experiments we focused on 1-hour CR2Crry treatment. Of note, brains from vehicle-treated animals were more likely to possess bilaterally enlarged ventricles occupying the majority of the intracranial compartment, coupled with relatively large lesions, as shown in the representative 3D reconstructed images of ventricle and lesion volume of all three conditions (Fig. 4e). The lateral ventricles in brains from CR2Crry-treated animals were closer to normal ventricular anatomy compared with the visually effaced ventricles observed in the vehicle group.
CR2Crry treatment decreases perilesional complement deposition, astrocytosis, and microgliosis We investigated the impact of complement inhibition on a perilesional cellular response in terms of posthemorrhage astrocyte and microglia/macrophage recruitment. For analysis of astrocytosis, brain sections from P14 mice were stained for Glial Fibrillary Acidic Protein (GFAP). Astrocytosis was examined in terms of the extent of astrocytic scar extending from the lesion border inward towards intact parenchymal tissue (Fig. 5a), and in terms of astrocyte density in the perilesional area at the interface with lesion (Fig. 5b). Compared to vehicle-treated animals, CR2Crry treated animals displayed reduced ipsilateral astrocytic scar formation within surrounding brain parenchyma (Fig. 5a, b). In addition, contralateral periventricular astrocytosis was also higher in vehicle animals compared to CR2Crry treated animals (Fig. 5c). Similarly, Iba-1 staining for microglia/macrophage in the perilesional region showed reduced microgliosis in CR2Crry-treated animals compared to vehicle treated animals (Fig. 5d). Correlating with reduced astrocytosis and microgliosis, there was also reduced C3 deposition in the perilesional area (Fig. 5e) and ipsilateral hippocampus (Fig. 5f) of CR2Crry treated mice.

Complement inhibition results in dendritic and neuronal preservation
To explore the role of complement activation in the neurodegenerative process occurring posthemorrhagic injury, we investigated the effect of complement inhibition on dendritic arborization (MAP2 stain) in ipsilateral and contralateral cortical hemispheres (Fig. 6a). Compared to naïve mice, vehicletreated animals displayed a decrease in dendritic arborization in both ipsi-and contralateral hemispheres, which was largely reversed with CR2Crry treatment; there was not a signi cant difference in MAP2 staining intensity between naïve and CR2Crry treated mice, suggesting a role for complement in dendritic loss post-injury. To further interrogate perilesional neurodegeneration, we immune-stained for neurons (NeuN). NeuN signal intensity, measured in terms of distance from lesion, was markedly higher in CR2Crry-treated mice compared to vehicle controls (Fig. 6b, upper panel). Brains from CR2Crry-treated mice showed high neuronal density in the immediate perilesional space compared to effacement of perilesional neurons in vehicle treated mice (Fig. 6b, lower panel). Notably, in vehicle-treated animals there is a presence of non-neuronal cells in the vicinity of the lesion as indicated by DAPI staining.
Neuronal loss is promoted by microglial/macrophage engulfment of complement opsonized neurons.
We next investigated a role for microglia in complement-mediated neuroin ammation that is associated with loss of neuronal density. In gure 5b, we analyzed perilesional neuronal density spatially extending from the lesion. Here, we analyzed overall neuronal density within perilesional elds and show that compared to naïve animals, vehicle-treated animals had a reduction in neuronal density in the perilesional area of microgliosis that colocalizes with C3 deposition. Compared to vehicle treatment, CR2Crry treatment reduced C3 deposition and microgliosis and preserved neuronal density (Fig. 7a-b). To investigate whether C3 opsonization may be responsible for microglial association with neurons and subsequent neuronal loss by promoting microglia-dependent engulfment, we rst quanti ed colocalization of microglia/macrophages with C3-tagged neurons. Within perilesional elds of view, C3/Iba-1 colocalization was observed on 62% of NeuN+ stained cells in vehicle-treated animals compared to 20% in CR2Crry-treated animals (Fig. 7c). We next demonstrated C3 deposition at the microglial/macrophage interface with neurons and quanti ed microglia/macrophage internalization of C3 and of neuronal (NeuN+) material. We found a higher number of microglia/macrophages with partially or fully internalized C3 in vehicle-treated animals compared to CR2Crry-treated animals (Fig. 7d). From calculations using the total number of NeuN+ cells within each eld as the denominator, we similarly found a higher percentage of microglia/macrophages with partially or fully internalized NeuN+ material in vehicle-treated animals compared to CR2Crry-treated animals (Fig. 7e). Two examples of microglia surface interaction with and internalization of a C3-tagged neuron are shown in gure 7f.
Example 1 shows a C3 tagged neuron engulfed within a microglia/macrophage, and example 2 shows a direct interaction between a C3 tagged neuron and a microglia/macrophage (see supplementary material for video demonstration). These data indicate a role for complement-dependent microglial phagocytosis in neuronal loss after GMH.

Complement inhibition improves overall weight gain and animal survival
Weight gain was monitored from P2 until sacri ce at P14. Compared to vehicle-treated mice, the overall weight gain in this period was signi cantly improved for mice treated with CR2Crry and was similar to percent weight gain in naïve mice ( Fig. 8a and b). In the two days prior to injury, all groups were growing at a comparable percent weight gain. 24 hours after GMH induction (shown by purple arrow, Fig. 8b), there was a deceleration in percent daily weight gain in both CR2Crry and vehicle animals until 4 days after injury (P8, shown by orange arrow). At that time point, CR2Crry animals began to exhibit an accelerated weight gain and approached the normal weight gain curve, as displayed by naïve animals. There was no difference between naïve and CR2Crry animal percent weight gains by P14. In a separate cohort of animals, survival was monitored for up to 41 days after collagenase-induced injury (P45). For this experiment, the same treatment paradigm used in the above studies was applied through P14, with subsequent CR2Crry or vehicle (PBS) treatments given weekly. Animal survival assessment began one day after injury (P5) to eliminate surgery-related deaths occurring within 24 hours. CR2Crry group mortality plateaued at P25, while vehicle animal mortality continued to increase through P45, at which time survival rate of CR2Crry-treated animals was 75% compared to 40% for vehicle treated animals (Fig.  8c). Within the vehicle cohort, 4 of 10 females and 4 of 9 males survived to P45 with no signi cant difference in gender. In the CR2Crry group, 4 of 5 females and 5 of 6 males survived, with no signi cant difference in gender.

Complement inhibition after germinal matrix hemorrhage enhances motor and cognitive performance at
adolescence.
An ongoing neuroin ammatory response has been linked to motor and cognitive dysfunction that is likely secondary to a loss of neurons. Our data above show an ongoing complement-dependent neuroin ammatory response and loss of neurons after GMH, and we therefore assessed whether this was linked to motor and cognitive performance at P30, when mice are able to physically perform behavioral tasks. Gait analysis (Noldus CatWalk XT) was performed at P30, and a CCI was computed using 100 plus different obtained values. Naïve and CR2Crry-treated mice had similar CCI scores, and their scores were signi cantly higher than vehicle-treated mice (Fig. 9a). Hippocampal integrity was assessed with fearconditioned memory retention using the passive avoidance task. CR2Crry-treated mice showed similar retention memory to naïve mice represented by a delayed time to enter the shock box of the task, which was signi cantly lower in vehicle-treated animals (Fig. 9b). The Barnes maze task was used to assess spatial learning and memory retention, and as with above tasks, CR2Crry-treated and naïve mice performed similarly and signi cantly better than vehicle-treated mice. CR2Crry-treated mice exhibited improved spatial learning ability throughout the learning phase of the task compared to vehicle-treated mice as shown by an improved total latency on the platform and latency until rst peek into the escape hole (Fig. 9c). Additionally, for both latency parameters, CR2Crry treatment signi cantly improved animal retention memory compared to vehicle on the nal day, in which animals performed the task after a 2-day break period. Heat maps depicting the movement of animals on the platform from representative experiments are shown in gure 9d. Thus, neuroin ammation and neuronal loss after GMH correlates with behavioral de cits as mice age, and these outcomes can be reversed by complement inhibition.

Discussion
The current work utilizes the targeted complement inhibitor CR2Crry in a novel murine model of GMH that mimics the natural mechanism of GMH in newborn humans. Unlike previously described animal models, the clostridium-derived collagenase-based murine model described here results in a high rate of posthemorrhagic hydrocephalus (Scale 5 lesion) in a high percentage of animals (about 60%). In comparison, human neonates with high scale GMH (Scale 3-4) are reported to develop PHH in up to 70% of cases. [29] Autologous intraventricular blood-injection models (ABM) have also been described, but they fail to mimic the natural mechanism of GMH. Those previous models also do not induce non-traumatic germinal matrix zone vessel rupture with disruption of the SVZ, BBB and parenchymal vasculature. [30,31] Cherian et al. described an ABM with a PHH rate of 65% following bilateral injection of autologous blood, but in the same study they showed that injection of arti cial CSF alone caused hydrocephalus in 50% of animals, indicating that the volume of injection likely contributed to the development of hydrocephalus. [30] Other ABM studies reported signi cantly lower resultant PHH with about 14% success. [31] In contrast, collagenase results in vascular collagen breakdown, leading to robust neurovascular destruction that closely mimics human GMH-IVH. This mimics disruption of the BBB and the long-lasting effects of immune cell in ltration and in ammation that occurs from blood leakage into the brain tissue. The collagenase model causes neurovascular injury that also potentiates hypoxia and ischemia, as well as local immune and in ammatory responses, which may represent a limitation of this model. Nevertheless, it is of note that similar responses can be seen in human GMH pathology. Current collagenase-based models (which produced minimal to no PHH) utilized slightly older, rat, models (P7), in contrast to our mouse model (P4). Notably, P4 induction of GMH in the mouse model equates to approximately 32-week-old premature human neonates in brain development, [32] making the model directly translatable to the human pathophysiology that results in post-hemorrhagic infarction, PHH, and periventricular leukomalacia. The P7 collagenase-injected rat models equate to approximately 2.5 monthold full term humans, in whom GMH is not encountered. [30,[33][34][35]. In our model it is possible that the diffusion of collagenase or blood across the ventricle can happen from the destruction of ependymal tissue on the ipsilateral side, with leakage of active collagenase, along with blood products, through the ventricle. To this point, an advantage of our model is the trajectory used to reach the SVZ and germinal matrix. Other models use a vertical, trans-ventricular approach to reach the SVZ, whereas the current model uses a horizontal injection to penetrate the SVZ while avoiding cross-penetration of the ventricular wall, which minimizes dispersion of collagenase through the ventricular system. This approach lowers the risk for tissue destruction resulting directly from extravasation of collagenase, with effects more likely to occur from tissue destruction and blood dispersion into the ventricles (IVH).
GMH pathophysiology is similar to other types of brain injury, such as TBI and stroke, in the pattern of primary injury which is then followed by secondary injury. Complement activation has been implicated in propagating secondary injury following TBI and stroke.[28, 36] However, the role of complement in GMH and the development of PHH has never been investigated. Multiple mechanisms have been described in the development of PHH, including iron deposition leading to in ammation and obstruction of the normal absorption pathways, [37,38] TLR-4-activation leading to hypersecretion of CSF, [39] and recruitment of in ammatory cells and the formation of an astrocytic scar. [34,40,41] Although complement has been independently associated with these pathways, any direct role for complement in post-hemorrhagic hydrocephalus has not been investigated. [28,42,43] Here we investigated the role of complement in post-GMH pathology and PHH development in a therapeutic paradigm. The complement inhibitor utilized, CR2Crry, is a fusion protein consisting of a CR2 targeting domain linked to Crry, an inhibitor of C3 activation which is a central step of the complement cascade. The CR2 moiety binds C3 activation fragments that become covalently attached to activating surfaces. [44] We initially investigated both 1-hour and 24-hour CR2Crry treatments post-GMH induction, since clinically delayed diagnosis of GMH is common. Treatment at both time points was protective, and there was no signi cant difference in outcomes between the different treatment times. Complement inhibition reduced the rate of PHH development and lesion volume, and increased brain tissue preservation. These improvements were associated with reduced perilesional C3 deposition, and reduced astrocytosis and microgliosis, occurrence of which has been shown to contribute to the secondary injury after neurotrauma. [45] Interestingly, we identi ed deposition of astrocytes in the contralateral periventricular region to be higher in vehicle compared to treated animals. This correlated with an increased rate of PHH in those animals. It is unclear whether periventricular in ltration of astrocytes contributed to hydrocephalus, but microgliosis and astrocytosis appear to be directly correlated, and both were reduced with complement inhibition. Furthermore, astrocytosis is known to play a role in minimizing expansion of injury.
[46] The reduction of astrocytosis following CR2Crry treatment in the current study may be attributed to the overall reduction in injury rather than a casual deleterious effect of astrocytes.
In addition to perilesional effects, complement inhibition reduced deposition of C3 in the ipsilateral hippocampus and preserved dendritic density globally throughout the cortex. Clinically, high grade GMH leads to major cognitive sequelae in up to 86% of human infants. [29] In ammation within the hippocampus has been linked to poor neurocognitive performance in memory-related tasks, both clinically and experimentally. [47] Prevention of global hippocampal in ammation with CR2Crry likely contributed to improved Barnes maze and passive avoidance tasks performed in early adulthood testing of treated animals (P30 and beyond). Secondary to lack of reliable ne motor and motor-related cognitive testing in younger animals, we evaluated motor and behavioral functions of pups at P30.
[48] Our results demonstrated preservation of motor function as well as cognitive function in CR2Crry-treated animals compared to vehicles. Maintaining larger regions of cortical tissue, both ipsilateral and contralateral to the injury site, are likely major contributors to improved motor and cognitive outcomes in CR2Crry-treated animals. CR2Crry treatment not only reduced rates of bilateral injury, but also the severity of unilateral injury, with more CR2Crry-treated animals having scale 1,2 and 3 hemorrhagic lesions relative to scale 4 and 5 in vehicle controls.
We also identi ed a probable mechanistic link between complement activation and neuronal loss. First, we identi ed a higher rate of colocalization of C3-opsonized neurons with microglia/macrophages in perilesional areas in the vehicle animals compared to CR2Crry-treated animals. Secondly, within perilesional areas we found higher a higher percentage of microglia/macrophages with internalized C3 and neuronal material in vehicle vs. CR2Crry-treated animals. This correlated with preservation of neuronal density in CR2Crry-treated animals that in turn was associated with improved motor and cognitive performance in adolescence. Together, the data indicate that following GMH, progression to PHH with neuronal loss and the associated behavioral de cits are mediated, at least in part, by complement receptor-mediated uptake of C3 opsonized neurons by microglia/macrophages. Although neurons appear to be phagocytosed at higher numbers in the vehicle animals, the underlying phenotype of those neurons remains unclear. Some studies have suggested neuronal damage occurs following hemorrhagic injury due to iron-induced ferroptosis [49]. However, there is also evidence of continued, pathologic complement targeting of neurons and neuronal progenitors secondary to continued activation of the alternative pathway [12,50].

Conclusions
In conclusion, PHH is a devastating pathology, currently managed exclusively through surgical CSF diversion procedures which carry life-long risks of repeated failure, infection and complications.
[51] Neonatal survival without surgical intervention for PHH is dismal, while those treated for hydrocephalus continue to suffer secondary brain injury-related neurological de cits, such as motor, cognitive, visual and psychological deterioration. [52] In this study, we demonstrated a survival rate of 75% at P45 following CR2Crry treatment, independent of surgical CSF diversion, compared to 40% in vehicle treated animals.
To our knowledge, this treatment paradigm with complement inhibition is the rst to demonstrate a successful preclinical pharmacologic therapy for this devastating neonatal pathology without surgery.
The data suggest that complement inhibition has the potential to also reduce the rate of neonates requiring PHH-related surgical intervention. On a translational note, a humanized CR2-targeted complement inhibitor has been shown to be safe and nonimmunogenic in humans.
[53] Also, in addition to FDA approved anti-C5 mAbs that inhibit downstream of the C3 inhibitor used here, [54] there are currently a multitude of companies developing complement inhibitors, with many in clinical trials. [44] Abbreviations GMH

Competing interests
ST is an inventor on a licensed patent for CR2-targeted complement inhibitors. The remaining 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.   Figure 1 Work ow of surgical procedure and treatment paradigms for all experimental groups. At P4, the surgical coordinate was marked and pups were cryo-anesthetized. 0.5 µl of collagenase was injected into the subventricular zone of the brain via a lateral transcortical approach and pups were returned to their mother after re-warming. Experimental groups were:   Fluorescence-tagged CR2Crry targets to the brain following injury. Fluorescent-tagged CR2Crry was administered i.p. to mice 1 hour after induction of GMH or to control mice (no injury). a) Representative live animal uorescence tomography images at indicated time points after CR2Crry administration, showing initial systemic distribution with subsequent retention of signal in the brain of GMH mice. b)

Figures
Quanti cation of uorescence intensity in brains of GMH mice and control mice at indicated time points after CR2Crry administration, showing the drug has a tissue half-life in the brain of about 3 days in GMH mice. Two-way ANOVA with Bonferroni's correction for multiple comparisons. *p<0.05. n=4 for GMH, no GMH, and control groups. Error bars = mean ± SEM.

Figure 4
Complement inhibition leads to a reduction in lesion size and hydrocephalus. a) Distribution of injury    Imaris to obtain colocalization analysis. b) Perilesional quanti cation of neurons in animals treated as indicated. Cortical images from naïve brains were used for comparison. *p<0.05, ****p<0.0001. Error bars= mean ± SEM. c) IF stain for C3 deposition, Iba-1, and NeuN colocalization, performed based on surface proximity and analyzed as a percent of total neurons/ eld. Association of microglia/macrophages (Iba-1) with complement-tagged neurons as a percentage of total neurons present was higher in vehicle (62%) compared to CR2Crry-treated animals (20%). ****p<0.0001. Error bars= mean ± SEM. n= 6 for naïve, n=15 for vehicle, and n=14 for CR2Crry. d) Quanti cation of microglia/macrophages within the perilesional space with partial or complete internalization of C3 material. e) Quanti cation of microglia/macrophages within the perilesional space with partial or complete internalization of NeuN+ material as a percent of total quanti ed neurons. ****p<0.0001. Error bars= mean ± SEM. n=15 for vehicle, and n=14 for CR2Crry. f) Example images showing microglia/macrophage association and internalization of C3-tagged neurons in vehicle animals. (see also supplementary material). No data points were excluded from the analysis.

Figure 8
CR2Crry treatment improves weight gain after GMH and promotes survival. a) Overall percent weight gain from day P2 to P14 in vehicle and CR2rry-treated mice. Unpaired Student's T-test. *p<0.05. Error bars= mean ± SEM. b) Daily weight percent gain over 12-day period following GMH induction. Deceleration in percent daily weight gain in CR2Crry and vehicle animals at P5 (purple arrow). Acceleration of weight gain in CR2Crry-treated animals (orange arrow). Two-way ANOVA with Bonferroni's correction for multiple comparisons. **p<0.01, ***p<0.001. n=11 for naïve, n= 18 for vehicle, and n=14 for CR2Crry. Error bars= mean ± SEM. c) Survival assessed over 41 days after injury (P45). Animal survival was assessed beginning at one day after injury (P5). Animals that died within 24 hours of injury were excluded from analysis. CR2Crry group deaths plateau around P25 while vehicle animal deaths continue until close to P45. P45 animal survival was 75% in the CR2Crry group compared to 45% in the vehicle (p<0.05). Logrank (Mantel-Cox) test. *p<0.05. n=20 for vehicle and n=12 for CR2Crry. Error bars= mean ± SEM.

Figure 9
CR2Crry treatment improves motor performance and cognitive performance of GMH injured mice. a) Gait analysis evaluated using Catwalk XT. The output variables for all four limbs were combined into a