Neural stem cell treatment for perinatal brain injury: A systematic review and meta‐analysis of preclinical studies

Abstract Perinatal brain injury can lead to significant neurological and cognitive deficits and currently no therapies can regenerate the damaged brain. Neural stem cells (NSCs) have the potential to engraft and regenerate damaged brain tissue. The aim of this systematic review was to evaluate the preclinical literature to determine whether NSC administration is more effective than controls in decreasing perinatal brain injury. Controlled interventional studies of NSC therapy using animal models of perinatal brain injury were identified using MEDLINE and Embase. Primary outcomes were brain infarct size, motor, and cognitive function. Data for meta‐analysis were synthesized and expressed as standardized mean difference (SMD) with 95% confidence intervals (CI), using a random effects model. We also reported secondary outcomes including NSC survival, migration, differentiation, and effect on neuroinflammation. Eighteen studies met inclusion criteria. NSC administration decreased infarct size (SMD 1.09; CI: 0.44, 1.74, P = .001; I 2 = 74%) improved motor function measured via the impaired forelimb preference test (SMD 2.27; CI: 0.85, 3.69, P = .002; I 2 = 86%) and the rotarod test (SMD 1.88; CI: 0.09, 3.67, P = .04; I 2 = 95%). Additionally, NSCs improved cognitive function measured via the Morris water maze test (SMD of 2.41; CI: 1.16, 3.66, P = .0002; I 2 = 81%). Preclinical evidence suggests that NSC therapy is promising for the treatment of perinatal brain injury. We have identified key knowledge gaps, including the lack of large animal studies and uncertainty regarding the necessity of immunosuppression for NSC transplantation in neonates. These knowledge gaps should be addressed before NSC treatment can effectively progress to clinical trial.


| INTRODUCTION
Injury to the developing brain during pregnancy or around the time of birth, termed perinatal brain injury, is a major cause of morbidity and mortality. There are various, complex causes of perinatal brain injury, including inflammation, hypoxia ischemia (HI), excitotoxicity, placental abnormalities and perinatal stroke. 1 Additionally, fetal growth restriction, chronic hypoxia, infection, and inflammation can render the brain more vulnerable to perinatal injury. [2][3][4][5] Perinatal brain injury can lead to cerebral palsy (CP), epilepsy, and other permanent neurological disorders. [6][7][8] Current therapies for many of these conditions are predominantly based on symptom management, including physical rehabilitation and anti-seizure medication. 9 Available medical interventions are limited but include therapeutic hypothermia for term-born infants with neonatal encephalopathy which is associated with significant improvements in neurological function.
Unfortunately, half of the neonates treated with hypothermia will still die or have serious adverse neurological outcomes. 10 Additionally, antenatal corticosteroids and magnesium sulfate provide neuroprotection, 9 but no treatments are available to repair the underlying brain injury.
Stem cells have been extensively researched to treat neurological conditions as some stem cell types have regenerative, anti-inflammatory, and neuroprotective properties. Over the last two decades a number of stem/progenitor cell types have been shown to have the capacity to differentiate into neural cell lineages in vitro, including embryonic stem cells (ESCs), induced pluripotent stem cells (iPSCs) and a number of fetal derived stem cells. 11,12 As such, many stem cell types have been tested and shown potential in preclinical studies to reduce brain injury, including mesenchymal stem cells (MSCs), neural stem cells (NSCs), and umbilical cord blood cells (UCBCs). 12,13 The mechanisms of action of MSCs and UCBCs for brain injury treatment are primarily trophic and anti-inflammatory, with no evidence of significant engraftment or neural lineage differentiation. [14][15][16][17] In contrast, ESCs, iPSCs or NSCs from fetal tissue origin, are anti-inflammatory and neurotrophic, and they can also engraft into the brain. Here, they can differentiate into the three primary cell types; neurons, astrocytes, and oligodendrocytes 18 and therefore hold promise to repair and regenerate damaged brain tissue. NSCs can be obtained from a number of sources including fetal or embryonic brain tissue, or they can be differentiated from embryonic stem cells (ESCs) or induced pluripotent stem cells (iPSCs). 19 NSCs are being investigated to treat adult neurological conditions, including Parkinson's disease, multiple sclerosis, Huntington's disease, spinal cord injury, and stroke. [20][21][22][23][24][25] Notably, in preclinical models of adult stroke, NSCs significantly reduce brain infarct size and alleviate behavioral deficits, 20,22 specifically in tests that assess motor function.
NSCs may hold the key to promoting brain repair and are therefore an appealing reparative therapy for perinatal brain injury. There is now a growing body of preclinical evidence investigating the efficacy of NSC therapy; however, there are often conflicting results in the literature. Therefore, we conducted a systematic review and metaanalysis to determine whether NSC administration is more effective than control/vehicle treatment. The primary outcomes of interest were reduced brain infarct and functional improvement in behavioral (motor and cognitive) tests. Secondary outcomes of interest were NSC survival, migration, and differentiation. Conducting this systematic review may identify preclinical gaps that should be addressed before NSC therapy moves toward clinical trials.

| METHODS
This systematic review and meta-analysis followed the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA, http://www.prisma-statement.org/). 26 The review protocol was registered on PROSPERO (CRD42021222952).

| Selection criteria
Preclinical studies must have utilized a neonatal or perinatal model of hypoxic, ischemic, or inflammatory, or excitotoxic brain injury to be eligible for inclusion. As the main objective of this systematic review was to determine the effectiveness of NSC therapy for perinatal brain injury, all studies must have included a group that were administered non-transfected or pretreated "neural stem cells," a term which we have defined to include NSCs, neural progenitor cells (NPC), neurosphere derived precursor cells (NDPCs), neural precursor cells, oligodendrocyte progenitor cells (OPCs), olfactory cells, ensheathing cells, or neuroepithelial cells. For simplicity, we use the term "NSCs" throughout this review to describe all eligible cell types. If a study included an adjuvant or concomitant therapy, it must also have included a non-transfected or pre-treated NSC group to be eligible. Cells could be given at any time point following brain injury and via any route of administration. Studies must have included an injured, non-cell-treated control group. Eligible studies must have included at least one of the following primary outcomes: brain volume or brain infarct volume; motor function; or cognitive function.

| Search strategy
We searched MEDLINE (1946 to September 24, 2020) and Embase (1947 to September 24, 2020) via Ovid using the following strategy: ([neonatal or perinatal or neonate or perinate or newborn].tw) AND ((brain or cerebr* or neuro*) AND (occlusion or stroke or hypoxic or hypoxia or ischemic or ischaemic or ischaemia or ischemia or injury)).
tw) AND ([neural stem or NSC* or neuroepithelial or neural progenitor* or NPC* or neuro-progenitor* or neuro-epithelial or oligodendrocyte progenitor* or OPC* or olfactory cell* or ensheathing cell* or OEC*].tw) AND ([transplantation or transplant or injection or inject or administration or administer or administered or intracerebral or intranasal or intraperitoneal or intravenous or intravenously or infusion or treatment or treat or treated].tw). Searches were limited to English language articles. To ensure no recent studies were missed, searches were rerun using the same parameters on May 4, 2021.

| Study selection process
Deduplicated results from Ovid were exported into EndNote (version X9.3.3). Additional deduplication was conducted both automatically using EndNote as well as manually by study authors. Preliminary title screening was conducted to remove reviews, protocols, conference abstracts and other ineligible study types before remaining studies were exported into Covidence Systematic Review Software (Veritas Health Innovation, Melbourne, Australia, available at http://www.covidence.org).
Using Covidence, titles or abstracts of retrieved studies were screened independently by two study authors (M.S. and C.M.) to identify studies that met the inclusion criteria. Any disagreements were resolved through discussion with an additional reviewer (M.P.). The full texts of potentially eligible studies were then retrieved and independently assessed for eligibility by two review authors (divided between M.S., C.M., M.P., and M.F.-E.), with any disagreements resolved by a third reviewer.

| Data extraction
Data were extracted from all eligible studies independently by study authors (divided between M.S., C.M., M.P., and M.F.-E.) into a spreadsheet (Microsoft Excel). Extracted information included the author and publication year, animal characteristics, including brain injury model, species, age, and the number of animals included for each outcome. Details of the intervention that were captured included cell type, donor source, dose, the use of immunosuppression, timing and route of administration and comparator. Additionally, outcome assessment details and data were extracted. We have classified all cells derived from embryonic or fetal brain tissue as "fetal tissue derived NSCs" and all neural lineage cells derived from iPSC, ESC, or adult cells as "iPSC-, ESC-, or adult tissue derived NPCs." Any identified discrepancies were resolved through discussion with an additional author. PlotDigitizer (version 2.6.9) was used to quantify the mean and standard deviation or standard error from figures if data were not provided in the text/tables. If relevant data were not available in the published manuscript or supplementary materials, authors were contacted twice, if required. If no response was received, the data were not included.

| Risk of bias
Three study authors (divided between M.S., S.M., and M.F.-E.) independently assessed the risk of bias for each included study using the Systematic Review Centre for Laboratory Animal Experimentation (SYRCLE) risk of bias tool. 27 SYRCLE assesses selection bias, performance bias, detection bias, attribution bias and reporting bias, reported as "Yes, No or Unclear." Disagreements were resolved through discussion with additional author/s.

| Data synthesis
Review Manager (RevMan) version 5.4 was used to conduct quantitative analysis for primary outcomes when three or more studies assessed brain volume or the same motor or cognitive test. Where studies included an adjuvant or concomitant therapy, only the NSC alone group was used for quantitative analysis. If multiple assessment time points for a single outcome were reported, only the last time point was included in the meta-analysis. If data were presented as brain volume (tissue remaining), we converted the data to percentage brain infarct volume (tissue loss) to allow assessment by metaanalysis.

| Data analysis
We used a random-effects, inverse variance model to evaluate the standardized mean difference (SMD) and 95% confidence interval (CI) for all continuous data. The effect of heterogeneity was assessed using the I 2 statistic, with values of 25%, 50%, and 75% considered to be low, moderate, and considerable heterogeneity, respectively. Subgroup analysis was performed where sufficient data were available (two or more studies in each subgroup) to determine potential sources of heterogeneity based on a priori nominated factors including brain injury model, hypoxia length, immunosuppression use, NSC donor species, NSC dose, NSC administration time point post-injury, and NSC route of administration.

| Characteristics of included studies
Characteristics of included studies are summarized in Table 1. All studies were performed in rats (n = 9, 50%) or mice (n = 9, 50%). The brain injury models used were hypoxic ischemic (HI) injury (n = 14, 78%), ischemic injury (n = 2, 11%), excitotoxic injury (n = 1, 6%), and HI plus inflammation injury (n = 1, 6%). During HI injury induction, the length of hypoxia ranged from 20 to 150 minutes. The age at which injury was induced ranged from postnatal day (PND) 2 to PND12.  Immunosuppression was used in two studies (11%), and both used cyclosporine. Three studies included either a concomitant therapy (hypothermia (n = 1) or chondroitinase ABC (n = 1)) or included an NSC group that was administered on a poly (glycolic acid)-based scaffold (n = 1). However, as required by our inclusion criteria, all of these studies included an NSC alone group.

| Primary outcomes
The results from the primary outcomes for this systematic review are summarized in Table 2. 3.3.1 | Effect of NSCs on brain infarct volume Ten of the 18 included studies assessed brain infarct volume using either infarct volume (n = 7), injured hemisphere volume (n = 2) or gray/white matter loss (n = 1), measured at time points between 5 and 77 days post-injury. Five studies showed that treatment with NSCs led to a reduction in brain infarct size, and five showed no significant difference compared with injured controls (Table 2). One study, 31 which looked at early (2 days) vs delayed (7 days) administration, showed that infarct size was reduced following administration at 2 days post-injury but not at 7 days.
After excluding one study that used an incomparable unit of measurement (median and interquartile range), 31 Table 2). In contrast, four studies failed to show any significant between-group difference for any motor function assessment. A range of motor assessments were used across the studies, and many studies assessed motor function via two or more different motor assessments.
Interestingly, most studies that used multiple assessments (n = 7/8) showed either a statistically significant improvement or an insignificant result across all of their chosen motor assessments. The most common assessment was impaired forelimb use, measured by the cylinder test (n = 7) or forelimb placement test (n = 1) between 7 and 77 days postinjury, followed by the rotarod test (n = 5) similarly measured between 7 and 84 days post-injury. One study included a repeat dose (four doses) group and a single dose group. 34 Figure 3).  Figure 3).

| Effect of NSCs on cognitive function
Less than half of the included studies assessed cognitive function Three studies showed that NSCs significantly improved the function on the Morris water maze test (Table 2). From the meta-analysis, NSCs significantly improved performance compared with injured controls on the Morris water maze test following brain injury by an SMD of 2.41 (CI: 1.16, 3.66, P = .0002; X 2 = 16.04, I 2 = 81%, P = .0002) with effect sizes ranging from 1.46 to 3.68 (Figure 3).

| Subgroup analysis
Subgroup analysis of relevant parameters and their contribution to outcome heterogeneity is presented in Supplementary Figures S1 and   S2. Due to the limited number of studies, we were unable to run analyses for NSC donor species, NSC dose, brain injury model and hypoxia duration. When a sufficient number of studies were available, subgroup analysis of brain infarct size showed no difference in effect size between cell administration time point (P = .17) or species of the donor NSCs (P = .67) (Supplementary Figure S1a,c). While studies that did not use immunosuppression had a larger reduction in infarct size compared with those that did (P = .03), this result should be interpreted with caution due to the small number of studies (n = 2) that used immunosuppression (Supplementary Figure S1b). Moreover, high heterogeneity was noted even within the group that did not use immunosuppression (I 2 = 72%). Additionally, studies that administered NSCs via intranasal delivery had a larger reduction in infarct size compared with both intracerebral and intraventricular delivery (P < .0001, P = .0007, respectively) (Supplementary Figure S1d)

| Secondary outcomes
3.4.1 | NSC survival, migration, differentiation, and neuroinflammation The secondary outcomes of interest are summarized in Table 3. Cell survival was always assessed by brain histology after animals were Contrastingly, one study found that NSC treatment increased the number of microglia in the striatum, indicating a pro-inflammatory response against transplanted NSCs.

| Modifications and concomitant therapies
Some studies included in this systematic review also tested the efficacy of modified donor cells, concomitant therapies or cell scaffolding, and their effect on the primary outcomes are summarized in

| Risk of bias assessment
The risk of bias across included studies is summarized in Figure 4. No studies were judged to have a low risk of bias across all domains.
Selection bias was low in most studies when examining randomization, but few studies described baseline characteristics of included animals. Additionally, no studies specifically described the method of random sequence generation. Across all studies, allocation concealment, random housing, blinding of caregivers, and random outcome F I G U R E 4 Risk of bias of the included studies: + = low risk of bias,? = unclear risk of bias, and -= high risk of bias

| Primary outcomes
This systematic review aimed to measure the efficacy of NSC administration for perinatal brain injury to identify the areas of research with insufficient data and key knowledge gaps. Primary outcomes assessed in this systematic review were brain infarct size and motor and cognitive outcomes, which have clinical relevance since it is well known that perinatal brain injury often leads to motor deficits such as cerebral palsy, 46 cognitive deficits, and neurodevelopmental delays. 47 There was variability in the behavioral outcomes assessed across studies, highlighting the need to standardize outcomes, and ensure that they closely measure relevant clinical outcomes of perinatal brain injury. Standardization of outcomes and use of measures that hold clinical relevance would provide more power to meta-analyses, so that robust efficacy data can progress NSC therapy along the translational research pipeline. We have found that motor function testing predominantly used impaired forelimb and rotarod tests, and given that 95% of included studies used a unilateral brain injury model that more closely mimics hemiplegic cerebral palsy, 48 these tests were clinically relevant and should be used more widely. In addition, the Morris water maze was the most common cognitive test used, which assesses learning and memory deficits that can occur in humans after perinatal brain injury. 47 Given that the key benefit of NSCs is the potential to repair the injured brain, 49 the effect of these cells on infarct size was also an important outcome. Overall, our meta-analysis showed that NSCs significantly decreased brain infarct volume and improved motor and cognitive outcomes, indicating that NSCs are effective in reducing the severity of perinatal brain injury and alleviating functional deficits. Our results are consistent with meta-analyses of preclinical models of adult stroke. 20,22 Both of these adult stroke meta-analyses reported a SMD of <1 in the cylinder test (impaired forelimb use), whereas we had a comparatively larger SMD of 2.27. SMD for infarct size were similar to that seen in adult stroke models. Additionally, the SMD reported for the rotarod test was 1.88, which was similar to the SMD reported in the adult stroke metaanalyses. The NSC field is more advanced in the area of adult diseases and phase I trials have shown safety in adults following stroke 50 and has now moved to phase II clinical trials. 51

| Secondary outcomes
The secondary outcomes of interest in this systematic review included the potential of NSC differentiation, survival, migration, and the effect of NSC therapy on neuroinflammation. Neuroregeneration is a proposed mechanism of action of NSC therapy, 49 which likely relies on NSC survival and differentiation into neural cells within the damaged brain. Encouragingly, in most studies that investigated NSC differentiation, NSCs were able to differentiate into neurons. This is consistent with studies demonstrating that NSCs first differentiate into neurons followed by glial cells. 54 Differentiation into astrocytes was observed, but not as commonly, and oligodendrocyte differentiation was generally observed at later cull time points (>28 days of transplantation).
Additionally, NSCs were present within the brain for up to 133 days after administration, which suggests these cells may have the ability engraft long-term in the neonatal brain. Studies in rodent models of spinal cord injury have shown that long-term engraftment of NSCs may be a necessary process to promote functional improvement, since selective ablation of exogenous NSCs reversed locomotor recovery. 55 Consequently, engraftment is likely necessary to elicit improvements in clinical outcomes following NSC administration in the injured perinatal brain and encouragingly this review has shown that NSC survival is possible.
Interestingly, while one of the main proposed mechanisms of action of NSCs is their immunomodulatory capacity, 56

| Knowledge gaps identified
Conducting this systematic review and meta-analysis enabled us to identify knowledge gaps that need to be addressed to progress this area of research. First, the mechanisms by which NSCs can elicit their beneficial effects remains unclear but likely includes cell replacement, immunomodulation, neurotrophic action, the acceleration of endogenous recovery, or a combination of these mechanisms. 58 If engraftment is required for brain regeneration, determining the interplay between the number of engrafted NSCs and efficacy will need to be established. In addition, the fate of these engrafted cells needs to be further investigated beyond identifying that a few cells can differenti- ate. This will include information such as the percentage of engrafted cells that differentiate and what cell type they become, that is neurons, oligodendrocytes, or astrocytes.
Engraftment is likely dependent on the immune response (or lack thereof) to NSCs by the host immune system. Therefore, the role of immunosuppression, particularly as we translate preclinical research, needs to be considered. It has been shown that transplanted NSC are recognized by the immune system and induce an immune response, 59 and it has been shown that immunosuppression is required for longterm engraftment in adult neurological animal models. 60

| Limitations
In this systematic review, we have specifically focused on the efficacy of NSCs for improving perinatal brain injury. We acknowledge that this is a limitation of the study and there are a number of other stem cell types that are being pursued for perinatal brain injury and have shown much promise. 15,69,70 In addition, there is interest in stem cell sources that possess the two-pronged mechanism of neural differentiation and growth factor secretion similar to NSCs, and these include ESCs and iPSCs. However, we are not aware of any studies that have tested the efficacy of undifferentiated cells in preclinical models of perinatal brain injury. It will be important for the field moving forward that head-to-head comparison studies of different stem cell types are performed so we can determine the optimal cell type to treat perinatal brain injury. Another limitation of this study was the meta-analyses were limited by the small number of studies and the high heterogeneity across outcomes, especially the behavioral assessments. Additionally, we were unable to determine the source of heterogeneity, subgroup analysis was limited and publication bias could not be investigated, due to the small number of studies, variation in study design, and intervention characteristics discussed above.
Through our risk of bias assessment, it was identified that there was minimal reporting across all domains, limiting the conclusions drawn from this meta-analysis. This weakness is widely recognized in preclinical animal research 71,72 and has likely contributed to the high risk of bias identified in this systematic review. The common areas of bias we identified included selection bias, including under reporting of allocation concealment and baseline characteristics of animals before the study, as well as attrition bias, with some studies having incomplete numbers in their outcome data and not accounting for the loss of animals from outcomes. While it is possible that studies had a low risk of bias in their study design and simply did not report clearly within the domains, especially due to strict word limits for publication, significant concerns to the validity of studies remain. This meta-analysis highlights the need to for preclinical scientists to rigorously design and report methodology and refer to the SYRCLE and ARRIVE guidelines 27,73 when publishing to allow future systematic reviews to rigorously interrogate preclinical research.

| Conclusions and future directions
The limited treatment options for perinatal brain injury and the subsequent life-long burden defines an urgent need to develop neuroregenerative treatments. From preclinical research to date, all performed in rodents, we show that NSC administration is an efficacious treatment for perinatal brain injury across neuropathological, motor, and cognitive domains. Before the commencement of clinical trials testing the efficacy of NSC transplantation for perinatal brain injury, we have identified important future directions that preclinical research needs to address first. Knowledge gaps identified include the lack of direct comparison of the route, dose, source, and timing of NSC administration in addition to the standardization of clinicallyrelevant behavioral outcomes. Studies in large animal studies are necessary to show the effectiveness of NSC transplantation for perinatal brain injury and further investigation is required into whether immunosuppression is necessary.

DATA AVAILABILITY STATEMENT
All supporting data and datasets generated for this study are available on request to the corresponding author.