Intravenous Route of Cell Delivery for Treatment of Neurological Disorders: A Meta-Analysis of Preclinical Results

Abstract

The last decade has been marked by a growing interest in an employment of intravenous cell delivery for treatment of neurological disorders. Numerous preclinical experimental studies have reported functional benefits, and have recently been followed by clinical trials. Some early clinical studies have indicated only modest positive effects, suggesting that the optimal conditions have not been defined yet. Thus, the evaluation of factors that influence outcomes, on the level of the whole population of preclinical studies by advanced statistical analysis, is warranted. PubMed search was conducted from the inception through 2006, and 60 preclinical studies were found and subjected to analysis. Categorical and continuous independent variables (IVs) were extracted. Three distinct outcomes of interest were selected as dependent variables (DVs) and named treatment effects: morphological, behavioral, and molecular, respectively. Mean outcomes, standard deviations (SDs), and animal numbers were retrieved and calculated by individual comparisons of experimental and control groups, based on the Hedges g formula, and were expressed as effect sizes (ESs) and variances. Publication bias and homogeneity were evaluated. The mainspring analyses were performed under a random effect model using Proc Mixed (SAS, version 9.2). A significant heterogeneity and publication bias were found. The ES pooling revealed large treatment effects. Univariate and multivariate meta-regression revealed that cell-related variables explained most of the heterogeneity. Cells retrieved from established lines and genetic modification of cells warrants the highest efficiency, in a dose-dependent manner. The stratified analysis of molecular effect measures revealed that apoptosis inhibition is the strongest brain tissue-positive change induced by cell therapy.

Introduction

There is a growing interest in an employment of intravenous route of cell delivery for treatment of neurological disorders. Numerous original reports from experimental studies revealed the potential of such an approach to overcome the limitation of standard therapeutic strategies (Table 1). Cell therapy for neurological disorders has also been discussed in numerous review articles [1 –5]. Intravenous route has been employed in several clinical trials conducted over the last several years [6 –12], and in another ongoing trial [13]. However, variable results have been reported.

Table 1.

List of Studies Used for Meta-Analysis

Id	First author	Last author	Year	Journal	Number	Pages
1	Li Y	Chopp M	2002	Neurology	59	514–523
2	Chen J	Chopp M	2003	J Neurosci Res	73	778–786
3	Chen J	Chopp M	2003	Circ Res	92	692–629
4	Seyfried D	Chopp M	2006	J Neurosurg	104	313–318
5	Shen LH	Chopp M	2006	Neuroscience	137	393–399
6	Shen LH	Chopp M	2006	J Cereb Blood Flow Metab	27	6–13
7	Mahmood A	Chopp M	2001	Neurosurgery	49	1196–1204
8	Lu D	Chopp M	2001	Neuroreport	12	559–563
9	Mahmood A	Chopp M	2004	Neurosurgery	55	1185–1193
10	Mahmood A	Chopp M	2006	J Neurosurg	104	272–277
11	Mahmood A	Chopp M	2003	Neurosurgery	53	697–703
12	Mahmood A	Chopp M	2004	J Neurotrauma	21	33–39
13	Li Y	Chopp M	2006	Exp Neurol	198	313–325
14	Zhang J	Chopp M	2006	J Neurosci Res	84	587–595
15	Willing AE	Sanberg PR	2003	J Neurosci Res	73	296–307
16	Willing AE	Sanberg PR	2003	Cell Transplant	12	449–454
17	Saporta S	Sanberg PR	2003	J Hematother Stem Cell Res	12	271–278
18	Vendrame M	Willing AE	2004	Stroke	35	2390–2395
19	Vendrame M	Willing AE	2005	Stem Cells Dev	14	595–604
20	Newcomb JD	Willing AE	2006	Cell Transplant	15	213–223
21	Vendrame JD	Willing AE	2006	Exp Neurol	199	191–200
22	Borlongan CV	Sanberg PR	2004	Stroke	35	2385–2389
23	Makinen S	Jolkkonen J	2006	Brain Res	1123	207–215
24	Chu K	Yoon BW	2003	Neurosci Lett	343	129–133
25	Jeong SW	Roh JK	2003	Stroke	34	2258–2263
26	Chu K	Roh JK	2004	Brain Res	1016	145–153
27	Chu K	Roh JK	2004	Brain Res	1023	213–221
28	Lee ST	Roh JK	2006	J Neurosci Methods	152	250–254
29	Chu K	Roh JK	2005	Neurosci Res	53	384–390
30	Lee ST	Kim M	2005	Neurosci Res	52	243–249
31	Chu K	Yoon BW	2004	Neurosci Res	50	459–465
32	Chen SH	Lin MT	2005	Crit Care Med	33	1377–1383
33	Chen SH	Lin MT	2006	Exp Neurol	199	67–76
34	Corti S	Comi GP	2005	FASEB J	19	1860–1862
35	Honma T	Kocsis JD	2006	Exp Neurol	199	56–66
36	Nomura T	Kocsis JD	2005	Neuroscience	136	161–169
37	Iihoshi S	Kocsis JD	2004	Brain Res	1007	1–9
38	Horita Y	Kocsis JD	2006	J Neurosci Res	84	1495–1504
39	Liu H	Kocsis JD	2006	Brain	129	2734–2745
40	Urdzikova L	Sykova E	2006	J Neurotrauma	23	1379–1391
41	Jendelova P	Sykova E	2004	J Neurosci Res	76	232–243
42	Kang SK	Kim CH	2006	Stem Cells Dev	15	583–594
43	Taguchi A	Matsuyama T	2004	J Clin Invest	114	330–338
44	Jin HK	Schuchman EH	2003	Mol Ther	8	876–885
45	Nakano K	Shimada T	2001	Transplantation	27	1735–1740
46	Young PP	Sands MS	2004	Exp Neurol	188	104–114
47	Jin K	Greenberg DA	2005	Neurobiol Dis	18	366–374
48	Park KW	Mouradian MM	2001	Neurosci Res	40	315–323
49	Lee RH	Prockop DJ	2006	Blood	107	2153–61
50	Fujiwara Y	Ochi M	2004	Neurosci Lett	366	287–291
51	Lu D	Chopp M	2001	J Neurotrauma	18	813–819
52	Pluchino S	Martino G	2003	Nature	422	688–694
53	Xiao J	Low WC	2005	Stem Cells Dev	14	722–733
54	Chen J	Chopp M	2001	Stroke	32	1005–1011
55	Garbuzova S	Sanberg PR	2003	J Hematother Stem Cell Res	12	255–270
56	Corti S	Comi GP	2004	Brain	127	2518–2532
57	Ende N	Ende M	2000	Life Sci	67	53–59
58	Chen J	Chopp M	2001	Stroke	32	2682–2688
59	Chen R	Ende N	2000	J Med	31	21–30
60	Ende N	Chen R	2001	J Med	32	231–240

To understand the discrepancy between the efficiency of intravenous route of cell delivery in preclinical studies and clinical trials, animal experimental data has been reviewed systematically in a problem-oriented fashion [14]. This review has revealed that many factors critical for the proper design of clinical trials were not addressed sufficiently on the level of individual studies. In addition, contradictory results between studies were reported. Thus, the effort to perform an advanced statistical analysis is justified in order to evaluate the factors that influence outcome on the level of the whole population of preclinical studies.

Such an approach was introduced by researchers many years ago, and was called “meta-analysis” [15]. After further methodological improvement with a random effects model, this approach has proven to be a very valuable method of the data aggregation and analysis that result from clinical trials [16]. Later, it was applied in animal experiments [17] and in the analysis of preclinical data [18]. Recently, it is being widely employed for animal studies that deal with neuroprotection [19 –26]. Meta-analytic methodology has also been employed to study the genetics of stem cells [27]. Several meta-analyses of cell therapy of solid organ diseases (myocardial infarction) have been published over the last 3 years [28 –32]. However, no meta-analysis of cell therapy for neurological disorders on both the preclinical and the clinical level has been conducted.

The power of meta-analysis has been recognized and influences the daily clinical routine. Bench-to-bedside translation should be preceded by a meta-analysis of animal experimental data [33]. Particularly because animal data often comes from many small studies, there is even more demand for pooling results by meta-analytic methods than for large clinical trials.

Materials and Methods

The PubMed database search was conducted from the inception through 2006, and 60 preclinical studies utilizing the intravenous route of cell delivery for neurological disorders were found and subjected to analysis (Table 1). The first author was responsible for data collection from the articles. Clinical studies usually involve complex statistical analyses to combat with patient heterogeneity and large quantities and these inherently are associated with the risk of disagreements between interpreters. In contrast, animal studies, at least in neurotransplantation field, involve small experimental groups that are usually homogeneous and for those simple statistical tests (expressed as mean and SD or SE) are used till now. Thus in our opinion in case of animal studies no obvious necessity for independent data extraction exists, as there is no necessity for animal behavioral test assessment by a minimum of 2 investigators.

Thirteen categorical and 7 continuous independent variables (IVs) were extracted. Their definitions are presented in Tables 2 and 3, respectively. Three distinct outcomes of interest (morphological, behavioral, and molecular) were selected as dependent variables (DV) and called “treatment effects.” Morphological effect refers to the measurements of lesion size. Behavioral effect presents functional recovery as evaluated by behavioral tests. Molecular effect describes changes in brain tissue on a molecular level, such as neurogenesis, apoptosis inhibition, and growth factors (GF) expression, and so on. The measures of molecular effect were introduced to the design of studies to explain the mechanisms that mediate morphological and behavioral cell-dependent effects. These factors are listed and defined in Table 4.

Table 2.

Description of Categorical IVs

Variable	Category	Description
Type of Disease—specifies the type of disease modeled experimentally	Vascular	The models of cerebral vascular incidents such as stroke and intracerebral hemorrhage (ICH)
	Trauma	The models of traumatic injuries of CNS such as TBI and SCI
	Deg/Infl	As inflammatory basis of brain degeneration is emphasized recently such disorders (SM, PD, HD, and enzymatic defects) were combined into one category
Type of Experimental Model—specifies characteristic of the model itself	Sudden	The model based on sudden and one-time induction of disease such as brain vessel closure or intracerebral injection of substances
	No sudden	The model based on gradual and progressive neurological deterioration by genetic alterations or repeated systemic application of substances
Donor—specifies the order/species of mammals used for collection of cells for grafting	Human	Cells of human origin
	Rodent	As only single studies employed mice as donor, they were combined with rats within the category—rodents
Recipient—specifies the species of cell recipient	Rat	Cells derived from rats
Recipient—specifies the species of cell recipient	Mice	Cells derived from mice
Relatedness—specifies relatedness of donor and recipient	Xeno	Donor and recipient come from different species
Relatedness—specifies relatedness of donor and recipient	Allo/Syn	Donor and recipient come from the same species. Due to data paucity in many studies the distinction between allo- and syngeneic transplants was not possible
Immunosuppression—notifies the use of immunosuppressant during the study	Yes	The use of immunosuppressant was reported in article
	No	No immunosuppressant was used or no information about immunosuppression was obtained
Cell Type—specifies the characteristic of cell population used for transplantation	MSC	Mesenchymal stem cells (also referred as marrow stromal cells)
	MNC	Mononuclear fraction obtained by Ficoll-Paque gradient
	NSC	Neural stem cells
	Other	The rest of cell type such as hematopoietic stem cells (HSCs) etc.
	Total	The whole population of cells derived from bone marrow or cord blood
Cell History—specifies the proceeding with cells between collection and grafting	Cultured	Cells were cultured and expanded in vitro prior to transplantation
	Non-cultured	Cells were used for transplantation directly after collection or were frozen but not put in culture
Cell Source—specifies the organ from which cells were derived	BM	Cells were derived from BM
	Blood	Cells were derived from cord or peripheral blood
	Brain	Cells were derived from fetal or adult brain
Cell Line—notifies whether cell line was used	Yes	Cell cultured as cell line with or without immortalization
Cell Line—notifies whether cell line was used	No	Cells not cultured as cell line or not cultured at all
Genetic Modification—notifies whether any genetic manipulation was performed	Modified	Cells subjected to genetic modification
	Non-modified	Cells without genetic modification
GF Modification—notifies whether insertion of gene for GF was performed	Modified	Cells with inserted gene for production of GF
	Non-modified	Rest of cells
Stem Cells—notifies cells with stem cells characteristics obtained by selection or culture	Yes	Cells with stem cells features or fraction with significant enrichment in stem cells
	No	Cells not selected or cultured to receive stem cell population

Abbreviations: BM, bone marrow; CNS, central nervous system; GF, growth factor; HSC, hematopoietic stem cell; ICH, intracerebral hemorrhage; IV, independent variable; TBI, traumatic brain injury.

Table 3.

Description of Continuous Independent Variables

Variable	Description
Cell dosage	The number of cell used for transplantation
Length of study	The time from starting experiments to the end of study
Timing	Time window between the induction of animal model of disease and the cell transplantation
Follow-up	The interval between cell transplantation and the end of the study
Cell engraftment	The number of cells engrafted within CNS
Engraftment rate	The percentage of engrafted cells in comparison to cell dose
Differentiation rate	The percentage of engrafted cells that presented neural or vascular phenotype in vivo

Abbreviation: CNS, central nervous system.

Table 4.

Mechanisms Mediating Cell-Dependent Effect

Mechanism	Description
Apoptosis	The inhibition of disease-induced apoptosis
Neurogenesis	Disease-induced formation of cells with both neural phenotypes: neuronal and glial
GF	The change in intraparenchymal concentration of GF following cell transplantation
Angiogenesis	Disease-induced formation of vessels
White matter function	White matter volume, remodeling and scar formation
Gray matter function	Gray matter volume, remodeling and count of neurons
Axonal function	Axonal density and conduction velocity
Immunomodulation	The count of inflammatory cells and cytokine expression
Enzyme supplementation	Intraparenchymal concentration of missing enzymes

Abbreviation: GF, growth factor.

Mean outcomes, standard deviations (SDs), and animal numbers were retrieved and calculated by individual comparisons of experimental and control groups, based on the Hedges g formula, for each treatment effect and then were expressed as effect sizes (ESs) and variances (Vs) (Excel, Microsoft) [34]. When only graphic presentation of data was available, the values were carefully read from graphs using Corel Draw 11 (Corel Corporate Communications, Ottawa, Canada). Each pair of an experimental and a control group, regardless of origin from the same or another study, was named an experimental unit (EU). If a study consisted of several experimental groups, they were considered as distinct EUs. Where multiple behavioral outcomes or molecular measures were reported within any EU, they were combined to give an overall behavioral or molecular score per EU, using a fixed effect model. Publication bias (file drawer problem) was addressed by a Begg rank correlation [35], an Egger regression [36], and a trim-and-fill method [37] using freeware MIX, version 1.7 [38]. Then, a test for homogeneity across EUs was performed to choose between a fixed and a random effect model for further analysis. The extent of heterogeneity among EUs was statistically quantified by Q, H, and I ² (Excel, Microsoft) [39,40].

If the criterion of homogeneity is fulfilled (there is no significant difference between variances of particular EUs), all EUs are considered as samples from the same population; thus within-studies variance is solely taken to influence uncertainty of results (confidence interval). It is implemented into statistical analysis by fixed effect model. The model is strongly influenced by EU sizes and robust narrowing of confidence interval with increase of number of EUs is observed.

The presence of heterogeneity assumes that samples come from various populations, thus both within-studies and between-studies variances have to be taken into account. It is put into practice by random effect model. The addition of second variance estimator results with widening of confidence intervals and limits the weight of study size. It prevents from claim of unjustified statistical significance. However, fixed effects model is a special case of random effects model and is equivalent when variance between studies equals zero.

The mainspring analyses were performed under a random effect model using Proc Mixed (SAS Institute, version 9.2). As the overall effect estimate has been found to be influenced by between-study variance (τ²—variation coefficient), a likelihood-based approach was used [41]. For evaluation of the global estimate of treatment effects, ESs from EUs were pooled. The pooled ES was considered large when it exceeded 0.8 according to Cohen’s rule of thumb. The potential source of heterogeneity was evaluated by meta-regression [42]. For reconnaissance of heterogeneity distribution within the population of IVs, a univariate analysis was performed. Then, a multivariate regression model was fitted by Bayesian information criterion (BIC) for each outcome separately. An additional hierarchical, stepwise forward and stepwise backward meta-regression was also conducted, based on F statistics and P values. During this process, the interrelationship of variables was explored and insignificant and redundant variables were removed from the model. Then, the stratified analysis of mechanisms that mediated cell-dependent effects was conducted. P values <0.05 were considered significant.

Results

Data characteristic

A total number of 2,145 animals were included in the meta-analysis. Of 60 articles, 138 distinct EUs were retrieved. Within these, 313 elementary comparisons of particular tests/measures were performed. After pooling of data within each treatment effect and within EUs, the results were as follows:

1. Morphological effect was evaluated in 57 EUs coming from 26 studies

2. Behavioral effect was evaluated in 69 EUs coming from 37 studies

3. Molecular effect was evaluated in 40 EUs coming from 24 studies.

A total of 109 measures of molecular effect were grouped into 9 mechanisms that mediate cell-dependent effects: apoptosis, neurogenesis, GFs, angiogenesis, white matter function, gray matter function, axonal function, immunomodulation, and enzyme supplementation to prepare data for stratified analysis.

A significant heterogeneity and publication bias was found within each treatment effect (Table 5). The ES pooling revealed a large treatment effect within each outcome of interest (morphological ES = 1.75, CI 1.25–2.26, P < 0.0001; behavioral ES = 2.07, CI 1.49–2.64, P < 0.0001; molecular ES = 2.18, CI 1.66–2.71, P < 0.0001) (Fig. 1). However, there was no difference between mean outcomes with regard to the magnitude of treatment effects (Type III test, F = 0.66; P = 0.52).

FIG. 1.

Overall estimates of treatment effects. Vertical continuous line separates positive effect of treatment (experimental group did better than control group) on the right and negative effect on the left side of chart. Vertical interrupted line denotes where the positive effect begins to be large (0.8 according to Cohen’s rule of thumb). Everything to the right of this line indicates a large positive effect.

Table 5.

Homogeneity Analysis and Publication Bias

	Morphological	Behavioral	Molecular
Homogeneity analysis
a) Statistics Q	341.62; P < 0.0001	991.72; P < 0.0001	256.77; P < 0.0001
b) Statistics H	2.47; CI 2.2–2.8	3.82; CI 3.5–4.1	2.53; CI 2.2–2.9
c) Statistics I2	83.61%	94.15%	84.42%
Publication bias
a) Begg rank test	0.58; P < 0.0001	0.39; P < 0.0001	0.39; P = 0.0002
b) Egger regression	5.74; P < 0.0001	5.86; P < 0.0001	3.21; P < 0.0001
c) Trim-and-fill
– Number of EUs	57	69	41
– Number of imputed EUs	15	19	13

Abbreviation: EU, experimental unit.

Random effect meta-regression

Univariate approach. The categorical IVs related to the characteristics of transplanted cells, such as cell source, cell type, cell history, cell line, and genetic modification of cells, have been found to significantly model DVs. On the contrary, species- and recipient-related IVs, such as type of disease, type of experimental model, species of donor and recipient, as well as immunosuppression, did not reveal such an influence on the treatment effects (Fig. 2).

FIG. 2.

Forest plot presenting univariate approach for categorical independent variables (IVs) conducted for all dependent variables (DVs). Vertical continuous lines separate positive effect of treatment (experimental group did better than control group) on the right and negative effect on the left side of chart. Vertical interrupted line denotes where the positive effect begins to be large (0.8 according to Cohen’s rule of thumb). Everything to the right of this line indicates a large positive effect. Abbreviations: BM, bone marrow; GF, growth factor; MNC, mononuclear cell; MSC, mesenchymal stem cell; NSC, neural stem cell.

The analysis of continuous IVs (Table 6) disclosed the existence of a dose–response association between injected cell number and treatment effects (Fig. 3). However, the information about the fate of cells following transplantation was underreported. Such data were found in 1/3 of studies, and only 9 studies included information about all of these 3 variables; thus, the conclusion is very limited. The variables related to the timeline of the studies, that is, length of study, timing, and follow-up, did not model DVs. However, the inspection of regression lines showed a consistently downward course (negative slopes), which indicates a trend toward smaller effects as the time window of cell application widens and follow-up is extended (data not shown).

FIG. 3.

The influence of cell dosage on treatment effects: scatter plots and regression lines. Abbreviations: DV, dependent variable; ES, effect size; IV, independent variable.

Table 6.

Univariate Approach—Continuous Variables

Continuous variable	Morphological	Behavioral	Molecular
Cell dosage	Num df = 1, den df = 55	Num df = 1, den df = 67	Num df = 1, den df = 38
Cell dosage	F = 5.47, P = 0.023	F = 4.46, P = 0.023	F = 12.50, P = 0.001
Timing	Num df = 1, den df = 55	Num df = 1, den df = 67	Num df = 1, den df = 38
Timing	F = 0.37, P = 0.54	F = 0.06, P = 0.81	F = 0.46, P = 0.50
Length of study	Num df = 1, den df = 55	Num df = 1, den df = 67	Num df = 1, den df = 38
Length of study	F = 0.38, P = 0.54	F = 0.13, P = 0.72	F = 0.63, P = 0.43
Length of follow up	Num df = 1, den df = 55	Num df = 1, den df = 67	Num df = 1, den df = 38
Length of follow up	F = 0.15, P = 0.69	F = 0.08, P = 0.77	F = 0.45, P = 0.51
Number of engrafted cells	Num df = 1, den df = 12	Num df = 1, den df = 21	Num df = 1, den df = 11
Number of engrafted cells	F = 4.03, P = 0.07	F = 0.09, P = 0.77	F = 1.68, P = 0.22
Engraftment rate	Num df = 1, den df = 12	Num df = 1, den df = 21	Num df = 1, den df = 11
Engraftment rate	F = 0.67, P = 0.43	F = 0.00, P = 0.97	F = 1.36, P = 0.27
Differentiation rate	Num df = 1, den df = 13	Num df = 1, den df = 20	Num df = 1, den df = 8
Differentiation rate	F = 0.13, P = 0.73	F = 1.91, P = 0.18	F = 2.75, P = 0.14

Hierarchical approach. The univariate approach pointed to cell characteristics as the major source of heterogeneity. The inspection of interrelationships between IVs revealed their nested structure. From detail to general direction of investigation has been undertaken. The genetic modification, GF modification, and cell line IVs significantly influenced outcome in the univariate analysis of all outcomes of interest. However, IV GF modification is entirely nested in IV genetic modification. The analysis of IV GF modification within the frame of IV genetic modification resulted in a lack of significance of the former IV in each treatment effect: morphological (df = 1, F = 1.55, P = 0.24); behavioral (df = 1, F = 4.06, P = 0.08); and molecular (df = 1, F = 3.09, P = 0.14). Thus, IV GF modification has been excluded from the model and not subjected to further analyses.

Further inspection disclosed that both the genetic modification and cell line IVs are nested within IV cell history. They proved to be significant source of heterogeneity within IV cell history, and no additive interaction of both IVs was found (Table 7). Furthermore, IV cell history was almost entirely within IV stem cells, but was not the source of significant heterogeneity (morphological: F = 0.54; P = 0.47, behavioral: F = 2.88; P = 0.1, molecular: F = 0.59; P = 0.45).

The visual assessment of the donor and relatedness IVs showed that they are the same or almost entirely overlap. As the donor IV seems to be more important from a conceptual point of view, the IV relatedness has been excluded from the model. The influence of immunosuppression on outcome within xenotransplants was added to the analysis, but no significance was found.

Table 7.

Hierarchical Approach: IVs: Genetic Modification and Cell Line Nested Within IV Cell History

Within IV: Cell history	Treatment effect
Within IV: Cell history	Morphological df = 1	Behavioral df = 1	Molecular df = 1
Genetic modification	F = 10.74; P = 0.0024	F = 37.86, P < 0.0001	F = 10.98, P = 0.0026
Cell line	F = 7.53; P = 0.0095	F = 6.54, P = 0.0146	F = 26.19, P < 0.0001
Interaction of IVs	F = 4.15; P = 0.0143	F = 11.44, P < 0.0001	F = 13.45, P < 0.0001

Abbreviation: IV, independent variable.

Multivariate approach. The constellations of significant IVs selected by both methods, stepwise forward and stepwise backward across treatment effects are presented in Table 8. Stepwise forward selection is more oriented toward the single strongest IV, while stepwise backward selection allows for identification of a well-fitted set of IVs. To summarize the overall predictive values the IVs for the whole meta-analysis, the frequency of IVs appearance in Table 8 was calculated. The most significant IVs are: cell line and cell dosage (both 5 times), followed by genetic modification (3 times) and cell source (2 times). The comparison with the univariate approach revealed a reduction in the number of significant IVs.

Table 8.

Multivariate Approach

Selection method	Treatment effects
Selection method	Morphological	Behavioral	Molecular
Stepwise forward	–Genetic modification	–Genetic modification	–Cell dosage
	df = 1, F = 13.80; P = 0.0005	df = 1, F = 39.61; P < 0.0001	df = 1, F = 12.17; P = 0.0013
	–Cell dosage	–Cell line	–Cell line
	df = 1, F = 6.79; P = 0.0118	df = 1, F = 5.46; P = 0.0225	df = 1, F = 23.06; P < 0.0001
Stepwise backward	–Cell source	–Cell history	–Cell type
	df = 2, F = 6.47; P = 0.0031	df = 1, F = 8.70; P = 0.0044	df = 3, F = 3.83; P = 0.0189
	–Cell line	–Genetic modification	–Cell source
	df = 1, F = 19.27; P < 0.0001	df = 1, F = 30.56; P < 0.0001	df = 1, F = 6.70; P < 0.0144
	–Cell dosage	–Cell line	–Cell history
	df = 1, F = 9.97; P = 0.0026	df = 1, F = 5.89; P = 0.0118	df = 1, F = 8.51; P = 0.0064
		–Cell dosage	–Cell line
		df = 1, F = 4.84; P = 0.03115	df = 1, F = 11.74; P = 0.0017
			–Cell dosage df = 1, F = 15.97; P = 0.0004

The analysis of IVs related to the fate of transplanted cells revealed a positive interaction between the number of engrafted cells and their differentiation rate in vivo (df = 1; F = 1.91; P = 0.028).

Stratified analysis of mechanisms that mediate the cell-dependent effect

The analysis of molecular effect measures revealed that apoptosis inhibition is the strongest brain tissue-positive change evoked by cell therapy (Fig. 4). In addition to the above measure, there were also 3 measures for which significance exceeded 0.0001: neurogenesis, GFs, and angiogenesis. Two measures were not significantly changed following cell transplantation: immunomodulation and axonal function.

FIG. 4.

Forest plot presenting the range of molecular measure changes induced by cell therapy.

Discussion

The focus of our search was on PubMed database. Relying on single database would be considered as significant weakness by clinicians; however for animal researchers, the PubMed is an absolutely crucial platform of knowledge exchange and they do not routinely use other ones.

The meta-analysis was open to all experiments using the intravenous route of cell delivery for neurological disorders; thus, it was plagued by significant variability of study designs and conditions. However, we were interested in common determinants that influenced outcomes across all applications of this approach.

This approach revealed large treatment effects in a preclinical setup. Cell-related factors are the most powerful predictors of results. The culture of stem cells as cell line or genetic modification of cells (no additive interaction) revealed the highest efficiency, in a dose-dependent manner. This emphasizes the role of cell processing in preparing cells before a transplantation procedure. It seems that adequate cell adjustment to the recipient requirements will be superior to transplantation of “raw” cells, directly after cell collection from the donor, without advanced cell processing. However, long in vitro culture of cells prior to clinical transplantation is currently not recommended due to the risk of malignant transformation. It may be that this restriction should be challenged to enable further development of therapeutic options for neurological disorders.

The focus on cell features, not on cell donor and recipient characteristics or timeline of the study, raises the issue of the still-unidentified holistic mechanisms of central nervous system (CNS) healing by cell transplantation. Tissue changes on a molecular level, that is, inhibition of apoptosis, increase of neurogenesis, and angiogenesis seem to mediate a positive effect.

To date, transplanted cells were believed to replace dead or dysfunctional endogenous counterparts or, more frequently, to enhance CNS function by secreting molecules, such as GFs. But, the surprisingly high impact of cell therapy on apoptosis inhibition encourages the consideration of an alternative hypothesis about the overall mechanism of cell-based therapy. In addition to the above-mentioned mechanisms, one may speculate that transplanted cells also participate in the utilization of toxic waste products from diseased brain tissue, perhaps by acting as scavengers. A less toxic environment may favor the survival of endogenous cells through the inhibition of apoptosis.

In addition to the heterogeneity of the study designs, there is another weakness of this meta-analysis—many studies used in this report were plagued by missing data or the data presented had little or no value (no information on data dispersion or animal number used for each test). The information regarding cell fate following transplantation has been especially underreported, but data on this topic could be critical for the cut of Gordian knot in the search for mechanisms that mediate cell-dependent effects. It is worth mentioning that, in general, data from human studies are much more complete and even individual patient data (IPD) meta-analyses are recommended [43], although patient data would be more difficult to place into the public domain due to personal data protection laws. Thus, more comprehensive reporting of experimental results by researchers would address this issue. One way to attempt to gather missing data in an article being included in a meta-analysis is to simply contact the author(s) of the study. However, no all authors respond and others may have original database missed what we experienced while contacting with them. Thus we wish to propose a more restrictive solution. As all animal experiments worldwide are (or should be) approved by Ethical Commissions, these commissions could require researchers to share individual animal data and basic statistics as soon as the study ends, regardless of whether more advanced statistics are performed subsequently and the article is submitted for publication. Such a special platform would provide better control of preclinical research, particularly in such a demanding and incompletely understood field as stem cell-based therapy for neurological disorders. In addition, it would enable individual animal data (IAD) meta-analysis in the future and protect against publication bias.

It is also worth mentioning that the cell therapy is equally effective in animal models of sudden brain damage as well with models of a slow progression of symptoms. Immunosuppression does not have an impact on outcome, with a tendency toward a decreasing therapeutic effect. As cyclosporine inhibits calcineurin, which is involved in the process of cell migration, it can also inhibit the migration of transplanted cells toward CNS tissue, thus decreasing their therapeutic potential.

The comparison of results between different meta-analyses is difficult due to methodological issues. As we were working with continuous DVs, we chose to calculate standardized mean differences with a Hedges adjustment, without any data pre-processing. Thus, our results are expressed as original ESs, which should be interpreted according to Cohen’s rule of thumb. Other researchers have used ratios of the results for similar continuous data that required logarithmic processing [19]. Other groups have normalized data to outcomes in the control group [20]. The results of both groups were expressed as percentages, but these results cannot be directly compared. Thus, manipulation of data enables the presentation of data in a more reader-friendly form, but it does not allow for direct comparisons of data between meta-analyses. Thus, we do not support such an approach. The DVs used in clinical trials are usually categorical (binary) and, in such circumstances, the calculation of ratios is necessary [44].

Conclusions

Intravenous route of cell delivery improves outcome in animal models of neurological disorders. This effect is most pronounced in stem cell lines and genetically modified cells. The existence of a dose–response association between injected cell number and treatment effects has also been revealed. Apoptosis inhibition is the most significant change, on a molecular level, in diseased brain tissue following cell therapy. However, significant variability and publication bias limit the strength of these conclusions. Significant findings were summarized in Table 9.

Table 9.

Summary of Significant Findings

Investigated issue	Significant	Nonsignificant
Heterogeneity of studies	YES
Publication bias	YES
Overall positive	YES
neurotransplantation effect
Continuous variables	Cell dosage	Timing
		Length of study
		Length of follow up
		Number of engrafted cells
		Engraftment rate
		Differentiation rate
Categorical variables	Cell source	Type of disease
	Cell type	Type of experimental model
	Cell history	Species of donor
	Cell line	Species of recipient
	Genetic modification	Immunosuppression
Mechanisms mediating cell-dependent effect	Apoptosis inhibition	Immunomodulation
	Neurogenesis	Axonal function
	GFs
	Angiogenesis White matter function

Abbreviation: GF, growth factor.

Footnotes

Acknowledgments

We are grateful to Mary McAllister from Johns Hopkins University for editorial assistance. This study was supported by the Polish Ministry of Scientific Research and Higher Education, grant no. 142/B/P01/2008/35.

Author Disclosure Statement

No competing financial interests exist.

References

Liu

, Lang

, Baskaya

, Dempsey

and Vemuganti

. (2009). The potential of neural stem cells to repair stroke-induced brain damage. Acta Neuropathol, 117:469–480.

Torrente

and Polli

. (2008). Mesenchymal stem cell transplantation for neurodegenerative diseases. Cell Transplant, 17:1103–1113.

Guzman

, Choi

, Gera

, De Los Angeles

, Andres

and Steinberg

. (2008). Intravascular cell replacement therapy for stroke. Neurosurg Focus, 24:E15.

Hess

and Borlongan

. (2008). Stem cells and neurological diseases. Cell Prolif, 41,(Suppl 1):94–114.

Yasuhara

, Borlongan

and Date

. (2006). Ex vivo gene therapy: transplantation of neurotrophic factor-secreting cells for cerebral ischemia. Front Biosci, 11:760–775.

Bang

, Lee

and Lee

. (2005). Autologous mesenchymal stem cell transplantation in stroke patients. Ann Neurol, 57:874–882.

, Xu

, Ji

, Wan

, Lu

, Dong

, Yu

and Lu

. (2006). Autologous peripheral blood stem cell transplantation for severe multiple sclerosis. Int J Hematol, 84:276–281.

Burt

, Loh

, Cohen

, Stefosky

, Balabanov

, Katsamakis

, Oyama

, Russell

, Stern

, Muraro

, Rose

, Testori

, Bucha

, Jovanovic

, Milanetti

, Storek

, Voltarelli

and Burns

. (2009). Autologous non-myeloablative haemopoietic stem cell transplantation in relapsing-remitting multiple sclerosis: a phase I/II study. Lancet Neurol, 8:244–253.

Appel

, Engelhardt

, Henkel

, Siklos

, Beers

, Yen

, Simpson

, Luo

, Carrum

, Heslop

, Brenner

and Popat

. (2008). Hematopoietic stem cell transplantation in patients with sporadic amyotrophic lateral sclerosis. Neurology, 71:1326–1334.

10.

Lee

, Kim

, Bang

, Ahn

, Joo

and Huh

. (2008). Autologous mesenchymal stem cell therapy delays the progression of neurological deficits in patients with multiple system atrophy. Clin Pharmacol Ther, 83:723–730.

11.

Sykova

, Homola

, Mazanec

, Lachmann

, Konradova

, Kobylka

, Padr

, Neuwirth

, Komrska

, Vavra

, Stulik

and Bojar

. (2006). Autologous bone marrow transplantation in patients with subacute and chronic spinal cord injury. Cell Transplant, 15:675–687.

12.

, Ji

, Su

, Dong

, Sun

and Liu

. (2006). Clinical outcomes after autologous haematopoietic stem cell transplantation in patients with progressive multiple sclerosis. Chin Med J (Engl), 119:1851–1855.

13.

Christie

. (2009). Study on stem cells for stroke patients to start in Scotland later in 2009. BMJ, 338:b245.

14.

Janowski

, Gornicka-Pawlak

, Kozlowska

, Domanska-Janik

, Gielecki

and Lukomska

. (2008). Structural and functional characteristic of a model for deep-seated lacunar infarct in rats. J Neurol Sci, 273:40–48.

15.

Smith

and Glass

. (1977). Meta-analysis of psychotherapy outcome studies. Am Psychol, 32:752–760.

16.

DerSimonian

and Laird

. (1986). Meta-analysis in clinical trials. Control Clin Trials, 7:177–188.

17.

Freedman

and Clifford

. (1994). Meta-analysis of animal experiments: elucidating relationships between dietary fat and mammary tumor development in rodents. Adv Exp Med Biol, 364:93–100.

18.

Bouzom

, Laveille

, Merdjan

and Jochemsen

. (2000). Use of nonlinear mixed effect modeling for the meta-analysis of preclinical pharmacokinetic data: application to S 20342 in the rat. J Pharm Sci, 89:603–613.

19.

Minnerup

, Heidrich

, Wellmann

, Rogalewski

, Schneider

and Schabitz

. (2008). Meta-analysis of the efficacy of granulocyte-colony stimulating factor in animal models of focal cerebral ischemia. Stroke, 39:1855–1861.

20.

Wheble

, Sena

and Macleod

. (2008). A systematic review and meta-analysis of the efficacy of piracetam and piracetam-like compounds in experimental stroke. Cerebrovasc Dis, 25:5–11.

21.

van der Worp

, Sena

, Donnan

, Howells

and Macleod

. (2007). Hypothermia in animal models of acute ischaemic stroke: a systematic review and meta-analysis. Brain, 130:3063–3074.

22.

Sena

, Wheble

, Sandercock

and Macleod

. (2007). Systematic review and meta-analysis of the efficacy of tirilazad in experimental stroke. Stroke, 38:388–394.

23.

Willmot

, Gibson

, Gray

, Murphy

and Bath

. (2005). Nitric oxide synthase inhibitors in experimental ischemic stroke and their effects on infarct size and cerebral blood flow: a systematic review. Free Radic Biol Med, 39:412–425.

24.

Willmot

, Gray

, Gibson

, Murphy

and Bath

. (2005). A systematic review of nitric oxide donors and l-arginine in experimental stroke; effects on infarct size and cerebral blood flow. Nitric Oxide, 12:141–149.

25.

Macleod

, O’Collins

, Horky

, Howells

and Donnan

. (2005). Systematic review and metaanalysis of the efficacy of FK506 in experimental stroke. J Cereb Blood Flow Metab, 25:713–721.

26.

Macleod

, O’Collins

, Horky

, Howells

and Donnan

. (2005). Systematic review and meta-analysis of the efficacy of melatonin in experimental stroke. J Pineal Res, 38:35–41.

27.

Glover

, Marin

, Eaves

, Helgason

, Piret

and Bryan

. (2006). Meta-analysis of differentiating mouse embryonic stem cell gene expression kinetics reveals early change of a small gene set. PLoS Comput Biol, 2:e158.

28.

Martin-Rendon

, Brunskill

, Doree

, Hyde

, Watt

, Mathur

and Stanworth

. (2008). Stem cell treatment for acute myocardial infarction. Cochrane Database Syst Rev:CD006536.

29.

Zhang

, Sun

, Ge

, Yao

, Huang

, Wang

and Zou

. (2008). Intracoronary autologous bone marrow stem cells transfer for patients with acute myocardial infarction: a meta-analysis of randomised controlled trials. Int J Cardiol, 136:178–185.

30.

Lipinski

, Biondi-Zoccai

, Abbate

, Khianey

, Sheiban

, Bartunek

, Vanderheyden

, Kim

, Kang

, Strauer

and Vetrovec

. (2007). Impact of intracoronary cell therapy on left ventricular function in the setting of acute myocardial infarction: a collaborative systematic review and meta-analysis of controlled clinical trials. J Am Coll Cardiol, 50:1761–1767.

31.

Abdel-Latif

, Bolli

, Tleyjeh

, Montori

, Perin

, Hornung

, Zuba-Surma

, Al-Mallah

and Dawn

. (2007). Adult bone marrow-derived cells for cardiac repair: a systematic review and meta-analysis. Arch Intern Med, 167:989–997.

32.

Kang

, Yang

, Li

and Gao

. (2008). Effects of intracoronary autologous bone marrow cells on left ventricular function in acute myocardial infarction: a systematic review and meta-analysis for randomized controlled trials. Coron Artery Dis, 19:327–335.

33.

Macleod

, O’Collins

, Howells

and Donnan

. (2004). Pooling of animal experimental data reveals influence of study design and publication bias. Stroke, 35:1203–1208.

34.

Hedges

. (1984). Research synthesis: the state of the art. Int J Aging Hum Dev, 19:85–93.

35.

Begg

and Mazumdar

. (1994). Operating characteristics of a rank correlation test for publication bias. Biometrics, 50:1088–1101.

36.

Egger

, Davey Smith

, Schneider

and Minder

. (1997). Bias in meta-analysis detected by a simple, graphical test. BMJ, 315:629–634.

37.

Duval

and Tweedie

. (2000). Trim and fill: A simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics, 56:455–463.

38.

Bax

, Yu

, Ikeda

, Tsuruta

and Moons

. (2006). Development and validation of MIX: comprehensive free software for meta-analysis of causal research data. BMC Med Res Methodol, 6:50.

39.

Higgins

, Thompson

, Deeks

and Altman

. (2003). Measuring inconsistency in meta-analyses. BMJ, 327:557–560.

40.

Higgins

and Thompson

. (2002). Quantifying heterogeneity in a meta-analysis. Stat Med, 21:1539–1558.

41.

, Platt

, Luo

, Wei

and Fraser

. (2008). Exploring heterogeneity in meta-analyses: needs, resources and challenges. Paediatr Perinat Epidemiol, 22,(Suppl 1):18–28.

42.

Thompson

and Higgins

. (2002). How should meta-regression analyses be undertaken and interpreted?. Stat Med, 21:1559–1573.

43.

Thompson

and Higgins

. (2005). Treating individuals 4: can meta-analysis help target interventions at individuals most likely to benefit?. Lancet, 365:341–346.

44.

Robenshtok

, Gafter-Gvili

, Goldberg

, Weinberger

, Yeshurun

, Leibovici

and Paul

. (2007). Antifungal prophylaxis in cancer patients after chemotherapy or hematopoietic stem-cell transplantation: systematic review and meta-analysis. J Clin Oncol, 25:5471–5489.